Сканирование и форматирование: Янко Слава (Библиотека Fort/Da) || slavaaa@yandex.ru || yanko_slava@yahoo.com || http://yanko.lib.ru || Icq# 75088656 || Библиотека: http://yanko.lib.ru/gum.html ||

update 04.05.05 Номера страниц вверху. Степень степени в эл.версии книги я обозначил через "[...]"верхнего индекса, типа: 101000[10] - это означает степень степени, второй уровень верхнего регистра

PHILOSOPHY OF MIND SERIES Series Editor: Owen Flanagan, Duke University


Mind, Morals, and the Meaning of Life Owen Flanagan



In Search of a Fundamental Theory

David J. Chalmers

THE Conscious Mind


David J. Chalmers

New York Oxford

Oxford University Press



What is consciousness? How do physical processes in the brain give rise to the sub-jective life of a conscious mind? These questions are among the most hotly debated issues in science and philoso-phy today. Now, in The Conscious Mind, philosopher David J. Chalmers offers a cogent analysis of this debate as he lays out a major new theory of consciousness, one that rejects the prevailing reductionist trend of science, but is still compatible with a scien-tific view of the world.

Writing in a rigorous, thought-provoking style, the author takes us on a far-reaching tour through the philosophical ramifications of consciousness. Chalmers convincingly establishes that contemporary cognitive science and neuroscience do not begin to explain how subjective experience emerges from neural processes in the brain. He pro-poses that conscious experience must instead be understood in a new light-as an irre-ducible entity (like such physical properties as time, mass, and space) that exists at a funda-mental level and cannot be understood as the sum of simpler physical parts. In the second half of the book, he sets out on a quest for a "fundamental theory"- a theory of the basic laws governing the structure and character of conscious experience-and shows how this reconception of the mind could lead us to a new science of consciousness.

Throughout the book, Chalmers provides fascinating thought experiments that vividly illustrate his ideas. For example, in exploring the possibility that consciousness could be experienced by machines as well as humans, Chalmers asks us to imagine a thinking brain in which neurons are slowly replaced by sili-con chips - as the neurons are replaced, will consciousness gradually fade away? The book also features thoughtful discussions of how

the author's ideas might be applied to sub-jects as diverse as artificial intelligence and the interpretation of quantum mechanics.

All of us have pondered the mysteries of consciousness. Engaging and penetrating, The Conscious Mind adds a fresh new perspective that will spark debate about our understand-ing of the mind for years to come.

About the Author

David J. Chalmers is a professor of philoso-phy at the University of California, Santa Cruz. Born in Sydney, Australia, he has been a Rhodes Scholar at the University of Oxford and a McDonnell Fellow at Washington Uni-versity. His article "The Puzzle of Conscious Experience" appeared in the December 1995 issue of Scientific American.



Oxford New York

Athens Auckland Bangkok Bombay Calcutta Cape Town Dar es Salaam Delhi

Florence Hong Kong Istanbul Karachi

Kuala Lumpur Madras Madrid Melbourne

Mexico City Nairobi Paris Singapore

Taipei Tokyo Toronto

and associated companies in Berlin Ibadan

Copyright © 1996 by David J. Chalmers

Published by Oxford University Press, Inc. 198 Madison Avenue, New York, New York 10016

Oxford is a registered trademark of Oxford University Press

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press.

Library of Congress Cataloging-in-Publication Data

Chalmers, David John

The conscious mind : in search of a fundamental theory p. cm. (Philosophy of mind series)

Includes bibliographical references and index.

ISBN 0-19-510553-2

1. Philosophy of mind. 2. Consciousness. 3. Mind and body. 4. Dualism. I. Title. II. Series.

BD418.3.C43 1996 128'.2-dc20 95-36036


Printed in the United States of America on acid-free paper.


Acknowledgments. 8

Contents. 9

Introduction. 11

Taking Consciousness Seriously. 11

PART I. Foundations. 14

1. Two Concepts of Mind. 14

1. What Is Consciousness?. 14

A catalog of conscious experiences 16

Visual experiences. 16

Auditory experiences. 16

Figure 1.1. Effability and ineffability in olfactory experience. 17

Calvin and Hobbes by Bill Watterson. 17

Tactile experiences. 17

Olfactory experiences. 17

Taste experiences. 17

Experiences of hot and cold. 18

Pain. 18

Other bodily sensations. 18

Mental imagery. 18

Conscious thought. 18

Emotions. 18

The sense of self. 18

2. The Phenomenal and the Psychological Concepts of Mind. 19

A potted history. 20

3. The Double Life of Mental Terms. 22

The co-occurrence of phenomenal and psychological properties 24

4. The Two Mind-Body Problems. 25

5. Two Concepts of Consciousness. 26

Varieties of psychological consciousness 27

Awakeness. 27

Introspection. 27

Reportability. 27

Self-consciousness. 27

Attention. 27

Voluntary control. 27

Knowledge. 27

Consciousness and awareness 28

Explaining consciousness versus explaining awareness 28

2. Supervenience and Explanation. 30

1. Supervenience. 30

Local and global Supervenience. 30

Logical and natural Supervenience. 31

A problem with logical Supervenience*. 33

Supervenience and materialism.. 34

2. Reductive Explanation. 35

Reductive explanation via functional analysis 36

Reductive explanations in cognitive science. 37

3. Logical Supervenience and Reductive Explanation. 38

Further notes on reductive explanation. 39

4. Conceptual Truth and Necessary Truth*. 40

Definitions 40

Figure 2.1. Two ways in which a property P might depend on properties A and B. 41

Figure 2.2. The round curve 2x2 + 3y2 = 1 and nonround friend. 41

Revisability. 42

A posteriori necessity. 42

Logical necessity, conceptual truth, and conceivability. 47

Logical necessity and logical Supervenience. 49

5. Almost Everything is Logically Supervenient on the Physical*. 50

Conceivability. 51

Epistemology. 51

Analyzability. 53

Some problem cases 55

Consciousness-dependent properties. 55

Intentionality. 56

Moral and aesthetic properties. 56

Names. 57

Indexicals. 57

Negative facts. 57

Physical laws and causation. 58

Recap. 58

Epistemological myth. 59

PART II. The Irreducibility of consciousness. 60

3. Can Consciousness Be Reductively Explained?. 60

1. Is Consciousness Logically Supervenient on the Physical?. 60

Argument 1: The logical possibility of zombies 60

Figure 3.1. Calvin and Hobbes on zombies. 60

Calvin and Hobbes by Bill Watterson. 61

Argument 2: The inverted spectrum.. 63

Argument 3: From epistemic asymmetry. 64

Argument 4: The knowledge argument 65

Argument 5: From the absence of analysis 66

2. The Failure of Reductive Explanation. 67

Objection 1: Are we setting the standards too high?. 67

Objection 2: Couldn't a vitalist have said the same thing about life?. 68

Objection 3: Is conceivability a guide to possibility?. 68

Objection 4: Isn't this a collection of circular intuitions?. 69

Objection 5: Doesn't all explanation have to stop somewhere?. 69

3. Cognitive Modeling. 69

Figure 3.2. Dennett's cognitive model of consciousness. 70

4. Neurobiological Explanation. 72

Figure 3.3. Edelman's scheme for higher-order consciousness. 73

5. The Appeal to New Physics. 74

6. Evolutionary Explanation. 75

7. Whither Reductive Explanation?. 75

4. Naturalistic Dualism.. 76

1. An Argument Against Materialism.. 76

What sort of dualism?. 76

Objections 79

2. Objections From A Posteriori Necessity*. 80

An alternative strategy. 82

Strong metaphysical necessity. 83

Cognitive limitations 84

3. Other Arguments for Dualism*. 84

Jackson's argument 85

Kripke's argument 88

4. Is This Epiphenomenalism?*. 89

Strategies for avoiding epiphenomenalism.. 90

1. Regularity-based causation. 90

152 The Irreducibility of Consciousness. 90

2. Causal overdetermination. 91

3. The nonsupervenience of causation. 91

Interactionist dualism?. 93

The problems of epiphenomenalism.. 94

5. The Logical Geography of the Issues. 95

i. Eliminativism. 95

ii. Reductive functionalism. 96

iii. Nonfunctionalist reductive materialism. 96

iv. New-physics materialism. 96

v. Nonreductive materialism. 96

vi. Interactionist dualism. 96

vii. Naturalistic dualism. 96

viii. Don't-have-a-clue materialism. 96

Type A, type B, and type C. 98

6. Reflections on Naturalistic Dualism.. 99

5.The Paradox of Phenomenal Judgment 101

1. Consciousness and Cognition. 101

Phenomenal judgments 102

Three kinds of phenomenal judgment 103

2. The Paradox of Phenomenal Judgment 103

Facing up to the paradox. 106

3. On Explaining Phenomenal Judgments. 107

Is explaining the judgments enough?. 108

Dennett on phenomenal judgments 110

4. Arguments Against Explanatory Irrelevance. 111

5. The Argument from Self-Knowledge*. 111

What justifies phenomenal judgments?. 113

Answering the arguments 114

6. The Argument from Memory*. 115

7. The Argument from Reference*. 116

The content of phenomenal beliefs 117

PART III. Toward a Theory of Consciousness. 120

6. The Coherence Between Consciousness and Cognition. 120

1. Toward a Nonreductive Theory. 120

How might we build a theory of consciousness?. 121

2. Principles of Coherence. 123

The coherence between consciousness and awareness 123

The principle of structural coherence. 125

3. More on the Notion of Awareness. 126

Relationship to functionalist theories of consciousness*. 128

Figure 6.1. Zippy the Pinhead on higher-order theories of consciousness. 128

First-order judgments and first-order registrations*. 130

4. The Explanatory Role of Coherence Principles. 131

Coherence principles as epistemic levers 132

The physical correlates of consciousness 133

5. Coherence as a Psychophysical Law.. 135

7. Absent Qualia, Fading Qualia, Dancing Qualia. 137

1. The Principle of Organizational Invariance. 137

Absent qualia and inverted qualia. 138

2. Absent Qualia. 139

3. Fading Qualia. 140

Objection 1: Neural replacement would be impossible in practice. 144

Objection 2: Some systems are massively mistaken about their experience. 144

Objection 3: Sorites arguments are suspect 144

Objection 4: Similar arguments could establish behavioral invariance. 145

4. Inverted Qualia. 145

Though one Man's Idea of Blue should be different from another's. 146

5. Dancing Qualia. 147

Objection 1: Loopholes concerning speed and history. 149

Objection 2: What about mild inversions?. 150

Objection 3: Unattended qualia. 150

Objection 4: Double switching. 150

6. Nonreductive Functionalism.. 151

8. Consciousness and Information: Some Speculation. 152

1. Toward a Fundamental Theory. 152

2. Aspects of Information. 153

Physically realized information. 154

Figure 8.1. Shannon's diagram of an information channel. 155

Phenomenally realized information. 156

The double-aspect principle. 156

3. Some Supporting Arguments. 158

Explaining phenomenal judgments 158

4. Is Experience Ubiquitous?. 161

What is it like to be a thermostat?. 161

Whither panpsychism?. 163

Constraining the double-aspect principle. 164

5. The Metaphysics of Information. 165

It from bit 165

Grounding information in phenomenology. 166

What about macroscopic phenomenology?. 167

6. Open Questions. 169

PART IV. Applications. 170

9. Strong Artificial Intelligence. 170

1. Machine Consciousness. 170

Figure 9.1. Bloom County on strong AI. 171

2. On Implementing a Computation. 171

3. In Defense of Strong AI 173

4. The Chinese Room and Other Objections. 175

The Chinese room.. 175

Syntax and semantic. 177

A simulation is just a simulation. 177

5. External Objections. 178

Objections from rule following. 178

Objections from Gödel's theorem.. 179

Objections from uncomputability and continuity. 179

6. Conclusion. 180

10. The Interpretation of Quantum Mechanics. 180

1. Two Mysteries. 180

2. The Framework of Quantum Mechanics. 181

3. Interpreting Quantum Mechanics. 182

Option 1: Take the calculus literally. 182

Option 2: Try to get the measurement postulate for free. 184

Option 3: Whereof one cannot speak... 185

Option 4: Postulate further basic physical principles 185

Option 5: The Schrödinger equation is all 186

4. The Everett Interpretation. 187

5. Objections to the Everett Interpretation. 189

Objections based on "splitting". 189

Objections to a preferred basis 189

What about superposed minds?. 190

Objections based on personal identity. 190

The interpretation of probabilities 191

6. Conclusion. 192

Notes. 193

Chapter 1. 193

Chapter 2. 194

Chapter 3. 199

Chapter 4. 200

34. Biological materialism. 203

35. Physicalist-functionalism. 203

36. Psychofunctionalism. 203

37. Anomalous monism. 204

38. Representationalism. 204

39. Consciousness as higher-order thought. 205

40. Reductive teleofunctionalism. 205

41. Emergent causation. 205

42. Mysterianism. 205

Chapter 5. 206

Chapter 6. 207

Chapter 7. 210

Chapter 8. 211

Chapter 9. 212

Chapter 10. 212

Bibliography. 213

Index. 221



I first became excited by consciousness and the mind-body problem as an undergraduate studying mathematics at the University of Adelaide. Conversations with a number of people, especially Paul Barter, Jon Baxter, Ben Hambly, and Paul McCann, helped form my ideas. Even then, the subject seemed about as fascinating a problem as there could be. It seemed faintly unreasonable that somebody could be occupied full-time thinking about something that was so much fun.

Later, as a graduate student at Oxford, I found that the mind was always occupying my thoughts where mathematics should have been, and I decided to switch fields and eventually to switch continents. Many people were patient and supportive during this difficult time, especially Michael Atiyah, Michael Dummett, and Robin Fletcher. Thanks also to all those who were subjected to hearing about whatever my latest theory of consciousness happened to be; the ideas in this book are a distant descendant.

My decision to move to Indiana University to gain a grounding in philosophy, cognitive science, and artificial intelligence was one of the best that I have made. I owe special thanks to Doug Hofstadter; it was his writing that first introduced me to the mysteries of the mind when I was young, and it was the stimulating and comfortable environment of his research lab, the Center for Research on Concepts and Cognition, that allowed these ideas to develop. Although he disagrees with many of the ideas in this book, I would like to think that at some level what I have written remains true to the intellectual spirit of his work.

I wrote the first version of this work (then known as Toward a Theory of Consciousness) in a heady six-month period in 1992 and 1993. I had useful discussions with a number of people at Indiana around this time: everybody at CRCC, especially Bob French and Liane Gabora, and many in other departments, including Mike Dunn, Rob Goldstone, Anil Gupta, Jim Hettmer, Jerry Seligman, and Tim van Gelder. Thanks also to members of the consciousness discussion group in the back room at Nick's for many enjoyable Monday afternoon conversations.

A two-year McDonnell fellowship in philosophy, neuroscience, and psychology at Washington University has provided another stimulating environ-

vi Acknowledgments

ment, as well as a chance to experience Zeno's paradox in finishing this book. I am grateful to the James S. McDonnell Foundation for their support, to all the participants in my graduate seminar on consciousness for discussions that helped to refine the book, and to a number of people for conversation and comments, including Morten Christiansen, Andy Clark, Jason Clevenger, Peggy DesAutels, Pepa Toribio, and Tad Zawidzki.

In the last couple of years, I have had an enormous amount of helpful conversation and correspondence about the material in this book. Among many others, thanks are due to Jon Baxter, Ned Block, Alex Byrne, Francis Crick, Dan Dennett, Eric Dietrich, Avi Elitzur, Matthew Elton, Owen Flanagan, Stan Franklin, Liane Gabora, Gьven Gьzeldere, Chris Hill, Terry Horgan, Steve Horst, Frank Jackson, Jaegwon Kim, Christof Koch, Martin Leckey, Dave Leising, Kerry Levenberg, Joe Levine, David Lewis, Barry Loewer, Bill Lycan, Paul McCann, Daryl McCullough, Brian McLaughlin, Thomas Metzinger, Robert Miller, Andrew Milne, John O'Leary-Hawthorne, Joseph O'Rourke, Calvin Ostrum, Rhett Savage, Aaron Sloman, Leopold Stubenberg, and Red Watson. I am grateful to too many others to mention for interesting conversations about consciousness in general. A special note of thanks to Norton Nelkin, who returned his copy of the manuscript covered with many helpful comments not long before he died of lymphoma. He will be missed.

My broader philosophical debts are many. I developed my initial views on consciousness largely on my own, but these have been greatly enriched by my reading on the subject. One discovers quickly that any given idea has likely been expressed already by someone else. Among recent thinkers, Thomas Nagel, Frank Jackson, and Joseph Levine have done much to emphasize the perplexities of consciousness; their work covers much of the same territory as my early chapters. My work also overlaps with work by Ned Block, Robert Kirk, and Michael Lockwood at a number of points. The metaphysical framework that I develop in Chapter 2 owes much to the work of Terry Horgan, Saul Kripke, and David Lewis, among others, and Frank Jackson has independently developed a similar framework, presented in his marvelous 1995 John Locke lectures. The ideas of Daniel Dennett, Colin McGinn, John Searle, and Sydney Shoemaker have provided stimulating challenges throughout.

My greatest debts are to Gregg Rosenberg, for memorable conversations and valuable feedback; to Lisa Thomas, for a book on zombies and moral support; to Sharon Wahl, for expert editing and warm friendship; and above all, to' all three of my parents, for their support and encouragement. And thanks to all my qualia, and to the environment responsible for producing them, for constant inspiration.

As I was finishing this book, I received a fortune cookie in a restaurant, saying "Your life will be full of delightful mysteries." So far it has been, and I am very grateful.

Contents      vii


Introduction: Taking Consciousness Seriously xi


1 Two Concepts of Mind 3

1. What is consciousness? 3

2. The phenomenal and the psychological concepts of mind 11

3. The double life of mental terms 16

4. The two mind-body problems 24

5. Two concepts of consciousness 25

2 Supervenience and Explanation 32

1. Supervenience 32

2. Reductive explanation 42

3. Logical Supervenience and reductive explanation 47

4. Conceptual truth and necessary truth* 52

5. Almost everything is logically supervenient on the physical* 71


3 Can Consciousness Be Reductively Explained? 93

1. Is consciousness logically supervenient on the physical? 93

2. The failure of reductive explanation 106

3. Cognitive modeling 111

4. Neurobiological explanation 115

5. The appeal to new physics 118

6. Evolutionary explanation 120

7. Whither reductive explanation? 121

viii Contents

4 Naturalistic Dualism 123

1. An argument against materialism 123

2. Objections from a posteriori necessity* 131

3. Other arguments for dualism* 140

4. Is this epiphenomenalism?* 150

5. The logical geography of the issues 161

6. Reflections on naturalistic dualism 168

5 The Paradox of Phenomenal Judgment 172

1. Consciousness and cognition 172

2. The paradox of phenomenal judgment 177

3. On explaining phenomenal judgments 184

4. Arguments against explanatory irrelevance 191

5. The argument from self-knowledge* 192

6. The argument from memory* 200

7. The argument from reference* 201


6 The Coherence Between Consciousness and Cognition 213

1. Toward a nonreductive theory 213

2. Principles of coherence 218

3. More on the notion of awareness 225

4. The explanatory role of coherence principles 233

5. Coherence as a Psychophysical law 242

7 Absent Qualia, Fading Qualia, Dancing Qualia 247

1. The principle of organizational invariance 247

2. Absent qualia 251

3. Fading qualia 253

4. Inverted qualia 263

5. Dancing qualia 266

6. Nonreductive functionalism 274

Contents      ix

8 Consciousness and Information: Some Speculation 276

1. Toward a fundamental theory 276

2. Aspects of information 277

3. Some supporting arguments 287

4. Is experience ubiquitous? 293

5. The metaphysics of information 301

6. Open questions 308


9 Strong Artificial Intelligence 309

1. Machine consciousness 309

2. On implementing a computation 315

3. In defense of strong AI 320

4. The Chinese room and other objections 322

5. External objections 328

6. Conclusion 331

10 The Interpretation of Quantum Mechanics 333

1. Two mysteries 333

2. The framework of quantum mechanics 334

3. Interpreting quantum mechanics 337

4. The Everett interpretation 346

5. Objections to the Everett interpretation 351

6. Conclusion 356

Notes 359

Bibliography 391

Index 405


Taking Consciousness Seriously

Consciousness is the biggest mystery. It may be the largest outstanding obstacle in our quest for a scientific understanding of the universe. The science of physics is not yet complete, but it is well understood; the science of biology has removed many ancient mysteries surrounding the nature of life. There are gaps in our understanding of these fields, but they do not seem intractable. We have a sense of what a solution to these problems might look like; we just need to get the details right.

Even in the science of the mind, much progress has been made. Recent work in cognitive science and neuroscience is leading us to a better understanding of human behavior and of the processes that drive it. We do not have many detailed theories of cognition, to be sure, but the details cannot be too far off. Consciousness, however, is as perplexing as it ever was. It still seems utterly mysterious that the causation of behavior should be accompanied by a subjective inner life.

We have good reason to believe that consciousness arises from physical systems such as brains, but we have little idea how it arises, or why it exists at all. How could a physical system such as a brain also be an experience? Why should there be something it is like to be such a system? Present-day scientific theories hardly touch the really difficult questions about consciousness. We do not just lack a detailed theory; we are entirely in the dark about how consciousness fits into the natural order.

Many books and articles on consciousness have appeared in the past few years, and one might think that we are making progress. But on a closer look, most of this work leaves the hardest problems about consciousness untouched. Often, such work addresses what might be called the "easy" problems of consciousness: How does the brain process environmental stimula-

xii Introduction

tion? How does it integrate information? How do we produce reports on internal states? These are important questions, but to answer them is not to solve the hard problem: Why is all this processing accompanied by an experienced inner life? Sometimes this question is ignored entirely; sometimes it is put off until another day; and sometimes it is simply declared answered. But in each case, one is left with the feeling that the central problem remains as puzzling as ever.

This puzzlement is not a cause for despair; rather, it makes the problem of consciousness one of the most exciting intellectual challenges of our time. Because consciousness is both so fundamental and so ill understood, a solution to the problem may profoundly affect our conception of the universe and of ourselves.

I am an optimist about consciousness: I think that we will eventually have a theory of it, and in this book I look for one. But consciousness is not just business as usual; if we are to make progress, the first thing we must do is face up to the things that make the problem so difficult. Then we can move forward toward a theory, without blinkers and with a good idea of the task at hand.

In this book, I do not solve the problem of consciousness once and for all, but I try to rein it in. I try to get clear about what the problems are, I argue that the standard methods of neuroscience and cognitive science do not work in addressing them, and then I try to move forward.


In developing my account of consciousness, I have tried to obey a number of constraints. The first and most important is to take consciousness seriously. The easiest way to develop a "theory" of consciousness is to deny its existence, or to redefine the phenomenon in need of explanation as something it is not. This usually leads to an elegant theory, but the problem does not go away. Throughout this book, I have assumed that consciousness exists, and that to redefine the problem as that of explaining how certain cognitive or behavioral functions are performed is unacceptable. This is what I mean by taking consciousness seriously.

Some say that consciousness is an "illusion," but I have little idea what this could even mean. It seems to me that we are surer of the existence of conscious experience than we are of anything else in the world. I have tried hard at times to convince myself that there is really nothing there, that conscious experience is empty, an illusion. There is something seductive about this notion, which philosophers throughout the ages have exploited, but in the end it is utterly unsatisfying. I find myself absorbed in an orange sensation, and something is going on. There is something that needs explaining, even after we have explained the processes of discrimination and action: there is the experience.

True, I cannot prove that there is a further problem, precisely because I cannot prove that consciousness exists. We know about consciousness more

Taking Consciousness Seriously xiii

directly than we know about anything else, so "proof" is inappropriate. The best I can do is provide arguments wherever possible, while rebutting arguments from the other side. There is no denying that this involves an appeal to intuition at some point; but all arguments involve intuition somewhere, and I have tried to be clear about the intuitions involved in mine.

This might be seen as a Great Divide in the study of consciousness. If you hold that an answer to the "easy" problems explains everything that needs to be explained, then you get one sort of theory; if you hold that there is a further "hard" problem, then you get another. After a point, it is difficult to argue across this divide, and discussions are often reduced to table pounding. To me, it seems obvious that there is something further that needs explaining here; to others, it seems acceptable that there is not. (Informal surveys suggest that the numbers run two or three to one in favor of the former view, with the ratio fairly constant across academics and students in a variety of fields.) We may simply have to learn to live with this basic division.

This book may be of intellectual interest to those who think there is not much of a problem, but it is really intended for those who feel the problem in their bones. By now, we have a fairly good idea of the sort of theory we get if we assume there is no problem. In this work, I have tried to explore what follows given that there is a problem. The real argument of the book is that if one takes consciousness seriously, the position I lay out is where one should end up.

The second constraint I have followed is to take science seriously. I have not tried to dispute current scientific theories in domains where they have authority. At the same time, I have not been afraid to go out on a limb in areas where scientists' opinions are as ungrounded as everyone else's. For example, I have not disputed that the physical world is causally closed or that behavior can be explained in physical terms; but if a physicist or a cognitive scientist suggests that consciousness can be explained in physical terms, this is merely a hope ungrounded in current theory, and the question remains open. So I have tried to keep my ideas compatible with contemporary science, but I have not restricted my ideas to what contemporary scientists find fashionable.

The third constraint is that I take consciousness to be a natural phenomenon, falling under the sway of natural laws. If so, then there should be some correct scientific theory of consciousness, whether or not we can arrive at such a theory. That consciousness is a natural phenomenon seems hard to dispute: it is an extraordinarily salient part of nature, arising throughout the human species and very likely in many others. And we have every reason to believe that natural phenomena are subject to fundamental natural laws; it would be very strange if consciousness were not. This is not to say that the natural laws concerning consciousness will be just like laws in other domains, or even that they will be physical laws. They may be quite different in kind.

xiv Introduction

The problem of consciousness lies uneasily at the border of science and philosophy. I would say that it is properly a scientific subject matter: it is a natural phenomenon like motion, life, and cognition, and calls out for explanation in the way that these do. But it is not open to investigation by the usual scientific methods. Everyday scientific methodology has trouble getting a grip on it, not least because of the difficulties in observing the phenomenon. Outside the first-person case, data are hard to come by. This is not to say that no external data can be relevant, but we first have to arrive at a coherent philosophical understanding before we can justify the data's relevance. So the problem of consciousness may be a scientific problem that requires philosophical methods of understanding before we can get off the ground.

In this book I reach conclusions that some people may think of as "antiscientific": I argue that reductive explanation of consciousness is impossible, and I even argue for a form of dualism. But this is just part of the scientific process. Certain sorts of explanation turn out not to work, so we need to embrace other sorts of explanation instead. Everything I say here is compatible with the results of contemporary science; our picture of the natural world is broadened, not overturned. And this broadening allows the possibility of a naturalistic theory of consciousness that might have been impossible without it. It seems to me that to ignore the problems of consciousness would be antiscientific; it is in the scientific spirit to face up to them directly. To those who suspect that science requires materialism, I ask that you wait and see.

I should note that the conclusions of this work are conclusions, in the strongest sense. Temperamentally, I am strongly inclined toward materialist reductive explanation, and I have no strong spiritual or religious inclinations. For a number of years, I hoped for a materialist theory; when I gave up on this hope, it was quite reluctantly. It eventually seemed plain to me that these conclusions were forced on anyone who wants to take consciousness seriously. Materialism is a beautiful and compelling view of the world, but to account for consciousness, we have to go beyond the resources it provides.

By now, I have grown almost happy with these conclusions. They do not seem to have any fearsome consequences, and they allow a way of thinking and theorizing about consciousness that seems more satisfactory in almost every way. And the expansion in the scientific worldview has had a positive effect, at least for me: it has made the universe seem a more interesting place.


This book has four parts. In the first, I lay out the problems, and set up a framework within which they can be addressed. Chapter 1 is an introduction to consciousness, teasing apart a number of different concepts in the vicinity, drawing out the sense in which consciousness is really interesting, and giving a preliminary account of its subtle relation to the rest of the mind. Chapter

Taking Consciousness Seriously xv

2 develops a metaphysical and explanatory framework within which much of the rest of the discussion is cast. What is it for a phenomenon to be reductively explained, or to be physical? This chapter gives an account of these things, centering on the notion of Supervenience. I argue that there is good reason to believe that almost everything in the world can be reductively explained; but consciousness may be an exception.

With these preliminaries out of the way, the second part focuses on the irreducibility of consciousness. Chapter 3 argues that standard methods of reductive explanation cannot account for consciousness. I also give a critique of various reductive accounts that have been put forward by researchers in neuroscience, cognitive science, and elsewhere. This is not just a negative conclusion: it follows that a satisfactory theory of consciousness must be a new sort of nonreductive theory instead. Chapter 4 takes things a step further by arguing that materialism is false and that a form of dualism is true, and outlines the general shape that a nonreductive theory of consciousness might take. Chapter 5 is largely defensive: it considers some apparent problems for my view, involving the relationship between consciousness and our judgments about consciousness, and argues that they pose no fatal difficulties.

In the third part, I move toward a positive theory of consciousness. Each of the three chapters here develops a component of a positive theory. Chapter 6 focuses on the "coherence" between consciousness and cognitive processes, drawing a number of systematic links between the two. I use these links to analyze and ground the central role that neuroscience and cognitive science play in explaining human consciousness. Chapter 7 discusses the relation between consciousness and functional organization, using thought experiments to argue that consciousness is an "organizational invariant": that is, that every system with the right functional organization will have the same sort of conscious experience, no matter what it is made of. Chapter 8 considers what a fundamental theory of consciousness might look like, and suggests that it may involve a close relation between consciousness and information. This is by far the most speculative chapter, but at this point some speculation is probably needed if we are to make progress.

The last two chapters are dessert. Here, I apply what has gone before to central questions in the foundations of artificial intelligence and quantum mechanics. Chapter 9 argues for the thesis of "strong artificial intelligence": that the implementation of an appropriate computer program will give rise to a conscious mind. Chapter 10 considers the baffling question of how quantum mechanics should be interpreted, and uses the ideas about consciousness developed in previous chapters to lend support to a "no-collapse" interpretation of the theory.

Perhaps the negative material will provoke the most reaction, but my real goal is positive: I want to see a theory of consciousness that works. When I first came into philosophy, I was surprised to find that most of the debate

xvi Introduction

over consciousness focused on whether there was a problem or not, or on whether it was physical or not, and that the business of building theories seemed to be left to one side. The only "theories" seemed to be put forward by those who (by my lights) did not take consciousness seriously. By now, I have come to enjoy the intricacies of the Ontological debate as much as anyone, but a detailed theory is still my major goal. If some of the ideas in this book are useful to others in constructing a better theory, the attempt will have been worthwhile.

This book is intended as a serious work of philosophy, but I have tried to make it accessible to nonphilosophers. In my notional audience at all times has been my undergraduate self of ten years ago: I hope I have written a book that he would have appreciated. There are a few sections that are philosophically technical. These are marked with an asterisk (*), and readers should feel free to skip them. The most technical material is in Chapter 2 and Chapter 4. Section 4 of the former and sections 2 and 3 of the latter involve intricate issues in philosophical semantics, as does the final section of Chapter 5. Other asterisked sections might be worth at least skimming, to get an idea of what is going on. Often, I have put especially technical material and comments on the philosophical literature in the endnotes. The one technical concept that is crucial to the book is that of Supervenience, introduced at the start of Chapter 2. This concept has an intimidating name but it expresses a very natural idea, and a good understanding of it will help central issues fall into place. Much of the material later in this chapter can be skipped on a first reading, although one might want to return to it later to clarify questions as they arise.

For a short tour that avoids technicalities, read Chapter 1, skim the early parts of Chapter 2 as background material, then read all of Chapter 3 (skimming section 1 where necessary) for the central arguments against reductive explanation, and the first and last sections of Chapter 4 for the central considerations about dualism. The beginning of Chapter 6 is worth reading for the basic shape of the positive approach. Of the positive material, Chapter 7 is perhaps the most self-contained chapter as well as the most fun, with easy-to-understand thought experiments involving silicon brains; and those who like wild and woolly speculation might enjoy Chapter 8. Finally, Chapters 9 and 10 should make sense to anyone with an interest in the issues involved.


A couple of philosophical notes. The philosophical literature on consciousness is quite unsystematic, with seemingly independent strands talking about related issues without making contact with each other. I have attempted to impose some structure on the sprawl by providing a unifying framework in which the various metaphysical and explanatory issues become clear. Much of the discussion in the literature can be translated into this framework

Taking Consciousness Seriously xvii

without loss, and I hope the structure brings out the deep relationships between a number of different issues.

This work is perhaps unusual in largely eschewing the philosophical notion of identity (between mental and physical states, say) in favor of the notion of Supervenience. I find that discussions framed in terms of identity generally throw more confusion than light onto the key issues, and often allow the central difficulties to be evaded. By contrast, Supervenience seems to provide an ideal framework within which the key issues can be addressed. To avoid loose philosophy, however, we need to focus on the strength of the Supervenience connection: Is it underwritten by logical necessity, natural necessity, or something else? It is widely agreed that consciousness supervenes on the physical in some sense; the real question is how tight the connection is. Discussions that ignore these modal issues generally avoid the hardest questions about consciousness. Those skeptical of modal notions will be skeptical of my entire discussion, but I think there is no other satisfactory way to frame the issues.


One of the delights of working on this book, for me, has come from the way the problem of consciousness has reached out to make contact with deep issues in many other areas of science and philosophy. But the scope and depth of the problem also make it humbling. I am acutely aware that at almost every point in this book there is more that could be said, and that in many places I have only scratched the surface. But I hope, minimally, to have suggested that it is possible to make progress on the problem of consciousness without denying its existence or reducing it to something it is not. The problem is fascinating, and the future is exciting.

No. Xia stopped, twirling toward him in slow motion. Her icy mint eyes grew wide. You're in danger here. Panic whitened her face as she stared toward the house. Go home now. Before it's too late. And find me the antidote.

What kind of antidote?

Xia disappeared beyond the junipers, yet her final message burst into Joey's mind like the pop of a firecracker: The antidote for zombie poison.

Dian Curtis Regan, My Zombie Valentine

PART I. Foundations

1. Two Concepts of Mind

1. What Is Consciousness?

Conscious experience is at once the most familiar thing in the world and the most mysterious. There is nothing we know about more directly than consciousness, but it is far from clear how to reconcile it with everything else we know. Why does it exist? What does it do? How could it possibly arise from lumpy gray matter? We know consciousness far more intimately than we know the rest of the world, but we understand the rest of the world far better than we understand consciousness.

Consciousness can be startlingly intense. It is the most vivid of phenomena; nothing is more real to us. But it can be frustratingly diaphanous: in talking about conscious experience, it is notoriously difficult to pin down the subject matter. The International Dictionary of Psychology does not even try to give a straightforward characterization:

Consciousness: The having of perceptions, thoughts, and feelings; awareness. The term is impossible to define except in terms that are unintelligible without a grasp of what consciousness means. Many fall into the trap of confusing consciousness with self-consciousness-to be conscious it is only necessary to be aware of the external world. Consciousness is a fascinating but elusive phenomenon: it is impossible to specify what it is, what it does, or why it evolved. Nothing worth reading has been written about it. (Sutherland 1989)

Almost anyone who has thought hard about consciousness will have some sympathy with these sentiments. Consciousness is so intangible that even this limited attempt at a definition could be disputed: there can arguably be perception and thought that is not conscious, as witnessed by the notions of subliminal perception and unconscious thought. What is central to conscious-

4 Foundations

ness, at least in the most interesting sense, is experience. But this is not definition. At best, it is clarification.

Trying to define conscious experience in terms of more primitive notions is fruitless. One might as well try to define matter or space in terms of something more fundamental. The best we can do is to give illustrations and characterizations that lie at the same level. These characterizations cannot qualify as true definitions, due to their implicitly circular nature, but they can help to pin down what is being talked about. I presume that every reader has conscious experiences of his or her own. If all goes well, these characterizations will help establish that it is just those that we are talking about.

The subject matter is perhaps best characterized as "the subjective quality of experience." When we perceive, think, and act, there is a whir of causation and information processing, but this processing does not usually go on in the dark. There is also an internal aspect; there is something it feels like to be a cognitive agent. This internal aspect is conscious experience. Conscious experiences range from vivid color sensations to experiences of the faintest background aromas; from hard-edged pains to the elusive experience of thoughts on the tip of one's tongue; from mundane sounds and smells to the encompassing grandeur of musical experience; from the triviality of a nagging itch to the weight of a deep existential angst; from the specificity of the taste of peppermint to the generality of one's experience of selfhood. All these have a distinct experienced quality. All are prominent parts of the inner life of the mind.

We can say that a being is conscious if there is something it is like to be that being, to use a phrase made famous by Thomas Nagel.1 Similarly, a mental state is conscious if there is something it is like to be in that mental state. To put it another way, we can say that a mental state is conscious if it has a qualitative feel-an associated quality of experience. These qualitative feels are also known as phenomenal qualities, or qualia for short.2 The problem of explaining these phenomenal qualities is just the problem of explaining consciousness. This is the really hard part of the mind-body problem.

Why should there be conscious experience at all? It is central to a subjective viewpoint, but from an objective viewpoint it is utterly unexpected. Taking the objective view, we can tell a story about how fields, waves, and particles in the spatiotemporal manifold interact in subtle ways, leading to the development of complex systems such as brains. In principle, there is no deep philosophical mystery in the fact that these systems can process information in complex ways, react to stimuli with sophisticated behavior, and even exhibit such complex capacities as learning, memory, and language. All this is impressive, but it is not metaphysically baffling. In contrast, the existence of conscious experience seems to be a new feature from this viewpoint. It is not something that one would have predicted from the other features alone.

Two Concepts of Mind 5

That is, consciousness is surprising. If all we knew about were the facts of physics, and even the facts about dynamics and information processing in complex systems, there would be no compelling reason to postulate the existence of conscious experience. If it were not for our direct evidence in the first-person case, the hypothesis would seem unwarranted; almost mystical, perhaps. Yet we know, directly, that there is conscious experience. The question is, how do we reconcile it with everything else we know?

Conscious experience is part of the natural world, and like other natural phenomena it cries out for explanation. There are at least two major targets of explanation here. The first and most central is the very existence of consciousness. Why does conscious experience exist? If it arises from physical systems, as seems likely, how does it arise? This leads to some more specific questions. Is consciousness itself physical, or is it merely a concomitant of physical systems? How widespread is consciousness? Do mice, for example, have conscious experience?

A second target is the specific character of conscious experiences. Given that conscious experience exists, why do individual experiences have their particular nature? When I open my eyes and look around my office, why do I have this sort of complex experience? At a more basic level, why is seeing red like this, rather than like that! It seems conceivable that when looking at red things, such as roses, one might have had the sort of color experiences that one in fact has when looking at blue things. Why is the experience one way rather than the other? Why, for that matter, do we experience the reddish sensation3 that we do, rather than some entirely different kind of sensation, like the sound of a trumpet?

When someone strikes middle C on the piano, a complex chain of events is set into place. Sound vibrates in the air and a wave travels to my ear. The wave is processed and analyzed into frequencies inside the ear, and a signal is sent to the auditory cortex. Further processing takes place here: isolation of certain aspects of the signal, categorization, and ultimately reaction. All this is not so hard to understand in principle. But why should this be accompanied by an experience? And why, in particular, should it be accompanied by that experience, with its characteristic rich tone and timbre? These are two central questions that we would like a theory of consciousness to answer.

Ultimately one would like a theory of consciousness to do at least the following: it should give the conditions under which physical processes give rise to consciousness, and for those processes that give rise to consciousness, it should specify just what sort of experience is associated. And we would like the theory to explain how it arises, so that the emergence of consciousness seems intelligible rather than magical. In the end, we would like the theory to enable us to see consciousness as an integral part of the natural world. Currently it may be hard to see what such a theory would be like, but without such a theory we could not be said to fully understand consciousness.

6 Foundations

Before proceeding, a note on terminology. The term "consciousness" is ambiguous, referring to a number of phenomena. Sometimes it is used to refer to a cognitive capacity, such as the ability to introspect or to report one's mental states. Sometimes it is used synonymously with "awakeness." Sometimes it is closely tied to our ability to focus attention, or to voluntarily control our behavior. Sometimes "to be conscious of something" comes to the same thing as "to know about something." All of these are accepted uses of the term, but all pick out phenomena distinct from the subject I am discussing, and phenomena that are significantly less difficult to explain. I will say more about these alternative notions of consciousness later, but for now, when I talk about consciousness, I am talking only about the subjective quality of experience: what it is like to be a cognitive agent.

A number of alternative terms and phrases pick out approximately the same class of phenomena as "consciousness" in its central sense. These include "experience," "qualia," "phenomenology," "phenomenal," "subjective experience," and "what it is like." Apart from grammatical differences, the differences among these terms are mostly subtle matters of connotation. "To be conscious" in this sense is roughly synonymous with "to have qualia," "to have subjective experience," and so on. Any differences in the class of phenomena picked out are insignificant. Like "consciousness," many of these terms are somewhat ambiguous, but I will never use these terms in the alternative senses. I will use all these phrases in talking about the central phenomenon of this book, but "consciousness" and "experience" are the most straightforward terms, and it is these terms that will recur.

A catalog of conscious experiences

Conscious experience can be fascinating to attend to. Experience comes in an enormous number of varieties, each with its own character. A far-from-complete catalog of the aspects of conscious experience is given in the following pretheoretical, impressionistic list. Nothing here should be taken too seriously as philosophy, but it should help focus attention on the subject matter at hand.


Visual experiences.

Visual experiences. Among the many varieties of visual experience, color sensations stand out as the paradigm examples of conscious experience, due to their pure, seemingly ineffable qualitative nature. Some color experiences can seem particularly striking, and so can be particularly good at focusing our attention on the mystery of consciousness. In my environment now, there is a particularly rich shade of deep purple from a book on my shelf; an almost surreal shade of green in a photograph of ferns on my wall; and a sparkling array of bright red, green, orange, and blue lights on a Christmas

Two Concepts of Mind 7

tree that I can see through my window. But any color can be awe-provoking if we attend to it, and reflect upon its nature. Why should it feel like that"? Why should it feel like anything at all? How could I possibly convey the nature of this color experience to someone who has not had such an experience?

Other aspects of visual experience include the experience of shape, of size, of brightness, and of darkness. A particularly subtle aspect is the experience of depth. As a child, one of my eyes had excellent vision, but the other was very poor. Because of my one good eye, the world looked crisp and sharp, and it certainly seemed three-dimensional. One day, I was fitted with glasses, and the change was remarkable. The world was not much sharper than before, but it suddenly looked more three-dimensional: things that had depth before somehow got deeper, and the world seemed a richer place. If you cover one eye and then uncover it, you can get an idea of the change. In my previous state, I would have said that there was no way for the depth of my vision to improve; the world already seemed as three-dimensional as it could be. The change was subtle, almost ineffable, but extremely striking. Certainly there is an intellectual story one can tell about how binocular vision allows information from each eye to be consolidated into information about distances, thus enabling more sophisticated control of action, but somehow this causal story does not reveal the way the experience felt. Why that change in processing should be accompanied by such a remaking of my experience was mysterious to me as a ten-year-old, and is still a source of wonder today.


Auditory experiences.

Auditory experiences. In some ways, sounds are even stranger than visual images. The structure of images usually corresponds to the structure of the world in a straightforward way, but sounds can seem quite independent. My telephone receives an incoming call, an internal device vibrates, a complex wave is set up in the air and eventually reaches my eardrum, and somehow, almost magically, I hear a ring. Nothing about the quality of the ring seems to correspond directly to any structure in the world, although I certainly know that it originated with the speaker, and that it is determined by a waveform. But why should that waveform, or even these neural firings, have given rise to a sound quality like that?

Musical experience is perhaps the richest aspect of auditory experience, although the experience of speech must be close. Music is capable of washing over and completely absorbing us, surrounding us in a way that a visual field can surround us but in which auditory experiences usually do not. One can analyze aspects of musical experience by breaking the sounds we perceive into notes and tones with complex interrelationships, but the experience of music somehow goes beyond this. A unified qualitative experience arises

8 Foundations

Figure 1.1. Effability and ineffability in olfactory experience.

Figure 1.1. Effability and ineffability in olfactory experience. (Calvin and Hobbes © Watterson. Distributed by Universal Press Syndicate. Reprinted with permission. All rights reserved)

Calvin and Hobbes by Bill Watterson

from a chord, but not from randomly selected notes. An old piano and a far-off oboe can combine to produce an unexpectedly haunting experience. As always, when we reflect, we ask the question: why should that feel like this?


Tactile experiences.

Tactile experiences. Textures provide another of the richest quality spaces that we experience: think of the feel of velvet, and contrast it to the texture of cold metal, or a clammy hand, or a stubbly chin. All of these have their own unique quality. The tactile experiences of water, of cotton candy, or of another person's lips are different again.


Olfactory experiences

Olfactory experiences. Think of the musty smell of an old wardrobe, the stench of rotting garbage, the whiff of newly mown grass, the warm aroma of freshly baked bread. Smell is in some ways the most mysterious of all the senses, due to the rich, intangible, indescribable nature of smell sensations. Ackermann (1990) calls it "the mute sense; the one without words." While there is something ineffable about any sensation, the other senses have properties that facilitate some description. Visual and auditory experiences have a complex combinatorial structure that can be described. Tactile and taste experiences generally arise from direct contact with some object, and a rich descriptive vocabulary has been built up by reference to these objects. Smell has little in the way of apparent structure and often floats free of any apparent object, remaining a primitive presence in our sensory manifold. (Perhaps animals might do better [Figure 1.1].) The primitiveness is perhaps partly due to the slot-and-key process by which our olfactory receptors are sensitive to various kinds of molecules. It seems arbitrary that a given sort of molecule should give rise to this sort of sensation, but give rise it does.

Two Concepts of Mind 9

Taste experiences.

Taste experiences. Psychophysical investigations tell us that there are only four independent dimensions of taste perception: sweet, sour, bitter, and salt. But this four-dimensional space combines with our sense of smell to produce a great variety of possible experiences: the taste of Turkish Delight, of curried black-eyed pea salad,4 of a peppermint Lifesaver, of a ripe peach.


Experiences of hot and cold.

Experiences of hot and cold. An oppressively hot, humid day and a frosty winter's day produce strikingly different qualitative experiences. Think also of the heat sensations on one's skin from being close to a fire, and the hot-cold sensation that one gets from touching ultracold ice.



Pain. Pain is a paradigm example of conscious experience, beloved by philosophers. Perhaps this is because pains form a very distinctive class of qualitative experiences, and are difficult to map directly onto any structure in the world or in the body, although they are usually associated with some part of the body. Because of this, pains can seem even more subjective than most sensory experiences. There are a great variety of pain experiences, from shooting pains and fierce burns through sharp pricks to dull aches.


Other bodily sensations.

Other bodily sensations. Pains are only the most salient kind of sensations associated with particular parts of the body. Others include headaches (which are perhaps a class of pain), hunger pangs, itches, tickles, and the experience associated with the need to urinate. Many bodily sensations have an entirely unique quality, different in kind from anything else in our experience: think of orgasms, or the feeling of hitting one's funny bone. There are also experiences associated with proprioception, the sense of where one's body is in space.


Mental imagery.

Mental imagery. Moving ever inward, toward experiences that are not associated with particular objects in the environment or the body but that are in some sense generated internally, we come to mental images. There is often a rich phenomenology associated with visual images conjured up in one's imagination, though not nearly as detailed as those derived from direct visual perception. There are also the interesting colored patterns that one gets when one closes one's eyes and squints, and the strong after-images that one gets after looking at something bright. One can have similar kinds of auditory "images" conjured up by one's imagination, and even tactile, olfactory, and gustatory images, although these are harder to pin down and their associated qualitative feel is usually fainter.


Conscious thought.

Conscious thought. Some of the things we think and believe do not have any particular qualitative feel associated with them, but many do. This ap-

10 Foundations

plies particularly to explicit, occurrent thoughts that one thinks to oneself, and to various thoughts that affect one's stream of consciousness. It is often hard to pin down just what the qualitative feel of an occurrent thought is, but it is certainly there. There is something it is like to be having such thoughts. When I think of a lion, for instance, there seems to be a whiff of leonine quality to my phenomenology: what it is like to think of a lion is subtly different from what it is like to think of the Eiffel tower. More obviously, cognitive attitudes such as desire often have a strong phenomenal flavor. Desire seems to exert a phenomenological "tug," and memory often has a qualitative component, as with the experience of nostalgia or regret.



Emotions. Emotions often have distinctive experiences associated with them. The sparkle of a happy mood, the weariness of a deep depression, the red-hot glow of a rush of anger, the melancholy of regret: all of these can affect conscious experience profoundly, although in a much less specific way than localized experiences such as sensations. These emotions pervade and color all of our conscious experiences while they last.

Other more transient feelings lie partway between emotions and the more obviously cognitive aspects of mind. Think of the rush of pleasure one feels when one gets a joke. Another example is the feeling of tension one gets when watching a suspense movie, or when waiting for an important event. The butterflies in one's stomach that can accompany nervousness also fall into this class.


The sense of self.

The sense of self. One sometimes feels that there is something to conscious experience that transcends all these specific elements: a kind of background hum, for instance, that is somehow fundamental to consciousness and that is there even when the other components are not. This phenomenology of self is so deep and intangible that it sometimes seems illusory, consisting in nothing over and above specific elements such as those listed above. Still, there seems to be something to the phenomenology of self, even if it is very hard to pin down.


This catalog covers a number of bases, but leaves out as much as it puts in. I have said nothing, for instance, about dreams, arousal and fatigue, intoxication, or the novel character of other drug-induced experiences. There are also rich experiences that derive their character from the combination of two or many of the components described above. I have mentioned the combined effects of smell and taste, but an equally salient example is the combined experience of music and emotion, which interact in a subtle, difficult-to-separate way. I have also left aside the unity of conscious experience-the way that all of these experiences seem to be tied together as the experience of a single experiencer. Like the sense of self, this unity sometimes

Two Concepts of Mind 11

seems illusory-it is certainly harder to pin down than any specific experiences-but there is a strong intuition that unity is there.

Sad to say, we will not again be involved this closely with the rich varieties of conscious experience. In addressing the philosophical mysteries associated with conscious experience, a simple color sensation raises the problems as deeply as one's experience of a Bach chorale. The deep issues cut across these varieties in a way that renders consideration of the nature of specific experiences not especially relevant. Still, this brief look at the rich varieties of conscious experience should help focus attention on just what it is that is under discussion, and provides a stock of examples that can be kept in mind during more abstract discussion.5

2. The Phenomenal and the Psychological Concepts of Mind

Conscious experience is not all there is to the mind. To see this, observe that although modern cognitive science has had almost nothing to say about consciousness, it has had much to say about mind in general. The aspects of mind with which it is concerned are different. Cognitive science deals largely in the explanation of behavior, and insofar as it is concerned with mind at all, it is with mind construed as the internal basis of behavior, and with mental states construed as those states relevant to the causation and explanation of behavior. Such states may or may not be conscious. From the point of view of cognitive science, an internal state responsible for the causation of behavior is equally mental whether it is conscious or not.

At the root of all this lie two quite distinct concepts of mind. The first is the phenomenal concept of mind. This is the concept of mind as conscious experience, and of a mental state as a consciously experienced mental state. This is the most perplexing aspect of mind and the aspect on which I will concentrate, but it does not exhaust the mental. The second is the psychological concept of mind. This is the concept of mind as the causal or explanatory basis for behavior. A state is mental in this sense if it plays the right sort of causal role in the production of behavior, or at least plays an appropriate role in the explanation of behavior. According to the psychological concept, it matters little whether a mental state has a conscious quality or not. What matters is the role it plays in a cognitive economy.

On the phenomenal concept, mind is characterized by the way it feels; on the psychological concept, mind is characterized by what it does. There should be no question of competition between these two notions of mind. Neither of them is the correct analysis of mind. They cover different phenomena, both of which are quite real.

12 Foundations

I will sometimes speak of the phenomenal and psychological "aspects" of mind, and sometimes of the "phenomenal mind" and the "psychological mind." At this early stage, I do not wish to beg any questions about whether the phenomenal and the psychological will turn out to be the same thing. Perhaps every phenomenal state is a psychological state, in that it plays a significant role in the causation and explanation of behavior, and perhaps every psychological state has an intimate relation to the phenomenal. For now, all that counts is the conceptual distinction between the two notions: what it means for a state to be phenomenal is for it to feel a certain way, and what it means for a state to be psychological is for it to play an appropriate causal role. These distinct notions should not be conflated, at least at the outset.

A specific mental concept can usually be analyzed as a phenomenal concept, a psychological concept, or as a combination of the two. For instance, sensation, in its central sense, is best taken as a phenomenal concept: to have a sensation is to have a state with a certain sort of feel. On the other hand, the concepts of learning and memory might best be taken as psychological. For something to learn, at a first approximation, is for it to adapt its behavioral capacities appropriately in response to certain kinds of environmental stimulation. In general, a phenomenal feature of the mind is characterized by what it is like for a subject to have that feature, while a psychological feature is characterized by an associated role in the causation and/or explanation of behavior.

Of course, this usage of the term "psychological" is a stipulation: it arises from identifying psychology with cognitive science as described above. The everyday concept of a "psychological state" is probably broader than this, and may well include elements of the phenomenal. But nothing will rest on my use of the term.

A potted history

The phenomenal and the psychological aspects of mind have a long history of being conflated. Rene Descartes may have been partly responsible for this. With his notorious doctrine that the mind is transparent to itself, he came close to identifying the mental with the phenomenal. Descartes held that every event in the mind is a cogitatio, or a content of experience. To this class he assimilated volitions, intentions, and every type of thought. In his reply to the Fourth Set of Objections, he wrote:


As to the fact that there can be nothing in the mind, in so far as it is a thinking thing, of which it is not aware, this seems to me to be self-evident. For there is nothing that we can understand to be in the mind, regarded in this way, that is not a thought or dependent on a thought. If it were not a thought nor

Two Concepts of Mind 13

dependent on a thought it would not belong to the mind qua thinking thing; and we cannot have any thought of which we are not aware at the very moment it is in us.


If Descartes did not actually identify the psychological with the phenomenal, he at least assumed that everything psychological that is worthy of being called mental has a conscious aspect.6 To Descartes, the notion of an unconscious mental state was a contradiction.

Progress in psychological theory rather than in philosophy was responsible for drawing the two aspects of mind apart. As recently as a century ago, psychologists such as Wilhelm Wundt and William James were recognizably Cartesian in that they used introspection to investigate the causes of behavior, and developed psychological theories on the basis of introspective evidence. In this fashion, phenomenology was made the arbiter of psychology. But developments soon after established the psychological as an autonomous domain.

Most notably, Sigmund Freud and his contemporaries solidified the idea that many activities of the mind are unconscious, and that there can be such things as unconscious beliefs and desires. The very fact that this notion seemed coherent is evidence that a nonphenomenal analysis of thought was being used. It appears that Freud construed the notions causally. Desire, very roughly, was implicitly construed as the sort of state that brings about a certain kind of behavior associated with the object of the desire. Belief was construed according to its causal role in a similar way. Of course Freud did not make these analyses explicit, but something along these lines clearly underlies his use of the notions. Explicitly, he recognized that accessibility to consciousness is not essential to a state's relevance in the explanation of behavior, and that a conscious quality is not constitutive of something's being a belief or a desire. These conclusions rely on a notion of mentality that is independent of phenomenal notions.

Around the same time, the behaviorist movement in psychology had thoroughly rejected the introspectionist tradition. A new "objective" brand of psychological explanation was developed, with no room for consciousness in its explanations. This mode of explanation had only partial success, but it established the idea that psychological explanation can proceed while the phenomenal is ignored. Behaviorists differed in their theoretical positions: some recognized the existence of consciousness but found it irrelevant to psychological explanation, and some denied its existence altogether. Many went further, denying the existence of any kind of mental state. The official reason for this was that internal states were supposed to be irrelevant in the explanation of behavior, which could be carried out entirely in external terms. Perhaps a deeper reason is that all mental notions were tainted with the disreputable odor of the phenomenal.

14 Foundations

In any case, these two developments established as orthodoxy the idea that explanation of behavior is in no way dependent on phenomenal notions. The move from behaviorism to computational cognitive science for the most part preserved this orthodoxy. Although the move brought back a role for internal states, which could even be called "mental" states, there was nothing particularly phenomenal about them. These states were admissible precisely on the grounds of their relevance in the explanation of behavior; any associated phenomenal quality was at best beside the point. The concept of the mental as psychological thus had center stage.

In philosophy, the shift in emphasis from the phenomenal to the psychological was codified by Gilbert Ryle (1949), who argued that all our mental concepts can be analyzed in terms of certain kinds of associated behavior, or in terms of dispositions to behave in certain ways.7 This view, logical behaviorism, is recognizably the precursor of much of what passes for orthodoxy in contemporary philosophy of psychology. In particular, it was the most explicit codification of the link between mental concepts and the causation of behavior.

Ryle did not put this theory forward as an analysis of just some mental concepts. He intended all of them to fall within its grasp. It seemed to many people, as it seems to me, that this view is a nonstarter as an analysis of our phenomenal concepts, such as sensation and consciousness itself. To many, it seemed clear that when we talk about phenomenal states, we are certainly not talking about our behavior, or about any behavioral disposition. But in any case, Ryle's analysis provided a suggestive approach to many other mental notions, such as believing, enjoying, wanting, pretending, and remembering.

Apart from its problems with phenomenal states, Ryle's view had some technical problems. First, it is natural to suppose that mental states cause behavior, but if mental states are themselves behavioral or behavioral dispositions, as opposed to internal states, then it is hard to see how they could do the job. Second, it was argued (by Chisholm [1957] and Geach [1957]) that no mental state could be defined by a single range of behavioral dispositions, independent of any other mental states. For example, if one believes that it is raining, one's behavioral dispositions will vary depending on whether one has the desire to get wet. It is therefore necessary to invoke other mental states in characterizing the behavioral dispositions associated with a given sort of mental state.

These problems were finessed by what has become known as functionalism, which was developed by David Lewis (1966) and most thoroughly by David Armstrong (1968).8 On this view, a mental state is defined wholly by its causal role: that is, in terms of the kinds of stimulation that tend to produce it, the kind of behavior it tends to produce, and the way it interacts with other mental states. This view made mental states fully internal and

Two Concepts of Mind 15

able to stand in the right kind of causal relation to behavior, answering the first objection, and it allowed mental states to be defined in terms of their interaction with each other, answering the second objection.

On this view, our mental concepts can be analyzed functionally: in terms of their actual or typical causes and effects. To give such an analysis for any given mental concept is highly nontrivial; Armstrong (1968) gives a number of analyses, but these are very incomplete. As an in-principle position, however, functionalism may provide a reasonable construal of many of our mental concepts, at least insofar as they play a role in the explanation of behavior. For instance, the notion of learning might be analyzed as the adaptation of one's behavioral capacities in response to environmental stimulation. To take a more complex state, a belief that it is raining might be very roughly analyzed as the sort of state that tends to be produced when it is raining, that leads to behavior that would be appropriate if it were raining, that interacts inferentially with other beliefs and desires in a certain sort of way, and so on. There is a lot of room for working out the details, but many have found the overall idea to be on the right track.

Like Ryle, however, Armstrong and Lewis did not put this forward as an analysis of some mental concepts. Rather, it was meant as an analysis of all mental concepts. In particular, they argued that the notions of experience, sensation, consciousness, and so on, could be analyzed in this fashion. This assimilation of the phenomenal to the psychological seems to me to be as great an error as Descartes's assimilation of the psychological to the phenomenal. It is simply a false analysis of what it means to be phenomenal. When we wonder whether somebody is having a color experience, we are not wondering whether they are receiving environmental stimulation and processing it in a certain way. We are wondering whether they are experiencing a color sensation, and this is a distinct question. It is a conceptually coherent possibility that something could be playing the causal role without there being an associated experience.

To put the point a different way, note that this analysis of phenomenal concepts leaves it unclear why anybody was ever bothered by the problem in the first place.9 There is no great mystery about how a state might play some causal role, although there are certainly technical problems there for science. What is mysterious is why that state should feel like something; why it should have a phenomenal quality. Why the causal role is played and why the phenomenal quality is present are two entirely different questions. The functionalist analysis denies the distinctness of these questions, and therefore seems to be unsatisfactory.

I will consider this matter in much more detail later, but for now we can note that even if the functionalist account gives an unsatisfactory analysis of phenomenal concepts, it may provide a good analysis of other mental notions, such as learning and memory, and perhaps belief. No parallel wor-

16 Foundations

ries come up in these cases. It seems no more mysterious that a system should be able to learn than that a system should be able to adapt its behavior in response to environmental stimulation; indeed, these seem to be more or less the same question. Similarly, when we wonder whether somebody has learned something, it seems reasonable to say that in doing this we are wondering whether they have undergone a change that will give rise to an improved capacity to deal with certain situations in the future. Of course, a thorough analysis of the concept of learning will be more subtle than this first approximation, but the further details of the analysis will be spelled out within the same framework.

Indeed, the functionalist account corresponds precisely to the definition I have given of psychological properties. Most nonphenomenal mental properties fall into this class, and can therefore be functionally analyzed. There is certainly room to argue over the details of a specific functionalist analysis. There are also significant framework questions about such matters as the role of the environment in characterizing psychological properties, and whether it is causation, explanation, or both that provides the defining link between psychological properties and behavior. These details are relatively unimportant here, though. What matters is that nonphenomenal mental states are largely characterized by their role in our cognitive economy.

The moral of this discussion is that both the psychological and the phenomenal are real and distinct aspects of mind. At a first approximation, phenomenal concepts deal with the first-person aspects of mind, and psychological concepts deal with the third-person aspects. One's approach to the mind will be quite different depending on what aspects of the mind one is interested in. If one is interested in the mind's role in bringing about behavior, one will focus on psychological properties. If one is interested in the conscious experience of mental states, one will focus on phenomenal properties. Neither the phenomenal nor the psychological should be defined away in terms of the other. Conceivably, some deep analysis might reveal a fundamental link between the phenomenal and the psychological, but this would be a nontrivial task, and is not something to be accomplished by prior stipulation. To assimilate the phenomenal to the psychological prior to some deep explanation would be to trivialize the problem of conscious experience; and to assimilate the psychological to the phenomenal would be to vastly limit the role of the mental in explaining behavior.

3. The Double Life of Mental Terms

It seems reasonable to say that together, the psychological and the phenomenal exhaust the mental. That is, every mental property is either a phenomenal

Two Concepts of Mind 17

property, a psychological property, or some combination of the two. Certainly, if we are concerned with those manifest properties of the mind that cry out for explanation, we find first, the varieties of conscious experience, and second, the causation of behavior. There is no third kind of manifest explanandum, and the first two sources of evidence-experience and behavior-provide no reason to believe in any third kind of nonphenomenal, nonfunctional properties (with perhaps a minor exception for relational properties, discussed shortly). There are certainly other classes of mental states of which we often speak-intentional states, emotional states, and so on-but it is plausible that these can be assimilated to the psychological, the phenomenal, or a combination of the two.

Things are complicated by the fact that many everyday mental concepts straddle the fence, having both a phenomenal and a psychological component. Pain provides a clear example. The term is often used to name a particular sort of unpleasant phenomenal quality, in which case a phenomenal notion is central. But there is also a psychological notion associated with the term: roughly, the concept of the sort of state that tends to be produced by damage to the organism, tends to lead to aversion reactions, and so on. Both of these aspects are central to the commonsense notion of pain. We might say that the notion of pain is ambiguous between the phenomenal and the psychological concept, or we might say that both of these are components of a single rich concept.

One can tie oneself into all kinds of knots by worrying about whether the phenomenal quality or the functional role is more essential to pain. For instance, would a hypothetical system in which all the functional criteria were satisfied but in which the conscious experience were not present be truly in pain? One might be tempted to say no, but what of the fact that we speak of pains that last for a day, even though there are times when they are not conscious? There is little point trying to legislate matters one way or the other. Nothing important rests on the semantic decision as to whether some phenomenal quality is really essential for something to count as pain. Instead, we can recognize the different components associated with a concept and explicitly distinguish them, speaking for example of "phenomenal pain" and "psychological pain." Our everyday concept of pain presumably combines the two in some subtle weighted combination, but for philosophical discussion things are clearer if we keep them separate.

The reason why phenomenal and psychological properties are often run together is clear: it is because the relevant properties tend to co-occur. Generally, when the processes resulting from tissue damage and leading to aversion reaction take place, some sort of phenomenal quality is instantiated. That is, when psychological pain is present, phenomenal pain is usually also present. It is not a conceptual truth that the process should be accompanied

18 Foundations

by the phenomenal quality, but it is a fact about the world. Once we have this sort of co-occurrence of properties in everyday situations, it is natural that our everyday concepts will bind them together.

Many mental concepts lead this sort of double life. For example, the concept of perception can be taken wholly psychologically, denoting the process whereby cognitive systems are sensitive to environmental stimulation in such a way that the resulting states play a certain role in directing cognitive processes. But it can also be taken phenomenally, involving the conscious experience of what is perceived. The possibility of subliminal perception counts against the latter construal, but some would argue that this qualifies as perception only in a weakened sense of the term. Once again, however, the issue is terminological. When we want to be clear, we can simply stipulate whether it is the psychological property, the phenomenal property, or a combination that we are concerned with.

Still, some of these dual concepts lean more strongly toward the phenomenal, and some lean toward the psychological. Take the concept of sensation, which is closely related to the concept of perception and which also has both phenomenal and psychological components. The phenomenal component is much more prominent in "sensation" than in "perception," as witnessed by the fact that the idea of unconscious perception seems to make more sense than that of unconscious sensation. Things are still somewhat gray-there remains a sense of "perception" that requires conscious experience, and a sense of "sensation" that does not-but these senses seem less central than the alternatives. Perhaps it is most natural to use "perception" as a psychological term, and "sensation" as a phenomenal term. This way, we can see sensation as something like perception's phenomenal counterpart.

A good test for whether a mental notion M is primarily psychological is to ask oneself: Could something be an instance of M without any particular associated phenomenal quality? If so, then M is likely psychological. If not, then M is phenomenal, or at least a combined notion that centrally involves phenomenology. The latter possibility cannot be ruled out, as some concepts may require both an appropriate sort of phenomenal quality and an appropriate cognitive role; perhaps a central sense of "sensation" has this combined character, for example. But we can at least separate those notions that involve phenomenology from those that do not.

The test suggests that a concept such as learning, for example, is largely psychological. To a first approximation, to learn is just for one's cognitive capacities to adapt in a certain way to various new circumstances and stimuli. No particular phenomenal quality is required for a cognitive process to be an instance of learning; such a quality may be present, but it is not what makes the process count as an example of learning. There may be a slight phenomenal tinge inherited from a link with concepts such as belief, dis-

Two Concepts of Mind 19

cussed below, but this is faint at best. In explaining learning, the central thing we have to explain is how the system manages to adapt in the appropriate way. Something similar goes for concepts such as those of categorization and memory, which seem to be largely psychological notions, in that what is central is the playing of an appropriate cognitive role.

Emotions have a much clearer phenomenal aspect. When we think of happiness and sadness, a distinct variety of conscious experience comes to mind. It is not quite obvious whether the phenomenal aspect is essential for a state to be an emotion, however; there is clearly a strong associated psychological property as well. As usual, we need not make any decision on this matter. We can simply talk about the psychological and phenomenal aspects of emotion, and observe that these exhaust the aspects of emotion that require explanation.

The most complex case is that of mental states such as belief, often called "prepositional attitudes" because they are attitudes to propositions concerning the world. When I believe that Bob Dylan will tour Australia, for example, I endorse a certain proposition concerning Dylan; when I hope that Dylan will tour Australia, I have a different attitude toward the same proposition. The central feature of these mental states is their semantic aspect, or Intentionality: the fact that they are about things in the world. That is, a belief has semantic content: the content of my belief cited above is something like the proposition that Dylan will tour Australia (although there is room for debate here).

Belief is most often regarded as a psychological property. On this view, at a rough first approximation, to believe that a proposition is true is to be in a state wherein one acts in a way that would be appropriate if it were true, a state that tends to be brought about by its being the case, and a state in which one's cognitive dynamics of reasoning reflect the appropriate interaction of the belief with other beliefs and desires. The functional criteria for belief are very subtle, however, and no one has yet produced anything like a complete analysis of the relevant criteria. All the same, there is reason to believe that this view captures much of what is significant about belief. It is related to the idea that belief is something of an explanatory construct: we attribute beliefs to others largely in order to explain their behavior.

Some would argue that this leaves something out, and that something over and above the relevant sort of psychological process is required for belief. In particular, it leaves out the experiential aspects of believing, which some have argued are essential for anything to count as a belief. For example, Searle (1990a) has argued that the intentional content of a belief depends entirely on the associated state of consciousness, or on a state of consciousness that the belief can bring about. Without consciousness, all that is present is "as-if" intentionality.10

20 Foundations

Certainly, there is often conscious experience in the vicinity of belief: there is something it is like when one has an occurrent (i.e., conscious) belief, and most nonoccurrent beliefs can at least bring about a conscious belief. The crucial questions, though, are whether this conscious quality is what makes the state a belief, and whether it is what gives it the content it has. This may be more plausible for some beliefs than for others: for example, one might argue that a conscious quality is required to truly have beliefs about one's experiences, and perhaps also certain sorts of experiences are required to have certain sorts of perceptual beliefs about the external world (perhaps one needs red experiences to believe that an object is red?). In other cases, this seems more problematic. For example, when I think that Don Bradman is the greatest cricketer of all time, it seems plausible to say that I would have had the same belief even if I had had a very different conscious experience associated with it. The phenomenology of the belief is relatively faint, and it is hard to see how it could be this phenomenal quality that makes the belief a belief about Bradman. What seems more central to the belief's content is the connection between the belief and Bradman, and the role it plays in my cognitive system.

As a weaker position, it might be suggested that although no particular phenomenal quality is required to have a particular belief, a being must at least be capable of conscious experience to believe anything at all.11 There is a certain plausibility in the idea that a being with no conscious inner life would not truly be a believer; at best, it would be only a pseudobeliever. All the same, this would make the role of the phenomenal in intentional concepts quite thin. The most substantial requirements for having a specific belief will lie elsewhere than in the phenomenal. One could even subtract any phenomenal component out, leaving a concept of pseudobelief that resembles belief in most important respects except that it does not involve the concept of consciousness. Indeed, it is plausible that pseudobelief could do most of the explanatory work that is done by the concept of belief.

In any case, I will not try to adjudicate these difficult issues about the relationship between intentionality and consciousness here. We can note that there is at least a deflationary concept of belief that is purely psychological, not involving conscious experience; if a being is in the right psychological state, then it is in a state that resembles belief in many important ways, except with respect to any phenomenal aspects. And there is an inflationary concept of belief, on which conscious experience is required for truly believing, and perhaps even on which a specific sort of conscious experience is required for truly believing a specific proposition. Which of these is the "true" concept of belief will not matter too much for my purposes.

What is central is that there is not any feature of belief that outstrips the phenomenal and the psychological. Perhaps a small qualification needs to

Two Concepts of Mind 21

be made to this: one may need to add a relational element, to account for the fact that certain beliefs may depend on the state of the environment as well as the internal state of the thinker. It has been argued, for example, that to believe that water is wet, a subject must be related in an appropriate way to water in the environment. This relation is usually taken to be a causal relation, so it is possible that one could build this into the characterization of the relevant psychological property, where the causal roles in question stretch outside the head into the environment. If so, then no extra component would be required. But in any case, it is not much of a burden to note that there might also be a relational component to certain mental states, over and above the psychological and phenomenal components. Either way, no deep further mystery arises.

To see that there is no deep further aspect over and above the phenomenal and the psychological/relational aspects of intentional states, note that the manifest phenomena in the vicinity that need explaining fall into two classes: those we have third-person access to, and those we have first-person access to. Those in the former class ultimately come down to behavior, relations to the environment, and so on, and can be subsumed into the class of the psychological and the relational. Those in the latter class come down to the experience associated with believing-for example, the way our concepts seem to reach out into a phenomenal world-and thus constitute part of the problem of consciousness, not a separate mystery. The reasons for believing in any given aspect of belief (including semantic content, "aboutness," and so on) will derive from one of these two classes; there is no independent third class of phenomena forcing itself on us to be explained.

Another way to see this is to note that once we have fixed the psychological, phenomenal, and relational properties of an individual, there seems to be nothing mental that can be independently varied. We cannot even imagine someone identical to me in the three respects mentioned above but who believes something different, in the way that we can arguably imagine someone psychologically identical to me who experiences something different. There is simply not enough room in conceptual space for the possibility. Intentional concepts are in some ways less primitive than psychological and phenomenal concepts, in that they cannot be varied independently of the latter.12

Everything that I have said here about belief applies equally to other intentional states, such as desire, hope, and so on. All of these states have a psychological and a phenomenal aspect, and we need not legislate which is primary, although a strong case might be made for a psychological analysis. What counts is that there is no aspect of this state that outstrips both the psychological and the phenomenal (with perhaps a relational component

22 Foundations

thrown in). Psychology and phenomenology together constitute the central aspects of the mind.

The co-occurrence of phenomenal and psychological properties

It is a fact about the human mind that whenever a phenomenal property is instantiated, a corresponding psychological property is instantiated. Conscious experience does not occur in a vacuum. It is always tied to cognitive processing, and it is likely that in some sense it arises from that processing. Whenever one has a sensation, for example, there is some information processing going on: a corresponding perception, if you like. Similarly, whenever one has the conscious experience of happiness, the functional role associated with happiness is generally being played by some internal state. Perhaps it is logically possible that one could have the experience without the causation, but it seems to be an empirical fact that they go together.

In the face of this co-occurrence, the faint-hearted may be tempted to worry whether any real distinction is being made. But it is clear that there is at least a conceptual distinction here, even if the extensions of the concepts involved seem to go together. One can wonder how to explain the phenomenal quality, and one can wonder how to explain the playing of the causal role, and these are two distinct wonderings.

That being said, the co-occurrence of phenomenal and psychological properties reflects something deep about our phenomenal concepts. We have no independent language for describing phenomenal qualities. As we have seen, there is something ineffable about them. Although greenness is a distinct sort of sensation with a rich intrinsic character, there is very little that one can say about it other than that it is green. In talking about phenomenal qualities, we generally have to specify the qualities in question in terms of associated external properties, or in terms of associated causal roles. Our language for phenomenal qualities is derivative on our nonphenomenal language. As Ryle said, there are no "neat" sensation words.

If one looks at the catalog of conscious experience that I presented earlier, the experiences in question are never described in terms of their intrinsic qualities. Rather, I used expressions such as "the smell of freshly baked bread," "the patterns one gets when closing one's eyes," and so on. Even with a term like "green sensation," reference is effectively pinned down in extrinsic terms. When we learn the term "green sensation," it is effectively by ostension-we learn to apply it to the sort of experience caused by grass, trees, and so on. Generally, insofar as we have communicable phenomenal categories at all, they are defined with respect either to their typical external associations or to an associated kind of psychological state. For instance, when one speaks of the phenomenal quality of happiness, the reference of the term "happiness" is implicitly fixed via some causal role-the state where

Two Concepts of Mind 23

one judges all to be good, jumps for joy, and so on. Perhaps this is one interpretation of Wittgenstein's famous remark, "An inner process stands in need of outward criteria."

This dependence of phenomenal concepts on causal criteria has led some (including Wittgenstein and Ryle, in some of their moods) to suggest that there is nothing to the meaning of our mental concepts beyond the associated causal criteria. There is a certain plausibility to this: if a phenomenal property is always picked out by invoking a psychological property, why not suppose that there is only one property involved? But this temptation should be resisted. When we talk of a green sensation, this talk is not equivalent simply to talk of "a state that is caused by grass, trees, and so on." We are talking about the phenomenal quality that generally occurs when a state is caused by grass and trees. If there is a causal analysis in the vicinity, it is something like "the kind of phenomenal state that is caused by grass, trees, and so on."13 The phenomenal element in the concept prevents an analysis in purely functional terms.

In general, when a phenomenal property is picked out with the aid of a psychological property P, the phenomenal notion is not just "P." It is "the sort of conscious experience that tends to accompany P." And importantly, the very notion of "phenomenal quality" or "conscious experience" is not defined in psychological terms. Rather, the notion of conscious experience is something of a primitive, as we saw earlier. If there were a functional analysis of the notion of experience or phenomenal quality, then the analysis in question would yield functional analyses of specific phenomenal properties, but in the absence of such an analysis we cannot draw such a conclusion.

We cannot identify the notion "phenomenal P" with that of "psychological P" for all the usual reasons: there are two distinct concepts here, as witnessed by the fact that there are two distinct explananda. Although "phenomenal P" is picked out as "the experience that tends to accompany psychological P," we can coherently imagine a situation in which phenomenal P occurs without psychological P, and vice versa. A Rolls-Royce icon can be roughly analyzed as the kind of icon that is generally found on Rolls-Royce cars, but this does not mean that to be a Rolls-Royce icon is to be a Rolls-Royce car.

This gives us some insight into the sparseness of our specifically phenomenal vocabulary compared to our psychological vocabulary, and it also helps us understand why phenomenal and psychological properties have so often been conflated. For most everyday purposes this conflation does not matter: when one claims that someone is happy, one need not be talking specifically about either the phenomenal quality or the functional role, as they usually go together. However, for philosophical purposes and in particular for the purposes of explanation, to conflate these properties is fatal. The conflation

24 Foundations

can be tempting, as collapsing the distinction makes the problem of explaining conscious experience suddenly very straightforward; but it is utterly unsatisfactory for the same reason. The problem of consciousness cannot be spirited away on purely verbal grounds.

4. The Two Mind-Body Problems

The division of mental properties into phenomenal and psychological properties has the effect of dividing the mind-body problem into two: an easy part and a hard part. The psychological aspects of mind pose many technical problems for cognitive science, and a number of interesting puzzles for philosophical analysis, but they pose no deep metaphysical enigmas. The question "How could a physical system be the sort of thing that could learn, or that could remember?" does not have the same bite as the corresponding question about sensations, or about consciousness in general. The reason for this is clear. By our analysis in section 3, learning and memory are functional properties, characterized by causal roles, so the question "How could a physical system have psychological property P?" comes to the same thing as "How could a state of a physical system play such-and-such a causal role?" This is a question for the sciences of physical systems. One simply needs to tell a story about the organization of the physical system that allows it to react to environmental stimulation and produce behavior in the appropriate sorts of ways. While the technical problems are enormous, there is a clearly defined research program for their answer. The metaphysical problems are relatively few.

This is not to say that psychological properties pose no philosophical difficulties. There are significant problems in coming up with the correct analyses of these notions, for instance. Even if it is widely accepted that these are functional concepts, there can be significant disagreement about just how the requisite functional analyses should run. Intentional properties such as belief and desire, for example, provide fertile grounds for argument. In particular, the question of just what constitutes the content of a given intentional state is still poorly understood. There are also technical problems concerning just how high-level constructs such as these can play a real causal role in the production of behavior, especially if these are partly constituted by properties of the environment, or if there are no strict laws connecting psychological states with behavior. Then there are semi-empirical problems in the foundations of cognitive science concerning just how these properties might be instantiated in existing cognitive systems, or even concerning whether they are instantiated at all.

These problems are all serious, but they have the character of puzzles rather than mysteries. The situation here is analogous to that in the philosophy

Two Concepts of Mind 25

of biology, where there is no pressing life-body problem; there are merely a host of technical problems about evolution, selection, adaptation, fitness, and species. Just as most of the apparent metaphysical mysteries surrounding biology were disposed of long ago, it is fair to say that the mind-body problem for psychological properties is for all intents and purposes dissolved. What remains is a collection of smaller technical problems with which the normal course of scientific and philosophical analysis can grapple.

The phenomenal aspects of mind are a different matter. Here, the mind-body problem is as baffling as it ever was. The impressive progress of the physical and cognitive sciences has not shed significant light on the question of how and why cognitive functioning is accompanied by conscious experience. The progress in the understanding of the mind has almost entirely centered on the explanation of behavior. This progress leaves the question of conscious experience untouched.

If we like, we can view the psychological-phenomenal distinction not so much as splitting the mind-body problem as factoring it into two separate parts. The hardest part of the mind-body problem is the question: how could a physical system give rise to conscious experience? We might factor the link between the physical and conscious experience into two parts: the link between the physical and the psychological, and the link between the psychological and the phenomenal. As we saw above, we now have a pretty good idea of how a physical system can have psychological properties: the psychological mind-body problem has been dissolved. What remains is the question of why and how these psychological properties are accompanied by phenomenal properties: why all the stimulation and reaction associated with pain is accompanied by the experience of pain, for instance. Following Jackendoff (1987), we can call this residue the mind-mind problem. Current physical explanations take us as far as the psychological mind. What remains ill understood is the link between the psychological mind and the phenomenal mind.14

It is conceivable that the link between the phenomenal and the physical might be independent of that between the psychological and the physical, so that factoring would be impossible, but it seems unlikely. The close correlation that we have seen between phenomenal and psychological properties suggests a deep link. In later chapters, I will argue that the link is an extremely strong one and that the factoring strategy is valuable in approaching the mind-body problem. If so, then understanding the link between the psychological and the phenomenal is crucial to understanding conscious experience.

5. Two Concepts of Consciousness

Given that so many mental terms have a dual nature, it will not be surprising to learn that even "consciousness" has both phenomenal and psychological

26 Foundations

senses. So far, I have been focusing on the phenomenal sense, which itself subsumes all the previously mentioned phenomenal aspects of mind. To be conscious in this sense is just to instantiate some phenomenal quality. This is the key sense of "consciousness," or at least the one that poses the major explanatory problems. But it is not the only sense of the term. "Consciousness" can also be used to refer to a variety of psychological properties, such as reportability or introspective accessibility of information. We can group psychological properties of this sort under the label of psychological consciousness, as opposed to the phenomenal consciousness on which I have been concentrating.

This ambiguity can lead to much confusion in the discussion of consciousness. Frequently, someone putting forward an explanation of consciousness will start by investing the problem with all the gravity of the problem of phenomenal consciousness, but will end by giving an explanation of some aspect of psychological consciousness, such as the ability to introspect. This explanation might be worthwhile in its own right, but one is left with the sense that more has been promised than has been delivered.

Varieties of psychological consciousness

There are numerous psychological notions for which the term "consciousness" is sometimes used. These include the following:



Awakeness. Sometimes we say that a person is conscious as another way of saying that they are not asleep. It makes sense to suppose that we have experiences while we are asleep, so this notion clearly does not coincide with phenomenal consciousness. Awakeness can plausibly be analyzed in functional terms-perhaps, at a first approximation, in terms of an ability to process information about the world and deal with it in a rational fashion.


Introspection. This is the process by which we can become aware of the contents of our internal states. If you ask me about my mental states, it is by introspection that I determine my answer. This access to one's mental states is an important component of the everyday concept of consciousness, and it is at least partly a functional notion. One might analyze it in terms of one's rational processes being sensitive to information about one's internal states in the right sort of way, and one's being able to use this information appropriately.


Reportability. This is our ability to report the contents of our mental states. It presupposes the ability to introspect, but is more constrained than that ability, as it presupposes a capacity for language. This concept of consciousness has often been the central target of philosophers and psychologists of an operationalist bent.

Two Concepts of Mind 27


Self-consciousness. This refers to our ability to think about ourselves, our awareness of our existence as individuals and of our distinctness from others. My self-consciousness might be analyzed in terms of my access to a self-model, or my possession of a certain sort of representation that is associated in some way with myself. It may well be that self-consciousness is limited to humans and a few species of animals.


Attention. We often say that someone is conscious of something precisely when they are paying attention to it; that is, when a significant portion of their cognitive resources is devoted to dealing with the relevant information. We can be phenomenally conscious of something without attending to it, as witnessed by the fringes of a visual field.

Voluntary control.

Voluntary control. In another sense, we say that a behavioral act is conscious when that act is performed deliberately; that is, where the action is caused in the appropriate sort of way by an element of prior thought.


Knowledge. In another everyday sense, we say that someone is conscious of a fact precisely when they know the fact, and that they are conscious of a thing precisely when they know about that thing. This notion is rarely the focus of technical discussion of consciousness, but it is probably as central to the everyday usage of the term as anything else.


That these are all largely functional notions can be seen from how one would explain the phenomena in question. If one were to try to explain attention, one might devise a model of the cognitive processes that lead to resources being concentrated on one aspect of available information rather than another. If one were to try to explain introspection, one would try to explain the processes by which one is sensitive to one's internal states in the appropriate way. Similar stories apply to explanation of the other properties. In each case, a functional explanation seems to capture what is central.

Although these concepts have a psychological core, many or all of them are associated with phenomenal states. There is a certain sort of phenomenal state associated with self-consciousness, for example. The same goes for introspection, attention, and the voluntary control of behavior. As with the other dual-aspect terms that I have discussed, terms such as "introspection" and "self-consciousness" are sometimes used to refer to the phenomenal state, which can lead to confusion. Indeed, some might argue that a phenomenal aspect is required for a process to truly qualify as "introspection," "attention," or whatever. As before, however, this issue is largely verbal. It is clear that there is a phenomenal and a psychological property in the vicinity of each of these concepts. Those who do not like to dignify the psychological property with a mental term such as "attention" can use the term "pseudo-attention" instead. The substantial philosophical issues remain the same, no matter what the properties are called.

28 Foundations

The phenomenal and the psychological properties in the vicinity of these notions tend to occur together, but as with other mental concepts, they should not be conflated. We should also be careful not to conflate the phenomenal senses of these terms with phenomenal consciousness in general.

Consciousness and awareness

We have seen that there is a psychological property associated with the experience of emotion, a psychological property associated with the experience of self-consciousness, a psychological property associated with the experience of sensation, and so on. It is natural to suppose that there might be a psychological property associated with experience itself, or with phenomenal consciousness. In fact, I think there is such a property in the vicinity; we can call it "awareness." This is the most general brand of psychological consciousness.

Awareness can be broadly analyzed as a state wherein we have access to some information, and can use that information in the control of behavior. One can be aware of an object in the environment, of a state of one's body, or of one's mental state, among other things. Awareness of information generally brings with it the ability to knowingly direct behavior depending on that information. This is clearly a functional notion. In everyday language, the term "awareness" is often used synonymously with "consciousness," but I will reserve the term for the functional notion I have described here.

In general, wherever there is phenomenal consciousness, there seems to be awareness. My phenomenal experience of the yellow book beside me is accompanied by my functional awareness of the book, and indeed by my awareness of the yellow color. My experience of a pain is accompanied by an awareness of the presence of something nasty, which tends to lead to withdrawal and the like, where possible. The fact that any conscious experience is accompanied by awareness is made clear by the fact that a conscious experience is reportable. If I am having an experience, I can talk about the fact that I am having it. I may not be paying attention to it, but I at least have the ability to focus on it and talk about it, if I choose. This reportability immediately implies that I am aware in the relevant sense. Of course, an animal or a prelinguistic human might have conscious experience without the ability to report, but such a being would still plausibly have a degree of awareness. Awareness does not entail the ability to report, although the two tend to go together in creatures with language.

Consciousness is always accompanied by awareness, but awareness as I have described it need not be accompanied by consciousness. One can be aware of a fact without any particular associated phenomenal experience, for instance. However, it may be possible to constrain the notion of awareness so that it turns out to be coextensive with phenomenal consciousness,

Two Concepts of Mind 29

or nearly so. I will not attempt that project here, but I discuss it further in Chapter 6.

The notion of awareness subsumes most or all of the various psychological notions of consciousness just enumerated. Introspection can be analyzed as awareness of some internal state. Attention can be analyzed as a particularly high degree of awareness of an object or event. Self-consciousness can be understood as awareness of oneself. Voluntary control is trickier, although it might be partly analyzed as requiring attention to the behavior one is performing. Awakeness might be roughly characterized as a state in which one is able to deal rationally with one's environment to some extent, and so implies a particular sort of awareness.

The idea that there is a functional notion of consciousness that can be explicated in terms of access has been fleshed out by Block (1995), who talks about the distinction between "phenomenal consciousness" and "access consciousness." Block's notion of access consciousness corresponds closely to the notion of awareness that I have been describing (I discuss the relationship further in Chapter 6). In a similar fashion, Newell (1992) explicitly distinguishes between "awareness" and "consciousness." He describes awareness as "the ability of a subject to make its behavior depend on some knowledge," and goes on to spell out the distinction between this notion and consciousness, which he says is a nonfunctional phenomenon. Similar distinctions have been made by other philosophers and cognitive scientists.15

Explaining consciousness versus explaining awareness

Awareness, like other psychological properties, poses few metaphysical problems. The problems posed by the psychological varieties of consciousness are of the same order of magnitude as those posed by memory, learning, and belief. Certainly, the notion of awareness is not crystal-clear, so there is room for significant philosophical analysis of just what it comes to. Further, there is room for an enormous amount of research in cognitive science, studying how natural and artificial cognitive systems might function in such a way that they are aware. But the outlines of these research programs are reasonably clear. There is little reason to suppose that the normal course of cognitive science, backed by appropriate philosophical analysis, should not eventually succeed.

Insofar as consciousness is the really difficult problem for a science of the mind, it is phenomenal consciousness that is central. The problems here are of a different order of magnitude. Even after we have explained the physical and computational functioning of a conscious system, we still need to explain why the system has conscious experiences. Some dispute this claim, of course, and I will discuss it at greater length soon. For now, though, we can simply note the prima facie difference in the problems that the phenomenal and

30 Foundations

psychological varieties present. It is phenomenal consciousness that poses the worrying problem of consciousness.

Given the differences between the psychological and phenomenal notions of consciousness, it is unfortunate that they are often conflated in the literature. This conflation matters little in everyday speech, as awareness and phenomenal consciousness usually go together. But for the purposes of explanation, the conceptual distinction is crucial. Insofar as any remotely satisfactory explanations of "consciousness" have been put forward, it is usually a psychological aspect that is explained. The phenomenal aspects generally go untouched.

Many recent philosophical analyses of consciousness have concerned themselves primarily with the nonphenomenal aspects. Rosenthal (1996) argues that a mental state is conscious precisely when there is a higher-order thought about that mental state. This might be a useful analysis of introspective consciousness, and perhaps of other aspects of awareness, but it does not appear to explain phenomenal experience.16 Similarly, Dennett (1991) spends much of his book outlining a detailed cognitive model, which he puts forward as an explanation of consciousness. On the face of it, the model is centrally a model of the capacity of a subject to verbally report a mental state. It might thus yield an explanation of reportability, of introspective consciousness, and perhaps of other aspects of awareness, but nothing in the model provides an explanation of phenomenal consciousness (although Dennett would put things differently).

Armstrong (1968), confronted by consciousness as an obstacle for his functionalist theory of mind, analyzes the notion in terms of the presence of some self-scanning mechanism. This might provide a useful account of self-consciousness and introspective consciousness, but it leaves the problem of phenomenal experience to the side. Armstrong (1981) talks about both perceptual consciousness and introspective consciousness, but is concerned with both only as varieties of awareness, and does not address the problems posed by the phenomenal qualities of experience. Thus the sense in which consciousness is really problematic for his functionalist theory is sidestepped, by courtesy of the ambiguity in the notion of consciousness.

Others writing on the topic of "consciousness" have been primarily concerned with self-consciousness or introspective consciousness. Van Gulick (1988), in suggesting that consciousness should be analyzed as the possession of "reflexive metapsychological information," is at best providing an analysis of these psychological notions, and indeed concedes that the phenomenal aspects may be left out by such an analysis. Similarly, Jaynes's (1976) elaborate theory of consciousness is concerned only with our awareness of our own thoughts. It says nothing about phenomena associated with perception and therefore could not hope to be a theory of awareness in general, let alone a theory of phenomenal consciousness. Hofstadter (1979) has some

Two Concepts of Mind 31

interesting things to say about consciousness, but he is more concerned with introspection, free will, and the sense of self than with experience per se.

Insofar as consciousness has been a topic for discussion among psychologists, the phenomenal and psychological notions have not often been carefully distinguished. Usually it is some aspect of awareness, such as introspection, attention, or self-consciousness, that psychological studies address. Even the psychological aspects of consciousness have had something of a bad name in psychology, at least until recently. Perhaps this is because of some unclear-ness in those notions, and the difficulties associated with high-level phenomena such as introspection. One might speculate that to a larger extent this bad reputation is due to their sharing a name with phenomenal consciousness, giving the appearance of partnership in crime.

One sometimes hears that psychological research has been "returning to consciousness" in recent years. The reality seems to be that the psychological aspects of consciousness have been an active subject of research, and that researchers have not been afraid to use the term "consciousness" for the phenomena. For the most part, however, phenomenal consciousness remains off to the side. Perhaps this is understandable. While one can see how the methods of experimental psychology might lead to an understanding of the various kinds of awareness, it is not easy to see how they could explain phenomenal experience.17

Cognitive models are well suited to explaining psychological aspects of consciousness. There is no vast metaphysical problem in the idea that a physical system should be able to introspect its internal states, or that it should be able to deal rationally with information from its environment, or that it should be able to focus its attention first in one place and then in the next. It is clear enough that an appropriate functional account should be able to explain these abilities, even if discovering the correct account takes decades or centuries. But the really difficult problem is that of phenomenal consciousness, and this is left untouched by the explanations of psychological consciousness that have been put forward so far.

In what follows, I revert to using "consciousness" to refer to phenomenal consciousness alone. When I wish to use the psychological notions, I will speak of "psychological consciousness" or "awareness." It is phenomenal consciousness with which I will mostly be concerned.

2. Supervenience and Explanation

What is the place of consciousness in the natural order? Is consciousness physical? Can consciousness be explained in physical terms? To come to grips with these issues, we need to build a framework; in this chapter, I build one. The centerpiece of this framework is the concept of Supervenience: I give an account of this concept and apply it to clarify the idea of reductive explanation. Using this account, I sketch a picture of the relationship between most high-level phenomena and physical facts, one that seems to cover everything except, perhaps, for conscious experience.

1. Supervenience

It is widely believed that the most fundamental facts about our universe are physical facts, and that all other facts are dependent on these. In a weak enough sense of "dependent," this may be almost trivially true; in a strong sense, it is controversial. There is a complex variety of dependence relations between high-level facts and low-level facts in general, and the kind of dependence relation that holds in one domain, such as biology, may not hold in another, such as that of conscious experience. The philosophical notion of Supervenience provides a unifying framework within which these dependence relations can be discussed.

The notion of Supervenience formalizes the intuitive idea that one set of facts can fully determine another set of facts.1 The physical facts about the world seem to determine the biological facts, for instance, in that once all the physical facts about the world are fixed, there is no room for the biological facts to vary. (Fixing all the physical facts will simultaneously fix which objects are alive.) This provides a rough characterization of the sense in

Supervenience and Explanation 33

which biological properties supervene on physical properties. In general, Supervenience is a relation between two sets of properties: B-properties-intuitively, the high-level properties-and A-properties, which are the more basic low-level properties.

For our purposes, the relevant A-properties are usually the physical properties: more precisely, the fundamental properties that are invoked by a completed theory of physics. Perhaps these will include mass, charge, spatio-temporal position; properties characterizing the distribution of various spatio-temporal fields, the exertion of various forces, and the form of various waves; and so on. The precise nature of these properties is not important. If physics changes radically, the relevant class of properties may be quite different from those I mention, but the arguments will go through all the same. Such high-level properties as juiciness, lumpiness, giraffehood, and the like are excluded, even though there is a sense in which these properties are physical. In what follows, talk of physical properties is implicitly restricted to the class of fundamental properties unless otherwise indicated. I will sometimes speak of "microphysical" or "low-level physical" properties to be explicit.

The A-facts and B-facts about the world are the facts concerning the instantiation and distribution of A-properties and B-properties.2 So the physical facts about the world encompass all facts about the instantiation of physical properties within the spatiotemporal manifold. It is also useful to stipulate that the world's physical facts include its basic physical laws. On some accounts, these laws are already determined by the totality of particular physical facts, but we cannot take this for granted.

The template for the definition of Supervenience is the following:


B-properties supervene on A-properties if no two possible situations are identical with respect to their A-properties while differing in their B-properties.


For instance, biological properties supervene on physical properties insofar as any two possible situations that are physically identical are biologically identical. (I use "identical" in the sense of indiscernibility rather than numerical identity here. In this sense, two separate tables might be physically identical.) More precise notions of Supervenience can be obtained by filling in this template. Depending on whether we take the "situations" in question to be individuals or entire worlds, we arrive at notions of local and global Supervenience, respectively. And depending on how we construe the notion of possibility, we obtain notions of logical Supervenience, natural Supervenience, and perhaps others. I will flesh out these distinctions in what follows.

Local and global Supervenience

B-properties supervene locally on A-properties if the A-properties of an individual determine the B-properties of that individual-if, that is, any two

34 Foundations

possible individuals that instantiate the same A-properties instantiate the same B-properties. For example, shape supervenes locally on physical properties: any two objects with the same physical properties will necessarily have the same shape. Value does not supervene locally on physical properties, however: an exact physical replica of the Mona Lisa is not worth as much as the Mona Lisa. In general, local Supervenience of a property on the physical fails if that property is somehow context-dependent-that is, if an object's possession of that property depends not only on the object's physical constitution but also on its environment and its history. The Mona Lisa is more valuable than its replica because of a difference in their historical context: the Mona Lisa was painted by Leonardo, whereas the replica was not.3

B-properties supervene globally on A-properties, by contrast, if the A-facts about the entire world determine the B-facts: that is, if there are no two possible worlds identical with respect to their A-properties, but differing with respect to their B-properties.4 A world here is to be thought of as an entire universe; different possible worlds correspond to different ways a universe might be.

Local Supervenience implies global Supervenience, but not vice versa. For example, it is plausible that biological properties supervene globally on physical properties, in that any world physically identical to ours would also be biologically identical. (There is a small caveat here, which I discuss shortly.) But they probably do not supervene locally. Two physically identical organisms can arguably differ in certain biological characteristics. One might be fitter than the other, for example, due to differences in their environmental contexts. It is even conceivable that physically identical organisms could be members of different species, if they had different evolutionary histories.

The distinction between global and local Supervenience does not matter too much when it comes to conscious experience, because it is likely that insofar as consciousness supervenes on the physical at all, it supervenes locally. If two creatures are physically identical, then differences in environmental and historical contexts will not prevent them from having identical experiences. Of course, context can affect experience indirectly, but only by virtue of affecting internal structure, as in the case of perception. Phenomena such as hallucination and illusion illustrate the fact that it is internal structure rather than context that is directly responsible for experience.

Logical and natural Supervenience

A more important distinction for our purposes is between logical (or conceptual) Supervenience, and mere natural (or nomic, or empirical) Supervenience.

Supervenience and Explanation 35

B-properties supervene logically on A-properties if no two logically possible situations are identical with respect to their A-properties but distinct with respect to their B-properties. I will say more about logical possibility later in this chapter. For now, one can think of it loosely as possibility in the broadest sense, corresponding roughly to conceivability, quite unconstrained by the laws of our world. It is useful to think of a logically possible world as a world that it would have been in God's power (hypothetically!) to create, had he so chosen.5 God could not have created a world with male vixens, but he could have created a world with flying telephones. In determining whether it is logically possible that some statement is true, the constraints are largely conceptual. The notion of a male vixen is contradictory, so a male vixen is logically impossible; the notion of a flying telephone is conceptually coherent, if a little out of the ordinary, so a flying telephone is logically possible.

It should be stressed that the logical Supervenience is not defined in terms of deducibility in any system of formal logic. Rather, logical Supervenience is defined in terms of logically possible worlds (and individuals), where the notion of a logically possible world is independent of these formal considerations. This sort of possibility is often called "broadly logical" possibility in the philosophical literature, as opposed to the "strictly logical" possibility that depends on formal systems.6

At the global level, biological properties supervene logically on physical properties. Even God could not have created a world that was physically identical to ours but biologically distinct. There is simply no logical space for the biological facts to independently vary. When we fix all the physical facts about the world-including the facts about the distribution of every last particle across space and time-we will in effect also fix the macroscopic shape of all the objects in the world, the way they move and function, the way they physically interact. If there is a living kangaroo in this world, then any world that is physically identical to this world will contain a physically identical kangaroo, and that kangaroo will automatically be alive.

We can imagine that a hypothetical superbeing-Laplace's demon, say, who knows the location of every particle in the universe-would be able to straightforwardly "read off" all the biological facts, once given all the microphysical facts. The microphysical facts are enough for such a being to construct a model of the microscopic structure and dynamics of the world throughout space and time, from which it can straightforwardly deduce the macroscopic structure and dynamics. Given all that information, it has all the information it needs to determine which systems are alive, which systems belong to the same species, and so on. As long as it possesses the biological concepts and has a full specification of the microphysical facts, no other information is relevant.

36 Foundations

In general, when B-properties supervene logically on A-properties, we can say that the A-facts entail the B-facts, where one fact entails another if it is logically impossible for the first to hold without the second. In such cases, Laplace's demon could read off the B-facts from a specification of the A-facts, as long as it possesses the B-concepts in question. (I will say much more about the connections between these different ways of understanding logical Supervenience later in the chapter; the present discussion is largely for illustration.) In a sense, when logical Supervenience holds, all there is to the B-facts being as they are is that the A-facts are as they are.

There can be Supervenience without logical Supervenience, however. The weaker variety of Supervenience arises when two sets of properties are systematically and perfectly correlated in the natural world. For example, the pressure exerted by one mole of a gas systematically depends on its temperature and volume according to the law pV = KT, where K is a constant (I pretend for the purposes of illustration that all gases are ideal gases). In the actual world, whenever there is a mole of gas at a given temperature and volume, its pressure will be determined: it is empirically impossible that two distinct moles of gas could have the same temperature and volume, but different pressure. It follows that the pressure of a mole of gas supervenes on its temperature and volume in a certain sense. (In this example, I am taking the class of A-properties to be much narrower than the class of physical properties, for reasons that will become clear.) But this Supervenience is weaker than logical Supervenience. It is logically possible that a mole of gas with a given temperature and volume might have a different pressure; imagine a world in which the gas constant K is larger or smaller, for example. Rather, it is just a fact about nature that there is this correlation.

This is an example of natural Supervenience of one property on others: in this instance, pressure supervenes naturally on temperature, volume, and the property of being a mole of gas. In general, B-properties supervene naturally on A-properties if any two naturally possible situations with the same A-properties have the same B-properties.

A naturally possible situation is one that could actually occur in nature, without violating any natural laws. This is a much stronger constraint than mere logical possibility. The scenario with a different gas constant is logically possible, for example, but it could never occur in the real world, so it is not naturally possible. Among naturally possible situations, any two moles of gas with the same temperature and volume will have the same pressure.

Intuitively, natural possibility corresponds to what we think of as real empirical possibility-a naturally possible situation is one that could come up in the real world, if the conditions were right. These include not just actual situations but counterfactual situations that might have come up in the world's history, if boundary conditions had been different, or that might come up in the future, depending on how things go. A mile-high skyscraper

Supervenience and Explanation 37

is almost certainly naturally possible, for example, even though none has actually been constructed. It is even naturally possible (although wildly improbable) that a monkey could type Hamlet. We can also think of a naturally possible situation as one that conforms to the laws of nature of our world.7 For this reason, natural possibility is sometimes called nomic or nomological possibility,8 from the Greek term nomos for "law."

There are a vast number of logically possible situations that are not naturally possible. Any situation that violates the laws of nature of our world falls into this class: a universe without gravity, for example, or with different values of fundamental constants. Science fiction provides many situations of this sort, such as antigravity devices and perpetual-motion machines. These are easy to imagine, but almost certainly could never come to exist in our world.

In the reverse direction, any situation that is naturally possible will be logically possible. The class of natural possibilities is therefore a subset of the class of logical possiblities. To illustrate this distinction: both a cubic mile of gold and a cubic mile of uranium-235 seem to be logically possible, but as far as we know, only the first is naturally possible-a (stable) cubic mile of uranium-235 could not exist in our world.

Natural Supervenience holds when, among all naturally possible situations, those with the same distribution of A-properties have the same distributon of B-properties: that is, when the A-facts about a situation naturally necessitate the B-facts. This happens when the same clusters of A-properties in our world are always accompanied by the same B-properties, and when this correlation is not just coincidental but lawful: that is, when instantiating the A-properties will always bring about the B-properties, wherever and whenever this happens. (In philosophical terms, the dependence must support counterfactuals.) This co-occurrence need not hold in every logically possible situation, but it must hold in every naturally possible situation.

It is clear that logical Supervenience implies natural Supervenience. If any two logically possible situations with the same A-properties have the same B-properties, then any two naturally possible situations will also. The reverse does not hold, however, as the gas law illustrates. The temperature and volume of a mole of gas determine pressure across naturally but not logically possible situations, so pressure depends naturally but not logically on temperature and volume. Where we have natural Supervenience without logical Supervenience, I will say that we have mere natural Supervenience.

For reasons that will become clear, it is hard to find cases of natural Supervenience on the set of physical properties without logical Supervenience, but consciousness itself can provide a useful illustration. It seems very likely that consciousness is naturally supervenient on physical properties, locally or globally, insofar as in the natural world, any two physically identical creatures will have qualitatively identical experiences. It is not at all clear that consciousness is logically supervenient on physical properties, however.

38 Foundations

It seems logically possible, at least to many, that a creature physically identical to a conscious creature might have no conscious experiences at all, or that it might have conscious experiences of a different kind. (Some dispute this, but I use it for now only as an illustration.) If this is so, then conscious experience supervenes naturally but not logically on the physical. The necessary connection between physical structure and experience is ensured only by the laws of nature, and not by any logical or conceptual force.

The distinction between logical and natural Supervenience is vital for our purposes.9 We can intuitively understand the distinction as follows. If B-properties supervene logically on A-properties, then once God (hypothetically) creates a world with certain A-facts, the B-facts come along for free as an automatic consequence. If B-properties merely supervene naturally on A-properties, however, then after making sure of the A-facts, God has to do more work in order to make sure of the B-facts: he has to make sure there is a law relating the A-facts and the B-facts. (I borrow this image from Kripke 1972.) Once the law is in place, the relevant A-facts will automatically bring along the B-facts; but one could, in principle, have had a situation where they did not.

One also sometimes hears talk of metaphysical Supervenience, which is based on neither logical nor natural necessity, but on "necessity tout court," or "metaphysical necessity" as it is sometimes known (drawing inspiration from Kripke's [1972] discussion of a posteriori necessity). I will argue later that the metaphysically possible worlds are just the logically possible worlds (and that metaphysical possibility of statements is logical possibility with an a posteriori semantic twist), but for now it is harmless to assume there is a notion of metaphysical Supervenience, to be spelled out by analogy with the notions of logical and natural Supervenience above. A notion of "weak" Supervenience is also mentioned occasionally, but seems too weak to express an interesting dependence relation between properties.10

The logical-natural distinction and the global-local distinction cut across each other. It is reasonable to speak of both global logical Supervenience and local logical Supervenience, although I will more often be concerned with the former. When I speak of logical Supervenience without a further modifier, global logical Supervenience is intended. It is also coherent to speak of global and local natural Supervenience, but the natural Supervenience relations with which we are concerned are generally local or at least localizable, for the simple reason that evidence for a natural Supervenience relation generally consists in local regularities between clusters of properties.11

A problem with logical Supervenience*

A technical problem with the notion of logical Supervenience needs to be dealt with. This problem arises from the logical possibility of a world physi-

Supervenience and Explanation 39

cally identical to ours, but with additional nonphysical stuff that is not present in our own world: angels, ectoplasm, and ghosts, for example. There is a conceivable world just like ours except that it has some extra angels hovering in a non-physical realm, made of ectoplasm. These angels might have biological properties of their own, if they reproduced and evolved. Presumably the angels could have all sorts of beliefs, and their communities might have complex social structure.

The problem these examples pose is clear. The angel world is physically identical to ours, but it is biologically distinct. If the angel world is logically possible, then according to our definition biological properties are not supervenient on physical properties. But we certainly want to say that biological properties are supervenient on physical properties, at least in this world if not in the angel world (assuming there are no angels in the actual world!). Intuitively, it seems undesirable for the mere logical possibility of the angel world to stand in the way of the determination of biological properties by physical properties in our own world.

This sort of problem has caused some (e.g., Haugeland 1982; Petrie 1987) to suggest that logical possibility and necessity are too strong to serve as the relevant sort of possibility and necessity in Supervenience relations, and that a weaker variety such as natural possibility and necessity should be used instead. But this would render useless the very useful distinction between logical and natural Supervenience outlined above, and would also ignore the fact that there is a very real sense in which the biological facts about our world are logically determined by the physical facts. Others (e.g., Teller 1989) have bitten the bullet by stipulating that worlds with extra nonphysical stuff are not logically or metaphysically possible, despite appearances, but this makes logical and metaphysical possibility seem quite arbitrary. Fortunately, such moves are not required. It turns out that it is possible to retain a useful notion of logical Supervenience compatible with the possibility of these worlds, as long as we fix the definition appropriately.12

The key to the solution is to turn Supervenience into a thesis about our world (or more generally, about particular worlds). This accords with the intuition that biological facts are logically determined by the physical facts in our world, despite the existence of bizarre worlds where they are not so determined. According to a revised definition, B-properties are logically supervenient on A-properties if the B-properties in our world are logically determined by the A-properties in the following sense: in any possible world with the same A-facts, the same B-facts will hold.13 The existence of possible worlds with extra B-facts will thus not count against logical Supervenience in our world, as long as at least the B-facts true in our world are true in all physically identical worlds. And this they generally will be (with an exception discussed below). If there is a koala eating in a gum tree in this world, there will be an identical koala eating in a gum tree in any physically identical world, whether or not that world has any angels hanging around.

40 Foundations

There is a minor complication. There is a certain sort of biological fact about our world that does not hold in the angel world: the fact that our world has no living ectoplasm, for example, and the fact that all living things are based on DNA. Perhaps the angel world might even be set up with ectoplasm causally dependent on physical processes, so that wombat copulation on the physical plane sometimes gives rise to baby ectoplasmic wombats on the nonphysical plane. If so, then there might be a wombat that is childless (in a certain sense) in our world, with a counterpart that is not childless in the physically identical angel world. It follows that the property of being childless does not supervene according to our definition, and nor do the world-level properties such as that of having no living ectoplasm. Not all the facts about our world follow from the physical facts alone.

To analyze the problem, note that these facts all involve negative existence claims, and so depend not only on what is going on in our world but on what is not. We cannot expect these facts to be determined by any sort of localized facts, as they depend not just on local goings-on in the world but on the world's limits. Supervenience theses should apply only to positive facts and properties, those that cannot be negated simply by enlarging a world. We can define a positive fact in W as one that holds in every world that contains W as a proper part;14 a positive property is one that if instantiated in a world W, is also instantiated by the corresponding individual in all worlds that contain W as a proper part.15 Most everyday facts and properties are positive-think of the property of being a kangaroo, or of being six feet tall, or of having a child. Negative facts and properties will always involve negative existence claims in one form or another. These include explicitly negative existential facts such as the nonexistence of ectoplasm, universally quantified facts such as the fact that all living things are made of DNA, negative relational properties such as childlessness, and superlatives such as the property of being the most fecund organism in existence.

In future, the Supervenience relations with which we are concerned should be understood to be restricted to positive facts and properties. When claiming that biological properties supervene on physical properties, it is only the positive biological properties that are at issue. All the properties with which we are concerned are positive-local physical and phenomenal properties, for instance-so this is not much of a restriction.

The definition of global logical Supervenience of B-properties on A-properties therefore comes to this: for any logically possible world W that is A-indiscernible from our world, then the B-facts true of our world are true of W. We need not build in a clause about positiveness, but it will usually be understood that the only relevant B-facts and properties are positive facts and properties. Similarly, B-properties supervene locally and logically on A-properties when for every actual individual x and every logically possible

Supervenience and Explanation 41

individual y, if y is A-indiscernible from x, then the B-properties instantiated by x are instantiated by y. More briefly and more generally: B-properties supervene logically on A-properties if the B-facts about actual situations are entailed by the A-facts, where situations are understood as worlds and individuals in the global and local cases respectively. This definition captures the idea that Supervenience claims are usually claims about our world, while retaining the key role of logical necessity.16

Supervenience and materialism

Logical and natural Supervenience have quite different ramifications for ontology: that is, for the matter of what there is in the world. If B-properties are logically supervenient on A-properties, then there is a sense in which once the A-facts are given, the B-facts are a free lunch. Once God (hypothetically) made sure that all the physical facts in our world held, the biological facts came along for free. The B-facts merely redescribe what is described by the A-facts. They may be different facts (a fact about elephants is not a microphysical fact), but they are not further facts.

With mere natural Supervenience, the ontology is not so straightforward. Contingent lawful connections connect distinct features of the world. In general, if B-properties are merely naturally supervenient on A-properties in our world, then there could have been a world in which our A-facts held without the B-facts. As we saw before, once God fixed all the A-facts, in order to fix the B-facts he had more work to do. The B-facts are something over and above the A-facts, and their satisfaction implies that there is something new in the world.

With this in mind we can formulate precisely the widely held doctrine of materialism (or physicalism), which is generally taken to hold that everything in the world is physical, or that there is nothing over and above the physical, or that the physical facts in a certain sense exhaust all the facts about the world. In our language, materialism is true if all the positive facts about the world are globally logically supervenient on the physical facts. This captures the intuitive notion that if materialism is true, then once God fixed the physical facts about the world, all the facts were fixed.

(Or at least, all the positive facts were fixed. The restriction to positive facts is needed to ensure that worlds with extra ectoplasmic facts do not count against materialism in our world. Negative existential facts such as "There are no angels" are not strictly logically supervenient on the physical, but their nonsupervenience is quite compatible with materialism. In a sense, to fix the negative facts, God had to do more than fix the physical facts; he also had to declare, "That's all." If we wanted, we could add a second-order "That's all" fact to the Supervenience base in the definition of materialism, in which case the positive-fact constraint could be removed.)

42 Foundations

According to this definition, materialism is true if all the positive facts about our world are entailed by the physical facts.17 That is, materialism is true if for any logically possible world W that is physically indiscernible from our world, all the positive facts true of our world are true of W. This is equivalent in turn to the thesis that any world that is physically indiscernible from our world contains a copy of our world as a (proper or improper) part, which seems an intuitively correct definition.18 (This matches the definition of physicalism given by Jackson [1994], whose criterion is that every minimal physical duplicate of our world is a duplicate simpliciter of our world.19)

I will discuss this matter at much greater length in Chapter 4, where this definition of materialism will be further justified. Some may object to the use of logical possibility rather than possibility tout court or "metaphysical possibility." Those people may substitute metaphysical possibility for logical possibility in the definition above. Later, I will argue that it comes to the same thing.

2. Reductive Explanation

The remarkable progress of science over the last few centuries has given us good reason to believe that there is very little that is utterly mysterious about the world. For almost every natural phenomenon above the level of microscopic physics, there seems in principle to exist a reductive explanation: that is, an explanation wholly in terms of simpler entities. In these cases, when we give an appropriate account of lower-level processes, an explanation of the higher-level phenomenon falls out.

Biological phenomena provide a clear illustration. Reproduction can be explained by giving an account of the genetic and cellular mechanisms that allow organisms to produce other organisms. Adaptation can be explained by giving an account of the mechanisms that lead to appropriate changes in external function in response to environmental stimulation. Life itself is explained by explaining the various mechanisms that bring about reproduction, adaptation, and the like. Once we have told the lower-level story in enough detail, any sense of fundamental mystery goes away: the phenomena that needed to be explained have been explained.

One can tell a similar story for most natural phenomena. In physics, we explain heat by telling an appropriate story about the energy and excitation of molecules. In astronomy, we explain the phases of the moon by going into the details of orbital motion and optical reflection. In geophysics, earthquakes are explained via an account of the interaction of subterranean masses. In cognitive science, to explain a phenomenon such as learning, all we have to do is explain various functional mechanisms-the mechanisms that give rise to appropriate changes in behavior in response to environmen-

Supervenience and Explanation 43

tal stimulation, at a first approximation (any worries about the experience of learning aside). Many of the details of these explanations currently evade our grasp, and are likely to prove very complex, but we know that if we find out enough about the low-level story, the high-level story will eventually come along.

I will not precisely define the notion of reductive explanation until later. For now, it remains characterized by example. However, I can issue some caveats about what reductive explanation is not. A reductive explanation of a phenomenon need not require a reduction of that phenomenon, at least in some senses of that ambiguous term. In a certain sense, phenomena that can be realized in many different physical substrates-learning, for example-might not be reducible in that we cannot identify learning with any specific lower-level phenomenon. But this multiple realizability does not stand in the way of reductively explaining any instance of learning in terms of lower-level phenomena.20 Reductive explanation of a phenomenon should also not be confused with a reduction of a high-level theory. Sometimes a reductive explanation of a phenomenon will provide a reduction of a preexisting high-level theory, but other times it will show such theories to be on the wrong track. Often there might not even be a high-level theory to reduce.

Reductive explanation is not the be-all and end-all of explanation. There are many other sorts of explanation, some of which may shed more light on a phenomenon than a reductive explanation in a given instance. There are historical explanations, for instance, explaining the genesis of a phenomenon such as life, where a reductive explanation only gives a synchronic account of how living systems function. There are also all sorts of high-level explanations, such as the explanation of aspects of behavior in terms of beliefs and desires. Even though this behavior might in principle be explainable reductively, a high-level explanation is often more comprehensible and enlightening. Reductive explanations should not be seen as displacing these other sorts of explanation. Each has its place.

Reductive explanation via functional analysis

What is it that allows such diverse phenomena as reproduction, learning, and heat to be reductively explained? In all these cases, the nature of the concepts required to characterize the phenomena is crucial. If someone objected to a cellular explanation of reproduction, "This explains how a cellular process can lead to the production of a complex physical entity that is similar to the original entity, but it doesn't explain reproduction," we would have little patience-for that is all that "reproduction" means. In general, a reductive explanation of a phenomenon is accompanied by some rough-and-ready analysis of the phenomenon in question, whether implicit or explicit. The notion of reproduction can be roughly analyzed in terms of

44 Foundations

the ability of an organism to produce another organism in a certain sort of way. It follows that once we have explained the processes by which an organism produces another organism, we have explained that instance of reproduction.

The point may seem trivial, but the possibility of this kind of analysis undergirds the possibility of reductive explanation in general. Without such an analysis, there would be no explanatory bridge from the lower-level physical facts to the phenomenon in question. With such an analysis in hand, all we need to do is to show how certain lower-level physical mechanisms allow the analysis to be satisfied, and an explanation will result.

For the most interesting phenomena that require explanation, including phenomena such as reproduction and learning, the relevant notions can usually be analyzed functionally. The core of such notions can be characterized in terms of the performance of some function or functions (where "function" is taken causally rather than teleologically), or in terms of the capacity to perform those functions. It follows that once we have explained how those functions are performed, then we have explained the phenomena in question. Once we explain how an organism performs the function of producing another organism, we have explained reproduction, for all it means to reproduce is to perform that function. The same goes for an explanation of learning. All it means for an organism to learn, roughly, is for its behavioral capacities to adapt appropriately in response to environmental stimulation. If we explain how the organism is able to perform the relevant functions, then we have explained learning.

(At most, we may have failed to explain any phenomenal aspects of learning, which I leave aside here for obvious reasons. If there is a phenomenal element to the concept of learning, then that part of learning may go unexplained; but I concentrate on the psychological aspects of learning here, which are plausibly the core of the concept.)

Explaining the performance of these functions is quite straightforward, in principle. As long as the results of such functions are themselves characterizable physically, and all physical events have physical causes, then there should be a physical explanation for the performance of any such function. One need only show how certain sorts of states are responsible for the production of appropriate resultant states, by a causal process in accord with the laws of nature. Of course the details of this kind of physical explanation can be nontrivial. Indeed, the details constitute the vast bulk of any reductive explanation, while the analysis component is often trivial. But once the relevant details are in, a story about low-level physical causation will explain how the relevant functions are performed, and will therefore explain the phenomenon in question.

Even a physical notion such as heat can be construed functionally: roughly, heat is the kind of thing that expands metals, is caused by fire, leads to a

Supervenience and Explanation 45

particular sort of sensation, and the like. Once we have an account of how these various causal relations are fulfilled, then we have an account of heat. Heat is a causal-role concept, characterized in terms of what it is typically caused by and of what it typically causes, under appropriate conditions. Once empirical investigation shows how the relevant causal role is played, the phenomenon is explained.

There are some technical complications here, but they are inessential. For example, Kripke (1980) has pointed out a difference between a term such as "heat" and the associated description of a causal role: given that heat is realized by the motion of molecules, then the motion of molecules might qualify as heat in a counterfactual world, whether or not those molecules play the relevant causal role. It remains the case, however, that explaining heat involves explaining the fulfillment of the causal role, rather than explaining the motion of molecules. To see this, note that the equivalence of heat with the motion of molecules is known a posteriori: we know this as a result of explaining heat. The concept of heat that we had a priori-before the phenomenon was explained-was roughly that of "the thing that plays this causal role in the actual world." Once we discover how that causal role is played, we have an explanation of the phenomenon. As a bonus, we know what heat is. It is the motion of molecules, as the motion of molecules is what plays the relevant causal role in the actual world.

A second minor complication is that many causal-role concepts are somewhat ambiguous between the state that plays a certain causal role and the actual performance of that role. "Heat" can be taken to denote either the molecules that do the causal work or the causal process (heating) itself. Similarly, "perception" can be used to refer to either the act of perceiving or the internal state that arises as a result. Nothing important turns on this ambiguity, however. An explanation of how the causal role is played will explain heat or perception in either of these senses.

A third complication is that many causal-role concepts are partly characterized in terms of their effect on experience: for example, heat is naturally construed as the cause of heat sensations. Does this mean that we have to explain heat sensations before we can explain heat? Of course, we have no good account of heat sensations (or of experience generally), so what happens in practice is that that part of the phenomenon is left unexplained. If we can explain how molecular motion comes about in certain conditions, and causes metals to expand, and stimulates our skin in certain ways, then the observation that this motion is correlated with heat sensations is good enough. From the correlation, we infer that there is almost certainly a causal connection. To be sure, no explanation of heat will be complete until we have an account of how that causal connection works, but the incomplete account is good enough for most purposes. It is somewhat paradoxical that we end up explaining almost everything about a phenomenon except for the details of

46 Foundations

how it affects our phenomenology, but it is not a problem in practice. It would not be a happy state of affairs if we had to put the rest of science on hold until we had a theory of consciousness.

Reductive explanations in cognitive science

The paradigm of reductive explanation via functional analysis works beautifully in most areas of cognitive science, at least in principle. As we saw in the previous chapter, most nonphenomenal mental concepts can be analyzed functionally. Psychological states are characterizable in terms of the causal role they play. To explain these states, we explain how the relevant causation is performed.

In principle, one can do this by giving an account of the underlying neurophysiology. If we explain how certain neurophysiological states are responsible for the performance of the functions in question, then we have explained the psychological state. We need not always descend to the neurophysiological level, however. We can frequently explain some aspect of mentality by exhibiting an appropriate cognitive model-that is, by exhibiting the details of the abstract causal organization of a system whose mechanisms are sufficient to perform the relevant functions, without specifying the physiochemical substrate in which this causal organization is implemented. In this way, we give a how-possibly explanation of a given aspect of psychology, in that we have shown how the appropriate causal mechanisms might support the relevant mental processes. If we are interested in explaining the mental states of an actual organism or type of organism (e.g., learning in humans, as opposed to the possibility of learning in general), this sort of explanation must be supplemented with a demonstration that the causal organization of the model mirrors the causal organization of the organism in question.

To explain the possibility of learning, we can exhibit a model whose mechanisms lead to the appropriate changes in behavioral capacity in response to various kinds of environmental stimulation-a connectionist learning model, for example. To explain human learning, we must also demonstrate that such a model reflects the causal organization responsible for the performance of such functions in humans. The second step is usually difficult: we cannot exhibit such a correspondence directly, due to our ignorance of neurophysiology, so we usually have to look for indirect evidence, such as qualitative similarities in patterns of response, measurements of timing, and the like. This is one reason why cognitive science is currently in an undeveloped state. But as usual, the in-principle possibility of such explanation is a straightforward consequence of the functional nature of psychological concepts.

Unfortunately, the kind of functional explanation that works so well for psychological states does not seem to work in explaining phenomenal states.

Supervenience and Explanation 47

The reason for this is straightforward. Whatever functional account of human cognition we give, there is a further question: Why is this kind of functioning accompanied by consciousness? No such further question arises for psychological states. If one asked about a given functional model of learning, "Why is this functioning accompanied by learning?" the appropriate answer is a semantic answer: "Because all it means to learn is to function like this." There is no corresponding analysis of the concept of consciousness. Phenomenal states, unlike psychological states, are not defined by the causal roles that they play. It follows that explaining how some causal role is played is not sufficient to explain consciousness. After we have explained the performance of a given function, the fact that consciousness accompanies the performance of the function (if indeed it does) remains quite unexplained.

One can put the point the following way. Given an appropriate functional account of learning, it is simply logically impossible that something could instantiate that account without learning (except perhaps insofar as learning requires consciousness). However, no matter what functional account of cognition one gives, it seems logically possible that that account could be instantiated without any accompanying consciousness. It may be naturally impossible-consciousness may in fact arise from that functional organization in the actual world-but the important thing is that the notion is logically coherent.

If this is indeed logically possible, then any functional and indeed any physical account of mental phenomena will be fundamentally incomplete. To use a phrase coined by Levine (1983), there is an explanatory gap between such accounts and consciousness itself. Even if the appropriate functional organization always gives rise to consciousness in practice, the question of why it gives rise to consciousness remains unanswered. This point will be developed at length later.

If this is so, it follows that there will be a partial explanatory gap for any mental concept that has a phenomenal element. If conscious experience is required for belief or learning, for example, we may not have a fully reductive explanation for belief or learning. But we at least have reason to believe that the psychological aspects of these mental features-which are arguably at the core of the relevant concepts-will be susceptible to reductive explanation in principle. If we leave worries about phenomenology aside, cognitive science seems to have the resources to do a good job of explaining the mind.

3. Logical Supervenience and Reductive Explanation

The epistemology of reductive explanation meets the metaphysics of Supervenience in a straightforward way. A natural phenomenon is reductively explainable in terms of some low-level properties precisely when it is logically

48 Foundations

supervenient on those properties. It is reductively explainable in terms of physical properties-or simply "reductively explainable"-when it is logically supervenient on the physical.

To put things more carefully: A natural phenomenon is reductively explainable in terms of some lower-level properties if the property of instantiating that phenomenon is globally logically supervenient on the low-level properties in question. A phenomenon is reductively explainable simpliciter if the property of exemplifying that phenomenon is globally logically supervenient on physical properties.

This can be taken as an explication of the notion of reductive explanation, with perhaps an element of stipulation. That our prior notion of reductive explanation implies (global) logical Supervenience should be clear from the earlier discussion. If the property of exemplifying a phenomenon fails to supervene logically on some lower-level properties, then given any lower-level account of those properties, there will always be a further unanswered question: Why is this lower-level process accompanied by the phenomenon? Reductive explanation requires some kind of analysis of the phenomenon in question, where the low-level facts imply the realization of the analysis. So reductive explanation requires a logical Supervenience relation. For example, it is precisely because reproduction is logically supervenient on lower-level facts that it is reductively explainable in terms of those facts.

That logical Supervenience suffices for reductive explainability is somewhat less clear. If a phenomenon P supervenes logically on some lower-level properties, then given an account of the lower-level facts associated with an instance of P, the exemplification of P is a logical consequence. An account of the lower-level facts will therefore automatically yield an explanation of P. Nevertheless, such an explanation can sometimes seem unsatisfactory, for two reasons. First, the lower-level facts might be a vast hotchpotch of arbitrary-seeming details without any clear explanatory unity. An account of all the molecular motions underlying an instance of learning might be like this, for example. Second, it is possible that different instances of P might be accompanied by very different sets of low-level facts, so that explanations of particular instances do not yield an explanation of the phenomenon as a type.

One option is to hold that logical Supervenience is merely necessary for reductive explanation, rather than sufficient. This is all that is required for my arguments about consciousness in the next chapter. But it is more useful to note that there is a useful notion of reductive explanation such that logical Supervenience is both necessary and sufficient. Instead of taking the problems above as indicating that the accounts in question are not explanations, we can instead take them to indicate that a reductive explanation is not necessarily an illuminating explanation. Rather, a reductive explanation is a mystery-removing explanation.

Supervenience and Explanation 49

As I noted earlier, reductive explanation is not the be-all and end-all of explanation. Its chief role is to remove any deep sense of mystery surrounding a high-level phenomenon. It does this by reducing the bruteness and arbitrariness of the phenomenon in question to the bruteness and arbitrariness of lower-level processes. Insofar as the low-level processes may themselves be quite brute and arbitrary, a reductive explanation may not give us a deep understanding of a phenomenon, but it at least eliminates any sense that there is something "extra" going on.

The gap between a reductive explanation and an illuminating explanation can generally be closed much further than this, however. This is due to two basic facts about the physics of our world: autonomy and simplicity. Microphysical causation and explanation seem to be autonomous, in that every physical event has a physical explanation; the laws of physics are sufficient to explain the events of physics on their own terms. Further, the laws in question are reasonably simple, so that the explanations in question have a certain compactness. Both of these things might have been otherwise. We might have lived in a world in which there were brutely emergent fundamental laws governing the behavior of high-level configurations such as organisms, with an associated downward causation that overrides any relevant microphysical laws. (The British emergentists, such as Alexander [1920] and Broad [1925], believed our world to be something like this.) Alternatively, our world might have been a world in which the behavior of microphysical entities is governed only by a vast array of baroque laws, or perhaps a world in which microphysical behavior is lawless and chaotic. In worlds like these, there would be little hope of achieving an illuminating reductive explanation, as the bruteness of low-level accounts might never be simplified.

But the actual world, with its low-level autonomy and simplicity, seems to allow that sense can generally be made even of complex processes. The low-level facts underlying a high-level phenomenon often have a basic unity that allows for a comprehensible explanation. Given an instance of high-level causation, such as a released trigger causing a gun to fire, we can not only isolate a bundle of lower-level facts that fix this causation; we can also tell a fairly simple story about how the causation is enabled, by encapsulating those facts under certain simple principles. This may not always work. It may be the case that some domains, such as those of sociology and economics, are so far removed from the simplicity of low-level processes that illuminating reductive explanation is impossible, even if the phenomena are logically supervenient. If so, then so be it: we can content ourselves with high-level explanations of those domains, while noting that logical Supervenience implies that there is a reductive explanation in principle, although perhaps one that only a superbeing could understand.

50 Foundations

Note also that on this account reductive explanation is fundamentally particular, accounting for particular instances of a phenomenon, without necessarily accounting for all instances together. This is what we should expect. If a property can be instantiated in many different ways, we cannot expect a single explanation to cover all the instances. Temperature is instantiated quite differently in different media, for example, and there are different explanations for each. At a much higher level, it is most unlikely that there should be a single explanation covering all instances of murder. Still, there is frequently a certain unity across the explanation of particulars, in that a good explanation of one is often an explanation of many. This is again a consequence of the underlying simplicity of our world, rather than a necessary property of explanation. In our world, the simple unifying stories that one can tell about lower-level processes often apply across the board, or at least across a wide range of particulars. It is also frequently the case, especially in the biological sciences, that the particulars have a common ancestry that leads to a similarity in the low-level processes involved. So the second problem mentioned, that of unifying the explanations of specific instances of a phenomenon, is not as much of a problem as it might be. In any case, it is the explanation of particulars that is central.

There is much more that could be said about closing the gap between reductive explanation and illuminating explanation, but the matter deserves a lengthy treatment in its own right and is not too important for my purposes. What is most important is that if logical Supervenience fails (as I will argue it does for consciousness), then any kind of reductive explanation fails, even if we are generous about what counts as explanation. Also important is that logical Supervenience removes any residual metaphysical mystery about a high-level phenomenon, by reducing any brutality in that phenomenon to brutality in lower-level facts. Of secondary importance is that if logical Supervenience holds, then some sort of reductive explanation is possible. Although such explanations can fail to be illuminating or useful, this failure is not nearly as fundamental as the failure of explanation in domains where logical Supervenience does not hold.

Further notes on reductive explanation

A few further notes: First, a practical reductive explanation of a phenomenon does not usually go all the way to the microphysical level. To do this would be enormously difficult, giving rise to all the brutality problems just discussed. Instead, high-level phenomena are explained in terms of some properties at a slightly more basic level, as when reproduction is explained in terms of cellular mechanisms, or the phases of the moon are explained in terms of orbital motion. In turn, one hopes that the more basic phenomena

Supervenience and Explanation 51

will themselves be reductively explainable in terms of something more basic still. If all goes well, biological phenomena may be explainable in terms of cellular phenomena, which are explainable in terms of biochemical phenomena, which are explainable in terms of chemical phenomena, which are explainable in terms of physical phenomena. As for the physical phenomena, one tries to unify these as far as possible, but at some level physics has to be taken as brute: there may be no explanation of why the fundamental laws or boundary conditions are the way they are. This ladder of explanation is little more than a pipe dream at the moment, but significant progress has been made. Given logical Supervenience, along with the simplicity and autonomy of the lowest level, this sort of explanatory connection between the sciences ought to be possible in principle. Whether the complexities of reality will make it practically infeasible is an open question.

Second, it is at least conceivable that a phenomenon might be reductively explainable in terms of lower-level properties without being reductively explainable simpliciter. This might happen in a situation where C-properties are logically supervenient on B-properties, and are therefore explainable in terms of B-properties, but where B-properties themselves are not logically supervenient on the physical. There is clearly one sense in which such an explanation is reductive and another sense in which it is not. For the most part, I will be concerned with reductive explanation in terms of the physical, or in terms of properties that are themselves explainable in terms of the physical, and so on. Even if the C-properties here are reductively explainable in a relative sense, their very existence implies the failure of reductive explanation in general.

Third, local logical Supervenience is too stringent a requirement for reductive explanation. One can reductively explain even context-dependent properties of an individual by giving an account of how relevant environmental relations come to be satisfied. As long as a phenomenon is globally supervenient, it will be reductively explainable in terms of some lower-level facts, even if these are spread widely in space and time.

Fourth, in principle there are two projects in reductive explanation of a phenomenon such as life, learning, or heat. There is first a project of explication, where we clarify just what it is that needs to be explained, by means of analysis. Learning might be analyzed as a certain kind of adaptational process, for example. Second, there is a project of explanation, where we see how that analysis comes to be satisfied by the low-level facts. The first project is conceptual, and the second is empirical. For many or most phenomena, the conceptual stage will be quite trivial. For some phenomena, however, such as belief, explication can be a major hurdle in itself. In practice, of course, there is never a clean separation between the projects, as explication and explanation take place in parallel.

52 Foundations

4. Conceptual Truth and Necessary Truth*

In my account of Supervenience and explanation, I have relied heavily on the notions of logical possibility and necessity. It is now time to say something more about this. The basic way to understand the logical necessity of a statement is in terms of its truth across all logically possible worlds. This requires some care in making sense of both the relevant class of worlds and the way that statements are evaluated in worlds; I will discuss this at some length later in this section. It is also possible to explicate the logical necessity of a statement as truth in virtue of meaning: a statement is logically necessary if its truth is ensured by the meaning of the concepts involved. But again, this requires care in understanding just how the "meanings" should be taken. I will discuss both of these ways of looking at things, and their relation, later in this section.

(As before, the notion of logical necessity is not to be identified with a narrow notion involving derivability in first-order logic, or some other syntactic formalism. Indeed, it is arguable that the justification of the axioms and rules in these formalisms depends precisely on their logical necessity in the broader, more primitive sense.)

All this requires taking seriously, at least to some extent, the notion of conceptual truth-that is, the notion that some statements are true or false simply by virtue of the meanings of the terms involved. Key elements of my discussion so far have depended on characterizations of various concepts. I have accounted for the reductive explanation of reproduction, for example, by arguing that low-level details entail that certain functions are performed, and that performance of these functions is all there is to the concept of reproduction.

The notion of conceptual truth has had a bad name in some circles since the critique by Quine (1951), who argued that there is no useful distinction between conceptual truths and empirical truths. The objections to these notions usually cluster around the following points:

1. Most concepts do not have definitions giving necessary and sufficient conditions (this observation has been made many times but is often associated with Wittgenstein 1953).

2. Most apparent conceptual truths are in fact revisable, and could be withdrawn in the face of sufficient empirical evidence (a point raised by Quine).

3. Considerations about a posteriori necessity, outlined by Kripke (1972), show that application-conditions of many terms across possible worlds cannot be known a priori.

These considerations count against an overly simplistic view of conceptual truth, but not against the way I am using these notions. In particular, it turns

Supervenience and Explanation 53

out that the class of Supervenience conditionals-"If the A-facts about a situation are X, then the B-facts are Y," where the A-facts fully specify a situation at a fundamental level-are unaffected by these considerations. These are the only conceptual truths that my arguments need, and we will see that none of the considerations above count against them. I will also analyze the relationship between conceptual truth and necessary truth in more detail, and spell out the role these play in understanding logical Supervenience.


The absence of cut-and-dried definitions is the least serious of the difficulties with conceptual truth. None of my arguments depend on the existence of such definitions. I occasionally rely on analyses of various notions, but these analyses need only be rough and ready, without any pretense at providing precise necessary and sufficient conditions. Most concepts (e.g., "life") are somewhat vague in their application, and there is little point trying to remove that vagueness by arbitrary precision. Instead of saying "A system is alive if and only if it reproduces, adapts with utility 800 or greater, and metabolizes with efficiency 75 percent, or exhibits these in a weighted combination with such-and-such properties," we can simply note that if a system exhibits these phenomena to a sufficient degree then it will be alive, by virtue of the meaning of the term. If an account of relevant low-level facts fixes the facts about a system's reproduction, utility, metabolism, and so on, then it also fixes the facts about whether the system is alive, insofar as that matter is factual at all.

We can sum this up with a schematic diagram (Figure 2.1) showing how a high-level property P might depend on two low-level parameters A and B, each of which can take on a range of values. If we had a crisp definition in terms of necessary and sufficient conditions, then we would have something like the picture at left, where the dark rectangle represents the region in which property P is instantiated. Instead, the dependence is invariably something like the picture at right, where the boundaries are vague and there is a large area in which the matter of P-hood is indeterminate, but there is also an area in which the matter is clear. (It may be indeterminate whether bacteria or computer viruses are alive, but there is no doubt that dogs are alive.) Given an example in the determinate area, exemplifying A and B to sufficient degrees that P is exemplified, the conditional "If x is A and B to this degree, then x is P" is a conceptual truth, despite the lack of a clean definition of P. Any indeterminacy in such conditionals, in the gray areas, will reflect indeterminacy in the facts of the matter, which is as it should be. The picture can straightforwardly be extended to dependence of a property

54 Foundations

Figure 2.1. Two ways in which a property P might depend on properties A and B.

on an arbitrary number of factors, and to Supervenience conditionals in general.

Importantly, then, one set of facts can entail another set without there being a clean definition of the latter notions in terms of the former. The above case provides an example: there is no simple definition of P in terms of A and B, but the facts about A and B in an instance entail the facts about P. For another example, think about the roundness of closed curves in two-dimensional space (Figure 2.2). There is certainly no perfect definition of roundness in terms of simpler mathematical notions. Nevertheless, take the figure at left, specified by the equation 2x2 + 3y2 = 1. There is a fact of the matter-this figure is round-insofar as there are ever facts about roundness at all (compare to the figure at right, which is certainly not round). Further, this fact is entailed by the basic description of the figure in mathematical terms-given that description, and the concept of roundness, the fact that the figure is round is determined. Given that A-facts can entail B-facts without a definition of B-facts in terms of A-facts, the notion of logical Supervenience is unaffected by the absence of definitions. (In thinking about more complex issues and objections concerning logical Supervenience, it may be worthwhile to keep this example in mind.)

We can put the point by saying that the sort of "meaning" of a concept that is relevant in most cases is not a definition, but an intension: a function specifying how the concept applies to different situations. Sometimes an intension might be summarizable in a definition, but it need not be, as these cases suggest. But as long as there is a fact of the matter about how concepts apply in various situations, then we have an intension; and as I will discuss shortly, this will generally be all the"meaning" that my arguments will need.

Supervenience and Explanation 55

Figure 2.2. The round curve 2x2 + 3y2 = 1 and nonround friend.



The second objection, raised by Quine (1951), is that purported conceptual truths are always subject to revision in the face of sufficient empirical evidence. For instance, if evidence forces us to revise various background statements in a theory, it is possible that a statement that once appeared to be conceptually true might turn out to be false.

This is so for many purported conceptual truths, but it does not apply to the Supervenience conditionals that we are considering, which have the form "If the low-level facts turn out like this, then the high-level facts will be like that." The facts specified in the antecedent of this conditional effectively include all relevant empirical factors. Empirical evidence could show us that the antecedent of the conditional is false, but not that the conditional is false. In the extreme case, we can ensure that the antecedent gives a full specification of the low-level facts about the world. The very comprehensiveness of the antecedent ensures that empirical evidence is irrelevant to the conditional's truth-value. (This picture is somewhat complicated by the existence of a posteriori necessities, which I discuss shortly. Here, I am only concerned with epistemic conditionals about ways the actual world might turn out.)

While considerations about revisability provide a plausible argument that there are not many short conceptual truths, nothing in these considerations counts against the constrained, complex sort of conceptual truth that I have been concerned with. The upshot of these observations is that the truth-conditions of a high-level statement may not be easily localizable, as all sorts of factors might have some kind of indirect relevance; but the global truth conditions provided by a Supervenience conditional are not threatened. Indeed, if meaning determines a function from possible worlds to reference classes (an intension), and if possible worlds are finitely describable (in

56 Foundations

terms of arrangement of basic qualities in those worlds, say), then there will automatically be a vast class of conceptually true conditionals that result.

A posteriori necessity

It has traditionally been thought that all conceptual truths are knowable a priori, as are all necessary truths, and that the classes of a priori truths, necessary truths, and conceptual truths are closely related or even coextensive. Saul Kripke's book Naming and Necessity (1972) threw a wrench into this picture by arguing that there is a large class of necessarily true statements whose truth is not knowable a priori. An example is the statement "Water is H2O." We cannot know this to be true a priori; for all we know (or for all we knew at the beginning of inquiry), water is made out of something else, perhaps XYZ. Kripke argues that nevertheless, given that water is H2O in the actual world, then water is H2O in all possible worlds. It follows that "Water is H2O" is a necessary truth despite its a posteriori nature.

This raises a few difficulties for the framework I have presented. For example, on some accounts these necessary truths are conceptual truths, implying that not all conceptual truths are knowable a priori. On alternative accounts, such statements are not conceptual truths, but then the link between conceptual truth and necessity is broken. At various points in this book, I use a priori methods to gain insight into necessity; this is the sort of thing that Kripke's account is often taken to challenge.

On analysis, I think it can be seen that these complications do not change anything fundamental to my arguments; but it worth taking the trouble to get clear about what is going on. I will spend some time setting up a systematic framework for dealing with these issues, which will recur. In particular, I will present a natural way of capturing Kripke's insights in a two-dimensional picture of meaning and necessity. This framework is a synthesis of ideas suggested by Kripke, Putnam, Kaplan, Stalnaker, Lewis, Evans, Davies and Humberstone, and others who have addressed these two-dimensional phenomena.

On the traditional view of reference, derived from Frege although cloaked here in modern terminology, a concept determines a function f: W -> R from possible worlds to referents. Such a function is often called an intension; together with a specification of a world w, it determines an extension f(w). In Frege's own view, every concept had a sense, which was supposed to determine the reference of the concept depending on the state of the world; so these senses correspond closely to intensions. The sense was often thought of as the meaning of the concept in question.

More recent work has recognized that no single intension can do all the work that a meaning needs to do. The picture developed by Kripke complicates things by noting that reference in the actual world and in counterfactual

Supervenience and Explanation 57

possible worlds is determined by quite different mechanisms. In a way, the Kripkean picture can be seen to split the Fregean picture into two separate levels.

Kripke's insight can be expressed by saying that there are in fact two intensions associated with a given concept. That is, there are two quite distinct patterns of dependence of the referent of a concept on the state of the world. First, there is the dependence by which reference is fixed in the actual world, depending on how the world turns out: if it turns out one way, a concept will pick out one thing, but if it turns out another way, the concept will pick out something else. Second, there is the dependence by which reference in counterfactual worlds is determined, given that reference in the actual world is already fixed. Corresponding to each of these dependencies is an intension, which I will call the primary and secondary intensions, respectively.

The primary intension of a concept is a function from worlds to extensions reflecting the way that actual-world reference is fixed. In a given world, it picks out what the referent of the concept would be if that world turned out to be actual. Take the concept "water." If the actual world turned out to have XYZ in the oceans and lakes, then "water" would refer to XYZ,21 but given that it turns out to have H2O in the oceans and lakes, "water" refers to H2O. So the primary intension of "water" maps the XYZ world to XYZ, and the H2O world to H2O. At a rough approximation, we might say that the primary intension picks out the dominant clear, drinkable liquid in the oceans and lakes; or more briefly, that it picks out the watery stuff in a world.

However, given that "water" turns out to refer to H2O in the actual world, Kripke notes (as does Putnam [1975]) that it is reasonable to say that water is H2O in every counterfactual world. The secondary intension of "water" picks out the water in every counterfactual world; so if Kripke and Putnam are correct, the secondary intension picks out H2O in all worlds.22

It is the primary intension of a concept that is most central for my purposes: for a concept of a natural phenomenon, it is the primary intension that captures what needs explaining. If someone says, "Explain water," long before we know that water is in fact H2O, what they are asking for is more or less an explanation of the clear, drinkable liquid in their environment. It is only after the explanation is complete that we know that water is H2O. The primary intension of a concept, unlike the secondary intension, is independent of empirical factors: the intension specifies how reference depends on the way the external world turns out, so it does not itself depend on the way the external world turns out.

Of course, any brief characterization of the primary intension of a concept along the lines of "the dominant clear, drinkable liquid in the environment" will be a simplification. The true intension can be determined only from detailed consideration of specific scenarios: What would we say if the world

58 Foundations

turned out this way? What would we say if it turned out that way? For example, if it had turned out that the liquid in lakes was H2O and the liquid in oceans XYZ, then we probably would have said that both were water; if the stuff in oceans and lakes was a mixture of 95 percent A and 5 percent B, we would probably have said that A but not B was water; if it turned out that a substance neither clear not drinkable bore an appropriate microphysical relation to the clear, drinkable liquid in our environment, we would probably call that substance "water" too (as we do in the case of ice or of "dirty water"). The full conditions for what it takes to qualify as "water" will be quite vague at the edges and need not be immediately apparent on reflection, but none of this makes much difference to the picture I am describing. I will use "watery stuff" as a term of art to encapsulate the primary intension, whatever it is.23

In certain cases, the decision about what a concept refers to in the actual world involves a large amount of reflection about what is the most reasonable thing to say; as, for example, with questions about the reference of "mass" when the actual world turned out to be one in which the theory of general relativity is true,24 or perhaps with questions about what qualifies as "belief" in the actual world. So consideration of just what the primary intension picks out in various actual-world candidates may involve a corresponding amount of reflection. But this is not to say that the matter is not a priori: we have the ability to engage in this reasoning independently of how the world turns out. Perhaps the reports of experiments confirming relativity are disputed, so we are not sure whether the actual world has turned out to be a relativistic world: either way, we have the ability to reason about what "mass" will refer to if that state of affairs turns out to be actual.

(Various intricacies arise in analyzing the primary intensions of concepts used by individuals within a linguistic community. These might be handled by noting that an individual's concept may have a primary intension that involves deference to a surrounding community's concept-so my concept "elm" might pick out what those around me call "elms"; but in any case this sort of problem is irrelevant to the issues I will be concerned with, for which we might as well assume that there is just one person in the community, or that all individuals are equally well informed, or even that the community is a giant individual. There are also a few technical problems that might come up in using primary intensions to build a general semantic theory-for example, is the reference of a concept essential to the concept? Might different speakers associate different primary intensions with the same word? But I am not trying to build a full semantic theory here, and we can abstract away from this sort of concern.

Sometimes philosophers are suspicious of entities such as primary intensions because they see them as reminiscent of a "description" theory of reference. But descriptions play no essential part in this framework; I use

Supervenience and Explanation 59

them merely to flesh out some of the character of the relevant functions from possible worlds to extensions. It is the function itself, rather than any summarizing description, that is truly central. This picture is quite compatible with the "causal" theory of reference: we need simply note that the primary intension of a concept such as "water" may require an appropriate causal connection between the referent and the subject. Indeed, we are led to believe in a causal theory of reference in the first place precisely by considering various ways the actual world might turn out, and noting what the referent of the concept would turn out to be in those cases; that is, by evaluating the primary intension of a concept at those worlds.)

Given that the actual-world reference of "water" is fixed by picking out the watery stuff, one might think that water is watery stuff in all possible worlds. Kripke and Putnam pointed out that this is not so: if water is H2O in the actual world, then water is H2O in all possible worlds. In a world (Putnam's "Twin Earth") in which the dominant clear, drinkable liquid is XYZ rather than H2O, this liquid is not water; it is merely watery stuff. All this is captured by the secondary intension of "water," which picks out the water in all worlds: that is, it picks out H2O in all worlds.

The secondary intension of a concept such as "water" is not determined a priori, as it depends on how things turn out in the actual world. But it still has a close relation to the primary intension above. In this case, the secondary intension is determined by first evaluating the primary intension at the actual world, and then rigidifying this evaluation so that the same sort of thing is picked out in all possible worlds. Given that the primary intension ("watery stuff") picks out H2O in the actual world, it follows from rigidification that the secondary intension picks out H2O in all possible worlds.

We can sum this up by saying "water" is conceptually equivalent to "dthat (watery stuff)," where dthat is a version of Kaplan's rigidifying operator, converting an intension into a rigid designator by evaluation at the actual world (Kaplan 1979).The single Fregean intension has fragmented into two: a primary intension ("watery stuff") that fixes reference in the actual world, and a secondary intension ("H2O ") that picks out reference in counterfactual possible worlds, and which depends on how the actual world turned out.

(There is sometimes a tendency to suppose that a posteriori necessity makes a priori conceptual analysis irrelevant, but this supposition is ungrounded. Before we even get to the point where rigid designation and the like become relevant, there is a story to tell about what makes an actual-world X qualify as the referent of "X" in the first place. This story can only be told by an analysis of the primary intension. And this project is an a priori enterprise, as it involves questions about what our concept would refer to if the actual world turned out in various ways. Given that we have the ability to know what our concepts refer to when we know how the actual world turns out, then we have the ability to know what our concepts would

60 Foundations

refer to if the actual world turned out in various ways. Whether or not the actual world does turn out a certain way makes little difference in answering this question, except in focusing our attention.)

Both the primary and secondary intensions can be seen as functions f: W -> R from possible worlds to extensions, where the possible worlds in question are seen in subtly different ways. We might say that the primary intension picks out the referent of a concept in a world when it is considered as actual-that is, when it is considered as a candidate for the actual world of the thinker-whereas the secondary intension picks out the referent of a concept in a world when it is considered as counterfactual, given that the actual world of the thinker is already fixed. When the XYZ world is considered as actual, my term "water" picks out XYZ in the world, but when it is considered as counterfactual, "water" picks out H2O.

The distinction between these two ways of looking at worlds corresponds closely to Kaplan's (1989) distinction between the context of utterance of an expression and the circumstances of evaluation. When we consider a world w as counterfactual, we keep the actual world as the context of utterance, but use w as a circumstance of evaluation. For example, if I utter "There is water in the ocean" in this world and evaluate it in the XYZ world, "water" refers to H2O and the statement is false. But when we consider w as actual, we think of it as a potential context of utterance, and wonder how things would be if the context of the expression turned out to be w. If the context of my sentence "There is water in the ocean" turned out to be the XYZ world, then the statement would be true when evaluated at that world. The primary intension is therefore closely related to what Kaplan calls the character of a term, although there are a few differences,25 and the secondary intension corresponds to what he calls a term's content.

There is a slight asymmetry in that a context of utterance but not the circumstance of evaluation is what Quine (1969) calls a centered possible world. This is an ordered pair consisting of a world and a center representing the viewpoint within that world of an agent using the term in question: the center consists in (at least) a "marked" individual and time. (This suggestion comes from Lewis 1979; Quine suggests that the center might be a point in space-time.) Such a center is necessary to capture the fact that a term like "water" picks out a different extension for me than for my twin on Twin Earth, despite the fact that we live in the same universe.26 It is only our position in the universe that differs, and it is this position that makes a relevant difference to the reference-fixing process.

This phenomenon arises in an especially obvious way for indexical terms such as "I", whose reference clearly depends on who is using the term and not just on the overall state of the world: the primary intension of "I" picks out the individual at the center of a centered world. (The secondary intension of my concept "I" picks out David Chalmers in all possible worlds.)

Supervenience and Explanation 61

There is a less overt indexical element in notions such as "water," however, which can be roughly analyzed as "dthat(the dominant clear, drinkable liquid in our environment)."27 It is this indexical element that requires primary intensions to depend on centered worlds. Once actual-world reference is fixed, however, no center is needed to evaluate reference in a counterfactual world. The circumstance of evaluation can therefore be represented by a simple possible world without a center.

All this can be formalized by noting that the full story about reference in counterfactual worlds is not determined a priori by a singly indexed function f: W -> R. Instead, reference in a counterfactual world depends both on that world and on the way the actual world turns out. That is, a concept determines a doubly indexed function

F:W* x W -> R

where W* is the space of centered possible worlds, and W is the space of ordinary possible worlds. The first parameter represents contexts of utterance, or ways the actual world might turn out, whereas the second parameter represents circumstances of evaluation, or counterfactual possible worlds. Equivalently, a concept determines a family of functions

Fv: W -> R

for each v Î W* representing a way the actual world might turn out, where Fv(w) = F(v, w). For "water," if a is a world in which watery stuff is H2O, then Fa picks out H2O in any possible world. Given that in our world water did turn out to be H2O, this Fa specifies the correct application conditions for "water" across counterfactual worlds. If our world had turned out to be a different world b in which watery stuff was XYZ, then the relevant application conditions would have been specified by Fb, a different intension which picks out XYZ in any possible world.

The function F is determined a priori, as all a posteriori factors are included in its parameters. From F we can recover both of our singly indexed intensions. The primary intension is the function f: W* -> R determined by the "diagonal" mapping f: w F(w, w'), where w' is identical to w except that the center is removed. This is the function whereby reference in the actual world is fixed. The secondary intension is the mapping Fa: w  F(a, w), where a is our actual world. This intension picks out reference in counterfactual worlds. An immediate consequence is that the primary intension and secondary intension coincide in their application to the actual world: f(a) = Fa(a') = F(a, a').

In the reverse direction, the doubly indexed function F and therefore the secondary intension Fa can usually be derived from the primary intension f,

62 Foundations

with the aid of a "rule" about how the secondary intension depends on the primary intension and the actual world a. This rule depends on the type of concept. For a concept that is a rigid designator, the rule is that in a world w, the secondary intension picks out in w whatever the primary intension picks out in a (or perhaps, for natural-kind terms, whatever has the same underlying structure as what the primary intension picks out in a). More formally, let D : R X W -> R be a "projection" operator that goes from a class picked out in some world to members of "that" class in another possible world. Then the secondary intension Fa is just the function D(f(a),-), which we can think of as dthat applied to the intension given by f.

For other concepts, derivation of the secondary intension from the primary intension will be easier. With "descriptive" expressions such as "doctor," "square," and "watery stuff," rigid designation plays no special role: they apply to counterfactual worlds independently of how the actual world turns out. In these cases, the secondary intension is a simple copy of the primary intension (except for differences due to centering). The framework I have outlined can handle both sorts of concepts.

Property terms, such as "hot," can be represented in one of two ways in an intensional framework. We can see the intension of a property as a function from a world to a class of individuals (the individuals that instantiate the property), or from a world to properties themselves. Either way of doing things is compatible with the current framework: we can easily find a primary and a secondary intension in either case, and it is easy to move back and forth between the two frameworks. I will usually do things the first way, however, so that the primary intension of "hot" picks out the entities that qualify as "hot" things in the actual world, depending on how it turns out, and the secondary intension picks out the hot things in a counterfactual world, given that the actual world has turned out as it has.


Both the primary and the secondary intensions can be thought of as candidates for the "meaning" of a concept. I think there is no point choosing one of these to qualify as the meaning; the term "meaning" here is largely an honorific. We might as well think of the primary and secondary intensions as the a priori and a posteriori aspects of meaning, respectively.

If we make this equation, both of these intensions will back a certain kind of conceptual truth, or truth in virtue of meaning. The primary intension backs a priori truths, such as "Water is watery stuff." Such a statement will be true no matter how the actual world turns out, although it need not hold in all nonactual possible worlds. The secondary intension does not back a priori truths, but backs truths that hold in all counterfactual possible worlds, such as "Water is H2O." Both varieties qualify as truths in virtue of meaning; they are simply true in virtue of different aspects of meaning.

Supervenience and Explanation 63

It is also possible to see both as varieties of necessary truth. The latter corresponds to the more standard construal of a necessary truth. The former, however, can also be construed as truth across possible worlds, as long as these possible worlds are construed as contexts of utterance, or as ways the actual world might turn out. On this subtly different construal, a statement S is necessarily true if no matter how the actual world turns out, it would turn out that S was true. If the actual world turns out to be a world in which watery stuff is XYZ, then my statement "XYZ is water" will turn out to be true. So, according to this construal on which possible worlds are considered as actual, "Water is watery stuff" is a necessary truth.

This kind of necessity is what Evans (1979) calls "deep necessity," as opposed to "superficial" necessities like "Water is H2O." It is analyzed in detail by Davies and Humberstone (1980) by means of a modal operator they call "fixedly actually." Deep necessity, unlike superficial necessity, is unaffected by a posteriori considerations. These two varieties of possibility and necessity apply always to statements. There is only one relevant kind of possibility of worlds; the two approaches differ on how the truth of a statement is evaluated in a world.

We can see this in a different way by noting that there are two sets of truth conditions associated with any statement. If we evaluate the terms in a statement according to their primary intensions, we arrive at the primary truth conditions of the statement; that is, a set of centered possible worlds in which the statement, evaluated according to the primary intensions of the terms therein, turns out to be true. The primary truth conditions tell us how the actual world has to be for an utterance of the statement to be true in that world; that is, they specify those contexts in which the statement would turn out to be true. For instance, the primary truth conditions of "Water is wet" specify roughly that such an utterance will be true in the set of worlds in which watery stuff is wet.

If instead we evaluate the terms involved according to their secondary intensions, we arrive at the more familiar secondary truth conditions. These conditions specify the truth-value of a statement in counterfactual worlds, given that the actual world has turned out as it did. For instance, the secondary truth conditions of "Water is wet" (uttered in this world) specifies those worlds in which water is wet: so given that water is H2O, it specifies those worlds in which H2O is wet. Note that there is no danger of an ambiguity in actual-world truth: the primary and secondary truth conditions will always specify the same truth-value when evaluated at the actual world.

If we see a proposition as a function from possible worlds to truth-values, then these two sets of truth conditions yield two propositions associated with any statement. Composing the primary intensions of the terms involved yields a primary proposition, which holds in precisely those contexts of

64 Foundations

utterance in which the statement would turn out to express a truth. (This is the' "diagonal proposition" of Stalnaker 1978. Strictly speaking, it is a centered proposition, or a function from centered worlds to truth-values.) The secondary intensions yield a secondary proposition, which holds in those counter-factual circumstances in which the statement, as uttered in the actual world, is true. The secondary proposition is Kaplan's "content" of an utterance and is more commonly seen as the proposition expressed by a statement, but the primary proposition is also central.

The two kinds of necessary truth of a statement correspond precisely to the necessity of the two kinds of associated proposition. A statement is necessarily true in the first (a priori) sense if the associated primary proposition holds in all centered possible worlds (that is, if the statement would turn out to express a truth in any context of utterance). A statement is necessarily true in the a posteriori sense if the associated secondary proposition holds in all possible worlds (that is, if the statement as uttered in the actual world is true in all counterfactual worlds). The first corresponds to Evans's deep necessity, and the second to the more familiar superficial necessity.

To illustrate, take the statement "Water is H2O." The primary intensions of "water" and "H2O" differ, so that we cannot know a priori that water is H2O; the associated primary proposition is not necessary (it holds in those centered worlds in which the watery stuff has a certain molecular structure). Nevertheless, the secondary intensions coincide, so that "Water is H2O" is true in all possible worlds when evaluated according to the secondary intensions-that is, the associated secondary proposition is necessary. Kripkean a posteriori necessity arises just when the secondary intensions in a statement back a necessary proposition, but the primary intensions do not.

Consider by contrast the statement "Water is watery stuff." Here the associated primary intensions of "water" and "watery stuff" are the same, so that we can know this statement to be true a priori, as long as we possess the concepts. The associated primary proposition is necessary, so that this statement is necessarily true in Evans's "deep" sense. However, the secondary intensions differ, as "water" is rigidified but "watery stuff" is not: in a world where XYZ is the clear, drinkable liquid, the secondary intension of "watery stuff" picks out XYZ but that of "water" does not. The associated secondary proposition is therefore not necessary, and the statement is not a necessary truth in the more familiar sense; it is an example of Kripke's "contingent a priori."

In general, many apparent "problems" that arise from these Kripkean considerations are a consequence of trying to squeeze the doubly indexed picture of reference into a single notion of meaning or of necessity. Such problems can usually be dissolved by explicitly noting the two-dimensional

Supervenience and Explanation 65

character of reference, and by taking care to explicitly distinguish the notion of meaning or of necessity that is in question.28

It is also possible to use this two-dimensional framework to give an account of the semantics of thought, as well as of language. I do this at much greater length elsewhere (Chalmers 1994c). This aspect of the framework will not be central here, but it is worth mentioning, as it will come up in one or two minor places. The basic idea is very similar: given an individual's concept in thought, we can assign a primary intension corresponding to what it will pick out depending on how the actual world turns out, and a secondary intension corresponding to what it picks out in counterfactual worlds, given that the actual world turns out as it has. Given a belief, we can assign a primary proposition and a secondary proposition in a similar way (what I elsewhere call the "notional" and "relational" content of the belief).

For example, concepts such as "Hesperus" and "Phosphorus" will have different primary intensions (one picks out the evening star in a given centered world, the other picks out the morning star), but the same secondary intensions (both pick out Venus in all worlds). The thought "Hesperus is Phosphorus" will have a primary proposition true in all centered worlds in which the evening star is the morning star: the fact that this thought is informative rather than trivial corresponds to the fact that the primary proposition is contingent, as the primary intensions of the two terms differ.

The primary proposition, more than the secondary proposition, captures how things seem from the point of view of the subject: it delivers the set of centered worlds which the subject, in having the belief, is endorsing as potential environments in which he or she might be living (in believing that Hesperus is Phosphorus, I endorse all those centered worlds in which the evening star and the morning star around the center are identical). It is also fairly easy to argue that the primary proposition, rather than the secondary proposition, governs the cognitive and rational relations between thoughts. For this reason it is natural to think of the primary proposition as the cognitive content of a thought.29

Logical necessity, conceptual truth, and conceivability

With this framework in hand, we can spell out the relationships among logical necessity, conceptual truth, and conceivability. Starting with logical necessity: this is just necessity as explicated above. A statement is logically necessary if and only if it is true in all logically possible worlds. Of course we have two varieties of logical necessity of statements, depending on whether we evaluate truth in a possible world according to primary and secondary intensions. We might call these varieties 1-necessity and 2-neces-sity, respectively.

66 Foundations

This analysis explicates the logical necessity and possibility of a statement in terms of (a) the logical possibility of worlds, and (b) the intensions determined by the terms involved in the statement. I have already discussed the intensions. As for the notion of a logically possible world, this is something of a primitive: as before, we can intuitively think of a logically possible world as a world that God might have created (questions about God himself aside). I will not engage the vexed question of the Ontological status of these worlds, but simply take them for granted as a tool, in the same way one takes mathematics for granted.30 As for the extent of the class, the most important feature is that every conceivable world is logically possible, a matter on which I will say more in a moment.

As for conceptual truth, if we equate meaning with intension (primary or secondary), it is easy to make the link between truth in virtue of meaning and logical necessity. If a statement is logically necessary, its truth will be an automatic byproduct of the intensions of the terms (and the compositional structure of the statement). We do not need to bring in the world in any further role, as the intensions in question will be satisfied in every possible world. Similarly, if a statement is true in virtue of its intensions, it will be true in every possible world.

As before, there are two varieties of conceptual truth, depending on whether we equate the "meanings" with primary or secondary intensions, paralleling the two varieties of necessary truth. As long as one makes parallel decisions in the two cases, a statement is conceptually true if and only if it is necessarily true. "Water is watery stuff" is conceptually true and necessarily true in the first sense; and "Water is H2O" is conceptually true and necessarily true in the second. Only the first variety of conceptual truth will in general be accessible a priori. The second variety will include many a posteriori truths, as the secondary intension depends on the way the actual world turns out.

(I do not claim that intensions are the correct way to think of meanings. Meaning is a many-faceted notion, and some of its facets may not be perfectly reflected by intensions, so one could resist the equation of the two at least in some cases.31 Rather, the equation of meaning and intension should here be thought of as stipulative: if one makes the equation, then one can make various useful connections. Not much rests on the use of the word "meaning." In any case, truth in virtue of intension is the only sort of truth in virtue of meaning that I will need.)

We can also make a link between the logical possibility of statements and the conceivability of statements, if we are careful. Let us say that a statement is conceivable (or conceivably true) if it is true in all conceivable worlds. This should not be confused with other senses of "conceivable." For example, there is a sense according to which a statement is conceivable if for all we know it is true, or if we do not know that it is impossible. In this sense, both

Supervenience and Explanation 67

Goldbach's conjecture and its negation are conceivable. But the false member of the pair will not qualify as conceivable in the sense I am using, as there is no conceivable world in which it is true (it is false in every world).

On this view of conceivability, the conceivability of a statement involves two things: first, the conceivability of a relevant world, and second, the truth of the statement in that world.32 It follows that in making conceivability judgments, one has to make sure that one describes the world that one is conceiving correctly, by properly evaluating the truth of a statement in the world. One might at first glance think it is conceivable that Goldbach's conjecture is false, by conceiving of a world where mathematicians announce it to be so; but if in fact Goldbach's conjecture is true, then one is misdescribing this world; it is really a world in which the conjecture is true and some mathematicians make a mistake.

In practice, to make a conceivability judgment, one need only consider a conceivable situation-a small part of a world-and then make sure that one is describing it correctly. If there is a conceivable situation in which a statement is true, there will obviously be a conceivable world in which the statement is true, so this method will give reasonable results while straining our cognitive resources less than conceiving of an entire world!

Sometimes it is said that examples such as "Water is XYZ" show that conceivability does not imply possibility, but I think the situation is subtler than this. In effect, there are two varieties of conceivability, which we might call 1-conceivability and 2-conceivability, depending on whether we evaluate a statement in a conceivable world according to the primary or secondary intensions of the terms involved. "Water is XYZ" is 1-conceivable, as there is a conceivable world in which the statement (evaluated according to primary intensions) is true, but it is not 2-conceivable, as there is no conceivable world in which the statement (evaluated according to secondary intension) is true. These two sorts of conceivability precisely mirror the two sorts of logical possibility mentioned previously.

Often, the conceivability of a statement is equated with 1-conceivability (the sense in which "Water is XYZ" is conceivable), as it is this sort of conceivability that is accessible a priori. And most often, the possibility of a statement is equated with 2-possibility (the sense in which "Water is XYZ" is impossible). Taken this way, conceivability does not imply possibility. But it remains the case that 1-conceivability implies 1-possibility, and 2-conceivability implies 2-possibility. One simply has to be careful not to judge 1-conceivability when 2-possibility is relevant. That is, one has to be careful not to describe the world that one is conceiving (the XYZ world, say) according to primary intensions, when secondary intensions would be more appropriate.33

It follows from all this that the oft-cited distinction between "logical" and "metaphysical" possibility stemming from the Kripkean cases-on which it

68 Foundations

is held to be logically possible but not metaphysically possible that water is XYZ-is not a distinction at the level of worlds, but at most a distinction at the level of statements. A statement is "logically possible" in this sense if it is true in some world when evaluated according to primary intensions; a statement is "metaphysically possible" if it is true in some world when evaluated according to secondary intensions. The relevant space of worlds is the same in both cases.37

Most importantly, none of the cases we have seen give reason to believe that any conceivable worlds are impossible. Any worries about the gap between conceivability and possibility apply at the level of statements, not worlds: either we use a statement to misdescribe a conceived world (as in the Kripkean case, and the second Goldbach case), or we claim that a statement is conceivable without conceiving of a world at all (as in the first Goldbach case). So there seems to be no reason to deny that conceivability of a world implies possibility. I will henceforth take this for granted as a claim about logical possibility; any variety of possibility for which conceivability does not imply possibility will then be a narrower class. Someone might hold that there is a narrower variety of "metaphysically possible worlds," but any reason for believing in such a class would have to be quite independent of the standard reasons I have considered here. In any case, it is logical possibility that is central to the issues about explanation. (A stronger "metaphysical" modality might at best be relevant to issues about ontology, materialism, and the like; I will discuss it when those issues become relevant in Chapter 4.)

An implication in the other direction, from logical possibility to conceivability, is trickier in that limits on our cognitive capacity imply that there are some possible situations that we cannot conceive, perhaps due to their great complexity. However, if we understand conceivability as conceivability-in-principle-perhaps conceivability by a superbeing-then it is plausible that logical possibility of a world implies conceivability of the world, and therefore that logical possibility of a statement implies conceivability of the statement (in the relevant sense). In any case, I will be more concerned with the other implication.

If a statement is logically possible or necessary according to its primary intension, the possibility or necessity is knowable a priori, at least in principle. Modality is not epistemically inaccessible: the possibility of a statement is a function of the intensions involved and the space of possible worlds, both of which are epistemically accessible in principle, and neither of which is dependent on a posteriori facts in this case. So matters of 1-possibility and 1-conceivability are in principle accessible from the armchair. By contrast, matters of 2-possibility and 2-conceivability will in many cases be accessible only a posteriori, as facts about the external world may play a role in determining the secondary intensions.

Supervenience and Explanation 69

The class of 1-necessary truths corresponds directly to the class of a priori truths. If a statement is true a priori, then it is true no matter how the actual world turns out; that is, it is true in all worlds considered as actual, so it is 1-necessary. And conversely, if a statement is 1-necessary, then it will be true no matter how the actual world turns out, so it will be true a priori. In most such cases, the statement's truth will be knowable by us a priori; the exceptions may be certain mathematical statements whose truth we cannot determine, and certain statements that are so complex that we cannot comprehend them. Even in these cases, it seems reasonable to say that they are knowable a priori at least in principle, although they are beyond our limited cognitive capacity. (I will return to this matter when it becomes relevant later.)

Logical necessity and logical Supervenience

We obtain two slightly different notions of logical Supervenience depending on whether we use the primary or secondary brands of logical necessity. If "gloop" has both a primary and a secondary intension associated with it, then gloopiness may supervene logically on physical properties according to either the primary or the secondary intension of "gloop". Supervenience according to secondary intension-that is, Supervenience with a posteriori necessity as the relevant modality-corresponds to what some call "metaphysical Supervenience," but we have now seen how this can be regarded as a variety of logical Supervenience.

(There is really only one kind of logical Supervenience of properties, just as there is only one kind of logical necessity of propositions. But we have seen that terms or concepts effectively determine two properties, one via a primary intension ["watery stuff"] and the other via a secondary intension ["H2O"]. So for a given concept ["water"], there are two ways in which properties associated with that concept might supervene. I will sometimes talk loosely of the primary and secondary intensions associated with a property, and of the two ways in which a property might supervene.)

I will discuss both the primary and secondary versions of logical Supervenience in specific cases, but the former will be more central. Especially when considering questions about explanation, primary intensions are more important than secondary intensions. As noted before, we have only the primary intension to work with at the start of inquiry, and it is this intension that determines whether or not an explanation is satisfactory. To explain water, for example, we have to explain things like its clarity, liquidity, and so on. The secondary intension ("H2O") does not emerge until after an explanation is complete, and therefore does not itself determine a criterion for explanatory success. It is logical Supervenience according to a primary intension that determines whether reductive explanation is possible. Where

70 Foundations

I do not specify otherwise, it is logical Supervenience according to primary intension that I will generally be discussing.

If we choose one sort of intension-say, the primary intension-and stick with it, then we can see that various ways of formulating logical Supervenience are equivalent. According to the definition given at the start of this chapter, B-properties are logically supervenient on A-properties if for any logically possible situation Y that is A-indiscernible from an actual situation X, then all the B-facts true of X are true of Y. Or more simply, B-properties are logically supervenient on A-properties if for any actual situation X, the A-facts about X entail the B-facts about X (where "P entails Q" is understood as "It is logically impossible that P and not Q")-

Sticking to global Supervenience, this means that B-properties supervene logically on A-facts if the B-facts about the actual world are entailed by the A-facts. Similarly, B-properties supervene logically on A-properties if there is no conceivable world with the same A-properties as our world but different B-properties. We can also say that logical Supervenience holds if, given the totality of A-facts A* and any B-fact B about our world W, "A*(W) -> B(W)" is true in virtue of the meanings of the A-terms and the B-terms (where meanings are understood as intensions).

Finally, if B-properties are logically supervenient on A-properties according to primary intensions, then the implication from A-facts to B-facts will be a priori. So in principle, someone who knows all the A-facts about an actual situation will be able to ascertain the B-facts about the situation from those facts alone, given that they possess the B-concepts in question. This sort of inference may be difficult or impossible in practice, due to the complexity of the situations involved, but it is at least possible in principle. For logical Supervenience according to secondary intensions, B-facts about a situation can also be ascertained from the A-facts in principle, but only a posteriori. The A-facts will have to be supplemented with contingent facts about the actual world, as those facts will play a role in determining the B-intensions involved.

There are therefore at least three avenues to establishing claims of logical Supervenience: these involve conceivability, epistemology, and analysis. To establish that B-properties logically supervene on A-properties, we can (1) argue that instantiation of A-properties without instantiation of the B-properties is inconceivable; (2) argue that someone in possession of the A-facts could come to know the B-facts (at least in cases of Supervenience via primary intension); or (3) analyze the intensions of the B-properties in sufficient detail that it becomes clear that B-statements follow from A-statements in virtue of these intensions alone. The same goes for establishing the failure of logical Supervenience. I will use all three methods in arguing for central claims involving logical Supervenience.

Supervenience and Explanation 71

Not everybody may be convinced that the various formulations of logical Supervenience are equivalent, so when arguing for important conclusions involving logical Supervenience I will run versions of the arguments using each of the different formulations. In this way it will be seen that the arguments are robust, with nothing depending on a subtle equivocation between different notions of Supervenience.

5. Almost Everything is Logically Supervenient on the Physical*

In the following chapter I will argue that conscious experience does not supervene logically on the physical, and therefore cannot be reductively explained. A frequent response is that conscious experience is not alone here, and that all sorts of properties fail to supervene logically on the physical. It is suggested that such diverse properties as tablehood, life, and economic prosperity have no logical relationship to facts about atoms, electromagnetic fields, and so on. Surely those high-level facts could not be logically entailed by the microphysical facts?

On a careful analysis, I think that it is not hard to see that this is wrong, and that the high-level facts in question are (globally) logically supervenient on the physical insofar as they are facts at all.35 Conscious experience is almost unique in its failure to supervene logically. The relationship between consciousness and the physical facts is different in kind from the standard relationship between high-level and low-level facts.

There are various ways to make it clear that most properties supervene logically on physical properties. Here I will only be concerned with properties that characterize natural phenomena-that is, contingent aspects of the world that need explaining. The property of being an angel might not supervene logically on the physical, but angels are not something that we have reason to believe in, so this failure need not concern us. I will also not concern myself with facts about abstract entities such as mathematical entities and propositions, which need to be treated separately.36

It should be noted that in claiming that most high-level properties supervene on the physical, I am not suggesting that high-level facts and laws are entailed by microphysical laws, or even by microphysical laws in conjunction with microphysical boundary conditions. That would be a strong claim, and although it might have some plausibility if qualified appropriately, the evidence is not yet in. I am making the much weaker claim that high-level facts are entailed by all the microphysical facts (perhaps along with microphysical laws). This enormously comprehensive set includes the facts

72 Foundations

about the distribution of every last particle and field in every last corner of space-time: from the atoms in Napoleon's hat to the electromagnetic fields in the outer ring of Saturn. Fixing this set of facts leaves very little room for anything else to vary, as we shall see.

Before moving to the arguments I should note some harmless reasons why logical Supervenience on the physical sometimes fails. First, some high-level properties fail to supervene logically because of a dependence on conscious experience. Perhaps conscious experience is partly constitutive of a property like love, for example. The primary (although not the secondary) intensions associated with some external properties such as color and heat may also be dependent on phenomenal qualities, as we will see. If so, then love and perhaps heat do not supervene logically on the physical. These should not be seen as providing counterexamples to my thesis, as they introduce no new failure of logical Supervenience. Perhaps the best way to phrase the claim is to say that all facts supervene logically on the combination of physical facts and phenomenal facts, or that all facts supervene logically on the physical facts modulo conscious experience. Similarly, a dependence on conscious experience may hinder the reductive explainability of some high-level phenomena, but we can still say that they are reductively explainable modulo conscious experience.

Second, an indexical element enters into the application of some primary intensions, although not secondary intensions, as we saw earlier. The primary intension of "water," for example, is something like "the clear, drinkable liquid in our environment," so that if there is watery H2O and watery XYZ in the actual universe, which of them qualifies as "water" depends on which is in the environment of the agent using the term. In principle we therefore need to add a center representing the location of an agent to the Supervenience base in some cases. This yields logical Supervenience and reductive explanation modulo conscious experience and indexicality.

Finally, cases where the high-level facts are indeterminate do not count against logical Supervenience. The claim is only that insofar as the high-level facts are determinate, they are determined by the physical facts. If the world itself does not suffice to fix the high-level facts, we cannot expect the physical facts to. Some might suggest that logical Supervenience would fail if there were two equally good high-level theories of the world that differed in their description of the high-level facts. One theory might hold that a virus is alive, for instance, whereas another might hold that it is not, so the facts about life are not determined by the physical facts. This is not a counterexample, however, but a case in which the facts about life are indeterminate. Given indeterminacy, we are free to legislate the terms one way or the other where it is convenient. If the facts are determinate-for example, if it is true that viruses are alive-then one of the descriptions is simply wrong. Either way,

Supervenience and Explanation 73

insofar as the facts about the situation are determinate at all, they are entailed by the physical facts.

I will argue for the ubiquity of logical Supervenience using arguments that appeal to conceivability, to epistemological considerations, and to analysis of the concepts involved.



Conceivability. The logical Supervenience of most high-level facts is most easily seen by using conceivability as a test for logical possibility. What kind of world could be identical to ours in every last microphysical fact but be biologically distinct? Say a wombat has had two children in our world. The physical facts about our world will include facts about the distribution of every particle in the spatiotemporal hunk corresponding to the wombat, and its children, and their environments, and their evolutionary histories. If a world shared those physical facts with ours, but was not a world in which the wombat had two children, what could that difference consist in? Such a world seems quite inconceivable. Once a possible world is fixed to have all those physical facts the same, then the facts about wombathood and parenthood are automatically fixed. These biological facts are not the sort of thing that can float free of their physical underpinnings even as a conceptual possibility.

The same goes for architectural facts, astronomical facts, behavioral facts, chemical facts, economic facts, meteorological facts, sociological facts, and so on. A world physically identical to ours, but in which these sort of facts differ, is inconceivable. In conceiving of a microphysically identical world, we conceive of a world in which the location of every last particle throughout space and time is the same. It follows that the world will have the same macroscopic structure as ours, and the same macroscopic dynamics. Once all this is fixed there is simply no room for the facts in question to vary (apart, perhaps, from any variation due to variations in conscious experience).

Furthermore, this inconceivability does not seem to be due to any contingent limits in our cognitive capacity. Such a world is inconceivable in principle. Even a superbeing, or God, could not imagine such a world. There is simply not anything for them to imagine. Once they imagine a world with all the physical facts, they have automatically imagined a world in which all the high-level facts hold. A physically identical world in which the high-level facts are false is therefore logically impossible, and the high-level properties in question are logically supervenient on the physical.



Epistemology. Moving beyond conceivability intuitions, we can note that if there were a possible world physically identical to ours but biologically distinct, then this would raise radical epistemological problems. How would

74 Foundations

we know that we were not in that world rather than this one? How would we know that the biological facts in our world are as they are? To see this, note that if I were in the alternative world, it would certainly look the same as this one. It instantiates the same distribution of particles found in the plants and animals in this world; indistinguishable patterns of photons are reflected from those entities; no difference would be revealed under even the closest examination. It follows that all the external evidence we possess fails to distinguish the possibilities. Insofar as the biological facts about our world are not logically supervenient, there is no way we can know those facts on the basis of external evidence.

In actuality, however, there is no deep epistemological problem about biology. We come to know biological facts about our world on the basis of external evidence all the time, and there is no special skeptical problem that arises. It follows that the biological facts are logically supervenient on the physical. The same goes for facts about architecture, economics, and meteorology. There is no special skeptical problem about knowing these facts on the basis of external evidence, so they must be logically supervenient on the physical.

We can back up this point by noting that in areas where there are epistemological problems, there is an accompanying failure of logical Supervenience, and that conversely, in areas where logical Supervenience fails, there are accompanying epistemological problems.

Most obviously, there is an epistemological problem about consciousness-the problem of other minds. This problem arises because it seems logically compatible with all the external evidence that beings around us are conscious, and it is logically compatible that they are not. We have no way to peek inside a dog's brain, for instance, and observe the presence or absence of conscious experience. The status of this problem is controversial, but the mere prima facie existence of the problem is sufficient to defeat an epistemological argument, parallel to those above, for the logical Supervenience of consciousness. By contrast, there is not even a prima facie problem of other biologies, or other economies. Those facts are straightforwardly publically accessible, precisely because they are fixed by the physical facts.

(Question: Why doesn't a similar argument force us to the conclusion that if conscious experience fails to supervene logically, then we can't know about even our own consciousness? Answer: Because conscious experience is at the very center of our epistemic universe. The skeptical problems about nonsupervenient biological facts arise because we only have access to biological facts by external, physically mediated evidence; external nonsupervenient facts would be out of our direct epistemic reach. There is no such problem with our own consciousness.)

Another famous epistemological problem concerns facts about causation. As Hume argued, external evidence only gives us access to regularities of

Supervenience and Explanation 75

succession between events; it does not give us access to any further fact of causation. So if causation is construed as something over and above the presence of a regularity (as I will assume it must be), it is not clear that we can know that it exists. Once again, this skeptical problem goes hand in hand with a failure of logical Supervenience. In this case, facts about causation fail to supervene logically on matters of particular physical fact. Given all the facts about distribution of physical entities in space-time, it is logically possible that all the regularities therein arose as a giant cosmic coincidence without any real causation. At a smaller scale, given the particular facts about any apparent instance of causation, it is logically possible that it is a mere succession. We infer the existence of causation by a kind of inference to the best explanation-to believe otherwise would be to believe in vast, inexplicable coincidences-but belief in causation is not forced on us in the direct way that belief in biology is forced on us.

I have sidestepped problems about the Supervenience of causation by stipulating that the Supervenience base for our purposes includes not just particular physical facts but all the physical laws. It is reasonable to suppose that the addition of laws fixes the facts about causation. But of course there is a skeptical problem about laws paralleling the problem about causation: witness Hume's problem of induction, and the logical possibility that any apparent law might be an accidental regularity.

As far as I can tell, these two problems exhaust the epistemological problems that arise from failure of logical Supervenience on the physical. There are some other epistemological problems that in a sense precede these, because they concern the existence of the physical facts themselves. First, there is Descartes's problem about the existence of the external world. It is compatible with our experiential evidence that the world we think we are seeing does not exist; perhaps we are hallucinating, or we are brains in vats. This problem can be seen to arise precisely because the facts about the external world do not supervene logically on the facts about our experience. (Idealists, positivists, and others have argued controversially that they do. Note that if these views are accepted the skeptical problem falls away.) There is also an epistemological problem about the theoretical entities postulated by science-electrons, quarks, and such. Their absence would be logically compatible with the directly observable facts about objects in our environment, and some have therefore raised skeptical doubts about them. This problem can be analyzed as arising from the failure of theoretical facts to supervene logically on observational facts. In both these cases, skeptical doubts are perhaps best quelled by a form of inference to the best explanation, just as in the case of causation, but the in-principle possibility that we are wrong remains.

In any case, I am bypassing this sort of skeptical problem by giving myself the physical world for free, and fixing all physical facts about the world in

76 Foundations

the Supervenience base (thereby assuming that the external world exists, and that there are electrons, and so on). Given that those facts are known, there is no room for skeptical doubts about most high-level facts, precisely because they are logically supervenient. To put the matter the other way around: All our sources of external evidence supervene logically on the microphysical facts, so that insofar as some phenomenon does not supervene on those facts, external evidence can give us no reason to believe in it. One might wonder whether some further phenomena might be posited via inference to the best explanation, as above, to explain the microphysical facts. Indeed, this process takes us from particular facts to simple underlying laws (and hence yields causation), but then the process seems to stop. It is in the nature of fundamental laws that they are the end of the explanatory chain (except, perhaps, for theological speculation). This leaves phenomena that we have internal evidence for-namely conscious experience-and that is all. Modulo conscious experience, all phenomena are logically supervenient on the physical.

We can also make an epistemological case for logical Supervenience more directly, by arguing that someone in possession of all the physical facts could in principle come to know all the high-level facts, given that they possess the high-level concepts involved. True, one could never in practice ascertain the high-level facts from the set of microphysical facts. The vastness of the latter set is enough to rule that out. (Even less am I suggesting that one could perform a formal derivation; formal systems are irrelevant for reasons canvased earlier.) But as an in-principle point, there are various ways to see that someone (a superbeing?) armed with only the microphysical facts and the concepts involved could infer the high-level facts.

The simplest way is to note that in principle one could build a big mental simulation of the world and watch it in one's mind's eye, so to speak. Say that a man is carrying an umbrella. From the associated microphysical facts, one could straightforwardly infer facts about the distribution and chemical composition of mass in the man's vicinity, giving a high-level structural characterization of the area. One could determine the existence of a male fleshy biped straightforwardly enough. For instance, from the structural information one could note that there was an organism atop two longish legs that were responsible for its locomotion, that the creature has male anatomy, and so on. It would be clear that he was carrying some device that was preventing drops of water, otherwise prevalent in the neighborhood, from hitting him. Doubts that this device is really an umbrella could be assuaged by noting from its physical structure that it can fold and unfold; from its history that it was hanging on a stand that morning, and was originally made in a factory with others of a similar kind, and so on. Doubts that the fleshy biped is really a human could be assuaged by noting the composition of his DNA, his evolutionary history, his relation to other beings, and so on. We

Supervenience and Explanation 77

need only assume that the being possesses enough of the concept involved to be able to apply it correctly to instances (that is, the being possesses the intension). If so, then the microphysical facts will give it all the evidence it needs to apply the concepts, and to determine that there really is a person carrying an umbrella here.

The same goes for almost any sort of high-level phenomena: tables, life, economic prosperity. By knowing all the low-level facts, a being in principle can infer all the facts necessary to determine whether or not this is an instance of the property involved. Effectively, what is happening is that a possible world compatible with the microphysical facts is constructed, and the high-level facts are simply read off that world using the appropriate intension (as the relevant facts are invariant across physically identical possible worlds). Hence the high-level facts are logically supervenient on the physical.



Analyzability. So far, I have argued that microphysical facts fix high-level facts without saying much explicitly about the high-level concepts involved. In any specific case, however, this entailment relationship relies on a concept's intension. If microphysical facts entail a high-level fact, this is because the microphysical facts suffice to fix those features of the world in virtue of which the high-level intension applies. That is, we should be able to analyze what it takes for an entity to satisfy the intension of a high-level concept, at least to a sufficient extent that we can see why those conditions for satisfaction could be satisfied by fixing the physical facts. It is therefore useful to look more closely at the intensions of high-level concepts, and to examine the features of the world in virtue of which they apply.

There are some obstacles to elucidating these intensions and to summarizing them in words. As we saw earlier, application conditions of a concept are often indeterminate in places. Is a cup-shaped object made of tissues a cup? Is a computer virus alive? Is a booklike entity that coagulates randomly into existence a book? Our ordinary concepts do not give straightforward answers to these questions. In a sense, it is a matter for stipulation. Hence there will not be determinate application conditions for use in the entailment process. But as we saw earlier, this indeterminacy precisely mirrors an indeterminacy about the facts themselves. Insofar as the intension of "cup" is a matter for stipulation, the facts about cups are also a matter for stipulation. What counts for our purposes is that the intension together with the microphysical facts determines the high-level facts insofar as they are really factual. Vagueness and indeterminacy can make discussion awkward, but they affect nothing important to the issues.


A related problem is that any short analysis of a concept will invariably fail to do justice to it. As we have seen, concepts do not usually have crisp definitions. At a first approximation, we can say something is a table if it

78 Foundations

has a flat horizontal surface with legs as support; but this lets in too many things (Frankenstein's monster on stilts?) and omits others (a table with no legs, sticking out from a wall?). One can refine the definition, adding further conditions and clauses, but we quickly hit the problems with indeterminacy, and in any case the product will never be perfect. But there is no need to go into all the details required to handle every special case: after a point the details are just more of the same. As long as we know what sort of properties the intension applies in virtue of, we will have enough to make the point.

As we saw before, we do not need a definition of B-properties in terms of A-properties in order for A-facts to entail B-facts. Meanings are fundamentally represented by intensions, not definitions. The role of analysis here is simply to characterize the intensions in sufficient detail that the existence of an entailment becomes clear. For this purpose, a rough-and-ready analysis will suffice. Intensions generally apply to individuals in a possible world in virtue of some of their properties and not others; the point of such an analysis is to see what sort of properties the intension applies in virtue of, and to make the case that properties of this sort are compatible with entailment by physical properties.

A third problem stems from the division between the a priori and a posteriori application conditions of many concepts. As long as we keep primary and secondary intensions separate, however, this is not much of a problem. The secondary intension associated with "water" is something like "H2O," which is obviously logically supervenient on the physical. But the primary intension, something like "the clear, drinkable liquid in our environment" is equally logically supervenient, as the clarity, drinkability, and liquidity of water is entailed by the physical facts.37 We can run things either way. As we have seen, it is the primary intension that enters into reductive explanation, so it is this that we are most concerned with. In general, if a primary intension I is logically supervenient on the physical, then so is a rigidified secondary intension dthat(I), as it will generally consist in a projection of some intrinsic physical structure across worlds.

Considerations about a posteriori necessity have led some to suppose that there can be no logical entailment from low-level facts to high-level facts. Typically one hears something like "Water is necessarily H2O, but that is not a truth of meaning, so there is no conceptual relation." But this is a vast oversimplification. For a start, the secondary intension "H2O" can be seen as part of the meaning of "water" in some sense, and it certainly supervenes logically. But more importantly, the primary intension ("the clear, drinkable liquid ...") which fixes reference also supervenes, perhaps modulo experience and indexicality. It is precisely in virtue of its satisfying this intension that we deemed that H2O was water in the first place. Given the primary intension I, the high-level facts are derivable unproblematically

Supervenience and Explanation 79

from the microphysical facts (modulo the contribution of experience and indexicality). The Kripkean observation that the concept is better represented as dthat(I) affects this derivability not at all. The semantic phenomenon of rigidification does not alone make an Ontological difference.

With these obstacles out of the way, we can look at the intensions associated with various high-level concepts. In most cases these are characterizable in functional or structural terms, or as a combination of the two. For example, the sorts of things relevant to something's being a table include (1) that it have a flat top and be supported by legs, and (2) that people use it to support various objects. The first of these is a structural condition: that is, a condition on the intrinsic physical structure of the object. The second is a functional condition: that is, it concerns the external causal role of an entity, characterizing the way it interacts with other entities. Structural properties are clearly entailed by microphysical facts. So are functional properties in general, although this is slightly less straightforward. Such properties depend on a much wider Supervenience base of microphysical facts, so that facts about an object's environment will often be relevant; and insofar as such properties are characterized dispositionally (something is soluble if it would dissolve if immersed in water), one needs to appeal to counterfactuals. But the truth-values of those counterfactuals are fixed by the inclusion of physical laws in the antecedent of our Supervenience conditionals, so this is not a problem.

To take another example, the conditions on life roughly come down to some combination of the ability to reproduce, to adapt, and to metabolize, among other things (as usual, we need not legislate the weights, or all other relevant factors). These properties are all characterizable functionally, in terms of an entity's relation to other entities, its ability to convert external resources to energy, and its ability to react appropriately to its environment. These functional properties are all derivable, in principle, from the physical facts. As usual, even if there is no perfect definition of life in functional terms, this sort of characterization shows us that life is a functional property, whose instantiation can therefore be entailed by physical facts.

A complication is raised by the fact that functional properties are often characterized in terms of a causal role relative to other high-level entities. It follows that logical Supervenience of the properties depends on the logical Supervenience of the other high-level notions involved, where these notions may themselves be characterized functionally. This is ultimately not a problem, as long as causal roles are eventually cashed out by nonfunctional properties: typically either by structural or phenomenal properties. There may be some circularity in the interdefinability of various functional properties-perhaps it is partly constitutive of a stapler that it deliver staples, and partly constitutive of staples that they are delivered by staplers. This circularity can be handled by cashing out the causal roles of all the properties simultaneously,38 as long as the analyses have a noncircular part that is

80 Foundations

ultimately grounded in structural or phenomenal properties. (The appeal to phenomenal properties may seem to count against logical Supervenience on the physical, but see below. In any case, it is compatible with logical Supervenience modulo conscious experience.)

Many properties are characterized relationally, in terms of relations to an entity's environment. Usually such relations are causal, so that the properties in question are functional, but this is not always so: witness the property of being on the same continent as a duck. Similarly, some properties are dependent on history (although these can usually be construed causally); to be a kangaroo, a creature must have appropriate ancestors. In any case these properties pose no problems for logical Supervenience, as the relevant historical and environmental facts will themselves be fixed by the global physical facts.

Even a complex social fact such as "There was economic prosperity in the 1950s"39 is characterizable in mostly functional terms, and so can be seen to be entailed by the physical facts. A full analysis would be very complicated and would be made difficult by the vagueness of the notion of prosperity, but to get an idea how it might go, one can ask why we say that there was economic prosperity in the 1950s? At a first approximation, because there was high employment, people were able to purchase unusually large amounts of goods, there was low inflation, much development in housing, and so on. We can in turn give rough-and-ready analyses of the notion of housing (the kind of place people sleep and eat in), of employment (organized labor for reward), and of monetary notions (presumably money will be roughly analyzable in terms of the systematic ability to exchange for other items, and its value will be analyzable in terms of how much one gets in exchange). All these analyses are ridiculously oversimplified, but the point is clear enough. These are generally functional properties that can be entailed by physical facts.

Many have been skeptical of the possibility of conceptual analysis. Often this has been for reasons that do not make any difference to the arguments I am making-because of indeterminacy in our concepts, for example, or because they lack crisp definitions. Sometimes this skepticism may have arisen for deeper reasons. Nevertheless, if what I have said earlier in this chapter is correct, and if the physical facts about a possible world fix the high-level facts, we should expect to be able to analyze the intension of the high-level concept in question, at least to a good approximation, in order to see how its application can be determined by physical facts. This is what I have tried to do in the examples given here. Other examples can be treated similarly.40

I am not advocating a program of performing such analyses in general. Concepts are too complex and unruly for this to do much good, and any explicit analysis is usually a pale shadow of the real thing. What counts is

Supervenience and Explanation 81

the general point that most high-level concepts are not primitive, unanalyzable notions. They are generally analyzable to the extent that their intensions can be seen to specify functional or structural properties. It is in virtue of this analyzability that high-level facts are in principle derivable from microphysical facts and reductively explainable in terms of physical facts.

Some problem cases

There are some types of properties that might be thought to provide particular difficulties for logical Supervenience, and therefore for reductive explanation. I will examine a number of such candidates, paying particular attention to the question of whether the associated phenomena pose problems for reductive explanation analogous to the problems posed by consciousness. It seems to me that with a couple of possible exceptions, no significant new problems arise here.


Consciousness-dependent properties.

Consciousness-dependent properties. As discussed already, some concepts' primary intensions involve a relation to conscious experience. An obvious example is redness, taken as a property of external objects. On at least some accounts, the primary intension associated with redness requires that for something to be red, it must be the kind of thing that tends to cause red experiences under appropriate conditions.41 So in its primary intension, redness is not logically supervenient on the physical, although it supervenes modulo conscious experience. On the other hand, its secondary intension almost certainly supervenes. If it turns out that in the actual world, the sort of thing that tends to cause red experience is a certain surface reflectance, then objects with that reflectance are red even in worlds in which there is no conscious being to see them. Redness is identified a posteriori with that reflectance, which is logically supervenient on the physical alone.

We saw earlier that failure of a primary intension to supervene logically is associated with a failure of reductive explanation. So, does reductive explanation fail for redness? The answer is yes, in a weak sense. If redness is construed as the tendency to cause red experiences, then insofar as experience is not reductively explainable, neither is redness. But one can come close. One can note that a certain physical quality causes red experiences; and one can even explain the causal relation between the quality and red-judgments. It is just the final step to experience that goes unexplained. In practice, our strictures on explanation are weak enough that this sort of thing counts. To explain a phenomenon to which reference is fixed by some experience, we do not require an explanation of experience. Otherwise we would wait a long time.

The same goes for phenomena such as heat, light, and sound. Although their secondary intensions determine structural properties (molecular mo-

82 Foundations

tion, the presence of photons, waves in air), their primary intensions involve a relation to conscious experience: heat is the thing that causes heat sensations, light causes visual experiences, and so on. But as Nagel (1974) and Searle (1992) have noted, we do not require an explanation of heat sensations when explaining heat. Explanation modulo experience is good enough.

Other properties depend even more directly on conscious experience, in that experience not only plays a role in reference fixation but is partly constitutive of the a posteriori notion as well. The property of standing next to a conscious person is an obvious example. On some accounts, mental properties such as love and belief, although not themselves phenomenal properties, have a conceptual dependence on the existence of conscious experience. If so, then in a world without consciousness, such properties would not be exemplified. Such properties therefore are not logically supervenient even a posteriori, and reductive explanation fails even more strongly than in the above cases. But they are logically supervenient and reductively explainable modulo conscious experience, so no further failure of reductive explanation arises here.



Intentionality. It is worth separately considering the status of intentionality, as this is sometimes thought to pose problems analogous to those posed by consciousness. It is plausible, however, that any failure of intentional properties to supervene logically is derivative on the nonsupervenience of consciousness. As I noted in Chapter 1, there seems to be no conceivable world that is physically and phenomenally identical to ours, but in which intentional contents differ.42 If phenomenology is partly constitutive of intentional content, as some philosophers suggest, then intentional properties may fail to supervene logically on the physical, but they will supervene modulo conscious experience. The claim that consciousness is partly constitutive of content is controversial, but in any case there is little reason to believe that intentionality fails to supervene in a separate, nonderivative way.

Leaving any phenomenological aspects aside, intentional properties are best seen as a kind of third-person construct in the explanation of human behavior, and should therefore be analyzable in terms of causal connections to behavior and the environment. If so, then intentional properties are straightforwardly logically supervenient on the physical. Lewis (1974) makes a thorough attempt at explicating the entailment from physical facts to intentional facts by giving an appropriate functional analysis. More recent accounts of intentionality, such as those by Dennett (1987), Dretske (1981), and Fodor (1987) can be seen as contributing to the same project. None of these analyses are entirely compelling, but it may be that a more sophisticated descendant might do the job. There is no argument analogous to the arguments against the Supervenience of consciousness showing that intentionality

Supervenience and Explanation 83

cannot supervene logically on physical and phenomenal properties.43 Indeed, conceivability arguments indicate that intentional properties must be logically supervenient on these if such properties are instantiated at all, and epistemological arguments lead us to a similar conclusion. So there is no separate Ontological problem of intentionality.


Moral and aesthetic properties.

Moral and aesthetic properties. It is often held that there is no conceptual connection from physical properties to moral and aesthetic properties. According to Moore (1922), nothing about the meaning of notions such as "goodness" allows that facts about goodness should be entailed by physical facts. In fact, Moore claimed that there is no conceptual connection from natural facts to moral facts, where the natural may include the mental as well as the physical (so Supervenience modulo conscious experience does not help here). Does this mean that moral properties are as problematic as conscious experience?

There are two disanalogies, however. First, there does not seem to be a conceivable world that is naturally identical to ours but morally distinct, so it is unlikely that moral facts are further facts in any strong sense. Second, moral facts are not phenomena that force themselves on us. When it comes to the crunch, we can deny that moral facts exist at all. Indeed, this reflects the strategy taken by moral antirealists such as Blackburn (1971) and Hare (1984). These antirealists argue that because moral facts are not entailed by natural facts and are not plausibly "queer" further facts, they have no objective existence and morality should be relativized into a construct or projection of our cognitive apparatus. The same strategy cannot be taken for phenomenal properties, whose existence is forced upon us.

For moral properties, there are at least two reasonable alternatives available. The first is antirealism of some sort, perhaps relativizing "objective moral facts" into "subjective moral facts,"44 or embracing a view on which moral discourse does not state facts at all. The second is to claim that there is an a priori connection from natural facts to moral facts, one that (contra Moore) can be seen to hold in virtue of an analysis and explication of moral concepts. If a concept such as "good" determines a stable nonindexical primary intension, then the second position follows: we will have an a priori function from naturally specified worlds to moral facts. If it only determines an indexical primary intension, or if different subjects can equally reasonably associate different primary intensions with the concept, or if it determines no primary intension at all, then a version of the first position will follow.

Some other positions are sometimes taken, but none seem tenable. Moore held that there is a nonconceptual a priori connection between natural and moral facts that we obtain through a mysterious faculty of "moral intuition," but this view is widely rejected (it is hard to see what could ground such intuitions' truth or falsity). A position on which moral properties supervene

84 Foundations

by a fundamental nomic link seems out of the question, as there is no conceivable world in which the natural facts are the same as ours but in which the moral facts are different. A popular position among contemporary moral realists (see, e.g., Boyd 1988; Brink 1989) is that moral facts supervene on natural facts with a posteriori necessity; that is, they supervene according to the secondary but not the primary intensions of moral concepts. This position is difficult to maintain, however, given that even a posteriori equivalences must be grounded in a priori reference fixation. Even though it is a posteriori that water is H2O, the facts about water follow from the microphysical facts a priori. Similarly, if moral concepts have a primary intension and if naturally identical centered worlds are morally identical, an a priori link from natural facts to moral facts would seem to follow. (Horgan and Timmons [1992a; 1992b] provide a critique along these lines.)

Aesthetic properties can be treated in a similar way. If anything, an anti-realist treatment is even more plausible here. In the final analysis, although there are interesting conceptual questions about how the moral and aesthetic domains should be treated, they do not pose metaphysical and explanatory problems comparable to those posed by conscious experience.



Names. On many accounts (e.g., Kaplan 1989), there is no analysis associated with a name such as "Rolf Harris," which simply picks out its referent directly. Does this mean that the property of being Rolf Harris fails to supervene logically on the physical? There is no problem about the Supervenience of the secondary intension (e.g., Rolf might be the person conceived from a given sperm and egg in all possible worlds), but the absence of a primary intension might be thought to pose problems for reductive explanation. Still, it is plausible that even though there is no primary intension that is shared across the community, every individual use of the name has a primary intension attached. When I use the name "Rolf Harris," there is some systematic way in which its referent depends on the way the world turns out; for me, the primary intension might be something like "the man called 'Rolf Harris' who bangs around on paint cans, and who bears the appropriate causal relation to me."45 Such an intension will supervene logically. Rather than justifying this in detail, however, it is easier to note that any failure of logical Supervenience will not be accompanied by an explanatory mystery. The property of being Rolf Harris does not constitute a phenomenon in need of explanation, as opposed to explication. What needs explaining is the existence of a person named "Rolf Harris" who bangs around on paint cans, and so on. These properties certainly supervene, and are explainable in principle in the usual way.



Indexicals. Reference fixation of many concepts, from "water" to "my dog," involves an indexical element. The reference of these notions is fixed

Supervenience and Explanation 85

on the basis of both physical facts and an agent-relative "indexical fact" representing the location of an agent using the term in question. Such a fact is determinate for any given agent, so reference fixation is determinate. Supervenience and explanation succeed modulo that indexical fact.

Does indexicality pose a problem for reductive explanation? For arbitrary speakers, perhaps not, as the "fact" in question can be relativized away. But for myself, it is not so easy. The indexical fact expresses something very salient about the world as I find it: that David Chalmers is me. How could one explain this seemingly brute fact? Indeed, is there really a fact here to be explained, as opposed to a tautology? The issue is extraordinarily difficult to get a grip on, but it seems to me that even if the indexical is not an objective fact about the world, it is a fact about the world as I find it, and it is the world as I find it that needs explanation. The nature of the brute indexical is quite obscure, though, and it is most unclear how one might explain it.46 (Of course, we can give a reductive explanation of why David Chalmers's utterance of "I am David Chalmers" is true. But this nonindexical fact seems quite different from the indexical fact that I am David Chalmers.)

It is tempting to look to consciousness. But while an explanation of consciousness might yield an explanation of "points of view" in general, it is hard to see how it could explain why a seemingly arbitrary one of those points of view is mine, unless solipsism is true. The indexical fact may have to be taken as primitive. If so, then we have a failure of reductive explanation distinct from and analogous to the failure with consciousness. Still, the failure is less worrying than that with consciousness, as the unexplained fact is so "thin" by comparison to the facts about consciousness in all its glory. Admitting this primitive indexical fact would require far less revision of our materialist worldview than would admitting irreducible facts about conscious experience.


Negative facts.

Negative facts. As we saw earlier, certain facts involving negative existentials and universal quantifiers are not logically determined by the physical facts, or indeed by any set of localized facts. Consider the following facts about our world: there are no angels; Don Bradman is the greatest cricketer; everything alive is based on DNA. All these could be falsified, consistently with all the physical facts about our world, simply by the addition of some new nonphysical stuff: cricket-playing angels made of ectoplasm, for instance. Even addition of facts about conscious experience or indexicality cannot help here.47

Does this mean that these facts are not reductively explainable? It seems so, insofar as there is no physical explanation of why there is no extra nonphysical stuff in our world. That is indeed a further fact. The best way to deal with this situation is to introduce a second-order fact that says of

86 Foundations

the set of basic particular facts, be they microphysical, phenomenal, indexical, or whatever: That's all. This fact says that all the particular facts about the world are included in or entailed by the given set of facts. From this second-order fact, in conjunction with all the basic particular facts, all the negative facts will follow.

This does not constitute a very serious failure of reductive explanation. Presumably there will be such a "That's all" fact true of any world, and such a fact will never be entailed by the particular facts. It simply expresses the bounded nature of our world, or of any world. It is a cheap way to bring all the negative existential and universally quantified facts within our grasp.


Physical laws and causation.

Physical laws and causation. On the most plausible accounts of physical laws, these are not logically supervenient on the physical facts, taken as a collection of particular facts about a world's spatiotemporal history. One can see this by noting the logical possibility of a world physically indiscernible from ours over its entire spatiotemporal history, but with different laws. For example, it might be a law of that world that whenever two hundred tons of pure gold are assembled in a vacuum, it will transmute into lead. Otherwise its laws are identical, with minor modifications where necessary. As it happens, in the spatiotemporal history of our world, two hundred tons of gold are never assembled in a vacuum. It follows that our world and the other world have identical histories, but their laws differ nevertheless.

Arguments like this suggest that the laws of nature do not supervene logically on the collection of particular physical facts.48 By similar arguments one can see that a causal connection between two events is something over and above a regularity between the events. Holders of various Humean views dispute these conclusions, but it seems to me that they have the worse of the arguments here.49 There is something irreducible in the existence of laws and causation.

I have bypassed these problems elsewhere by including physical laws in the Supervenience base, but this steps over the metaphysical puzzle rather than answering it. It is true that laws and causation lead to less significant failure of reductive explanation than consciousness. The laws and causal relations are themselves posited to explain existing physical phenomena, namely the manifold regularities present in nature, whereas consciousness is a brute explanandum. Nevertheless the very existence of such irreducible further facts raises deep questions about their metaphysical nature. Apart from conscious experience and perhaps indexicality, these constitute the only such further facts in which we have any reason to believe. It is not unnatural to speculate that these two nonsupervenient kinds, consciousness and causation, may have a close metaphysical relation.

Supervenience and Explanation 87


The position we are left with is that almost all facts supervene logically on the physical facts (including physical laws), with possible exceptions for conscious experience, indexicality, and negative existential facts. To put the matter differently, we can say that the facts about the world are exhausted by (1) particular physical facts, (2) facts about conscious experience, (3) laws of nature, (4) a second-order "That's all" fact, and perhaps (5) an indexical fact about my location. (The last two are minor compared to the others, and the status of the last is dubious, but I include them for completeness.) Modulo conscious experience and indexicality, it seems that all positive facts are logically supervenient on the physical. To establish this conclusively would require a more detailed examination of all kinds of phenomena, but what we have seen suggests that the conclusion is reasonable. We can sum up the Ontological and epistemological situations with a couple of fables. Perhaps there is a grain of truth in the shape of these stories, if not in the details.

Creation myth. Creating the world, all God had to do was fix the facts just mentioned. For maximum economy of effort, he first fixed the laws of nature-the laws of physics, and any laws relating physics to conscious experience. Next, he fixed the boundary conditions: perhaps a time-slice of physical facts, and maybe the values in a random-number generator. These combined with the laws to fix the remaining physical and phenomenal facts. Last, he decreed, "That's all."


Epistemological myth.

Epistemological myth. At first, I have only facts about my conscious experience. From here, I infer facts about middle-sized objects in the world, and eventually microphysical facts. From regularities in these facts, I infer physical laws, and therefore further physical facts. From regularities between my conscious experience and physical facts, I infer Psychophysical laws, and therefore facts about conscious experience in others. I seem to have taken the abductive process as far as it can go, so I hypothesize: that's all. The world is much larger than it once seemed, so I single out the original conscious experiences as mine.

Note the very different order involved from the two perspectives. One could almost say that epistemology recapitulates ontology backward. Note also that it seems beyond God's powers to fix my indexical fact. Perhaps this is another reason to be skeptical about it.


The logical Supervenience of most high-level phenomena is a conclusion that has not been as widely accepted as it might have been, even among those who discuss Supervenience. Although the matter is often not discussed, many have been wary about invoking the conceptual modality as relevant to super-

88 Foundations

venience relations. As far as I can tell there have been a number of separate reasons for this hesitation, none of which are ultimately compelling.

First, the problem with logically possible physically identical worlds with extra nonphysical stuff (angels, ectoplasm) has led some to suppose that Supervenience relations cannot be logical (Haugeland 1982; Petrie 1987); but we have seen how to fix this problem. Second, many have supposed that considerations about a posteriori necessity demonstrate that Supervenience relations cannot be underwritten by meanings (Brink 1989; teller 1984); but we have seen that Supervenience relations based on a posteriori necessity can be seen as a variety of logical Supervenience. Third, there is a general skepticism about the notion of conceptual truth, deriving from Quine; but we have seen that this is a red herring here. Fourth, worries about "reducibility" have led some to suppose that Supervenience is not generally a conceptual relation (Hellman and Thompson 1975); but it is unclear that there are any good arguments against reducibility that are also good arguments against logical Supervenience. Fifth, the very phenomenon of conscious experience is sometimes invoked to demonstrate that Supervenience relations cannot be logical in general (Seager 1988); but we have seen that conscious experience is almost unique in its failure to supervene logically. Finally, the claim that Supervenience relations are not generally logical is often stated without argument, presumably as something that any reasonable person must believe (Bacon 1986; Heil 1992).50

It is plausible that every Supervenience relation of a high-level property upon the physical is ultimately either (1) a logical Supervenience relation of either the primary or secondary variety, or (2) a contingent natural Supervenience relation. If neither of these holds for some apparent Supervenience relation, then we have good reason to believe that there are no objective high-level facts of the kind in question (as, perhaps, for moral facts). I will argue further in Chapter 4 that there is no deep variety of Supervenience intermediate between the logical and the natural.

This provides a unified explanatory picture, in principle. Almost every phenomenon is reductively explainable, in the weak sense outlined earlier, except for conscious experience and perhaps indexicality, along with the rock-bottom microphysical facts and laws, which have to be taken as fundamental.

It is worth taking a moment to answer a query posed by Blackburn (1985) and Horgan (1993): How do we explain the Supervenience relations themselves? For a logical Supervenience relation based on the primary intension of a concept, this is a simple matter of giving an appropriate analysis of the concept, perhaps in functional or structural terms, and noting that its reference is invariant across physically identical worlds. Here, the Supervenience conditional is itself an a priori conceptual truth. For a logical Supervenience relation based on a secondary intension, the Supervenience can be explained

Supervenience and Explanation 89

by noting that the primary intension of the concept picks out some actual-world referent that is projected (by rigidification) invariantly across physically identical worlds. All we need here for an explanation is an a priori conceptual analysis combined with contingent facts about the actual world.51 On the other hand, a mere natural Supervenience relation will itself be a contingent law. At best it will be explainable in terms of more fundamental laws; at worst, the Supervenience law will itself be fundamental. In either case, one explains certain regularities in the world by invoking fundamental laws, just as one does in physics, and as always, fundamental laws are where explanation must stop. Mere natural Supervenience is ontologically expensive, as we have seen, so it is fortunate that logical Supervenience is the rule and natural Supervenience the exception.

PART II. The Irreducibility of consciousness

3. Can Consciousness Be Reductively Explained?

1. Is Consciousness Logically Supervenient on the Physical?

Almost everything in the world can be explained in physical terms; it is natural to hope that consciousness might be explained this way, too. In this chapter, however, I will argue that consciousness escapes the net of reductive explanation. No explanation given wholly in physical terms can ever account for the emergence of conscious experience. This may seem to be a negative conclusion, but it leads to some strong positive consequences that I will bring out in later chapters.

To make the case against reductive explanation, we need to show that consciousness is not logically supervenient on the physical. In principle, we need to show that it does not supervene globally-that is, that all the microphysical facts in the world do not entail the facts about consciousness. In practice, it is easier to run the argument locally, arguing that in an individual, microphysical facts do not entail the facts about consciousness. When it comes to consciousness, local and global Supervenience plausibly stand and fall together, so it does not matter much which way we run the argument: if consciousness supervenes at all, it almost certainly supervenes locally. If this is disputed, however, all the arguments can be run at the global level with straightforward alterations.

How can we argue that consciousness is not logically supervenient on the physical? There are various ways. We can think about what is conceivable, in order to argue directly for the logical possibility of a situation in which the physical facts are the same but the facts about experience are different. We can appeal to epistemology, arguing that the right sort of link between

94 The Irreducibility of Consciousness

knowledge of physical facts and knowledge of consciousness is absent. And we can appeal directly to the concept of consciousness, arguing that there is no analysis of the concept that could ground an entailment from the physical to the phenomenal. In what follows I will give arguments using all three of these strategies. The first two are essentially arguments from conceivability, the second two are arguments from epistemology, and the fifth is an argument from analysis. There is some element of redundancy among the five arguments, but together they make a strong case.

One can also do things more directly, making the case against reductive explanation without explicitly appealing to logical Supervenience. I have taken that route elsewhere, but here I will give the more detailed analysis to allow a fuller case. All the same, the case against reductive explanation and the critique of existing reductive accounts (in section 2 onward) should make sense even without this analysis. Some readers might like to proceed there directly, at least on a first reading.

(A technical note: The burden of this chapter is to argue, in effect, that there is no a priori entailment from physical facts to phenomenal facts. The sort of necessity that defines the relevant Supervenience relation is the a priori version of logical necessity, where primary intensions are central. As we saw in Chapter 2, this is the relation that is relevant to issues about explanation; matters of a posteriori necessity can be set to one side. In the next chapter, issues of ontology rather than explanation are central, and I argue separately that there is no a posteriori necessary connection between physical facts and phenomenal facts.)

Argument 1: The logical possibility of zombies

The most obvious way (although not the only way) to investigate the logical Supervenience of consciousness is to consider the logical possibility of a zombie: someone or something physically identical to me (or to any other conscious being), but lacking conscious experiences altogether.1 At the global level, we can consider the logical possibility of a zombie world: a world physically identical to ours, but in which there are no conscious experiences at all. In such a world, everybody is a zombie.

So let us consider my zombie twin. This creature is molecule for molecule identical to me, and identical in all the low-level properties postulated by a completed physics, but he lacks conscious experience entirely. (Some might prefer to call a zombie "it," but I use the personal pronoun; I have grown quite fond of my zombie twin.) To fix ideas, we can imagine that right now I am gazing out the window, experiencing some nice green sensations from seeing the trees outside, having pleasant taste experiences through munching on a chocolate bar, and feeling a dull aching sensation in my right shoulder.

Can Consciousness Be Reductively Explained? 95

Figure 3.1. Calvin and Hobbes on zombies.

Figure 3.1. Calvin and Hobbes on zombies. (Calvin and Hobbes © Watterson. Distributed by Universal Press Syndicate. Reprinted with permission. All rights reserved)

Calvin and Hobbes by Bill Watterson

What is going on in my zombie twin? He is physically identical to me, and we may as well suppose that he is embedded in an identical environment. He will certainly be identical to me functionally: he will be processing the same sort of information, reacting in a similar way to inputs, with his internal configurations being modified appropriately and with indistinguishable behavior resulting. He will be psychologically identical to me, in the sense developed in Chapter 1. He will be perceiving the trees outside, in the functional sense, and tasting the chocolate, in the psychological sense. All of this follows logically from the fact that he is physically identical to me, by virtue of the functional analyses of psychological notions. He will even be "conscious" in the functional senses described earlier-he will be awake, able to report the contents of his internal states, able to focus attention in various places, and so on. It is just that none of this functioning will be accompanied by any real conscious experience. There will be no phenomenal feel. There is nothing it is like to be a zombie.

This sort of zombie is quite unlike the zombies found in Hollywood movies, which tend to have significant functional impairments (Figure 3.1). The sort of consciousness that Hollywood zombies most obviously lack is a psychological version: typically, they have little capacity for introspection and lack a refined ability to voluntarily control behavior. They may or may not lack phenomenal consciousness; as Block (1995) points out, it is reasonable to suppose that there is something it tastes like when they eat their victims. We can call these psychological zombies; I am concerned with phenomenal zombies, which are physically and functionally identical, but which lack experience. (Perhaps it is not surprising that phenomenal zombies have not been popular in Hollywood, as there would be obvious problems with their depiction.)

96 The Irreducibility of Consciousness

The idea of zombies as I have described them is a strange one. For a start, ' it is unlikely that zombies are naturally possible. In the real world, it is likely that any replica of me would be conscious. For this reason, it is most natural to imagine unconscious creatures as physically different from conscious ones-exhibiting impaired behavior, for example. But the question is not whether it is plausible that zombies could exist in our world, or even whether the idea of a zombie replica is a natural one; the question is whether the notion of a zombie is conceptually coherent. The mere intelligibility of the notion is enough to establish the conclusion.

Arguing for a logical possibility is not entirely straightforward. How, for example, would one argue that a mile-high unicycle is logically possible? It just seems obvious. Although no such thing exists in the real world, the description certainly appears to be coherent. If someone objects that it is not logically possible-it merely seems that way-there is little we can say, except to repeat the description and assert its obvious coherence. It seems quite clear that there is no hidden contradiction lurking in the description.

I confess that the logical possibility of zombies seems equally obvious to me. A zombie is just something physically identical to me, but which has no conscious experience-all is dark inside. While this is probably empirically impossible, it certainly seems that a coherent situation is described; I can discern no contradiction in the description. In some ways an assertion of this logical possibility comes down to a brute intuition, but no more so than with the unicycle. Almost everybody, it seems to me, is capable of conceiving of this possibility. Some may be led to deny the possibility in order to make some theory come out right, but the justification of such theories should ride on the question of possibility, rather than the other way around.

In general, a certain burden of proof lies on those who claim that a given description is logically impossible. If someone truly believes that a mile-high unicycle is logically impossible, she must give us some idea of where a contradiction lies, whether explicit or implicit. If she cannot point out something about the intensions of the concepts "mile-high" and "unicycle" that might lead to a contradiction, then her case will not be convincing. On the other hand, it is no more convincing to give an obviously false analysis of the notions in question-to assert, for example, that for something to qualify as a unicycle it must be shorter than the Statue of Liberty. If no reasonable analysis of the terms in question points toward a contradiction, or even makes the existence of a contradiction plausible, then there is a natural assumption in favor of logical possibility.

That being said, there are some positive things that proponents of logical possibility can do to bolster their case. They can exhibit various indirect arguments, appealing to what we know about the phenomena in question and the way we think about hypothetical cases involving these phenomena,

Can Consciousness Be Reductively Explained? 97

in order to establish that the obvious logical possibility really is a logical possibility, and really is obvious. One might spin a fantasy about an ordinary person riding a unicycle, when suddenly the whole system expands a thousandfold. Or one might describe a series of unicycles, each bigger than the last. In a sense, these are all appeals to intuition, and an opponent who wishes to deny the possibility can in each case assert that our intuitions have misled us, but the very obviousness of what we are describing works in our favor, and helps shift the burden of proof further onto the other side.

For example, we can indirectly support the claim that zombies are logically possible by considering nonstandard realizations of my functional organization.2 My functional organization-that is, the pattern of causal organization embodied in the mechanisms responsible for the production of my behavior-can in principle be realized in all sorts of strange ways. To use a common example (Block 1978), the people of a large nation such as China might organize themselves so that they realize a causal organization isomorphic to that of my brain, with every person simulating the behavior of a single neuron, and with radio links corresponding to synapses. The population might control an empty shell of a robot body, equipped with sensory transducers and motor effectors.

Many people find it implausible that a set-up like this would give rise to conscious experience-that somehow a "group mind" would emerge from the overall system. I am not concerned here with whether or not conscious experience would in fact arise; I suspect that in fact it would, as I argue in Chapter 7. All that matters here is that the idea that such a system lacks conscious experience is coherent. A meaningful possibility is being expressed, and it is an open question whether consciousness arises or not. We can make a similar point by considering my silicon isomorph, who is organized like me but who has silicon chips where I have neurons. Whether such an isomorph would in fact be conscious is controversial, but it seems to most people that those who deny this are expressing a coherent possibility. From these cases it follows that the existence of my conscious experience is not logically entailed by the facts about my functional organization.

But given that it is conceptually coherent that the group-mind set-up or my silicon isomorph could lack conscious experience, it follows that my zombie twin is an equally coherent possibility. For it is clear that there is no more of a conceptual entailment from biochemistry to consciousness than there is from silicon or from a group of homunculi. If the silicon isomorph without conscious experience is conceivable, we need only substitute neurons for silicon in the conception while leaving functional organization constant, and we have my zombie twin. Nothing in this substitution could force experience into the conception; these implementational differences are simply not the sort of thing that could be conceptually relevant to experience. So consciousness fails to logically supervene on the physical.

98 The Irreducibility of Consciousness

The argument for zombies can be made without an appeal to these non-standard realizations, but these have a heuristic value in eliminating a source of conceptual confusion. To some people, intuitions about the logical possibility of an unconscious physical replica seem less than clear at first, perhaps because the familiar co-occurrence of biochemistry and consciousness can lead one to suppose a conceptual connection. Considerations of the less familiar cases remove these empirical correlations from the picture, and therefore make judgments of logical possibility more straightforward.3 But once it is accepted that these nonconscious functional replicas are logically possible, the corresponding conclusion concerning a physical replica cannot be avoided.

Some may think that conceivability arguments are unreliable. For example, sometimes it is objected that we cannot really imagine in detail the many billions of neurons in the human brain. Of course this is true; but we do not need to imagine each of the neurons to make the case. Mere complexity among neurons could not conceptually entail consciousness; if all that neural structure is to be relevant to consciousness, it must be relevant in virtue of some higher-level properties that it enables. So it is enough to imagine the system at a coarse level, and to make sure that we conceive it with appropriately sophisticated mechanisms of perception, categorization, high-bandwidth access to information contents, reportability, and the like. No matter how sophisticated we imagine these mechanisms to be, the zombie scenario remains as coherent as ever. Perhaps an opponent might claim that all the unimagined neural detail is conceptually relevant in some way independent of its contribution to sophisticated functioning; but then she owes us an account of what that way might be, and none is available. Those implementational details simply lie at the wrong level to be conceptually relevant to consciousness.

It is also sometimes said that conceivability is an imperfect guide to possibility. The main way that conceivability and possibility can come apart is tied to the phenomenon of a posteriori necessity: for example, the hypothesis that water is not H2O seems conceptually coherent, but water is arguably H2O in all possible worlds. But a posteriori necessity is irrelevant to the concerns of this chapter. As we saw in the last chapter, explanatory connections are grounded in a priori entailments from physical facts to high-level facts. The relevant kind of possibility is to be evaluated using the primary intensions of the terms involved, instead of the secondary intensions that are relevant to a posteriori necessity. So even if a zombie world is conceivable only in the sense in which it is conceivable that water is not H2O, that is enough to establish that consciousness cannot be reductively explained.

Those considerations aside, the main way in which conceivability arguments can go wrong is by subtle conceptual confusion: if we are insufficiently

Can Consciousness Be Reductively Explained? 99

reflective we can overlook an incoherence in a purported possibility, by taking a conceived-of situation and misdescribing it. For example, one might think that one can conceive of a situation in which Fermat's last theorem is false, by imagining a situation in which leading mathematicians declare that they have found a counterexample. But given that the theorem is actually true, this situation is being misdescribed: it is really a scenario in which Fermat's last theorem is true, and in which some mathematicians make a mistake. Importantly, though, this kind of mistake always lies in the a priori domain, as it arises from the incorrect application of the primary intensions of our concepts to a conceived situation. Sufficient reflection will reveal that the concepts are being incorrectly applied, and that the claim of logical possibility is not justified.

So the only route available to an opponent here is to claim that in describing the zombie world as a zombie world, we are misapplying the concepts, and that in fact there is a conceptual contradiction lurking in the description. Perhaps if we thought about it clearly enough we would realize that by imagining a physically identical world we are thereby automatically imagining a world in which there is conscious experience. But then the burden is on the opponent to give us some idea of where the contradiction might lie in the apparently quite coherent description. If no internal incoherence can be revealed, then there is a very strong case that the zombie world is logically possible.

As before, I can detect no internal incoherence; I have a clear picture of what I am conceiving when I conceive of a zombie. Still, some people find conceivability arguments difficult to adjudicate, particularly where strange ideas such as this one are concerned. It is therefore fortunate that every point made using zombies can also be made in other ways, for example by considering epistemology and analysis. To many, arguments of the latter sort (such as arguments 3-5 below) are more straightforward and therefore make a stronger foundation in the argument against logical Supervenience. But zombies at least provide a vivid illustration of important issues in the vicinity.

Argument 2: The inverted spectrum

Even in making a conceivability argument against logical Supervenience, it is not strictly necessary to establish the logical possibility of zombies or a zombie world. It suffices to establish the logical possibility of a world physically identical to ours in which the facts about conscious experience are merely different from the facts in our world, without conscious experience being absent entirely. As long as some positive fact about experience in our world does not hold in a physically identical world, then consciousness does not logically supervene.

100 The Irreducibility of Consciousness

It is therefore enough to note that one can coherently imagine a physically identical world in which conscious experiences are inverted, or (at the local level) imagine a being physically identical to me but with inverted conscious experiences. One might imagine, for example, that where I have a red experience, my inverted twin has a blue experience, and vice versa. Of course he will call his blue experiences "red," but that is irrelevant. What matters is that the experience he has of the things we both call "red"-blood, fire engines, and so on-is of the same kind as the experience I have of the things we both call "blue," such as the sea and the sky.

The rest of his color experiences are systematically inverted with respect to mine, in order that they cohere with the red-blue inversion. Perhaps the best way to imagine this happening with human color experiences is to imagine that two of the axes of our three-dimensional color space are switched-the red-green axis is mapped onto the yellow-blue axis, and vice versa.4 To achieve such an inversion in the actual world, presumably we would need to rewire neural processes in an appropriate way, but as a logical possibility, it seems entirely coherent that experiences could be inverted while physical structure is duplicated exactly. Nothing in the neurophysiology dictates that one sort of processing should be accompanied by red experiences rather than by yellow experiences.

It is sometimes objected (Harrison 1973; Hardin 1987) that human color space is asymmetrical in a way that disallows such an inversion. For instance, certain colors have a warmth or coolness associated with them, and warmth and coolness appear to be directly associated with different functional roles (e.g., warmth is perceived as "positive," whereas coolness is perceived as "negative"). If a warm color and a cool color were switched, then the "warm" phenomenal feel would become dissociated from the "warm" functional role-a "cool" green experience would be reported as positive rather than negative, and so on. In a similar way, there seem to be more discriminable shades of red than of yellow, so swapping red experiences with yellow experiences directly might lead to the odd situation in which a subject could functionally discriminate more shades of yellow than are distinguishable phenomenologically. Perhaps there are enough asymmetries in color space that any such inversion would lead to a strange dissociation of phenomenal feel from the "appropriate" functional role.

There are three things we can say in response to this. First, there does not seem to be anything incoherent about the notion of such a dissociation (e.g., cool phenomenology with warm reactions), although it is admittedly an odd idea.5 Second, instead of mapping red precisely onto blue and vice versa, one can imagine that these are mapped onto slightly different colors. For example, red might be mapped onto a "warm" version of blue (as Levine [1991] suggests), or even onto color not in our color space at all. In the

Can Consciousness Be Reductively Explained? 101

red-yellow case, we might imagine that red is mapped onto an extended range of yellow experiences, in which more discrimination is available. There is no reason why spectrum inversion scenarios must involve colors drawn from the usual color space. Third, perhaps the most compelling response is to argue (with Shoemaker [1982]) that even if our own color space is asymmetrical, there certainly could be creatures whose color space is symmetrical. For example, there is probably a naturally possible creature who sees (and experiences) precisely two colors, A and B, which correspond to distinct, well-separated ranges of light wavelengths, and for which the distinction between the two exhausts the structure of the color space. It seems entirely coherent to imagine two such creatures that are physically identical, but whose experiences of A and B are inverted. That is enough to make the point.

Even many reductive materialists (e.g., Shoemaker [1982]) have conceded that it is coherent that one's color experiences might be inverted while one's functional organization stays constant. It is allowed that a system with different underlying neurophysiological properties, or with something like silicon in place of neurobiology, might have different color experiences. But once this is granted, it follows automatically that inversion of experiences in a physical replica is at least conceptually coherent. The extra neurophysiological properties that are constrained in such a case are again not the kind of thing that could logically determine the nature of the experience. Even if there is some sort of a posteriori identification between certain neurophysiological structures and certain experiences (as Shoemaker believes), we must still allow that a different pattern of associations is conceivable, in the sense of conceivability that is relevant to reductive explanation.

While the possibility of inverted spectra and the possibility of zombies both establish that consciousness fails to supervene logically, the first establishes a conclusion strictly weaker than the second. Somebody might conceivably hold that inverted spectra but not zombies are logically possible. If this were the case, then the existence of consciousness could be reductively explained, but the specific character of particular conscious experiences could not be.

Argument 3: From epistemic asymmetry

As we saw earlier, consciousness is a surprising feature of the universe. Our grounds for belief in consciousness derive solely from our own experience of it. Even if we knew every last detail about the physics of the universe-the configuration, causation, and evolution among all the fields and particles in the spatiotemporal manifold-that information would not lead us to postulate the existence of conscious experience. My knowledge of consciousness,

102 The Irreducibility of Consciousness

in the first instance, comes from my own case, not from any external observation. It is my first-person experience of consciousness that forces the problem

on me.

From all the low-level facts about physical configurations and causation, we can in principle derive all sorts of high-level facts about macroscopic systems, their organization, and the causation among them. One could determine all the facts about biological function, and about human behavior and the brain mechanisms by which it is caused. But nothing in this vast causal story would lead one who had not experienced it directly to believe that there should be any consciousness. The very idea would be unreasonable; almost mystical, perhaps.

It is true that the physical facts about the world might provide some indirect evidence for the existence of consciousness. For example, from these facts one could ascertain that there were a lot of organisms that claimed to be conscious, and said they had mysterious subjective experiences. Still, this evidence would be quite inconclusive, and it might be most natural to draw an eliminativist conclusion-that there was in fact no experience present in these creatures, just a lot of talk.

Eliminativism about conscious experience is an unreasonable position only because of our own acquaintance with it. If it were not for this direct knowledge, consciousness could go the way of the vital spirit. To put it another way, there is an epistemic asymmetry in our knowledge of consciousness that is not present in our knowledge of other phenomena.6 Our knowledge that conscious experience exists derives primarily from our own case, with external evidence playing at best a secondary role.

The point can also be made by pointing to the existence of a problem of other minds. Even when we know everything physical about other creatures, we do not know for certain that they are conscious, or what their experiences are (although we may have good reason to believe that they are). It is striking that there is no problem of "other lives," or of "other economies," or of "other heights." There is no epistemic asymmetry in those cases, precisely because those phenomena are logically supervenient on the physical.

The epistemic asymmetry in knowledge of consciousness makes it clear that consciousness cannot logically supervene. If it were logically supervenient, there would be no such epistemic asymmetry; a logically supervenient property can be detected straightforwardly on the basis of external evidence, and there is no special role for the first-person case. To be sure, there are some supervenient properties-memory, perhaps-that are more easily detected in the first-person case. But this is just a matter of how hard one has to work. The presence of memory is just as accessible from the third person, in principle, as from the first person. The epistemic asymmetry associated with consciousness is much more fundamental, and it tells us that no

Can Consciousness Be Reductively Explained? 103

collection of facts about complex causation in physical systems adds up to a fact about consciousness.

Argument 4: The knowledge argument

The most vivid argument against the logical Supervenience of consciousness is suggested by Jackson (1982), following related arguments by Nagel (1974) and others. Imagine that we are living in an age of a completed neuroscience, where we know everything there is to know about the physical processes within our brain responsible for the generation of our behavior. Mary has been brought up in a black-and-white room and has never seen any colors except for black, white, and shades of gray.7 She is nevertheless one of the world's leading neuroscientists, specializing in the neurophysiology of color vision. She knows everything there is to know about the neural processes involved in visual information processing, about the physics of optical processes, and about the physical makeup of objects in the environment. But she does not know what it is like to see red. No amount of reasoning from the physical facts alone will give her this knowledge.

It follows that the facts about the subjective experience of color vision are not entailed by the physical facts. If they were, Mary could in principle come to know what it is like to see red on the basis of her knowledge of the physical facts. But she cannot. Perhaps Mary could come to know what it is like to see red by some indirect method, such as by manipulating her brain in the appropriate way. The point, however, is that the knowledge does not follow from the physical knowledge alone. Knowledge of all the physical facts will in principle allow Mary to derive all the facts about a system's reactions, abilities, and cognitive capacities; but she will still be entirely in the dark about its experience of red.

A related way to make this point is to consider systems quite different from ourselves, perhaps much simpler-such as bats or mice-and note that the physical facts about these systems do not tell us what their conscious experiences are like, if they have any at all (Nagel focuses on this sort of issue). Once all the physical facts about a mouse are in, the nature of its conscious experience remains an open question: it is consistent with the physical facts about a mouse that it has conscious experience, and it is consistent with the physical facts that it does not. From the physical facts about a bat, we can ascertain all the facts about a bat, except the facts about its conscious experiences. Knowing all the physical facts, we still do not know what it is like to be a bat.

Along similar lines we can consider a computer, designed as a simple cognitive agent (perhaps it has the intelligence of a dog), but similar to us in certain respects, such as its capacity for perceptual discrimination. In

104 The Irreducibility of Consciousness

particular it categorizes color stimuli in a manner quite similar to ours, grouping things that we would call "red" under one category and things we would call "green" under another. Even if we know every detail about the computer's circuits, questions remain: (1) Is the computer experiencing anything at all when it looks at roses?; (2) If it is, is it experiencing the same sensory color quality that we have when we look at a rose, or some quite different quality? These are entirely meaningful questions, and knowing all the physical facts does not force one answer rather than another onto us. The physical facts therefore do not logically entail the facts about conscious experience.

Jackson put his argument forward as an argument against materialism rather than against reductive explanation. There have been many replies to the argument; I will discuss them in the next chapter, where materialism rather than reductive explanation will be at issue. But for now it is interesting to note that most of the objections to the argument against materialism have conceded the point that is relevant to the argument against reductive explanation: that knowledge of what red is like is factual knowledge that is not entailed a priori by knowledge of the physical facts. The only way that the conclusion can be evaded is to deny that knowing what red experience is like gives knowledge of a fact at all. This is the strategy taken by Lewis (1990) and Nemirow (1990), who argue that all Mary is lacking is an ability, such as the ability to recognize red things. I discuss this suggestion in the next chapter; here, I simply note that insofar as it seems clear that when she sees red for the first time, Mary is discovering something about the way the world is, it seems clear that the knowledge she is gaining is knowledge of a fact.

Argument 5: From the absence of analysis

If proponents of reductive explanation are to have any hope of defeating the arguments above, they will have to give us some idea of how the existence of consciousness might be entailed by physical facts. While it is not fair to expect all the details, one at least needs an account of how such an entailment might possibly go. But any attempt to demonstrate such an entailment is doomed to failure. For consciousness to be entailed by a set of physical facts, one would need some kind of analysis of the notion of consciousness-the kind of analysis whose satisfaction physical facts could imply-and there is no such analysis to be had.

The only analysis of consciousness that seems even remotely tenable for these purposes is a functional analysis. Upon such an analysis, it would be seen that all there is to the notion of something's being conscious is that it should play a certain functional role. For example, one might say that all there is to a state's being conscious is that it be verbally reportable, or that

Can Consciousness Be Reductively Explained? 105

it be the result of certain kinds of perceptual discrimination, or that it make information available to later processes in a certain way, or whatever. But on the face of it, these fail miserably as analyses. They simply miss what it means to be a conscious experience. Although conscious states may play various causal roles, they are not defined by their causal roles. Rather, what makes them conscious is that they have a certain phenomenal feel, and this feel is not something that can be functionally defined away.

To see how unsatisfactory these analyses are, note how they trivialize the problem of explaining consciousness. Suddenly, all we have to do to explain consciousness is explain our ability to make certain verbal reports, or to perform certain sorts of discrimination, or to manifest some other capacity. But on the face of it, it is entirely conceivable that one could explain all these things without explaining a thing about consciousness itself; that is, without explaining the experience that accompanies the report or the discrimination. To analyze consciousness in terms of some functional notion is either to change the subject or to define away the problem. One might as well define "world peace" as "a ham sandwich." Achieving world peace becomes much easier, but it is a hollow achievement.

Functional analyses of consciousness can also be argued against on more specific grounds. For example, any functionally analyzed concept will have a degree of semantic indeterminancy. Does a mouse have beliefs? Do bacteria learn? Is a computer virus alive? The best answer to these questions is usually in a sense yes, in a sense no. It all depends on how we draw the boundaries in the concepts, and in any high-level functional concepts the boundaries will be vague. But compare: Does a mouse have conscious experience? Does a virus? These are not matters for stipulation. Either there is something that it is like to be a mouse or there is not, and it is not up to us to define the mouse's experience into or out of existence. To be sure, there is probably a continuum of conscious experience from the very faint to the very rich; but if something has conscious experience, however faint, we cannot stipulate it away. This determinacy could not be derived from any functional analysis of the concepts in the vicinity of consciousness, as the functional concepts in the vicinity are all somewhat vague. If so, it follows that the notion of consciousness cannot be functionally analyzed.

Another objection is that the functionalist analysis collapses the important distinction, outlined in Chapter 1, between the notions of awareness and consciousness. Presumably if consciousness is to be functionally analyzed, it will be analyzed roughly as we analyzed awareness then: in terms of a certain accessibility of information in later processing and in the control of behavior. Awareness is a perfectly good concept, but it is quite distinct from the concept of conscious experience. The functionalist treatment collapses the two notions of consciousness and awareness into one, and therefore does not do justice to our conceptual system.

106 The Irreducibility of Consciousness

The alternatives to functional analysis look even worse. It is most unclear that there could be any other kind of analysis appropriate for reductive explanation. The only alternative might be a structural analysis-perhaps consciousness could be analyzed as some sort of biochemical structure-but that analysis would be even more clearly inadequate. Whether or not consciousness is a biochemical structure, that is not what "consciousness" means. To analyze consciousness that way again trivializes the explanatory problem by changing the subject. It seems that the concept of consciousness is irreducible, being characterizable only in terms of concepts that themselves involve consciousness.

Note that this is quite unlike the sort of irreducibility that is sometimes supposed to hold for high-level concepts in general. We have seen that many high-level notions have no crisp definitions, and no manageable analyses in terms of necessary and sufficient conditions. Nevertheless, as we saw in the last chapter, these concepts at least have rough-and-ready analyses that get us into the ballpark, although they will inevitably fail to do justice to the details. Most importantly, it is easy to see that properties such as life, learning, and so on can be analyzed as functional properties, even if spelling out the details of just which functional property is a difficult matter. Even though these properties lack crisp functional definitions, they are nevertheless quite compatible with entailment by the physical facts.

The problems with consciousness are in a different league. Here, the purported analyses do not even get into the ballpark. In a much starker way, they completely fail to characterize what needs to be explained. There is no temptation to even try to add epicycles to a purported functional analysis of consciousness in order to make it satisfactory, as there is with similar analyses of life and of learning. Consciousness is simply not to be characterized as a functional property in the first place. The same goes for analyses of consciousness as a structural property, or in other reductive terms. There is therefore no way for an entailment from physical facts to consciousness to get off the ground.

2. The Failure of Reductive Explanation

The failure of consciousness to logically supervene on the physical tells us that no reductive explanation of consciousness can succeed. Given any account of the physical processes purported to underlie consciousness, there will always be a further question: Why are these processes accompanied by conscious experience? For most other phenomena, such a question is easily answered: the physical facts about those processes entail the existence of the phenomena. For a phenomenon such as life, for example, the physical facts imply that certain functions will be performed, and the performance of those functions is

Can Consciousness Be Reductively Explained? 107

all we need to explain in order to explain life. But no such answer will suffice for consciousness.

Physical explanation is well suited to the explanation of structure and of function. Structural properties and functional properties can be straightforwardly entailed by a low-level physical story, and so are clearly apt for reductive explanation. And almost all the high-level phenomena that we need to explain ultimately come down to structure or function: think of the explanation of waterfalls, planets, digestion, reproduction, language. But the explanation of consciousness is not just a matter of explaining structure and function. Once we have explained all the physical structure in the vicinity of the brain, and we have explained how all the various brain functions are performed, there is a further sort of explanandum: consciousness itself. Why should all this structure and function give rise to experience? The story about the physical processes does not say.

We can put this in terms of the thought experiments given earlier. Any story about physical processes applies equally to me and to my zombie twin. It follows that nothing in that story says why, in my case, consciousness arises. Similarly, any story about physical processes applies equally to my inverted twin, who sees blue where I see red: it follows that nothing in that story says why my experience is of one variety rather than another. The very fact that it is logically possible that the physical facts could be the same while the facts about consciousness are different shows us that as Levine (1983) has put it, there is an explanatory gap between the physical level and conscious experience.

If this is right, the fact that consciousness accompanies a given physical process is a further fact, not explainable simply by telling the story about the physical facts. In a sense, the accompaniment must be taken as brute. We might try to systematize and explain these brute facts in terms of some simple underlying pattern, but there will always remain an element here that is logically independent of the physical story. Perhaps we might get some kind of explanation by combining the underlying physical facts with certain further bridging principles that link the physical facts with consciousness, but this explanation will not be a reductive one. The very need for explicit bridging principles shows us that consciousness is not being explained reductively, but is being explained on its own terms.

Of course nothing I have said implies that physical facts are irrelevant to the explanation of consciousness. We can still expect physical accounts to play a significant role in a theory of consciousness, giving information about the physical basis of consciousness, for example, and perhaps yielding a detailed correspondence between various aspects of physical processing and aspects of conscious experience. Such accounts may be especially useful in helping to understand the structure of consciousness: the patterns of similarity and difference between experiences, the geometric structure of phenome-

108 The Irreducibility of Consciousness

nal fields, and so on. I say much more about these and other things that physical explanation can tell us about experience in a nonreductive framework in Chapter 6. But a physical account, alone, is not enough. At this point, a number of objections naturally arise.

Objection 1: Are we setting the standards too high?

Some might argue that explanation of any high-level phenomena will postulate "bridge laws" in addition to a low-level account, and that it is only with the aid of these bridge laws that the details of the high-level phenomena are derived. However, as the discussion in the last chapter suggests (and as is carefully argued by Horgan [1978]), in such cases the bridge laws are not further facts about the world. Rather, the connecting principles themselves are logically supervenient on the low-level facts. The extreme case of such a bridging principle is a Supervenience conditional, which we have seen is usually a conceptual truth. Other more "localized" bridging principles, such as the link between molecular motion and heat, can at least be derived from the physical facts. For consciousness, by contrast, such bridging principles must be taken as primitive.

It is interesting to see how a typical high-level property-such as life, say-evades the arguments put forward in the case of consciousness. First, it is straightforwardly inconceivable that there could be a physical replica of a living creature that was not itself alive. Perhaps a problem might arise due to context-dependent properties (would a replica that forms randomly in a swamp be alive, or be human?), but fixing environmental facts eliminates even that possibility. Second, there is no "inverted life" possibility analogous to the inverted spectrum. Third, when one knows all the physical facts about an organism (and possibly about its environment), one has enough material to know all the biological facts. Fourth, there is no epistemic asymmetry with life; facts about life in others are as accessible, in principle, as facts about life in ourselves. Fifth, the concept of life is plausibly analyzable in functional terms: to be alive is roughly to possess certain capacities to adapt, reproduce, and metabolize. As a general point, most high-level phenomena come down to matters of physical structure and function, and we have good reason to believe that structural and functional properties are logically supervenient on the physical.

Objection 2: Couldn't a vitalist have said the same thing about life?

All this notwithstanding, a common reaction to the sort of argument I have given is to reply that a vitalist about life might have said the same things.8 For example, a vitalist might have claimed that it is logically possible that a physical replica of me might not be alive, in order to establish that life

Can Consciousness Be Reductively Explained? 109

cannot be reductively explained. And a vitalist might have argued that life is a further fact, not explained by any account of the physical facts. But the vitalist would have been wrong. By analogy, might not the opponent of reductive explanation for consciousness also be wrong?

I think this reaction misplaces the source of vitalist objections. Vitalism was mostly driven by doubt about whether physical mechanisms could perform all the complex functions associated with life: adaptive behavior, reproduction, and the like. At the time, very little was known about the enormous sophistication of biochemical mechanisms, so this sort of doubt was quite natural. But implicit in these very doubts is the conceptual point that when it comes to explaining life, it is the performance of various functions that needs to be explained. Indeed, it is notable that as physical explanation of the relevant functions gradually appeared, vitalist doubts mostly melted away. With consciousness, by contrast, the problem persists even when the various functions are explained.

Presented with a full physical account showing how physical processes perform the relevant functions, a reasonable vitalist would concede that life has been explained. There is not even conceptual room for the performance of these functions without life. Perhaps some ultrastrong vitalist would deny even this, claiming that something is left out by a functional account of life-the vital spirit, perhaps. But the obvious rejoinder is that unlike experience, the vital spirit is not something we have independent reason to believe in. Insofar as there was ever any reason to believe in it, it was as an explanatory construct-"We must have such a thing in order to be able to do such amazing stuff." But as an explanatory construct, the vital spirit can be eliminated when we find a better explanation of how the functions are performed. Conscious experience, by contrast, forces itself on one as an explanandum and cannot be eliminated so easily.

One reason a vitalist might think something is left out of a functional explanation of life is precisely that nothing in a physical account explains why there is something it is like to be alive. Perhaps some element of belief in a "vital spirit" was tied to the phenomena of one's inner life. Many have perceived a link between the concepts of life and experience, and even today it seems reasonable to say that one of the things that needs to be explained about life is the fact that many living creatures are conscious. But the existence of this sort of vitalist doubt is of no comfort to the proponent of reductive explanation of consciousness, as it is a doubt that has never been overturned.

Objection 3: Is conceivability a guide to possibility?

Philosophers are often suspicious of arguments that give a key role to conceivability, frequently responding that conceivability does not suffice for possibility. This is a subtle issue that I have discussed earlier and will dis-

110 The Irreducibility of Consciousness

cuss again: but here, the subtleties are not especially relevant. When it comes to matters of explanation, it is clear that conceivability is central. If on reflection we find it conceivable that all these physical processes could take place in the absence of consciousness, then no reductive explanation of consciousness will be satisfactory: the further question of why we exist and not zombies will always arise. Even if conceivability is tied to the limits of human capacity, explanation is tied to the limits of human capacity in a similar way.

Another way to put the point is to note that reductive explanation of a phenomenon in terms of the physical requires an a priori implication from the physical facts to the relevant high-level facts (logical Supervenience according to primary intension, as I put it earlier). If such a connection does not hold, then we will always be able to raise the further question of why the physical processes give rise to consciousness. We have seen that in almost all domains, the right sort of connection holds, making reductive explanation possible; but it does not seem to hold for conscious experience. One can question whether Ontological views such as materialism turn on these a priori links-I discuss that matter in the next chapter-but when it comes to reductive explanation, such links are crucial.

Objection 4: Isn't this a collection of circular intuitions?

It might be further objected that the arguments I have given consist, at bottom, in a collection of intuitions. There is certainly a sense in which all these arguments are based on intuition, but I have tried to make clear just how natural and plain these intuitions are, and how forced it is to deny them. The main intuition at work is that there is something to be explained-some phenomenon associated with first-person experience that presents a problem not presented by observation of cognition from the third-person point of view. Given the premise that some explanandum is forced on us by first-person experience that is not forced on us by third-person observation, most of the arguments above fall out. It follows immediately, for example, that what needs to be explained cannot be analyzed as the playing of some functional role, for the latter phenomenon is revealed to us by third-person observation and is much more straightforward.

The "intuition" at work here is the very raison d'кtre of the problem of consciousness. The only consistent way to get around the intuitions is to deny the problem and the phenomenon altogether. One can always, at least when speaking "philosophically," deny the intuitions altogether, and deny that there is anything (apart from the performance of various functions) that needs explaining. But if one takes consciousness seriously, the conclusions for which I am arguing must follow.

Can Consciousness Be Reductively Explained? 111

Objection 5: Doesn't all explanation have to stop somewhere?

A final objection is that no explanation gives one something for nothing: all explanation has to stop somewhere. In explaining the motion of the planets, for example, one takes the laws of gravity and the existence of mass for granted. Perhaps we should simply take something for granted in this case, too? I am sympathetic with this point; I think we do have to take something for granted in explaining consciousness. But in doing so we inevitably move beyond a reductive explanation. Indeed, this sort of analogy lends support to the nonreductive position I am advocating. We take the laws of physics for granted because they are fundamental laws. If we take a link between physical processes and conscious experience for granted, this suggests that the link should be taken as fundamental in the same way. I return to this point in the next chapter.

3. Cognitive Modeling

In this and the sections that follow, I illustrate the failure of reductive explanation by giving a critique of a number of accounts of consciousness that have been proposed by researchers in various disciplines. Not all of these proposals have been put forward as reductive explanations of conscious experience, although they have often been interpreted this way; but in any case, it is instructive to see just what these accounts can and cannot achieve. Along the way, it is interesting to note these researchers' varying attitudes to the hard questions about conscious experience.

First, I will consider accounts based on cognitive modeling. Cognitive modeling works well for most problems in cognitive science. By exhibiting a model of the causal dynamics involved in cognitive processes, one can explain the causation of behavior in a cognitive agent. This provides a valuable kind of explanation for psychological phenomena, such as learning, memory, perception, control of action, attention, categorization, linguistic behavior, and so on. If we have a model that captures the causal dynamics of someone who is learning, for example, it follows that anything instantiating those dynamics in the right environment will be learning. From the model we can see how certain functions are performed, and this is all we have to explain to explain learning. But this is insufficient to explain consciousness. For any model we exhibit, it remains a further question why realization of the model should be accompanied by consciousness. This is not a question that description and analysis of the model alone can answer.

It is sometimes objected that purported models of consciousness are untestable, as there is no way to verify whether or not instantiations of the model are conscious. This is a problem, but there is a deeper problem. Even if we

112 The Irreducibility of Consciousness

had (per impossibile) an "experience meter" that could peek in and tell us whether an instantiation was conscious, this would only establish a correlation. We would know that whenever the model is instantiated, consciousness goes along with it. But it would not explain consciousness, in the way that such models explain other mental phenomena.

Such models can certainly explain "consciousness" in the psychological senses thereof, where it is construed as some kind of cognitive or functional capacity. Many existing "models of consciousness" can be most charitably interpreted in this light. We can see these as providing explanations of reportability, or of attention, or of introspective abilities, and so on. None of them, however, gives us anything close to an explanation of why these processes should be accompanied by conscious experience. Some examples will illustrate this.

The first example is the cognitive model presented by Bernard Baars (1988), as part of a book-length treatment of consciousness from the standpoint of cognitive psychology. Baars brings all sorts of experimental evidence to bear in establishing his main thesis: consciousness is a kind of global workspace in a distributed system of intelligent information processors. When processors gain access to the global workspace, they broadcast a message to the entire system, as if they had written it on a blackboard. The contents of the global workspace are the contents of consciousness.

Baars uses this model to explain a remarkable number of properties of human processing. The model provides a very suggestive framework for explaining a subject's access to information, and its role in attention, reportability, voluntary control, and even the development of a self-concept. The global workspace framework is therefore well suited to explaining consciousness in its whole bundle of psychological senses. There is at least a general theory of awareness on offer.

But there is no reductive explanation of experience to be found here. The question of why these processes should give rise to experience is simply not addressed. One might suppose that according to the theory, the contents of experience are precisely the contents of the workspace. But even if this is so, nothing internal to the theory explains why it is that the information within the global workspace is experienced. The best the theory can do is to say that the information is experienced because it is globally accessible. But now the question arises in a different form: Why should global accessibility give rise to conscious experience? This bridging question is not addressed in Baars's work.

Baars mentions this sort of worry briefly: "A skeptical reader may ... wonder whether we are truly describing conscious experience, or whether, instead, we can only deal with incidental phenomena associated with it" (p. 27). His response is to note that scientific theories tend to at least approach

Can Consciousness Be Reductively Explained? 113

the "thing itself"; for instance, biology explains inheritance itself, and not just associated phenomena. But this is simply to ignore the ways in which consciousness is different in kind from these phenomena, as we have seen. With inheritance, various functions are all there is to explain. With consciousness, there is a further explanandum: experience itself. Baars's theory can therefore be seen as an interesting approach to the cognitive processes underlying consciousness, and one that gives us much indirect insight into consciousness, but it leaves the key questions-why is there consciousness and how does it arise from cognitive processing?-untouched.

Daniel Dennett has also put forward a cognitive model of consciousness. In fact, he has put forward at least two of them. The first (Dennett 1978c) is a "box-and-lines" model, consisting in an account of the flow of information between various modules (Figure 3.2). Central to the model are (1) a perceptual module, (2) a short-term memory store M, which receives

Figure 3.2. Dennett's cognitive model of consciousness.

Figure 3.2. Dennett's cognitive model of consciousness. (Redrawn from Figure 9.1, p. 155, from Daniel C. Dennett, Brainstorms: Philosophical Essays on Mind and Psychology, The MIT Press. Copyright © 1987 by Bradford Books, Publishers. By permission of The MIT Press)

114 The Irreducibility of Consciousness

information from the perceptual module, (3) a control system, which interacts with the memory store by a question-and-answer-process, and which can direct attention to the contents of the perceptual module, and (4) a "public relations" unit, which receives speech act commands from the control system and converts them into public-language utterances.

What might this model explain? Although it is in a very simplified form (as Dennett would concede), it might be fleshed out to provide an explanation of reportability; that is, of our ability to report the contents of our internal states. It also provides the skeleton of an explanation of our ability to bring perceptual information to bear on the control of behavior, to introspect our internal states, and so on. But it tells us nothing about why there should be something it is like to be a system undergoing these processes.

In Consciousness Explained (1991), Dennett puts forward a more sophisticated account that draws on much recent work in cognitive science. The model proposed here is essentially a "pandemonium" model, consisting in many small agents competing for attention, with the agent that shouts the loudest playing the primary role in the direction of later processing. On this model there is no central "headquarters" of control, but multiple channels exerting simultaneous influence. Dennett supplements this account with appeals to neuroscience, evolutionary biology, and connectionist models and production systems in artificial intelligence.

The complexity of this account notwithstanding, it is directed at largely the same phenomena as the earlier account. If successful, it would provide an explanation of reportability, and more generally of the influence of various sorts of information on the control of behavior. It also provides a potential explanation of the focus of attention. It gives a provocative account of some of our cognitive capacities, but it goes no further than the previous model in telling us why there should be conscious experience in the vicinity of these capacities.

Unlike most authors who put forward cognitive models, Dennett claims explicitly that his models are the sort of thing that could explain everything about experience that needs explaining. In particular, he thinks that to explain consciousness, one only needs to explain such functional phenomena as reportability and control; any phenomenon that is apparently omitted is a chimera. Sometimes he seems to take it as a basic premise that once one has explained the various functions, one has explained everything (see, e.g., Dennett [1993a], p. 210), but he occasionally puts forward arguments, some of which I will consider later.9

The same sort of critique could be directed at cognitive-model approaches to consciousness by Churchland (1995), Johnson-Laird (1988), Shallice (1972, 1988a, 1988b), and many others. All provide intriguing accounts of the performance of cognitive functions, but all leave the really hard questions untouched.

Can Consciousness Be Reductively Explained? 115

4. Neurobiological Explanation

Neurobiological approaches to consciousness have recently become popular. Like cognitive models, these have much to offer in explaining psychological phenomena, such as the varieties of awareness. They can also tell us some thing about the brain processes that are correlated with consciousness. But none of these accounts explains the correlation: we are not told why brain processes should give rise to experience at all. From the point of view of neuroscience, the correlation is simply a brute fact.

From a methodological standpoint, it is not obvious how one can begin to develop a neuroscientific theory. How does one perform the experiments that detect a correlation between some neural process and consciousness? What usually happens is that theorists implicitly rely on some psychological criterion for consciousness, such as the focus of attention, the control of behavior, and most frequently the ability to make verbal reports about an internal state. One then notes that some neurophysiological property instantiated when these criteria are present, and one's theory of consciousness is off the ground.

The very fact that such indirect criteria are relied upon, however, makes it clear that no reductive explanation of consciousness is being given. At best, a neurophysiological account might be able to explain why the relevant psychological property is instantiated. The question of why the psychological property in question should be accompanied by conscious experience is left unanswered. Because these theories gain their purchase by assuming a link between psychological properties and conscious experience, it is clear that they do nothing to explain that link. We can see this by examining some recent neuroscientific accounts of consciousness.

Much recent attention in neuroscience has focused on certain 40-hertz oscillations in the visual cortex and elsewhere. Francis Crick and Christof Koch (1990) have hypothesized that this sort of oscillation may be the fundamental neural feature responsible for conscious experience, and have advocated the development of a neurobiological theory along these lines.10

Why 40-hertz oscillations? Primarily because evidence suggests that these oscillations play an important role in the binding of various kinds of information into a unified whole. Two different kinds of information about a scene-the shape and location of an object, for instance-maybe represented quite separately, but this theory suggests that the separate neural representations may have a common frequency and phase in their oscillations, allowing the information to be bound together by later processes and stored in working memory. In this way all sorts of disparate information might be integrated into the "contents of consciousness."

Such a theory might indeed provide neurobiological insight into binding and working memory, and perhaps eventually could be elaborated into an

116 The Irreducibility of Consciousness

account of how information is brought to bear in an integrated way in the control of behavior. But the key question remains unanswered: Why should these oscillations be accompanied by conscious experience? The theory provides a partial answer: because these oscillations are responsible for binding. But the question of why binding itself should be accompanied by experience is not addressed. The theory gains its purchase by assuming a link between binding and consciousness, and therefore does nothing to explain it.

Crick and Koch seem sympathetic with the "big" problem of consciousness, calling it the "major puzzle confronting the neural view of the mind." They argue that pure cognitive-level approaches are doomed to be unsuccessful, and that neural-level theories are required. But they give us no reason to believe that their theory is better suited than cognitive theories when it comes to answering the really difficult questions. Indeed, they do not claim that their project handles the problem of experience. In a published interview, Koch is quite clear about the limitations of the approach:


Well, let's first forget about the really difficult aspects, like subjective feelings, for they may not have a scientific solution. The subjective state of play, of pain, of pleasure, of seeing blue, of smelling a rose-there seems to be a huge jump between the materialistic level, of explaining molecules and neurons, and the subjective level. Let's focus on things that are easier to study-like visual awareness. You're now talking to me, but you're not looking at me, you're looking at the cappuccino, and so you are aware of it. You can say, "It's a cup and there's some liquid in it." If I give it to you, you'll move your arm and you'll take it-you'll respond in a meaningful manner. That's what I call awareness.11


Another neurophysiological theory of consciousness has been outlined by Gerald Edelman in The Remembered Present (1989) and other books and articles. The central element of his theory involves re-entrant neural circuits by which perceptual signals can be conceptually categorized before they contribute to memory. Perceptual information and internal state interact in a subtle way (as depicted in Figure 3.3) to give rise to "primary consciousness." His model of "higher-order consciousness" brings in a new memory element through "semantic bootstrapping," which yields concepts of the self, past, and, future. All this is linked to language production through Broca's and Wernicke's areas.

Much of Edelman's work is devoted to the explanation of perception, memory, and language, rather than of consciousness. Insofar as it is devoted to consciousness, the discussion is often vague, but it seems that what ultimately might be explained by this sort of model is perceptual awareness- that is, the effects of perceptual processing on later processes and on the control of behavior-and aspects of self-consciousness, especially the origin of the concept of the self.

Can Consciousness Be Reductively Explained? 117

Figure 3.3. Edelman's scheme for higher-order consciousness.

Figure 3.3. Edelman's scheme for higher-order consciousness. (Redrawn from Figure 12-4, p. 132, from Bright Air, Brilliant Fire by Gerald M. Edelman. Copyright © 1992 by BasicBooks, Inc. By permission of BasicBooks, a division of HarperCollins, Inc.)

Edelman gives no account of how all this processing should give rise to conscious experience. He simply takes it that there is a correlation. He is up-front about this, noting that phenomenal experience is the hardest problem for a theory of consciousness, and that no physical theory will take us all the way to qualia:


This suggests an approach to the problem of qualia. As a basis for a theory of consciousness, it is sensible to assume that, just as in ourselves, qualia exist in other conscious human beings, whether they are considered as scientific observers or as subjects. ... We can then take human beings to be the best canonical referent for the study of consciousness. This is justified by the fact that human subjective reports (including those about qualia), actions, and brain structures and function can all be correlated. After building a theory based on the assumption that qualia exist in human beings, we can then look anew at some of the properties of qualia based on these correlations. It is our ability to report and correlate while individually experiencing qualia that opens up the possibility of a scientific investigation of consciousness. (Edelman 1992, p. 115)

118 The Irreducibility of Consciousness

As before, because this theory is based on the assumption of correlation, it is clear that a reductive explanation of experience is not on offer. Most of the time Edelman claims only to be explaining the processes that underlie conscious experience; he does not claim to be explaining experience itself.12

5. The Appeal to New Physics

Sometimes it is held that the key to the explanation of consciousness may lie in a new sort of physical theory. Perhaps, in arguing that consciousness is not entailed by the physics of our world, we have been tacitly assuming that the physics of our world is something like physics as we understand it today, consisting in an arrangement of particles and fields in the spatiotemporal manifold, undergoing complex processes of causation and evolution. An opponent might agree that nothing in this sort of physics entails the existence of consciousness, but argue that there might be a new kind of physical theory from which consciousness falls out as a consequence.

It is not easy to evaluate this claim in the absence of any detailed proposal. One would at least like to see an example of how such a new physics might possibly go. Such an example need not be plausible in the light of current theories, but there would have to be a sense in which it would recognizably be physics. The crucial question is: How could a theory that is recognizably a physical theory entail the existence of consciousness? If such a theory consists in a description of the structure and dynamics of fields, waves, particles, and the like, then all the usual problems will apply. And it is unclear that any sort of physical theory could be different enough from this to avoid the problems.

The trouble is that the basic elements of physical theories seem always to come down to two things: the structure and dynamics of physical processes. Different theories invoke different sorts of structure. Newtonian physics invokes a Euclidean space-time; relativity theory invokes a non-Euclidean differential manifold; quantum theory invokes a Hilbert space for wave functions. And different theories invoke different kinds of dynamics within those structures: Newton's laws, the principles of relativity, the wave equations of quantum mechanics. But from structure and dynamics, we can only get more structure and dynamics. This allows the possibility of satisfying explanations of all sorts of high-level structural and functional properties, but conscious experience will remain untouched. No set of facts about physical structure and dynamics can add up to a fact about phenomenology.

Of course, there is a sense in which the physics of the universe must entail the existence of consciousness, if one defines physics as the fundamental science from whose facts and laws everything else follows. This construal of physics, however, trivializes the question involved. If one allows physics to

Can Consciousness Be Reductively Explained? 119

include theories developed specifically to deal with the phenomenon of consciousness, unmotivated by more basic considerations, then we may get an "explanation" of consciousness, but it will certainly not be a reductive one. For our purposes, it is best to take physics to be the fundamental science developed to explain observations of the external world. If this kind of physics entailed the facts about consciousness, without invoking consciousness itself in a crucial role, then consciousness would truly be reductively explained. For the reasons I have given, however, there is good reason to believe that no such reductive explanation is possible.

Almost all existing proposals concerning the use of physics to explain consciousness focus on the most puzzling part of physics, namely quantum mechanics. This is understandable: for physics to explain consciousness would take something extraordinary, and quantum mechanics is by far the most extraordinary part of contemporary physics. But in the end it does not seem to be extraordinary enough.

For example, Penrose (1994) suggests that the key to understanding consciousness may lie in a theory that reconciles quantum theory with the theory of general relativity. He suggests that gravitational effects not yet understood may be responsible for the collapse of the quantum wave function, leading to a nonalgorithmic element in the laws of nature. Drawing on the ideas of Hameroff (1994), he suggests that human cognition may depend on quantum collapses in microtubules, which are protein structures found in the skeleton of a neuron. Indeed, Penrose and Hameroff suggest that quantum collapse in microtubules may be the physical basis of conscious experience.

These ideas are extremely speculative, but they could at least conceivably help to explain certain elements of human cognitive functioning. Penrose suggests that the nonalgorithmic element in collapse could explain certain aspects of our mathematical insight, which he believes goes beyond the capacity of any algorithmic system. Hameroff suggests that the collapse of a superposed wave function might help explain certain aspects of human decision making. But nothing here seems to help with the explanation of conscious experience. Why should quantum processes in microtubules give rise to consciousness? The question here is just as hard as the corresponding question about classical processes in a classical brain. When it comes to the problem of experience, nonalgorithmic and algorithmic processes are in the same boat.

Some have suggested that the nonlocality of quantum mechanics, as suggested by recent experiments bearing on the Einstein-Podolsky-Rosen paradox and Bell's theorem, might be the key to a theory of consciousness (see Lahav and Shanks 1992 for suggestions along these lines). But even if physics is nonlocal, it is hard to see how this should help in the explanation of consciousness. Even given a nonlocal physical process, it remains logically possible that the process could take place in the absence of consciousness. The explanatory gap is as wide as ever.

120 The Irreducibility of Consciousness

The most frequently noted connection between consciousness and quantum mechanics lies in the fact that on some interpretations of the latter, measurement by a conscious observer is required to bring about the collapse of the wave function. On this sort of interpretation, consciousness plays a central role in the dynamics of the physical theory. These interpretations are highly controversial, but in any case it is notable that they do nothing to provide an explanation of consciousness. Rather, they simply assume the existence of consciousness, and use it to help explain certain physical phenomena. Theories of consciousness that exploit this relationship are occasionally put forward (e.g., Hodgson 1988; Stapp 1993), but they are certainly not reductive theories.13

One cannot rule out the possibility that fundamental physical theories such as quantum mechanics will play a key role in a theory of consciousness. For example, perhaps consciousness will turn out to be associated with certain fundamental physical properties, or with certain configurations of those properties, or perhaps there will be a more subtle link. But all the same, there is little hope that this sort of theory will provide a wholly physical explanation of consciousness. When it comes to reductive explanation, physics-based theories are no better off than neurobiological and cognitive theories.

6. Evolutionary Explanation

Even those who take consciousness seriously are often drawn to the idea of an evolutionary explanation of consciousness. After all, consciousness is such a ubiquitous and central feature that it seems that it must have arisen during the evolutionary process for a reason. In particular, it is natural to suppose that it arose because there is some function that it serves that could not be achieved without it. If we could get a clear enough idea of the relevant function, then we would have some idea of why consciousness exists.

Unfortunately, this idea overestimates what an evolutionary explanation can provide us. The process of natural selection cannot distinguish between me and my zombie twin. Evolution selects properties according to their functional role, and my zombie twin performs all the functions that I perform just as well as I do; in particular he leaves around just as many copies of his genes. It follows that evolution alone cannot explain why conscious creatures rather than zombies evolved.

Some may be tempted to respond, "But a zombie couldn't do all the things that I can." But my zombie twin is by definition physically identical to me over its history, so it certainly produces indistinguishable behavior. Anyone wishing to question zombie capacity must therefore find something wrong with the arguments at the start of this chapter, rather than raising the objection here.

Can Consciousness Be Reductively Explained? 121

To see the point in a different way, note that the real problem with consciousness is to explain the principles in virtue of which consciousness arises from physical systems. Presumably these principles-whether they are conceptual truths, metaphysical necessities, or natural laws-are constant over space-time: if a physical replica of me had popped into existence a million years ago, it would have been just as conscious as I am. The connecting principles themselves are therefore independent of the evolutionary process. While evolution can be very useful in explaining why particular physical systems have evolved, it is irrelevant to the explanation of the bridging principles in virtue of which some of these systems are conscious.

7. Whither Reductive Explanation?

It is not uncommon for people to agree with critiques of specific reductive accounts, but to qualify this agreement: "Of course that doesn't explain consciousness, but if we just wait a while, an explanation will come along." I hope the discussion here has made it clear that the problems with this kind of explanation of consciousness are more fundamental than that. The problems with the models and theories presented here do not lie in the details; at least, we have not needed to consider the details in order to see what is wrong with them. The problem lies in the overall explanatory strategy. These models and theories are simply not the sort of thing that could explain consciousness.

It is inevitable that increasingly sophisticated reductive "explanations" of consciousness will be put forward, but these will only produce increasingly sophisticated explanations of cognitive functions. Even such "revolutionary" developments as the invocation of connectionist networks, nonlinear dynamics, artificial life, and quantum mechanics will provide only more powerful functional explanations. This may make for some very interesting cognitive science, but the mystery of consciousness will not be removed.

Any account given in purely physical terms will suffer from the same problem. It will ultimately be given in terms of the structural and dynamical properties of physical processes, and no matter how sophisticated such an account is, it will yield only more structure and dynamics. While this is enough to handle most natural phenomena, the problem of consciousness goes beyond any problem about the explanation of structure and function, so a new sort of explanation is needed.

It might be supposed that there could eventually be a reductive explanatory technique that explained something other than structure and function, but it is very hard to see how this could be possible, given that the laws of physics are ultimately cast in terms of structure and dynamics. The existence of

122 The Irreducibility of Consciousness

consciousness will always be a further fact relative to structural and dynamic facts, and so will always be unexplained by a physical account.

For an explanation of consciousness, then, we must look elsewhere. We certainly need not give up on explanation; we need only give up on reductive explanation. The possibility of explaining consciousness nonreductively remains open. This would be a very different sort of explanation, requiring some radical changes in the way we think about the structure of the world. But if we make these changes, the beginnings of a theory of consciousness may become visible in the distance.

4. Naturalistic Dualism

1. An Argument Against Materialism

In the last chapter, I was concerned with the explanatory question, "Can consciousness be explained by physical theories?" rather than the Ontological question, "Is consciousness itself physical?" But the two questions are closely related, and in this chapter I will draw out the Ontological consequences of the arguments in the last chapter. In particular, the failure of logical Supervenience directly implies that materialism is false: there are features of the world over and above the physical features. The basic argument for this goes as follows.

1. In our world, there are conscious experiences.

2. There is a logically possible world physically identical to ours, in which the positive facts about consciousness in our world do not hold.

3. Therefore, facts about consciousness are further facts about our world, over and above the physical facts.

4. So materialism is false.

If a physically identical zombie world is logically possible, it follows that the presence of consciousness is an extra fact about our world, not guaranteed by the physical facts alone. The character of our world is not exhausted by the character supplied by the physical facts; there is extra character due to the presence of consciousness. To use a phrase due to Lewis (1990), consciousness carries phenomenal information. The physical facts incompletely constrain the way the world is; the facts about consciousness constrain it further.

124 The Irreducibility of Consciousness

A similar conclusion can be drawn from the logical possibility of a world with inverted conscious experiences. Such a world is physically identical to ours, but some of the facts about conscious experience in our world do not hold in that world. It follows that the facts about conscious experience in our world are further facts over and above the physical facts, and that materialism is false.

Either way, if consciousness is not logically supervenient on the physical, then materialism is false. The failure of logical Supervenience implies that some positive fact about our world does not hold in a physically identical world, so that it is a further fact over and above the physical facts. As in Chapter 2, I take materialism to be the doctrine that the physical facts about the world exhaust all the facts, in that every positive fact is entailed by the physical facts. If zombie worlds or inverted worlds are possible, the physical facts do not entail all the positive facts about our world, and materialism is false.

We can use Kripke's image here. When God created the world, after ensuring that the physical facts held, he had more work to do. He had to ensure that the facts about consciousness held. The possibility of zombie worlds or inverted worlds shows that he had a choice. The world might have lacked experience, or it might have contained different experiences, even if all the physical facts had been the same. To ensure that the facts about consciousness are as they are, further features had to be included in the world.

What sort of dualism?

This failure of materialism leads to a kind of dualism: there are both physical and nonphysical features of the world. The falsity of logical Supervenience implies that experience is fundamentally different in kind from any physical feature. But there are many varieties of dualism, and it is important to see just where the argument leads us.

The arguments in the last chapter establish that consciousness does not supervene logically on the physical, but this is not to say that it does not supervene at all. There appears to be a systematic dependence of conscious experience on physical structure in the cases with which we are familiar, and nothing in the arguments of the last chapter suggests otherwise. It remains as plausible as ever, for example, that if my physical structure were to be replicated by some creature in the actual world, my conscious experience would be replicated, too. So it remains plausible that consciousness supervenes naturally on the physical. It is this view-natural Supervenience without logical Supervenience-that I will develop.

The arguments do not lead us to a dualism such as that of Descartes, with a separate realm of mental substance that exerts its own influence on physi-

Naturalistic Dualism 125

cal processes. The best evidence of contemporary science tells us that the physical world is more or less causally closed: for every physical event, there is a physical sufficient cause. If so, there is no room for a mental "ghost in the machine" to do any extra causal work. A small loophole may be opened by the existence of quantum indeterminacy, but I argue later that this probably cannot be exploited to yield a causal role for a nonphysical mind. In any case, for all the arguments in the previous chapter, it remains plausible that physical events can be explained in physical terms, so a move to a Cartesian dualism would be a stronger reaction than is warranted.

The dualism implied here is instead a kind of property dualism: conscious experience involves properties of an individual that are not entailed by the physical properties of that individual, although they may depend lawfully on those properties. Consciousness is a feature of the world over and above the physical features of the world. This is not to say it is a separate "substance"; the issue of what it would take to constitute a dualism of substances seems quite unclear to me. All we know is that there are properties of individuals in this world-the phenomenal properties-that are ontologically independent of physical properties.

There is a weaker sort of property dualism with which this view should not be confused. It is sometimes said that property dualism applies to any domain in which the properties are not themselves properties invoked by physics, or directly reducible to such properties. In this sense, even biological fitness is not a physical property. But this sort of "dualism" is a very weak variety. There is nothing fundamentally ontologically new about properties such as fitness, as they are still logically supervenient on microphysical properties. Property dualism of this variety is entirely compatible with materialism. By contrast, the property dualism that I advocate involves fundamentally new features of the world. Because these properties are not even logically supervenient on microphysical properties, they are nonphysical in a much stronger sense. When I speak of property dualism and nonphysical properties, it is this stronger view and the stronger sense of nonphysicality that I have in mind.

It remains plausible, however, that consciousness arises from a physical basis, even though it is not entailed by that basis. The position we are left with is that consciousness arises from a physical substrate in virtue of certain contingent laws of nature, which are not themselves implied by physical laws. This position is implicitly held by many people who think of themselves as materialists. It is common to hear, "Of course I'm a materialist; the mind certainly arises from the brain." The very presence of the word "arises" should be a tip-off here. One tends not to say "learning arises from the brain," for instance-and if one did, it would be in a temporal sense of "arises." Rather, one would more naturally say that learning is a process in the brain. The very fact that the mind needs to arise from the brain indi-

126 The Irreducibility of Consciousness

cates that there is something further going on, over and above the physical facts.1

Some people will think that the view should count as a version of materialism rather than dualism, because it posits such a strong lawful dependence of the phenomenal facts on the physical facts, and because the physical domain remains autonomous. Of course there is little point arguing over a name, but it seems to me that the existence of further contingent facts over and above the physical facts is a significant enough modification to the received materialist world view to deserve a different label. Certainly, if all that is required for materialism is that all facts be lawfully connected to the physical facts, then materialism becomes a weak doctrine indeed.

Although it is a variety of dualism, there is nothing antiscientific or supernatural about this view. The best way to think about it is as follows. Physics postulates a number of fundamental features of the world: space-time, mass-energy, charge, spin, and so on. It also posits a number of fundamental laws in virtue of which these fundamental features are related. Fundamental features cannot be explained in terms of more basic features, and fundamental laws cannot be explained in terms of more basic laws; they must simply be taken as primitive. Once the fundamental laws and the distribution of the fundamental features are set in place, however, almost everything about the world follows. That is why a fundamental theory in physics is sometimes known as a "theory of everything." But the fact that consciousness does not supervene on the physical features shows us that this physical theory is not quite a theory of everything. To bring consciousness within the scope of a fundamental theory, we need to introduce new fundamental properties and laws.

In his book Dreams of a Final Theory (1992), physicist Steven Weinberg notes that what makes a fundamental theory in physics special is that it leads to an explanatory chain all the way up, ultimately explaining everything. But he is forced to concede that such a theory may not explain consciousness. At best, he says, we can explain the "objective correlates" of consciousness. "That may not be an explanation of consciousness, but it will be pretty close" (p. 45). But it is not close enough, of course. It does not explain everything that is happening in the world. To be consistent, we must acknowledge that a truly final theory needs an additional component.

There are two ways this might go. Perhaps we might take experience itself as a fundamental feature of the world, alongside space-time, spin, charge, and the like. That is, certain phenomenal properties will have to be taken as basic properties. Alternatively, perhaps there is some other class of novel fundamental properties from which phenomenal properties are derived. Previous arguments have shown that these cannot be physical properties, but perhaps they are nonphysical properties of a new variety, on which phenomenal properties are logically supervenient. Such properties would be related

Naturalistic Dualism 127

to experience in the same way that basic physical properties are related to nonbasic properties such as temperature. We could call these properties protophenomenal properties, as they are not themselves phenomenal but together they can yield the phenomenal. Of course it is very hard to imagine what a protophenomenal property could be like, but we cannot rule out the possibility that they exist. Most of the time, however, I will speak as if the fundamental properties are themselves phenomenal.

Where we have new fundamental properties, we also have new fundamental laws. Here the fundamental laws will be Psychophysical laws, specifying how phenomenal (or protophenomenal) properties depend on physical properties. These laws will not interfere with physical laws; physical laws already form a closed system. Instead, they will be Supervenience laws, telling us how experience arises from physical processes. We have seen that the dependence of experience on the physical cannot be derived from physical laws, so any final theory must include laws of this variety.

Of course, at this stage we have very little idea what the relevant fundamental theory will look like, or what the fundamental Psychophysical laws will be. But we have reason to believe that such a theory exists. There is good reason to believe that there is a lawful relationship between physical processes and conscious experience, and any lawful relationship must be supported by fundamental laws. The case of physics tells us that fundamental laws are typically simple and elegant; we should expect the same of the fundamental laws in a theory of consciousness. Once we have a fundamental theory of consciousness to accompany a fundamental theory in physics, we may truly have a theory of everything. Given the basic physical and Psychophysical laws, and given the distribution of the fundamental properties, we can expect that all the facts about the world will follow. Developing such a theory will not be straightforward, but it ought to be possible in principle.

In a way, what is going on here with consciousness is analogous to what happened with electromagnetism in the nineteenth century. There had been an attempt to explain electromagnetic phenomena in terms of physical laws that were already understood, involving mechanical principles and the like, but this was unsuccessful. It turned out that to explain electromagnetic phenomena, features such as electromagnetic charge and electromagnetic forces had to be taken as fundamental, and Maxwell introduced new fundamental electromagnetic laws. Only this way could the phenomena be explained. In the same way, to explain consciousness, the features and laws of physical theory are not enough. For a theory of consciousness, new fundamental features and laws are needed.

This view is entirely compatible with a contemporary scientific worldview, and is entirely naturalistic. On this view, the world still consists in a network of fundamental properties related by basic laws, and everything is to be

128 The Irreducibility of Consciousness

ultimately explained in these terms. All that has happened is that the inventory of properties and laws has been expanded, as happened with Maxwell. Further, nothing about this view contradicts anything in physical theory; rather, it supplements that theory. A physical theory gives a theory of physical processes, and a Psychophysical theory tells us how those processes give rise to experience.

To capture the spirit of the view I advocate, I call it naturalistic dualism. It is naturalistic because it posits that everything is a consequence of a network of basic properties and laws, and because it is compatible with all the results of contemporary science. And as with naturalistic theories in other domains, this view allows that we can explain consciousness in terms of basic natural laws. There need be nothing especially transcendental about consciousness; it is just another natural phenomenon. All that has happened is that our picture of nature has expanded. Sometimes "naturalism" is taken to be synonymous with "materialism," but it seems to me that a commitment to a naturalistic understanding of the world can survive the failure of materialism. (If a reader doubts this, I point to the rest of this work as evidence.) Some might find a certain irony in the name of the view, but what is most important is that it conveys the central message: to embrace dualism is not necessarily to embrace mystery.

In some ways, those who hold this sort of dualism may be temperamentally closer to materialists than to dualists of other varieties. This is partly because of its avoidance of any transcendental element and its commitment to natural explanation, and partly because of its commitment to the physical causation of behavior. Conversely, by avoiding any commitment to a ghost in the machine, this view avoids the worst implausibilities of the traditional dualist views. One often hears that the successes of cognitive science and neuroscience make dualism implausible, but not all varieties of dualism are affected equally. These successes are all grounded in physical explanations of behavior and of other physical phenomena, and so do not distinguish between the materialist and the naturalistic dualist view.

Two final notes. Some will wonder why, if experience is fundamental, it does not qualify as a physical property. After all, is not physics just the science of what is truly fundamental? In reply: Certainly if we define physics that way, experience will indeed qualify as a physical property, and the Supervenience laws will count as laws of physics. But on a more natural reading of "physics" and "physical," experience does not qualify. Experience is not a fundamental property that physicists need to posit in their theory of the external world; physics forms a closed, consistent theory even without experience. Given the possibility of a zombie world, there is a clear sense in which experience is superfluous to physics as it is usually understood. It is therefore more natural to consider experience as a fundamental property

Naturalistic Dualism 129

that is not a physical property, and to consider the Psychophysical laws as fundamental laws of nature that are not laws of physics. But nothing much turns on the terminological issue, as long as the shape of the view is clear. I should also note that although I call the view a variety of dualism, it is possible that it could turn out to be a kind of monism. Perhaps the physical and the phenomenal will turn out to be two different aspects of a single encompassing kind, in something like the way that matter and energy turn out to be two aspects of a single kind. Nothing that I have said rules this out, and in fact I have some sympathy with the idea. But it remains the case that if a variety of monism is true, it cannot be a materialist monism. It must be something broader.


There are a number of objections that might be raised to the argument against materialism at the beginning of this chapter. Some of these are objections to premise (2), the denial of logical Supervenience; I have dealt with objections of that sort in the previous chapter. Here, I will deal with objections to the step from the failure of logical Supervenience to the falsity of materialism. The most serious objections of this sort are those that invoke a posteriori necessity. I will deal with these in the next section. Here I will deal with some more minor objections.

Sometimes it is argued that consciousness might be an emergent property, in a sense that is still compatible with materialism. In recent work on complex systems and artificial life, it is often held that emergent properties are unpredictable from low-level properties, but that they are physical all the same. Examples are the emergence of self-organization in biological systems, or the emergence of flocking patterns from simple rules in simulated birds (Lang-ton 1990; Reynolds 1987). But emergent properties of this sort are not analogous to consciousness. What is interesting about these cases is that the relevant properties are not obvious consequences of low-level laws; but they are still logically supervenient on low-level facts. If all the physical facts about such a system over time are given, then the fact that self-organization is occurring will be straightforwardly derivable. This is just what we would expect, as properties such as self-organization and flocking are straightforwardly functional and structural.

If consciousness is an emergent property, it is emergent in a much stronger sense. There is a stronger notion of emergence, used by the British emergentists (e.g., Broad [1925]), according to which emergent properties are not even predictable from the entire ensemble of low-level physical facts. It is reasonable to say (as the British emergentists did) that conscious experience is emergent in this sense. But this sort of emergence is best counted as a

130 The Irreducibility of Consciousness

variety of property dualism. Unlike the more "innocent" examples of emergence given above, the strong variety requires new fundamental laws in order that the emergent properties emerge.

Another objection is that consciousness and the physical might be two aspects of the same thing, in the way that the morning star and the evening star are two aspects of Venus. If so, consciousness might in a sense be physical. But again, we have to ask: Is the phenomenal aspect entailed by the physical aspect? If it is, we have a variety of materialism, but we are back to the arguments in Chapter 3. If it is not, then the phenomenal aspect provides further contingency in the world over and above the physical aspect, and the duality of the aspects gives us a kind of property dualism. Perhaps it may turn out that the duality of the physical and the phenomenal can be subsumed under a grander monism, but this will not be a monism of the physical alone.

A third objection is suggested by the work of Searle (1992). Like me, Searle holds that consciousness is merely naturally supervenient on the non-physical. He allows that a zombie replica is logically possible, holding that consciousness is merely caused by states of the brain. But he denies that this is a variety of dualism, even property dualism. This might seem to be a mere terminological issue, but Searle insists that the Ontological status of consciousness is the same as that of physical features such as liquidity, so the issue is not merely terminological. Searle's argument that the view is not dualistic is that a similar story holds elsewhere: for example, H2O causes liquidity, but no one is a dualist about liquidity.

It seems clear that this is a false analogy, however. Given all the microphysical facts about a particular batch of H2O, it is logically impossible that those facts could hold without liquidity being instantiated. The notion of a nonliquid replica of a batch of liquid H2O is simply incoherent. It follows that the relation between the microphysical facts and liquidity is much tighter than a simple causal relation. The microphysical features do not cause liquidity; they constitute it. This is entirely different from what is going on in the case of consciousness, so the analogy fails. Consciousness is ontologically novel in a much more significant way than liquidity.2

Finally, some will find the argument for dualism that I have given reminiscent of the argument given by Descartes. Descartes argued that he could imagine his mind existing separately from his body, so his mind could not be identical to his body. This sort of argument is generally regarded to be flawed: just because one can imagine that A and B are not identical, it does not follow that A and B are not identical (think of the morning star and the evening star, for example). Might not my argument make a similar mistake? The zombie world only shows that it is conceivable that one might have a physical state without consciousness; it does not show that a physical state and consciousness are not identical.

Naturalistic Dualism 131

This is to misunderstand the argument, however. It is crucial that the argument as I have put it does not turn on questions of identity but of Supervenience. The form of the argument is not, "One can imagine physical state P without consciousness, therefore consciousness is not physical state P." The form of the argument is rather, "One can imagine all the physical facts holding without the facts about consciousness holding, so the physical facts do not exhaust all the facts." This is an entirely different sort of argument. In general, modal arguments for dualism that are cast in terms of identity are less conclusive than modal arguments cast in terms of Supervenience; this is one reason why I have put things in terms of Supervenience throughout, and avoided talk of identity almost entirely. It seems to me that the issues about Supervenience are the most fundamental here.

One might nevertheless try to reply to this argument with a strategy analogous to the reply to Descartes. For example, one might note that my strategy still relies on a sort of inference from conceivability to possibility that might be questioned. I consider strategies along these lines in the next section.

2. Objections From A Posteriori Necessity*

A popular response to this sort of argument is to object that it only establishes that a zombie world is logically possible, which is quite different from being metaphysically possible. Whereas conceptual coherence suffices for logical possibility, metaphysical possibility is more constrained. The point is also often made by suggesting that there is a difference between conceivability and true possibility. Although it may be the case that a zombie world is conceivable, something more is required in order to show that it is possible in the metaphysical sense relevant to the falsity of materialism.

This objection is most often accompanied by an appeal to Kripke's Naming and Necessity (1980), which demonstrates the existence of necessary truths such as "Water is H2O" whose necessity is only knowable a posteriori. In the terms of these objectors, it is logically possible that water is not H2O, but it is not metaphysically possible. It is not unnatural to suppose that zombies might be logically possible but metaphysically impossible in a similar way. If so, this would arguably be enough to save materialism.

This is by far the most common strategy of materialists who are persuaded that there is no entailment between physical and phenomenal concepts. On this view, there can be a conceptual gap without a metaphysical gap. The view offers the enticing prospect of taking consciousness seriously while nevertheless holding on to materialism. Unfortunately, upon close examination the view can be seen quite straightforwardly to fail. The notion of a posteriori necessity cannot carry the burden that this argument requires, and in fact is something of a red herring in this context.3

132 The Irreducibility of Consciousness

We can best see this by using the two-dimensional framework for dealing with a posteriori necessity developed in Chapter 2, section 4. Recall that in this framework there are two intensions (functions from possible worlds to referents) associated with any concept: a primary intension (determined a priori) that fixes reference in the actual world, and a secondary intension (determined a posteriori) that picks out reference in counterfactual worlds. The primary intension associated with "water" is something like "watery stuff." The secondary intension is "H2O," which is derived from the primary intension by applying Kaplan's dthat operator: "dthat(watery stuff)" picks out H2O in all possible worlds, as watery stuff is H2O in the actual world.

"Logical possibility" comes down to the possible truth of a statement when evaluated according to the primary intensions involved (what I called 1-possibility in Chapter 2). The primary intensions of "water" and "H2O" differ, so it is logically possible in this sense that water is not H2O. "Metaphysical possibility" comes down to the possible truth of a statement when evaluated according to the secondary intensions involved (that is, 2-possibility). The secondary intensions of "water" and "H2O" are the same, so it is metaphysically necessary that water is H2O.

The objection therefore comes down to the point that in using arguments from conceivability and the like, we have demonstrated the possibility of a zombie world using the primary intensions of the notions involved, but not using the more appropriate secondary intensions. While the primary intension of phenomenal notions may not correspond to that of any physical notion, the secondary intensions may be the same. If so, then phenomenal and physical/functional concepts may pick out the same properties a posteriori despite the a priori distinction. Such an objection might be made by an advocate of "psychofunctionalism" (see Block 1980), which equates phenomenal properties with functional properties a posteriori, or by an advocate of a view that equates phenomenal properties with certain neurophysiological properties a posteriori.

The easiest way to see that none of this affects the argument for dualism is to note that the argument I have given goes through if we concentrate on the primary intension throughout and ignore the secondary intension. We saw in Chapter 2 that it is the primary intension that is most relevant to explanation, but it also serves us well in the argument for dualism. For note that whether or not the primary and secondary intensions coincide, the primary intension determines a perfectly good property of objects in possible worlds. The property of being watery stuff is a perfectly reasonable property, even though it is not the same as the property of being H2O. If we can show that there are possible worlds that are physically identical to ours but in which the property introduced by the primary intension is lacking, then dualism will follow.

This is just what has been done with consciousness. We have seen that there are worlds just like ours that lack consciousness, according to the

Naturalistic Dualism 133

primary intension thereof. This difference in worlds is sufficient to show that there are properties of our world over and above the physical properties. By analogy, if we could show that there were worlds physically identical to ours in which there was no watery stuff, we would have established dualism about water just as well as if we had established that there were worlds physically identical to ours in which there was no H2O. And importantly, the difference with respect to the primary intension can be established independently of a posteriori factors, so that considerations about a posteriori necessity are irrelevant.

(Two technical notes here: Strictly speaking, a primary intension determines a center-relative property of an object in a possible world (or a relation between objects and centers), as the primary intension applies to centered possible worlds. But this relativity cannot be exploited to help our objector. Once the location of a center is specified, a primary intension determines a perfectly good nonindexical property; and all the arguments of Chapter 3 go through even when the location of the center is included in the Supervenience base. For example, even if Mary's facts about the world include facts about where she is located, this will not enable her to know what it is like to see red.

One might also be worried by the fact that the concept of consciousness is arguably not present at the center of the zombie world, whereas the application of a primary intension might require the presence of the relevant concept at the center of the world. (One might even start worrying about the application of the zombie's concept!) I think the situation is more subtle than this-primary intensions need not require the presence of the original concept-but in any case, we can bypass this worry altogether simply by considering a partial zombie world: one in which I am at the center, conscious, with all the relevant concepts, but in which some other people are zombies.)

The irrelevance of a posteriori necessity can be further supported by the observation that with consciousness, the primary and secondary intensions coincide. What it takes for a state to be a conscious experience in the actual world is for it to have a phenomenal feel, and what it takes for something to be a conscious experience in a counterfactual world is for it to have a phenomenal feel. The difference between the primary and secondary intensions for the concept of water reflects the fact that there could be something that looks and feels like water in some counterfactual world that in fact is not water, but merely watery stuff. But if something feels like a conscious experience, even in some counterfactual world, it is a conscious experience. All it means to be a conscious experience, in any possible world, is to have a certain feel. (Kripke makes a similar point, although he puts the point in terms of essential properties rather than in terms of meaning.)

Even if someone insists that the primary and the secondary intensions differ, however, the argument still goes through. We simply focus on the primary intension used to fix reference, as above. For instance, if "con-

134 The Irreducibility of Consciousness

sciousness" comes to "dthat (has a phenomenal feel)", then we simply focus on the intension "has a phenomenal feel." The arguments in Chapter 3 establish that there is a possible world in which my replica lacks a phenomenal feel, so the property of having a phenomenal feel is a fact over and above the physical facts, and the argument for dualism is successful.4

The most general way to make the point is to note that nothing about Kripke's a posteriori necessity renders any logically possible worlds impossible. It simply tells us that some of them are misdescribed, because we are applying terms according to their primary intensions rather than the more appropriate secondary intensions. One might have thought it possible a priori that water is XYZ, rather than H2O. In conceiving this, one imagines something like a world in which XYZ is the liquid found in oceans and lakes. However, Kripke's analysis shows us that due to the way the actual world turns out, we are misdescribing this world as one in which XYZ is water, as we are describing it with the primary intension instead of the more appropriate secondary intension. Strictly speaking, it is a world in which XYZ is watery stuff. These considerations cannot show the impossibility of this apparently possible world; they simply show us the correct way to describe it.

As we saw in Chapter 2, Kripkean considerations show us that the secondary intension Fa: W -> R sometimes differs from the primary intension f : W* -> R. This puts some a posteriori constraints on the application conditions of concepts, but the relevant space of worlds stays constant throughout; the only difference between the arguments of the two functions involves the location of a center. So although there may be two kinds of possibility of statements, there is only one relevant kind of possibility of worlds.

It follows that if there is a conceivable world that is physically identical to ours but which lacks certain positive features of our world, then no considerations about the designation of terms such as "consciousness" can do anything to rule out the metaphysical possibility of the world. We can simply forget the semantics of these terms, and note that the relevant possible world clearly lacks something, whether or not we call it "consciousness." The Kripkean considerations might tell us at best how this world and the relevant features should be appropriately described, but they have no effect on its possibility; and the mere possibility of such a world, no matter how it is described, is all the argument for dualism needs to succeed.

An alternative strategy

There is a quite different way in which one might appeal to a posteriori necessity in order to avoid dualism. It might be argued that to claim that the zombie world is physically identical to ours is to misdescribe it. Just as the XYZ world seems to contain water but does not, the zombie world seems

Naturalistic Dualism 135

physically identical while being physically different. This may sound strange, but there is a way to cash it out. An opponent might argue that there are properties essential to the physical constitution of the world that are not accessible to physical investigation. In conceiving of a "physically identical" world, we are really only conceiving of a world that is identical from the standpoint of physical investigation, while differing in the inaccessible essential properties, which are also the properties that guarantee consciousness.

For example, it might be that for something to qualify as an electron in a counterfactual world, it is not sufficient that it be causally related to other physical entities in the way that an electron is. Some hidden essence of electronhood might also be required. On this view, the concept of an electron is something like "dthat(the entity that plays the electron role)." Reference to electrons is fixed by an extrinsic characterization, but is then rigidified so that entities with the same intrinsic nature are picked out in counterfactual worlds, regardless of whether they play the appropriate role, and so that entities that play the role in those worlds do not qualify as electrons unless they have the appropriate intrinsic nature. The same might go for properties such as mass, which might be understood as "dthat(the property that plays the mass role)." The essential nature of electrons or of mass would then be hidden to physical theory, which characterizes electrons and mass only extrinsically. If so, it might be that the relevant essential properties are themselves phenomenal or protophenomenal properties, so that their instantiation could guarantee the existence of consciousness in our world.

If this were the case, the zombie world that we are conceiving would lack these hidden essential properties and would therefore fail to be physically identical to our world. The zombie world would give the same results as our world when evaluated according to the primary intensions of physical predicates, which apply on the basis of extrinsic relations, but not when evaluated according to the secondary intensions, which require the hidden essence. Given this, conscious experience might supervene "metaphysically" on physical properties after all. (An argument very much like this is given by Maxwell [1978], and is also suggested by the approach in Lockwood [1989]. As Maxwell puts it, the basic idea is that even though phenomenal concepts cannot be given topic-neutral analyses that pick out underlying physical properties, physical concepts can be given topic-neutral analyses that might pick out underlying phenomenal properties.5)

This is in many ways a more interesting objection than the previous one. It certainly relies on a speculative metaphysics, but this does not prevent it from being a coherent position. A more direct reply is that it relies on an incorrect view of the semantics of physical terms. Arguably, physical predicates apply even a posteriori on the basis of extrinsic relations between physical entities, irrespective of any hidden properties. This is a purely conceptual question: if electrons in our world have hidden protophenomenal

136 The Irreducibility of Consciousness

properties, would we call an otherwise identical counterfactual entity that lacks those properties an electron? I think we would. Not only is reference to electrons fixed by the role that electrons play in a theory; the very concept of an electron is defined by that role, which determines the application of the concept across all worlds. The notion of an electron that has all the extrinsic properties of actual protons does not appear to be coherent, and neither does the notion that there is a world in which mass plays the role that charge actually plays. The semantic account given above predicts that these notions should be coherent, and so gives a false account of the concepts.

Semantic intuitions may differ, but as usual there is a reply that runs deeper than the semantic intuitions. Even if we allow that certain hidden properties could be constitutive of physical properties, the difference between this view and the property dualism that I have advocated is small. It remains the case that the world has phenomenal properties that are not fixed by the properties that physics reveals. After ensuring that a world is identical to ours from the standpoint of our physical theories, God has to expend further effort to make that world identical to ours across the board. The dualism of "physical" and "nonphysical" properties is replaced on this view by a dualism of "accessible" and "hidden" physical properties, but the essential point remains.

The view that physical entities have an intrinsic protophenomenal nature is one to which I will return, but the metaphysics of the view remains much the same regardless of the approach we take to the semantics of physical predicates. As before, secondary intensions and a posteriori necessity make only a semantic and not a metaphysical difference. However the view is spelled out, it admits phenomenal or protophenomenal properties as fundamental, and so remains closer to a version of dualism (or perhaps an idealism or a neutral monism, as I discuss later) than to a version of materialism.

Strong metaphysical necessity

The two-dimensional analysis just discussed establishes that an invocation of Kripkean a posteriori necessity has no force against the argument from Supervenience. This sort of necessity does not put a posteriori constraints on the space of possible worlds, but merely constrains the way in which certain terms are used to describe them; so if there is a logically possible world that is identical to ours in all physical respects but not in all positive respects, then these considerations cannot count against the world's metaphysical possibility.

Some may claim, however, that the relevant worlds might be metaphysically impossible nevertheless. It could be held that there is a modality of metaphysical possibility that is distinct from and more constrained than

Naturalistic Dualism 137

logical possibility, and that arises for reasons independent of the Kripkean considerations. On this view, there are fewer metaphysically possible worlds than there are logically possible worlds, and the a posteriori necessity of certain statements can stem from factors quite independent of the semantics of the terms involved. We can call this hypothesized modality strong metaphysical necessity, as opposed to the weak metaphysical necessity introduced by the Kripkean framework.

On this view, there are worlds that are entirely conceivable, even according to the strongest strictures on conceivability, but which are not possible at all. This is a gap between conceivability and possibility much stronger than any gap found elsewhere. There is a sense in which the truth of statements such as "Water is XYZ" is conceivable but not possible, but these examples never rule out the possibility of any conceivable world. They are merely instances in which such a world is misdescribed. Strong metaphysical necessity goes beyond this. On this position, "zombie world" may correctly describe the world that we are conceiving, even according to a secondary intension. It is just that the world is not metaphysically possible.6

The short answer to this objection is that there is no reason to believe that such a modality exists. Such "metaphysical necessities" will put constraints on the space of possible worlds that are brute and inexplicable. It may be reasonable to countenance brute, inexplicable facts about our world, but the existence of such facts about the space of possible worlds would be quite bizarre. The realm of the possible (as opposed to the realm of the natural) has no room for this sort of arbitrary constraint.

The position cannot be supported by analogy, as no analogies are available.7 We have already seen that analogies with the necessity of "Water is H2O," "Hesperus is Phosphorus," and so on fail, as these examples require only a single space of worlds. Indeed, if some worlds are logically possible but metaphysically impossible, it seems that we could never know it. By assumption the information is not available a priori, and a posteriori information only tells us about our world. This can serve to locate our world in the space of possible worlds, but it is hard to see how it could give information about the extent of that space. Any claims about the added constraints of metaphysical possibility would seem to be a matter of arbitrary stipulation; one might as well stipulate that it is metaphysically impossible that a stone could move upward when one lets go of it.

Further, the position leads to an ad hoc proliferation of modalities. If it were accepted, we would have to countenance four kind of possibility and necessity of statements, even leaving the natural modality aside: possibility and necessity according to primary or secondary intensions, over the space of logically possible or metaphysically possible worlds. And considering the possibility of worlds rather than statements, we would now have three objective classes of possible worlds: logically possible worlds, metaphysically pos-

138 The Irreducibility of Consciousness

sible worlds, and naturally possible worlds. We have good reason to believe in the first and the last of these classes, but we have very little reason to believe in a third, distinct class as a metaphysical given.

Someone who holds that a zombie world is logically possible but metaphysically impossible has to answer the key question: Why couldn't God have created a zombie world? Presumably it is in God's powers, when creating the world, to do anything that is logically possible. Yet the advocate of metaphysical necessity must say either the possibility is coherent, but God could not have created it, or God could have created it, but it is nevertheless metaphysically impossible. The first is quite unjustified, and the second is entirely arbitrary. If the second holds, in any case, an argument against materialism still goes through; after fixing the physical facts about the world, God still had more work to do.

Even if this view were accepted, it would look very much like the property dualism I advocate, in many crucial respects. On this view, it would still be the case that the existence of consciousness cannot be derived from physical knowledge, so that consciousness cannot be reductively explained. And it would remain the case that we would need certain primitive connecting principles to explain the Supervenience of the phenomenal on the physical. The only difference between the views is that the relevant Psychophysical principles are deemed to be brute "laws of necessity" rather than laws of nature. For all explanatory purposes in constructing a theory, we are left in the same position in which property dualism leaves us; the main difference is in an Ontological stipulation.

The only real motivation for this view would seem to be to save materialism at all costs, perhaps because of perceived problems with dualism. But this sort of materialism seems far more mysterious than the dualist alternative. The invocation of brute "metaphysically necessary" principles constraining the space of possible worlds introduces an element much more problematic, and indeed far less naturalistic, than the mere invocation of further natural laws postulated by property dualism. In the end, the invocation of a new degree of necessity is a sort of solution by ad hoc stipulation that raises as many problems as it answers. The view saves materialism only at the cost of making it entirely mysterious how consciousness could be physical.8

Cognitive limitations

There is a final position that might be taken by a materialist who finds the zombie world conceivable but still wants to save materialism. In the position discussed above, the materialist accepts that the zombie notion is entirely coherent, even to a maximally rational being, but nevertheless denies its metaphysical possibility, thus leading to a "two-layered" picture of logically

Naturalistic Dualism 139

and metaphysically possible worlds. But a materialist might also argue that the apparent conceivability arises from some sort of impaired rationality, so that if we were only more intelligent we would see that the description of the world is not coherent after all. On this view, the world is not really even logically possible; it is just that the limitations of human cognitive faculties mislead us into believing that it is. (This may be one interpretation of the position of McGinn [1991].)

One might try to support this position by analogy with the necessity of certain complex mathematical truths that lie beyond our powers of mathematical insight. If our mathematical powers are computable, such truths must exist (by Gödel's theorem), and even if not, they may well exist all the same. (Perhaps Goldbach's conjecture is an example, or perhaps the continuum hypothesis or its negation.) These truths are necessary even though they are not knowable a priori by us, nor are they grounded in a combination of a priori knowable and empirical factors in the manner of Kripkean necessities. Perhaps the implication from physical facts to phenomenal facts is a necessity of this form, somehow beyond our powers of modal comprehension?9

The analogy is imperfect, however. In the mathematical case, our modal reasoning leaves the matter open; our conceivability intuitions do not tell us anything one way or the other. There may be some weak sense in which it is "conceivable" that the statements are false-for example, they are false for all we know-but this is not a sense that delivers a conceivable world where they fail. In the zombie case, by contrast, the matter is not left open: there seems to be a clearly conceivable world in which the implication is false. To save materialism, the possibility of this world has to be ruled out despite the best evidence of our modal powers; but nothing in the mathematical case comes close to providing an example whereby an apparently possible world is ruled out in this way. Once again, any gap between conceivability and possibility that the materialist might invoke here must be sui generis, unsupported by relevant analogies elsewhere.10

Of course, a materialist might bite the bullet and make a case for a sui generis cognitive impairment. To do this, she would have to hold that the arguments in Chapter 3 all go wrong in ways that we cannot appreciate. Apart from requiring that imperfect rationality leads our conceivability intuitions to go massively astray, the view also requires that a smarter version of Mary really could know what it is like to see red on the basis of physical information, and that there is an analysis of phenomenal concepts to support the implication from physical to phenomenal facts (perhaps a structural or functional analysis), although one whose correctness lies beyond our powers of appreciation.

While it must be conceded that any philosophical argument could go wrong because of cognitive impairment, in the absence of any substantial reason

140 The Irreducibility of Consciousness

to believe this, this sort of objection seems quite ad hoc. As before, the main motivation would seem to be a desire to hang onto materialism at all costs. Such an option should always be the last option considered, only after we have given up on both substantial arguments pointing out where we have gone wrong and substantial attempts to develop an alternative to materialism. If we find a substantial alternative that is satisfactory, then any motivation for this view will disappear.11

3. Other Arguments for Dualism*

I am not the first to use the argument from logical possibility against materialism.12 Indeed, I think that in one form or another it is the fundamental antimaterialist argument in the philosophy of mind. Nevertheless, it has not received the careful attention it deserves. More attention has focused on two antimaterialist arguments by Jackson (1982) and Kripke (1972). These arguments strike me as related, but as perhaps less fundamental. Jackson's argument is important for the entry it provides to the argument from logical Supervenience, and the most compelling portion of Kripke's argument depends on the argument from logical Supervenience, as we will see.

Jackson's argument

I have already discussed Jackson's argument, the knowledge argument, in the context of establishing the failure of logical Supervenience, where it plays a supporting role. Recall that the argument is concerned with Mary, a neuroscientist brought up in a black-and-white room, who knows all the physical facts about color processing in the brain. Later, when she first sees a red object, she learns some new facts. In particular, she learns what it is like to see red. The argument concludes that the physical facts do not exhaust all the facts, and that materialism is false.

This argument is closely related to the arguments from zombies or inverted spectra, in that both revolve around the failure of phenomenal facts to be entailed by physical facts. In a way, they are flip sides of the same argument. As a direct argument against materialism, however, Jackson's argument is often seen as vulnerable due to its use of the intensional notion of knowledge. Many attacks on the argument have centered on this intensionality- arguing, for example, that the same fact can be known in two different ways. These attacks fail, I think, but the most straightforward way to see this is to proceed directly to the failure of Supervenience, which is cast in terms of metaphysics rather than epistemology. The framework I have developed helps bring out just why the various objections do not succeed. I will discuss some of these objections in what follows.

Naturalistic Dualism 141

First, various respondents have argued that although Mary gains new knowledge upon seeing red, this knowledge does not correspond to any new fact. She simply comes to know an old fact in a new way, under a new "mode of presentation," due to the intensionality of knowledge (Churchland 1985; Horgan 1984b; Lycan 1995; McMullen 1985; Papineau 1993; teller 1992; Tye 1986). For example, Tye and Lycan appeal to the intensional difference between "This liquid is water" and "This liquid is H2O": in a sense these express the same fact, but one can be known without the other. Similarly, Churchland appeals to the gap between knowledge of temperature and knowledge of mean kinetic energy, Horgan discusses the difference between knowledge of Clark Kent and knowledge of Superman, while McMullen points to Mark Twain and Samuel Clemens.

These gaps arise precisely because of the difference between primary and secondary intensions. One can know things about water without knowing things about H2O because the primary intensions differ-there is no a priori connection between water thoughts and H2O thoughts. Nevertheless, in a sense there is only one set of facts about the two: because of the a posteriori identity between water and H2O, the relevant secondary intensions coincide. (It is not obvious that one has to individuate facts this way, so that water facts and H2O facts are the same facts, but I will go along with this for the sake of argument.13) In the terminology used earlier, "If this is water, it is H2O" is logically contingent but metaphysically necessary. This objection therefore comes to precisely the same thing as the objection from the distinction between logical necessity and (Kripkean) metaphysical necessity discussed earlier, and the discussion there of primary and secondary intensions is sufficient to refute it.

We can also put the point a more direct way. Whenever one knows a fact under one mode of presentation but not under another, there will always be a different fact that one lacks knowledge of-a fact that connects the two modes of presentation.14 If one knows that Hesperus is visible but not that Phosphorus is visible (because one does not know that Hesperus is Phosphorus), then one does not know that one object is both the brightest star in the morning sky and the brightest star in the evening sky. This is a separate fact that one lacks knowledge of entirely. Similarly, if one knows that Superman can fly but not that Clark Kent can fly, then one does not know that there is an individual who is both the lead reporter at the Daily Planet and who wears a cape. If one knows that water is wet but not that H2O is wet, one does not know that the stuff in the lakes is made out of H2O molecules. And so on.

More formally: Say that "a is G" and "b is G" are the same fact in this sense, but one cannot connect the two facts a priori. This must be because a = b and the secondary intensions are the same, but the primary intensions are different: perhaps a is equivalent to dthat{P) and b to dthat(Q). If one

142 The Irreducibility of Consciousness

knows that a is G but not that b is G, then one lacks the factual knowledge that something is both P and Q. More generally, one lacks the factual knowledge that something is both P' and Q', where these are any identifying descriptions such that one knows that a is P' and that b is Q'. This fact is quite separate from the facts that one initially possessed. Even when interpreted according to secondary intensions, there will be a possible world in which a is F but in which nothing is both P and Q (or both P' and Q').

(As in section 2, there is the complication that P and Q may be index-relative properties, but this changes nothing fundamental. To make the unknown novel fact strictly nonindexical, one need only move to the fact "There exists a point [with property X] from which P and Q pick out the same thing." X is a backup just in case one knows of some other location from which P and Q pick out the same thing; in such a case, we simply make X specific enough to distinguish oneself from those other locations. The extreme case where one lacks any distinguishing self-knowledge reduces to the pure indexical case, discussed below.)

It follows that if Mary gains any factual knowledge that she previously lacked-even if it is only knowledge of an old fact under a different mode of presentation-then there must be some truly novel fact that she gains knowledge of. In particular, she must come to know a new fact involving that mode of presentation. Given that she already knew all the physical facts, it follows that materialism is false. The physical facts are in no sense exhaustive.

This rejoinder may seem less straightforward than the corresponding rejoinder to the argument from logical possibility. The Supervenience framework eliminates the less clear-cut question of how to individuate pieces of knowledge, and so makes discussion less confusing. All the same, close analysis shows that water-H2O analogies and related objections fail equally either way. Despite the fact that this is easily the most popular response to the knowledge argument, it is also easily the weakest of the major replies. It simply does not hold up to scrutiny.

A second, more sophisticated objection, due to Loar (1990), also holds that Mary gains new knowledge of old facts because of intensionality, but explicitly goes beyond the usual water-H2O analogies. Loar recognizes that analogies with the usual examples cannot do the job for the materialist, as (in our terminology) such analogies allow that physical and phenomenal notions have distinct primary intensions, and the antimaterialist can simply apply the argument to the property corresponding to the primary intension. As Loar puts it, even though "heat" and some statistical-mechanical predicate designate the same property (secondary intension), they nevertheless introduce distinct properties (primary intension). So he takes the argument further, and argues that two predicates can introduce the same property-that is, share the same primary intension-even when this sameness is not knowable a priori. If so, then Mary's knowledge of phenomenal proper-

Naturalistic Dualism 143

ties may just be knowledge of physical/functional properties, even though she could not have connected the two beforehand.

But how can two primary intensions coincide without our being able to know it a priori? Only if the space of possible worlds is smaller than we would have thought a priori. We think the intensions differ because we conceive of a world where they have different reference, such as a zombie world. Loar's position therefore requires this world is not really possible, despite the fact that we cannot rule it out on conceptual grounds, and despite the fact that Kripkean a posteriori necessity cannot do any work for us. This position therefore comes to precisely the same thing as the "strong metaphysical necessity" objection considered above. Like that objection, Loar's position requires that a conditional from physical facts to phenomenal facts be metaphysically necessary despite being logically contingent, where this gap cannot be explained by a difference in primary intensions. Like that objection, Loar's position requires a brute and arbitrary restriction on possible worlds. Loar offers no argument for this restriction, and his position is subject to precisely the same criticisms.15

One might expect there to be a third objection analogous to the intermediate "alternative strategy" from section 2. This would be cashed out in the claim that Mary does not really know all the physical facts. She knows all the facts couched in the terms of physics, but she lacks knowledge about the hidden (phenomenal or protophenomenal) essences of physical entities. If she had this knowledge, she would thereby know the phenomenal facts. As before, however, this view has only a very tenuous claim to the name "materialism." Like my own view, this view must take phenomenal or protophenomenal properties to be fundamental properties.

A fourth objection draws a connection between Mary's plight and a lack of indexical knowledge (Bigelow and Pargetter 1990; McMullen 1985; Papineau 1993). Although Mary gains new knowledge, it is argued that this is no more puzzling than other cases where someone who knows all the relevant objective facts discovers something new: for example, an omniscient amnesiac who discovers "I am Rudolf Lingens," or a well-informed insomniac who does not know that it is 3:49 a.m. now (see Perry 1979 and Lewis 1979). In these cases, there is gap between physical knowledge and indexical knowledge, just as there is a gap between physical knowledge and phenomenal knowledge in Mary's case.

The connection might be drawn in two ways. First, an objector might try to reduce phenomenal knowledge to indexical knowledge, arguing that all that Mary lacks is indexical knowledge. Second, one might try to draw an analogy between the two cases, arguing that in the indexical case the epistemic gap does not lead to an Ontological gap (indexicality does not falsify materialism), so that the phenomenal case need not lead to an Ontological gap either.

144 The Irreducibility of Consciousness

The reduction strategy clearly fails. As we saw in Chapter 2, indexicals accompany facts about conscious experience in their failure to supervene logically on physical facts, but they are all settled by the addition of a thin "indexical fact" about the location of the agent in question. But even when we give Mary perfect knowledge about her indexical relation to everything in the physical world, her knowledge of red experiences will not be improved in the slightest. In lacking phenomenal knowledge, she lacks far more than someone lacking indexical knowledge.

The analogy strategy is more interesting. One might respond by arguing for an Ontological gap even in the indexical case (see, e.g., Nagel 1983), but a more straightforward response is available. To see this, note that in the indexical case, an argument analogous to that in section 1 does not get off the ground: there is no conceivable uncentered world in which the physical facts are the same as ours, but in which the indexical facts differ. In uncentered worlds, indexical facts do not even apply. There is a relevant conceivable centered world, to be sure, but it is uncentered worlds that are relevant to the Ontological question. (If not, there is an Ontological gap in the indexical case as well, so the objector's argument does not get started.16) So in this case alone we can explain away the epistemic gap by noting that epistemic connections are determined by centered primary intensions, whereas Ontological connections are determined by properties corresponding to uncentered intensions. Indeed, this is reflected in the single loophole that was found in the argument of section 2, and in the analogous argument in this section: the fact that primary intensions determine only center-relative properties. This loophole allows through a single piece of irreducible indexical knowledge (the location of a centered world's center) without Ontological cost, but nothing further. Once the location of a center is specified, the loophole is closed. The phenomenal facts remain unsettled even when the location of a center is specified, so conscious experience remains out in the cold.17

If a materialist is to hold on to materialism, she really needs to deny that Mary makes any discovery about the world at all. Materialism requires logical Supervenience, which requires that Mary can gain no new factual knowledge of any sort when she first experiences red. Thus, in a fifth strategy, Lewis (1990) and Nemirow (1990) argue that at most Mary gains a new ability. For example, she gains the ability to imagine the sight of red things, and to recognize them when she sees them. But this is only knowledge how, not knowledge that. When she first experiences red, she learns no facts about the world.18

Unlike the previous options, this strategy does not suffer from internal problems. Its main problem is that it is deeply implausible. No doubt Mary does gain some abilities when she first experiences red, as she gains some abilities when she learns to ride a bicycle. But it certainly seems that she

Naturalistic Dualism 145

learns something else: some facts about the nature of experience. For all she knew before, the experience of red things might have been like this, or it might have been like that, or it might even have been like nothing at all. But now she knows that it is like this. She has narrowed down the space of epistemic possibilities. No such new knowledge comes along when an omniscient mechanic learns to ride a bicycle (except perhaps for knowledge about the phenomenology of bicycle riding). So this reply fails to come to grips with what goes on when Mary learns what it is like to see red.

We can also use more indirect methods to see that Mary's discovery involves factual knowledge. For example, Loar (1990) points out that this sort of knowledge can be embedded in conditionals: "If seeing red things is like this and seeing blue things is like this, then seeing purple things is probably like that"; "If it is like this for dogs to see red, then such-and-such follows"; and so on. Another example: as Lycan (1995) points out,19 what we imagine can turn out to be right or wrong; thus, after seeing a few colors, Mary might imagine what it is like to see another one, and her imagination might be correct or incorrect. If so, then to know what something is like is to know a truth about the world, and the ability analysis fails.

Dennett (1991) takes a related but more extreme position, arguing that Mary learns nothing at all. He notes that Mary could use her neurophysiological knowledge to recognize that a red object is red when she sees it, by noticing its effects on her reactions, which may differ from the effects of something blue. (If a team of experimenters tries to fool her by holding up a blue apple, she might not be fooled.) Perhaps this is so, but all that follows is that contra Lewis and Nemirow, Mary had certain abilities to recognize even before she had her first experience of red. It does nothing to show that she had the crucial knowledge: knowledge of what seeing red would be like. That would only follow if we had already accepted the ability analysis of "knowing what it is like"; but if we had accepted that analysis, the argument against materialism would already have been defeated. So Dennett's argument is a red herring here.

Ultimately, the strategy that a materialist must take is to deny that Mary gains knowledge about the world. And the only tenable way to do this seems to be via an ability analysis of "knowing what it is like." This is the only position with the internal coherence to ensure that it is not defeated by technical objections, just as analytic functionalism is ultimately the most coherent way for a materialist to resist the argument from logical Supervenience. But contraposing, the very implausibility of the denial that Mary gains knowledge about the world is evidence that materialism is doomed.20

We have seen that the modal argument (the argument from logical possibility) and the knowledge argument are two sides of the same coin. I think that in principle each succeeds on its own, but in practice they work best in tandem.21 Taking the knowledge argument alone: most materialists find it

146 The Irreducibility of Consciousness

hard to deny that Mary gains knowledge about the world, but often deny the step from there to the failure of materialism. Taking the modal argument alone: most materialists find it hard to deny the argument from the conceivability of zombies or inverted spectra to the failure of materialism, but often deny the premise. But taking the two together, the modal argument buttresses the knowledge argument where help is needed, and vice versa. In perhaps the most powerful combination of the two arguments, we can use the knowledge argument to compellingly establish the failure of logical Supervenience, and the modal argument to compellingly make the step from that failure to the falsity of materialism.

Kripke's argument

Kripke's argument was directed at the identity thesis put forward by Place (1957) and Smart (1959), but it can be seen to have a broad force against all forms of materialism. I will discuss the strengths and weaknesses of this argument in some detail, to support the conclusion that the parts that succeed are precisely those parts that correspond to the argument from logical Supervenience.

The argument goes roughly as follows: According to the identity thesis, certain mental states (such as pains) and brain states (such as C-fibers firing) are identical, even though "pain" and "C-fibers firing" do not mean the same thing. The identity here was originally supposed to be contingent rather than necessary, just as the identity between water and H2O is contingent. Against this, Kripke argues that all identities are necessary: If X is Y, then X is necessarily Y, as long as the terms X and Y designate rigidly, picking out the same individual or kind across worlds. Water is necessarily H2O, he argues; that is, water is H2O in every possible world. The identity may seem contingent-that is, it might seem that there is a possible world in which water is not H2O but XYZ-but this is illusory. In fact, the possible world that one is imagining contains no water at all. It is just a world in which there is some watery stuff-stuff that looks and behaves like water-made out of XYZ. In asserting that this watery stuff is water, one is misdescribing it.

Similarly, Kripke argues, if pains are identical to the firing of C-fibers, then this identity must be necessary. But the identity does not seem to be necessary. On the face of it, one can imagine a possible world where a pain occurs without any brain state whatsoever (disembodied pain), and one can imagine a world in which C-fibers fire without any accompanying pain (in a zombie, say). Further, he argues, these possibilities cannot be explained away as merely apparent possibilities, in the way that the possibility of water without H2O was explained away. For that to be the case, we would have to be misdescribing the "disembodied pain" world as one in which pain occurred, when really there was just "painy stuff" (something that feels like

Naturalistic Dualism 147

pain) going on. Similarly, we would have to be misdescribing the zombie as lacking pain, when all it really lacks is painy stuff. On such an account, the zombie would presumably have real pain, which is the firing of C-fibers; it is just that it doesn't feel like real pain.

But this cannot be the case, according to Kripke: all it is for something to be pain is for it to feel like pain. There is no distinction between pain and painy stuff, in the way that there is a distinction between water and watery stuff. One could have something that felt like water without it being water, but one could not have something that felt like pain without it being pain. Pain's feel is essential to it. So the possibility of the pains without the brain states (and vice versa) cannot be dismissed as before. Those possible worlds really are possible, and mental states are not necessarily identical to brain states. It follows that they cannot be identical to brain states at all.

Kripke runs the argument in two different ways, once against token-identity theories and once against type-identity theories. Token-identity theories hold that particular pains (such as my pain now) are identical to particular brain states (such as the C-fibers firing in my head now). Kripke argues in the above fashion that a particular pain could occur without the particular associated brain state, and vice versa, so they cannot be identical. Type-identity theories hold that mental states and brain states are identical as types: pain, for example, might be identical as a type to the firing of C-fibers. Kripke holds that this is straightforwardly refuted by the fact that one could instantiate the mental-state type without the brain-state type, and vice versa. Overall, we can count four separate arguments here, divided according to the target (token- or type-identity theories) and according to the method of argument (from the possibility of disembodiment or from the possibility of zombies).

There are some obvious differences between Kripke's argument and the argument I have given. For a start, Kripke's argument is couched entirely in terms of identity, whereas I have relied on the notion of Supervenience. Second, Kripke's argument is closely tied to his theoretical apparatus involving rigid designators and a posteriori necessity, whereas that apparatus plays only a secondary role in my argument, in answering certain objections. Third, Kripke's argument is usually seen to rely on a certain essentialism about various states, whereas no such doctrine is invoked in my argument. Fourth, my argument nowhere appeals to the possibility of disembodiment, as Kripke's does. Nevertheless there are obvious similarities. Both are modal arguments, involving necessity and possibility in key roles. And both appeal to the logical possibility of dissociating physical states from the associated phenomenal states.

I will now discuss what succeeds and what fails in Kripke's arguments, starting with those against token identity. These are generally held to be inconclusive. This is largely because they rely on intuitions about what counts

148 The Irreducibility of Consciousness

as that very thing across possible worlds, and such intuitions are notoriously unreliable. Kripke's claim that one could have that very pain state without that very brain state relies on the claim that what is essential to that pain state is its feel, and only its feel. But such claims about the essential properties of individuals are hard to justify. The token-identity theorist can respond by arguing that it is just as plausible that the firing of C-fibers is an essential property of the state. Of course, C-fiber firing does not seem to be essential to pain as a type, but who is to say that it is not essential to this particular pain token, especially if that token is identical to a brain state? If it is, then one simply could not have the particular pain in question without the particular brain state. (A line like this is taken by Feldman [1974], who argues that painfulness need not be essential to a particular pain, and by McGinn [1977], who in effect argues that both painfulness and C-fiber firing might be essential to a particular pain.) If so, then in imagining a disembodied version of my pain, one is not imagining that very pain but a separate, numerically distinct pain. The same goes for imagining my C-fiber firing without pain. So the arguments against token identity are inconclusive, although the arguments against type identity may survive.

Next, the argument from disembodiment does not establish a conclusive case against materialism. It might refute a type-identity thesis of the kind put forward by Place and Smart, but materialism does not require such a thesis.22 As Boyd (1980) notes, the materialist need not hold that mental states are physical states in all possible worlds-it is compatible with materialism that in some worlds mental states are constituted out of nonphysical stuff, as long as in this world they are physically constituted. The possibility of disembodiment only establishes the possibility of dualism, rather than its truth.23 To illustrate this, we can note that that few would argue that the possibility of nonphysical life implies dualism about biology. An argument against the identity thesis may be all that Kripke intended, but in any case the more general version of materialism survives.

This leaves the argument from the possibility of instantiating physical states without the corresponding phenomenal states-essentially an argument from the possibility of zombies. Curiously, this is the part of Kripke's argument that has received the least critical attention, with most commentators focusing on the possibility of disembodiment. As before, the argument that zombies yield against strong type-identity theses may be irrelevant, due to the fact that materialism does not require such a thesis, but there is a more general argument lurking here. The possibility of instantiating the relevant physical states without pain, Kripke argues (pp. 153-54), shows that even after God created all the physical stuff going on when one has a pain-perhaps a brain with C-fibers firing-he had to do more work in order that those firings be felt as pain. This is enough to establish that materialism is false.24

Naturalistic Dualism 149

This argument from physical states without phenomenal states corresponds directly to the argument I have given against materialism. Even the further maneuvers correspond. To the objection that this situation is merely conceivable and not truly possible, Kripke will respond: One cannot explain away the conceived situation as one that lacks the feeling of pain but not pain itself, as to be a pain is to feel like pain in any possible world. (That is, the secondary intension and the primary intension of "pain" coincide.) To this we might add (with Jackson [1980]) that even if the equivalence is disputed, the argument against materialism will succeed when applied to feelings of pain rather than pain. (That is, even if the intensions differ, the argument still goes through using the primary intension.) These are isomorphic to the responses that I gave to the same sort of objection earlier in this chapter.

(Note that with his thesis that an apparently-conceivable-but-impossible situation should be explained away as an epistemically possible situation that is misdescribed, Kripke is in effect endorsing the "weak" treatment of a posteriori necessity: the spaces of conceivable and possible worlds are the same, but a posteriori factors put constraints on their correct description.25 To see this, note that an advocate of "strong" metaphysical necessity, on which the space of possible worlds is a proper subset of the space of conceivable worlds, would not advocate such a thesis. On such a view, we might correctly describe an epistemically possible situation, but it might still be (brutely) metaphysically impossible. Kripke's reliance on the misdescription strategy, by contrast, suggests an implicit endorsement of the two-dimensional framework: indeed, all his examples of misdescription can be seen as cases in which a world is described under primary rather than secondary intensions.)

This argument from physical states without phenomenal states strikes me as the most conclusive part of Kripke's discussion. It is frequently overlooked amidst the discussion of identity theses, disembodiment, and the like; even Kripke assigns this aspect of his discussion a noncentral role. All the same, I think it is this part of the discussion that ultimately carries the burden of Kripke's argument.

To summarize, it seems to me that insofar as Kripke's argument against materialism succeeds, (1) the possibility of disembodiment is inconclusive as an argument against materialism but inessential to the case; (2) arguments phrased in terms of identity are similarly inconclusive but inessential; (3) an essentialist metaphysics is inessential, except insofar as the feel of pain is essential to pain as a type-but that is just a fact about what "pain" means; and (4) Kripke's apparatus of rigid designation and the like is not central, although it is required to answer a certain sort of objection.26 But his argument contains a sound core, in what is essentially an argument from the failure of logical Supervenience.

150 The Irreducibility of Consciousness

4. Is This Epiphenomenalism?*

A problem with the view I have advocated is that if consciousness is merely naturally supervenient on the physical, then it seems to lack causal efficacy. The physical world is more or less causally closed, in that for any given physical event, it seems that there is a physical explanation (modulo a small amount of quantum indeterminacy). This implies that there is no room for a nonphysical consciousness to do any independent causal work. It seems to be a mere epiphenomenon, hanging off the engine of physical causation, but making no difference in the physical world. It exists, but as far as the physical world is concerned it might as well not. Huxley (1874) advocated such a view, but many people find it counterintuitive and repugnant. Indeed, this consequence has been enough to cause some (e.g., Kirk [1979]; Seager [1991]) to question the conclusions of their arguments against materialism, and to consider the possibility that consciousness might be logically supervenient on the physical after all.

This argument has been formalized in different but related ways by Kirk (1979), Horgan (1987), and Seager (1991). If we assume that the physical world is causally closed and that consciousness causes some physical events, then it follows under certain natural assumptions about causation that consciousness must supervene logically (or metaphysically) on the physical.27 If so, then given that the physical world is causally closed, the mere natural Supervenience of consciousness implies that consciousness is epiphenomenal. The basic shape of the argument is clear: if it is possible to subtract the phenomenal from our world and still retain a causally closed world Z, then everything that happens in Z has a causal explanation that is independent of the phenomenal, as there is nothing phenomenal in Z. But everything that happens in Z also happens in our world, so the causal explanation that applies in Z applies equally here. So the phenomenal is causally irrelevant. Even if conscious experience were absent, the behavior might have been caused in exactly the same way.

In responding to this, I will pursue a two-pronged strategy. First, it is not obvious that mere natural Supervenience must imply epiphenomenalism in the strongest sense. It is clear that the picture it produces looks something like epiphenomenalism. Nevertheless, the very nature of causation itself is quite mysterious, and it is possible that when causation is better understood we will be in a position to understand a subtle way in which conscious experience may be causally relevant. (In effect, it may turn out that background assumptions in arguments above are false.) I will outline some ways in which such an analysis might be made below. On the second prong, I will consider the reasons why epiphenomenalism might be found unpalatable, and analyze their force as arguments. If these intuitions do not translate into compelling arguments, it may turn out the sort of epiphenomenalism that

Naturalistic Dualism 151

this position implies is only counterintuitive, and that ultimately a degree of epiphenomenalism can be accepted.

Strategies for avoiding epiphenomenalism

There are a number of ways in which one might try to preserve the failure of logical Supervenience while nevertheless avoiding epiphenomenalism. The most obvious of these is to deny the causal closure of the physical, and to embrace a strong form of interactionist dualism in which the mental fills causal gaps in physical processing. I think this strategy should be avoided, for reasons I discuss shortly. However, there are a number of more subtle options that depend on an appropriate view of metaphysics and especially of causation. I discuss four such options.

1. Regularity-based causation.

1. Regularity-based causation. The first option is to accept a strong Humean account of causation, upon which all it is for A to cause B is for there to be a uniform regularity between events of type A and events of type B. Such a view would allow a "causal" role for the phenomenal: the mere fact that pain sensations are generally followed by withdrawal reactions would imply that pain causes withdrawal reactions.

A related non-Humean option identifies a causal connection with any nomic (or lawful) connection, even if a nomic regularity is something more than a uniform regularity. The natural Supervenience view is entirely compatible with the existence of a nomic connection between experience and behavior (for example, there might be a lawful connection between experience and an underlying brain state, and a lawful connection between that brain state and behavior). One might claim that this is enough for causation. This might be supported by noting that the counterfactual "Behavior would have been the same even in the absence of experience" is false on the most natural interpretation: if the experience were absent, the brain state would have been different, and behavior would have been different. Here, the counterfactual is assessed by considering naturally possible worlds, rather than logically possible worlds.

I find both of these positions implausible. I have argued against Humean views of causation in Chapter 2, and even on the non-Humean view it is implausible that just any nomic connection suffices for causation-think of the correlation between the hair color of identical twins, for instance. Nevertheless, considerations like these at least give us an idea of why consciousness appears to play a causal role. There are all sorts of systematic regularities between conscious experiences and later physical events, each of which leads us to infer a causal connection. Faced with such regularities, we would expect people to infer a causal relation for broadly Humean reasons. This can therefore explain away some of our intuitions that con-

152 The Irreducibility of Consciousness

sciousness is causally efficacious, thus supporting the second prong of the strategy.

2. Causal overdetermination.

2. Causal overdetermination. Perhaps we might claim that a physical state and a phenomenal state, though wholly distinct, might both qualify as causing a later physical state. If physical state P1 is associated with phenomenal state Q1, then perhaps it is true both that P1 causes a later physical state P2 and that Q1 causes P2. This is counterintuitive: P1 is already a sufficient cause of P2, so Q1 would seem to be causally redundant. But it is not obvious that Q1 could not stand in a causal relation to P1 nevertheless. This may be especially reasonable if we adopt a nonreductive view of causation (of the sort

"advocated by Tooley 1987). Perhaps there is an irreducible causal connection between the two physical states, and a separate irreducible causal connection between the phenomenal state and the physical state.

This sort of causal overdetermination of events is often regarded with suspicion, but it is hard to demonstrate conclusively that there is something wrong with it. The nature of causation is sufficiently ill understood that overdetermination cannot yet be ruled out. I will not pursue this line myself, but it nevertheless deserves to be taken seriously.


3. The nonsupervenience of causation.

3. The nonsupervenience of causation. A third strategy rests with the very nature of causation itself. We saw in Chapter 2 that there are two classes of facts that do not supervene logically on particular physical facts: facts about consciousness and facts about causation. It is natural to speculate that these two failures might be intimately related, and that consciousness and causation have some deep metaphysical tie. Both are quite mysterious, after all, and two mysteries might be more neatly wrapped into one. Perhaps, for instance, experience itself is a kind of causal nexus; perhaps it somehow realizes Hume's "unknowable causal relation"; or perhaps the relationship is more complex. A relationship like this might suggest a role for experience in causation that is more subtle than the usual sort of causation, but nevertheless avoids the strongest form of epiphenomenalism.

A proposal like this has been developed by Rosenberg (1996), who argues that many of the problems of consciousness are precisely paralleled by problems about causation. He argues that because of these parallels, it may be that experience realizes causation, or some aspects of causation, in the actual world. On this view, causation needs to be realized by something in order to support its many properties, and experience is a natural candidate. If this is so, it may be that it is the very existence of experience that allows for causal relations to exist, so that there is a subtle sort of relevance for experience in causation.

Of course, this proposal is extremely speculative, and faces some problems. For a start, it seems to lead to a version of panpsychism, the view

Naturalistic Dualism 153

that everything is conscious, which many find counterintuitive. Further, the zombie world is still a problem-it seems that we can imagine all that causation going on without experience, so that experience might still seem epiphenomenal. A response might be that causation has to be realized by something; in the zombie world it is realized by something else, but experience is still relevant in this world in virtue of realizing causation here. It is not obvious to me that causation has to be realized by something with any further properties; if it need not be, then the phenomenal nature of causation might still be redundant. But again, the metaphysics of causation is as yet far from clear, and this proposal is certainly worth investigating.

4. The intrinsic nature of the physical. The strategy to which I am most drawn stems from the observation that physical theory only characterizes its basic entities relationally, in terms of their causal and other relations to other entities. Basic particles, for instance, are largely characterized in terms of their propensity to interact with other particles. Their mass and charge is specified, to be sure, but all that a specification of mass ultimately comes to is a propensity to be accelerated in certain ways by forces, and so on. Each entity is characterized by its relation to other entities, and these entities are characterized by their relations to other entities, and so on forever (except, perhaps, for some entities that are characterized by their relation to an observer). The picture of the physical world that this yields is that of a giant causal flux, but the picture tells us nothing about what all this causation relates. Reference to the proton is fixed as the thing that causes interactions of a certain kind, that combines in certain ways with other entities, and so on; but what is the thing that is doing the causing and combining? As Russell (1927) notes, this is a matter about which physical theory is silent.28

One might be attracted to the view of the world as pure causal flux, with no further properties for the causation to relate, but this would lead to a strangely insubstantial view of the physical world.29 It would contain only causal and nomic relations between empty placeholders with no properties of their own. Intuitively, it is more reasonable to suppose that the basic entities that all this causation relates have some internal nature of their own, some intrinsic properties, so that the world has some substance to it. But physics can at best fix reference to those properties by virtue of their extrinsic relations; it tells us nothing directly about what those properties might be. We have some vague intuitions about these properties based on our experience of their macroscopic analogs-intuitions about the very "massiveness" of mass, for example-but it is hard to flesh these intuitions out, and it is not clear on reflection that there is anything to them.

There is only one class of intrinsic, nonrelational property with which we have any direct familiarity, and that is the class of phenomenal properties.

154 The Irreducibility of Consciousness

It is natural to speculate that there may be some relation or even overlap between the uncharacterized intrinsic properties of physical entities, and the familiar intrinsic properties of experience. Perhaps, as Russell suggested, at least some of the intrinsic properties of the physical are themselves a variety of phenomenal property?30 The idea sounds wild at first, but on reflection it becomes less so. After all, we really have no idea about the intrinsic properties of the physical. Their nature is up for grabs, and phenomenal properties seem as likely a candidate as any other.

There is of course the threat of panpsychism. I am not sure that this is such a bad prospect-if phenomenal properties are fundamental, it is natural to suppose that they might be widespread-but it is not a necessary consequence. An alternative is that the relevant properties are protophenomenal properties. In this case the mere instantiation of such a property does not entail experience, but instantiation of numerous such properties could do so jointly. It is hard to imagine how this would work (we know that it cannot work for standard physical properties), but these intrinsic properties are quite foreign to our conception. The possibility cannot be ruled out a priori.

Either way, this sort of intimate link suggests a kind of causal role for the phenomenal. If there are intrinsic properties of the physical, it is instantiations of these properties that physical causation ultimately relates. If these are phenomenal properties, then there is phenomenal causation; and if these are protophenomenal properties, then phenomenal properties inherit causal relevance by their supervenient status, just as billiard balls inherit causal relevance from molecules. In either case, the phenomenology of experience in human agents may inherit causal relevance from the causal role of the intrinsic properties of the physical.

Of course, this would be a subtler kind of causal relevance than the usual kind. It remains the case, for example, that one can imagine removing the phenomenal properties, with the pattern of causal flux remaining the same. But now the response is that in imagining such a scenario, one is effectively altering the intrinsic properties of physical entities and replacing them by something else (of course, the trouble is that we are not used to imagining intrinsic properties of the physical at all). Thus one is simply moving to a world where something else is doing the causation. If there could be a world of pure causal flux, this argument would fail, but such a world is arguably logically impossible, as there is nothing in such a world for causation to relate.

This position is rather akin to the second position described in section 2, where electrons have a hidden essence to which physical descriptions merely fix reference. I think that for the reasons given there, the intrinsic properties should not be identified with physical properties such as mass. It seems reasonable to say that there is still mass in the zombie world, despite differences in its intrinsic nature. If so, then mass is an extrinsic property that can be "realized" by different intrinsic properties in different worlds. But

Naturalistic Dualism 155

whichever way we make this semantic decision, the position retains an essential duality between the properties that physics deals with directly and the hidden intrinsic properties that constitute phenomenology.

There is a sense in which this view can be seen as a monism rather than a dualism, but it is not a materialist monism. Unlike physicalism, this view takes certain phenomenal or protophenomenal properties as fundamental. What it finally delivers is a network of intrinsic properties, at least some of which are phenomenal or protophenomenal, and which are related according to certain causal/dynamic laws. These properties "realize" the extrinsic physical properties, and the laws connecting them realize the physical laws. In the extreme case in which all the intrinsic properties are phenomenal, the view might be best seen as a version of idealism. It is an idealism very unlike Berkeley's, however. The world is not supervenient on the mind of an observer, but rather consists in a vast causal network of phenomenal properties underlying the physical laws that science postulates. A less extreme case in which intrinsic properties are protophenomenal, or in which some are neither phenomenal nor protophenomenal, is perhaps best regarded as a version of Russell's neutral monism. The basic properties of the world are neither physical nor phenomenal, but the physical and the phenomenal are constructed out of them. From their intrinsic natures in combination, the phenomenal is constructed; and from their extrinsic relations, the physical is constructed.

On this view, the most basic laws will be those that connect the basic intrinsic properties. The familiar physical laws capture the relational shape of these laws, while abstracting away from the intrinsic properties. Psychophysical laws can be reinterpreted as laws that connect intrinsic properties (or properties constructed out of these) to their relational profiles (or to complex relational structures). Thus these laws do not "dangle" ontologically from physical laws. Rather, both are consequences of the truly basic laws. But the epistemological order differs from the Ontological order: we are led first to the relational structure of the causal network, and only slowly to the underlying intrinsic properties. For everyday explanatory purposes, it is therefore most useful to continue to think of this view in terms of a network of physical laws, with further principles connecting the physical to the phenomenal.


All this metaphysical speculation may need to be taken with a pinch of salt, but it shows that the issue of epiphenomenalism is not cut and dried. There are a number of subtle issues about causation and about the nature of experience that will need to be understood better before we can say for certain whether experience is epiphenomenal. In any case, I will now set aside the metaphysical speculation and return to a less lofty plane (although I will return to some of these issues in Chapter 8).

156 The Irreducibility of Consciousness

It remains the case that natural Supervenience feels epiphenomenalistic. We might say that the view is epiphenomenalistic to a first approximation: if it allows some causal relevance for experience, it does so in a subtle way. I think we can capture this first-approximation sense by noting that the view makes experience explanatorily irrelevant. We can give explanations of behavior in purely physical or computational terms, terms that neither involve nor imply phenomenology. If experience is tied in some intimate way to causation, it is in a way that these explanations can abstract away from. One might find even explanatory irrelevance troubling; I will say much more about it in the next chapter.

Some have been tempted to avoid epiphenomenalism by leaping into the "strong metaphysical necessity" position of section 2 of this chapter. If experience does not supervene logically on the physical, it has seemed to some that the only way to preserve its causal role is to declare it brutely identical to or metaphysically supervenient on some physical property or properties. Apart from the problems that I have already mentioned, however, the view still has serious problems with explanatory irrelevance. The very conceivability of a zombie shows that on this view, behavior can be explained in terms that neither involve nor imply the existence of experience. Explanatory relations are conceptual relations, so that strong metaphysical necessity is irrelevant here. The view still leaves behavior independent of experience in a strong sense and has to face up to most of the same difficulties as a property dualism. There is therefore not much to be gained by taking such a position.

Interactionist dualism?

Some people, persuaded by the arguments for dualism but convinced that phenomenal consciousness must play a significant causal role, may be tempted by an interactionist variety of dualism, in which experience fills causal gaps in physical processes. Giving in to this temptation raises more problems than it solves, however. For a start, it requires a hefty bet on the future of physics, one that does not currently seem at all promising; physical events seem inexorably to be explained in terms of other physical events. It also requires a large wager on the future of cognitive science, as it suggests that the usual kinds of physical/functional models will be insufficient to explain behavior. But the deepest problem is that this view may be no better at getting around the problems with epiphenomenalism than the view with causal closure, for reasons I will discuss shortly.

The only form of interactionist dualism that has seemed even remotely tenable in the contemporary picture is one that exploits certain properties of quantum mechanics. There are two ways this might go. First, some have appealed to the existence of quantum indeterminacy, and have suggested

Naturalistic Dualism 157

that a nonphysical consciousness might be responsible for filling the resultant causal gaps, determining which values some physical magnitudes might take within an apparently "probabilistic" distribution (e.g., Eccles 1986). Although these decisions would have only a tiny proximate effect, perhaps nonlinear dynamics could amplify these tiny fluctuations into significant macroscopic effects on behavior.

This is an audacious and interesting suggestion, but it has a number of problems. First, the theory contradicts the quantum-mechanical postulate that these microscopic "decisions" are entirely random, and in principle it implies that there should be some detectable pattern to them-a testable hypothesis. Second, in order that this theory allows that consciousness does any interesting causal work, it needs to be the case that the behavior produced by these microscopic decisions is somehow different in kind than that produced by most other sets of decisions that might have been made by a purely random process. Presumably the behavior is more rational than it would have been otherwise, and it leads to remarks such as "I am seeing red now" that the random processes would not have produced. This again is testable in principle, by running a simulation of a brain with real random processes determining those decisions. Of course we do not know for certain which way this test would come out, but to hold that the random version would lead to unusually degraded behavior would be to make a bet at long odds.

A second way in which quantum mechanics bears on the issue of causal closure lies with the fact that in some interpretations of the quantum formalism, consciousness itself plays a vital causal role, being required to bring about the so-called "collapse of the wave-function." This collapse is supposed to occur upon any act of measurement; and in one interpretation, the only way to distinguish a measurement from a nonmeasurement is via the presence of consciousness. This theory is certainly not universally accepted (for a start, it presupposes that consciousness is not itself physical, surely contrary to the views of most physicists), and I do not accept it myself, but in any case it seems that the kind of causal work consciousness performs here is quite different from the kind required for consciousness to play a role in directing behavior.31 It is unclear how a collapse in external perceived objects allows consciousness to affect physical processing within the brain; such theories are usually silent on what happens to the brain during collapse. And even if consciousness somehow manages to collapse the brain state, then all the above remarks about apparently random processes and their connection with behavior still apply.

In any case, all versions of interactionist dualism have a conceptual problem that suggests that they are less successful in avoiding epiphenomenalism than they might seem; or at least that they are no better off than the view I have advocated. Even on these views, there is a sense in which the phenomenal is irrelevant. We can always subtract the phenomenal component from

158 The Irreducibility of Consciousness

any explanatory account, yielding a purely causal component. Imagine (with Eccles) that "psychons" in the nonphysical mind push around physical processes in the brain, and that psychons are the seat of experience. We can tell a story about the causal relations between psychons and physical processes, and a story about the causal dynamics among psychons, without ever invoking the fact that psychons have phenomenal properties. Just as with physical processes, we can imagine subtracting the phenomenal properties of psychons, yielding a situation in which the causal dynamics are isomorphic. It follows that the fact that psychons are the seat of experience plays no essential role in a causal explanation, and that even in this picture experience is explanatorily irrelevant.

Some might object that psychons (or ectoplasm, or whatever) are entirely constituted by their phenomenal properties. Even so, there is a sense in which their phenomenal properties are irrelevant to the explanation of behavior; it is only their relational properties that matter in the story about causal dynamics. If one objects that still, they have further intrinsic properties that are causally relevant, we have a situation like the one that arose above with phenomenal properties intrinsic to physical entities. Either way, we have a sort of causal relevance but explanatory irrelevance. Indeed, nothing especially is gained by moving away from the causal closure of the physical. We still have a broader causal network that is closed, and it remains the case that the phenomenal nature of entities in the network is explanatorily superfluous.

We can even imagine that if interactionism is true, then for reasons quite independent of conscious experience we would be led eventually to postulate psychons in order to explain behavior, to fill the observed causal gaps and account for the data. If so, psychons would have the status of a kind of theoretical entity like the theoretical entities of physics. Nothing in this story would involve or imply experience, which would be as explanatorily superfluous as in the usual case; we could still tell a zombie story involving psychons, and so on. The additional observation that these psychons might have phenomenal properties works no better or worse as a response to epiphenomenalism than the analogous observation that physical entities (perhaps basic entities, perhaps quite complex ones) might have phenomenal properties over and above their extrinsic features. The denial of the causal closure of the physical therefore makes no significant difference in the avoidance of epiphenomenalism.32

The problems of epiphenomenalism

Any view that takes consciousness seriously will at least have to face up to a limited form of epiphenomenalism. The very fact that experience can be coherently subtracted from any causal account implies that experience is

Naturalistic Dualism 159

superfluous in the explanation of behavior, whether or not it has some subtle causal relevance. It is possible that it will turn out to be causally irrelevant in a stronger sense; that question is open. We therefore need to pursue the second prong of the strategy, and see just what are the problems with the causal or explanatory irrelevance of experience, and whether they are ultimately fatal. I will do this at much greater length in Chapter 5, but here we can briefly survey the field.

The most common objection to epiphenomenalism is simply that it is counterintuitive or even "repugnant." Finding a conclusion counterintuitive or repugnant is not sufficient reason to reject the conclusion, however, especially if it is the conclusion of a strong argument. Epiphenomenalism may be counterintuitive, but it is not obviously false, so if a sound argument forces it on us, we should accept it. Of course, a counterintuitive conclusion may give us reason to go back and reexamine the argument, but we still need to find something wrong with the argument on independent grounds. If it turns out that the falsity of logical Supervenience implies epiphenomenalism, then logical Supervenience may be desirable, but we cannot simply assert it as a brute fact. To hold logical Supervenience, one needs some account of how the physical facts might entail the facts about consciousness, and this is precisely what I have argued cannot be given.

More detailed objections to epiphenomenalism fall into three classes: those concerning the relationship of experience to ordinary behavior, those concerning the relationship of experience to judgments about experience, and those concerning the overall picture of the world that it gives rise to.

Take the first class first. Many find it simply obvious that their feelings of pain cause them to withdraw their hand from a flame, or that my experience of a headache cannot be irrelevant to the explanation of why I take pills. There is certainly a strong intuition to this effect. On the other hand, we can easily explain away the source of this intuition, in terms of the systematic regularities between these events. We are much more directly aware of experience and of behavior than we are of an underlying brain state; upon exposure to systematic regularities between experience and behavior, it is natural that a strong causal connection should be inferred. Even if the connection were only an indirect nomic connection due to relations to the underlying brain state, we would still expect the inference to be made. So this intuition can be explained away. In any case, this sort of objection cannot be fatal to the view, as it is an intuition that does not extend directly into an argument. It is an instance of the merely counterintuitive.

The second class of objections is more worrying. It seems very strange that our experiences should be irrelevant to the explanation of why we talk about experiences, for instance, or perhaps even to our internal judgments about experiences; this seems much stranger than the mere irrelevance of my pain to the explanation of my hand's withdrawal. Some claim

160 The Irreducibility of Consciousness

that this sort of problem is not merely counterintuitive but fatal. For example, it might be claimed that this is incompatible with our knowledge of experience, or with our ability to refer to experiences. I believe that when these arguments are spelled out they do not ultimately gain their conclusion, but these questions are certainly challenging. I devote Chapter 5 to these issues.

Objections in the third class concern the overall structure of the view. One objection is that the picture is ugly and implausible, with experience hanging off the physical by "nomological danglers" that are not integrated with the other laws of nature. I think this can be combated by developing a theory that leads to a more integrated picture. The label "epiphenomenalism", tends to suggest a view on which experience is dangling "up there," floating free of processing in some way; a better picture that is still compatible with natural Supervenience is a picture of experience sitting down among the causal cracks. At the very least, we can try to make the Psychophysical laws as simple and elegant as possible. Also falling into this class is a worry about how consciousness might evolve in an epiphenomenalist account, but it is not hard to see that this poses no problem for the view I advocate; I discuss this further at the end of this chapter.

On examination, there are not many arguments that do serious damage to epiphenomenalism. The main class of worrying arguments are those concerning judgments about experience, which I will discuss in the next chapter. Arguments aside, some have the intuition that epiphenomenalism must be wrong, but the intuition does not suffice to reject the position in the face of strong arguments in its favor.

I do not describe my view as epiphenomenalism. The question of the causal relevance of experience remains open, and a more detailed theory of both causation and of experience will be required before the issue can be settled. But the view implies at least a weak form of epiphenomenalism, and it may end up leading to a stronger sort. Even if it does, however, I think the arguments for natural Supervenience are sufficiently compelling that one should accept them. Epiphenomenalism is counterintuitive, but the alternatives are more than counterintuitive. They are simply wrong, as we have already seen and will see again. The overall moral is that if the arguments suggest that natural Supervenience is true, then we should learn to live with natural Supervenience.

Some will find that nevertheless the epiphenomenalist nature of this position is a fatal flaw. I have some sympathy with this position, which can be seen as an expression of the paradox of consciousness: when it comes to consciousness, it can seem that all the alternatives are bad. However, I think the problems with all other views are fatal in a much stronger way than the counterintuitiveness of this one. Given that some option in logical space has to be correct, this view seems to be the only reasonable candidate.

Naturalistic Dualism 161

5. The Logical Geography of the Issues

The argument for my view is an inference from roughly four premises:

1. Conscious experience exists.

2. Conscious experience is not logically supervenient on the physical.

3. If there are phenomena that are not logically supervenient on the physical facts, then materialism is false.

4. The physical domain is causally closed.

Premises (1), (2), and (3) clearly imply the falsity of materialism. This, taken in conjunction with premise (4) and the plausible assumption that physically identical beings will have identical conscious experiences, implies the view that I have called natural Supervenience: conscious experience arises from the physical according to some laws of nature, but is not itself physical. The various alternative positions can be catalogued according to whether they deny premises (1), (2), (3), or (4). Of course, some of these premises can be denied in more than one way.

Denying premise (1):

i. Eliminativism.

i. Eliminativism. On this view, there are no positive facts about conscious experience. Nobody is conscious in the phenomenal sense.

Denying premise (2):

Premise (2) can be denied in various ways, depending on how the entailment in question proceeds-that is, depending on what sort of physical properties are centrally responsible for entailing consciousness. I call all of these views "reductive materialist" views, as they all suppose an analysis of the notion of consciousness that is compatible with reductive explanation.

ii. Reductive functionalism.

ii. Reductive functionalism. This view takes consciousness to be conceptually entailed by the physical in virtue of functional or dispositional properties. On this view, what it means for a state to be conscious is for it to play a certain causal role. In a world physically identical to ours, all the relevant causal roles would be played, and therefore the conscious states would all be the same. The zombie world is therefore logically impossible.

iii. Nonfunctionalist reductive materialism.

iii. Nonfunctionalist reductive materialism. On this view, the facts about consciousness are conceptually entailed by the physical facts in virtue of some nonfunctional property. Possible candidates might include biochemical and quantum properties, or properties yet to be determined.

iv. New-physics materialism.

iv. New-physics materialism. According to this view, we have no current idea of how physical facts could explain consciousness, but

162 The Irreducibility of Consciousness

that is because our current conception of physical facts is too narrow. When one argues that a zombie world is logically possible, one is really arguing that all the fields and particles interacting in the space-time manifold, postulated by current physics, could exist in the absence of consciousness. But with a new physics, things might be different. The entities in a radically different theoretical framework might be sufficient to entail and explain consciousness.

Denying premise (3):

v. Nonreductive materialism.

v. Nonreductive materialism. This is the view that although there may be no logical entailment from the physical facts to the facts about consciousness, and therefore no reductive explanation of consciousness, consciousness just is physical. The physical facts "metaphysically necessitate" the facts about consciousness. Even though the idea of a zombie world is quite coherent, such a world is metaphysically impossible.

Denying premise (4):

vi. Interactionist dualism.

vi. Interactionist dualism. This view accepts that consciousness is nonphysical, but denies that the physical world is causally closed, so that consciousness can play an autonomous causal role.

Then there is my view, which accepts premises (1), (2), (3), and (4):

vii. Naturalistic dualism.

vii. Naturalistic dualism. Consciousness supervenes naturally on the physical, without supervening logically or "metaphysically."

There is also an eighth common view, which is generally underspecified:

viii. Don't-have-a-clue materialism.

viii. Don't-have-a-clue materialism. "I don't have a clue about consciousness. It seems utterly mysterious to me. But it must be physical, as materialism must be true." Such a view is held widely, but rarely in print (although see Fodor 1992).

To quickly summarize the situation as I see it, option (i) seems to be manifestly false; (ii) and (iii) rely on false analyses of the notion of consciousness and therefore change the subject; (iv) and (vi) place large and implausible bets on the way that physics will turn out, and also have fatal conceptual problems; and (v) either makes an invalid appeal to Kripkean a posteriori necessity or relies on a bizarre metaphysics. I have a certain amount of sympathy with (viii), but it presumably must eventually reduce to some more specific view, and none of these seem to work. This leaves (vii) as the only tenable option.

More slowly, starting with options (iv) and (vi): Option (vi), interactionist dualism, requires that physics will turn out to have gaps that can be filled

Naturalistic Dualism 163

by the action of a nonphysical mind. Current evidence suggests that this is unlikely. Option (iv) requires that the shape of physics will be transformed so radically that it could entail facts about conscious experience; but nobody has an idea of how any physics could do this. Indeed, given that physics ultimately deals in structural and dynamical properties, it seems that all physics will ever entail is more structure and dynamics, which (unless one of the other reductive options is embraced) will never entail the existence of experience.

The deepest reason to reject options (iv) and (vi) is that they ultimately suffer from the same problem as a more standard physics: the phenomenal component can be coherently subtracted from the causal component. On the interactionist view, we have seen that even if the nonphysical entities have a phenomenal aspect, we can coherently imagine subtracting the phenomenal component, leaving a purely causal/dynamic story characterizing the interaction and behavior of the relevant entities. On the new physics view, even if it explicitly incorporates phenomenal properties, the fact that these properties are phenomenal can play no essential role in the causal/dynamic story; we would be left with a coherent physics even if that aspect were subtracted. Either way, the dynamics is all we need to explain causal interactions, and no set of facts about dynamics adds up to a fact about phenomenology. A zombie story can therefore still be told.

Various moves can be made in reply, but each of these moves can also be made on the standard physical story. For example, perhaps the abstract dynamics misses the fact that the nonphysical stuff in the interactionist story is intrinsically phenomenal, so that phenomenal properties are deeply involved in the causal network. But equally, perhaps the abstract dynamics of physics misses the fact that its basic entities are intrinsically phenomenal (physics characterizes them only extrinsically, after all), and the upshot would be the same. Either way, we have the same kind of explanatory irrelevance of the intrinsic phenomenal properties to the causal/dynamic story. The move to interactionism or new physics therefore does not solve any problems inherent in the property dualism I advocate. At the end of the day, they can be seen as more complicated versions of the same sort of view.

As for option (iii), the most tempting version is the one that gestures toward unknown properties that we have so far overlooked as the key to the entailment. But ultimately the problem is the same: physics only gives us structure and dynamics, and structure and dynamics does not add up to phenomenology. The only available properties would seem to be those characterizing physical structure or function, or properties constructed out of the two. But structural properties are obviously inappropriate analyses of the concept of experience, and functional properties are not much better (although I consider them below). Any view of this sort will ultimately change the subject.

164 The Irreducibility of Consciousness

This leaves options (i), (ii), (v), and (vii), which correspond to the options taken most seriously in the contemporary literature: eliminativism, reductive functionalism, nonreductive materialism, and property dualism. Of these I reject option (i) as being in conflict with the manifest facts. Perhaps an extraordinary argument could establish that conscious experience does not exist, but I have never seen an argument that comes remotely close to making this case. In the absence of such an argument, to take option (i) is simply to evade the problem by denying the phenomenon.

Option (v) is often attractive to those who want to take consciousness seriously and also retain materialism. But I have argued that it simply does not work. The nonreductive materialism advocated by Searle turns out to have internal problems and collapses into one of the other views (most likely property dualism). Other proponents of this view rely on an appeal to Kripke's a posteriori necessity, but the sort of a posteriori necessity demon-strated by Kripke cannot save materialism. The only consistent way to take option (v) is to appeal to a strong a posteriori necessity that goes well beyond Kripke's, and to invoke brute constraints on the space of "metaphysically possible" worlds. We have seen that there is no reason to believe in such constraints, or to believe in such a third, intermediate grade of the possibility of worlds. This metaphysics gains no support from any other phenomena, and it is hard to see how it could be supported.

Even if this metaphysics of necessity is accepted, for most explanatory purposes the view ends up looking like the view I advocate. It implies that consciousness cannot be reductively explained. It implies that conscious experience is explanatorily irrelevant to the physical domain. And it implies that a theory of consciousness must invoke bridging principles to connect the physical and phenomenal domains, principles that are not themselves entailed by physical laws. This view calls these principles "metaphysically necessary," but for all practical purposes the upshot is the same. This sort of theory will have the same shape as the dualist theories I advocate, and almost everything I say in developing a nonreductive theory in the next few chapters will apply equally here.

Option (ii), reductive functionalism, is the most serious materialist option. Leaving aside various wild options, if materialism is true, then consciousness is logically supervenient, and the only remotely reasonable way for it to be logically supervenient is via a functional analysis. On this view, then, all it means for something to be a conscious experience is for it to play a certain causal role in a system. Phenomenal properties are treated exactly the same way as psychological properties, such as learning or categorization.

The problem with this view, of course, is that it misrepresents what it means to be a conscious experience, or to be conscious. When I wonder whether other beings are conscious, I am not wondering about their abilities or their internal mechanisms, which I may know all about already; I am

Naturalistic Dualism 165

wondering whether there is something it is like to be them. This point can be supported in various familiar ways. One way is to note that even once we have explained various functional capacities, the problem of explaining experience may still remain. Another rests on the observation that we can imagine any functional role being played in the absence of conscious experi-ence. A third derives from the fact that knowledge of functional roles does not automatically yield knowledge of consciousness. There are also the ob-jections, made earlier, that a functionalist analysis cannot account for the semantic determinacy of attributions of consciousness and that it collapses the conceptual distinction between consciousness and awareness.

At the end of the day, reductive functionalism does not differ much from eliminativism. Both of these views hold that there is discrimination, categori-zation, accessibility, reportability, and the like; and both deny that there is anything else that even needs to be explained. The main difference is that the reductive line holds that some of these explananda deserve the name "experience," whereas the eliminative line holds that none of them do. Apart from this terminological issue, the substance of the views is largely the same. It is often noted that the line between reductionism and eliminativism is blurry, with reduction gradually sliding into elimination the more we are forced to modify the relevant concepts in order to perform a reduction. In allowing that consciousness exists only insofar as it is defined as some func-tional capacity, the reductive functionalist view does sufficient violence to the concept of consciousness that it is probably best viewed as a version of eliminativism. Neither is a view that takes consciousness seriously.

This leaves view (vii), the property dualism that I have advocated, as the only tenable option. Certainly it seems to be a consequence of well-justified premises. In some ways it is counterintuitive, but it is the only view without a fatal flaw. Some will find its dualistic nature unpalatable; but I will argue shortly that dualism of this variety is not as unreasonable as many have thought, and that it is open to few serious objections. The biggest worry about this view is that it implies a certain irrelevance of phenomenal properties in the explanation of behavior, and may lead to epiphenomenalism, although this is not automatic. I will argue in the next chapter, however, that this explanatory irrelevance has no fatal consequences. Ultimately, this view gives us a coherent, naturalistic, unmysterious view of consciousness and its place in the natural order.

Type A, type B, and type C

Taking a broader view of the logical geography, we can say that there are three main classes of views about conscious experience. Type-A views hold that consciousness, insofar as it exists, supervenes logically on the physical, for broadly functionalist or eliminativist reasons. Type-B views accept that

166 The Irreducibility of Consciousness

consciousness is not logically supervenient, holding that there is no a priori implication from the physical to the phenomenal, but maintain materialism all the same. Type-C views deny both logical Supervenience and materialism.

Type-A views come in numerous varieties-eliminativism, behaviorism, various versions of reductive functionalism-but they have certain things in common. A type-A theorist will hold that (1) physical and functional dupli-cates that lack the sort of experience that we have are inconceivable; (2) Mary learns nothing about the world when she first sees red (at best she gains an ability); and (3) everything there is to be explained about conscious-ness can be explained by explaining the performance of various functions. Archetypal type-A theorists include Armstrong (1968), Dennett (1991), Lewis (1966), and Ryle (1949). Others may include Dretske (1995), Rey (1982), Rosenthal (1996), Smart (1959), White (1986), and Wilkes (1984).

Type-B views, or nonreductive versions of materialism, usually fall prey to internal difficulties. The only type-B view that seems to be even internally coherent is the view that invokes strong metaphysical necessity in a crucial role. Taking this view, a type-B theorist must hold that (1) zombies and inverted spectra are conceivable but metaphysically impossible; (2) Mary learns something when she sees red, but that this learning can be explained away with a Loar-style analysis; and (3) consciousness cannot be reductively explained, but is physical nevertheless. The central type-B view has never received a definitive statement, but the closest thing to such a statement is given by Levine (1983,1993) and Loar (1990). Others who appear to endorse physicalism without logical Supervenience include Byrne (1993), Flanagan (1992), Hill (1991), Horgan (1984b), Lycan (1995), Papineau (1993), Tye (1995), and van Gulick (1992).

Type-C positions include various kinds of property dualism, in which materialism is taken to be false and some sort of phenomenal or protopheno-menal properties are taken as irreducible. On such a view, (1) zombies and inverted spectra are logically and metaphysically possible; (2) Mary learns something new, and her knowledge is of nonphysical facts; and (3) conscious-ness cannot be reductively explained, but might be nonreductively explained in terms of further laws of nature. Type-C positions are taken by Campbell (1970), Honderich (1981), Jackson (1982), H. Robinson (1982), W. Robinson (1988), Sprigge (1994), and in the present work.

It is perhaps worth mentioning separately the position discussed earlier in which phenomenal properties are identified with the intrinsic properties of physical entities. This sort of view is endorsed by Feigl (1958), Lockwood (1989), Maxwell (1978), and Russell (1926), and I have some sympathy with it myself. I include this as a version of type C, as it takes phenomenal or protophenomenal properties as fundamental, but it has its own metaphysical shape. In particular, it is more of a monism than the natural interpretation

Naturalistic Dualism 167

of type C. Perhaps we can call this position type C, but I will usually include it under type C.

There are two main choice points between types A, B, and C. First, is consciousness logically supervenient (type A versus the rest)? Second, is physicalism true (type B versus type C)? Taking the second choice point first, I have little difficulty in rejecting type B. While it has the virtue of taking consciousness seriously, it relies on a metaphysics that is either incoherent or obscure, and one that is largely unmotivated; the main motivation is simply to avoid dualism at all costs. In the end, this view shares the same explanatory shape as type C, but with an added dose of metaphysical mystery. Type C is straightforward by comparison.

The central choice is the choice between type A and the rest. For myself, reductive functionalism and eliminativism seem so clearly false that I find it hard to fathom how anyone could accept a type-A view. To me, it seems that one could only accept such a view if one believed that there was no significant problem about consciousness in the first place. Nevertheless, expe-rience indicates that almost one-third of the population are willing to accept a type-A position and do not budge. This indicates the Great Divide men-tioned in the preface: the divide between views that take consciousness se-riously and those that do not.

In many ways, the divide between type A and the others is deeper than that between type B and type C. The latter division involves relatively subtle issues of metaphysics, but the former involves some very basic intuitions. Even though type B and type A are both "materialist" views, type-B views are much closer to type-C views in their spirit. Both these views acknowledge the depth of the problem of consciousness where type-A views do not.

Ultimately, argument can take us only so far in settling this issue. If someone insists that explaining access and reportability explains everything, that Mary discovers nothing about the world when she first has a red experi-ence, and that a functional isomorph differing in conscious experience is inconceivable, then I can only conclude that when it comes to experience we are on different planes. Perhaps our inner lives differ dramatically. Per-haps one of us is "Cognitively closed" to the insights of the other. More likely, one of us is confused or is in the grip of a dogma. In any case, once the dialectic reaches this point, it is a bridge that argument cannot cross. Rather, we have reached a brute clash of intuitions of a sort that is com-mon in the discussion of deep philosophical questions. Explicit argument can help us to isolate and characterize the clash, but not to resolve it.

At the beginning of this work, I said that my approach was premised on taking consciousness seriously. We can now see just what this comes to. To take consciousness seriously is to accept just this: that there is something interesting that needs explaining, over and above the performance of various

168 The Irreducibility of Consciousness

functions.33 This has the status of a prima facie premise that only an extremely strong argument could overturn. No argument that I have ever seen comes close to overturning the premise. Indeed, type-A theorists do not usually argue against the premise, but simply deny it. Conversely, beyond a certain point it is almost impossible to argue for the premise, any more than one can argue that conscious experience exists. At best, one can try to clarify the issues in the hope that enlightenment sets in.

With the issues clarified, readers can decide for themselves whether to take consciousness seriously. All I claim is that if one takes consciousness seriously, then property dualism is the only reasonable option. Once we reject reductive functionalism and eliminativism, it follows inexorably that consciousness is not logically supervenient on the physical. And once we reject logical Supervenience, the path to property dualism is unswerving. Type-B views are popular, but do not appear to stand up to close philosophi-cal scrutiny. The main metaphysical choice that remains open is whether to accept a standard type-C view or a type-C view. This is not a question that we have to settle immediately-I do not have a settled opinion on it myself-but in any case, it follows either way that if we want to take con-sciousness seriously, we must admit phenomenal or protophenomenal prop-erties as fundamental.

Some other views found in the philosophical literature do not fall explicitly into the framework I have outlined. With this framework in place, however, it is not hard to locate them and to analyze their problems. I briefly discuss nine such positions in the endnotes: biological materialism,34 physicalist-func-tionalism,35 psychofunctionalism,36 anomalous monism,37 representational-ism,38 consciousness as higher-order thought,39 reductive teleofunctionalism,40 emergent causation,40 and mysterianism.42

6. Reflections on Naturalistic Dualism

Many people, including a past self of mine, have thought that they could simultaneously take consciousness seriously and remain a materialist. In this chapter I have argued that this is not possible, and for straightforward rea-sons. The moral is that those who want to come to grips with the phenomenon must embrace a form of dualism. One might say: You can't have your ma-terialist cake and eat your consciousness too.

All the same, many will be searching for an alternative to the position I have put forward, because they find its dualistic nature unacceptable. This reaction is natural, given the various negative associations of dualism, but I suspect that it is not grounded in anything more solid than contemporary dogma. To see this, it is worthwhile to consider the various reasons that one

Naturalistic Dualism 169

might have for rejecting dualism in favor of materialism, and to measure the force of these reasons as things stand.

The first reason to prefer materialism is simplicity. This is a good reason. Other things being equal, one should prefer a simpler theory over one that is ontologically profligate. Ockham's razor tells us that we should not multiply entities without necessity. But other things are not equal, and in this case there is necessity. We have seen that materialism cannot account for the phenomena that need to be explained. Just as Maxwell sacrificed a simple mechanistic worldview by postulating electromagnetic fields in order to ex-plain certain natural phenomena, we need to sacrifice a simple physicalistic worldview in order to explain consciousness. We have paid due respect to Ockham by recognizing that for materialism to be overthrown, one will need good arguments. But when the arguments against materialism are there, Ockham's razor cannot save it.

The second and perhaps the most pervasive reason to believe in material-ism is inductive: materialism has always worked elsewhere. With phenomena such as life, cognition, and the weather, we either have materialist accounts already or we have good reason to suppose that they are not far off. Why should consciousness be any different?

But this reason is easy to defeat. As we have seen, there is a simple explanation for the success of materialist accounts in various external do-mains. With phenomena such as learning, life, and the weather, all that needs to be explained are structures and functions. Given the causal closure of the physical, one should expect a physical account of this structure and function. But with consciousness, uniquely, we need to explain more than structures and functions, so there is little reason to expect an explanation to be similar in kind.

Indeed, we saw in Chapter 2 that given the nature of our access to external phenomena, we should expect a materialist account of any such phenomena to succeed. Our knowledge of these phenomena is physically mediated, by light, sound, and other perceptual media. Given the causal closure of the physical, we should expect phenomena that we observe by these means to be logically supervenient on the physical-otherwise we would never know about them. But our epistemic access to conscious experience is of an entirely different kind. Consciousness is at the very center of our epistemic universe, and our access to it is not perceptually mediated. The reasons for expect-ing a materialist account of external phenomena therefore break down in the case of consciousness, and any induction from those phenomena will be shaky at best.

Third, many have preferred materialism in order to take science seriously. It has been thought that a dualist view would challenge science on its own grounds. According to Churchland (1988), "[D]ualism is inconsistent with

170 The Irreducibility of Consciousness

evolutionary biology and modern physics and chemistry." But this is quite false. Nothing about the dualist view I advocate requires us to take the physical sciences at anything other than their word. The causal closure of the physical is preserved; physics, chemistry, neuroscience, and cognitive science can proceed as usual. In their own domains, the physical sciences are entirely successful. They explain physical phenomena admirably; they simply fail to explain conscious experience.

Churchland suggests a number of other reasons to reject dualism: (1) the systematic dependence of mental phenomena on neurobiological phenom-ena; (2) modern computational results that suggest that complex results can be achieved without a nonphysical homunculus; and (3) a lack of evidence, explanation, or methodology for dualism. The first two reasons offer no evidence against my view. As for the third, arguments for dualism have already been presented, while dualist explanation and methodology will be illustrated in the remainder of this work.

A fourth motivation to avoid dualism, for many, has arisen from various spiritualistic, religious, supernatural, and other antiscientific overtones of the view. But those are quite inessential. On the view I advocate, conscious-ness is governed by natural law, and there may eventually be a reasonable scientific theory of it. There is no a priori principle that says that all natural laws will be physical laws; to deny materialism is not to deny naturalism. A naturalistic dualism expands our view of the world, but it does not invoke the forces of darkness.

In a related concern, many have thought that to accept dualism would be to give up on explanation. In the words of Dennett (1991), "[G]iven the way that dualism wallows in mystery, accepting dualism is giving up" (p. 37). Perhaps some dualist views have this feature, but it is far from an automatic corollary, as I hope the remainder of this work will make clear.

One occasionally hears a fifth objection to dualism, which is that it cannot explain how the physical and the nonphysical interact. But the answer to this is simple on the natural Supervenience framework: they interact by virtue of Psychophysical laws. There is a system of laws that ensures that a given physical configuration will be accompanied by a given experience, just as there are laws that dictate that a given physical object will gravitationally affect others in a certain way.

It might be objected that this does not tell us what the connection is, or how a physical configuration gives rise to experience. But the search for such a connection is misguided. Even with fundamental physical laws, we cannot find a "connection" that does the work. Things simply happen in accordance with the law; beyond a certain point, there is no asking "how.' As Hume showed, the quest for such ultimate connections is fruitless. If there are indeed such connections, they are entirely mysterious in both the physical and Psychophysical cases, so the latter poses no special problem here.

Naturalistic Dualism 171

It is notable that Newton's opponents made a similar objection to his theory of gravitation: How does one body exert a force on another far away? But the force of the question dissolved over time. We have learned to live with taking certain things as fundamental.

There is also a worry, raised occasionally, about how consciousness might have evolved on a dualist framework: did a new element suddenly pop into nature, as if by magic? But this is not a problem. Like the fundamental laws of physics, Psychophysical laws are eternal, having existed since the beginning of time. It may be that in the early stages of the universe there was nothing that satisfied the physical antecedents of the laws, and so no consciousness, although this depends on the nature of the laws. In any case, as the universe developed, it came about that certain physical systems evolved that satisfied the relevant conditions. When these systems came into exis-tence, conscious experience automatically accompanied them by virtue of the laws in question. Given that Psychophysical laws exist and are timeless, as naturalistic dualism holds, the evolution of consciousness poses no spe-cial problem.

In short, very few of the usual reasons for rejecting dualism have any force against the view I am advocating. The main residual motivation to reject dualism may simply lie in the term's negative connotations, and the fact that it goes against what many of us have been brought up to believe. But once we see past these associations, we see that there is no reason why dualism cannot be a reasonable and palatable view. Indeed, I think that the position I have outlined is one that those who think of themselves as materialists, but who want to take conscious experience seriously, can learn to live with and even to appreciate.

Indeed, mine is a view that many who think of themselves as "materialists" may already implicitly share. All I have done is bring the Ontological impli-cations of a natural view-that consciousness "arises" from the physical, for example-out into the open. Some dualists may even find my view all too materialistic for their tastes, in which case so be it. Ideally, it is a view that takes the best of both worlds and the worst of neither.

This dualism, then, requires us to give up little that is important about our current scientific worldview. It merely requires us to give up a dogma. Otherwise, the view is merely a supplement to the worldview; it is a necessary broadening in order to bring consciousness within its scope. Our credo: If this is dualism, then we should learn to love dualism.

5.The Paradox of Phenomenal Judgment

1. Consciousness and Cognition

So far, the distinctions and divisions between consciousness and cognition have been stressed above all else. Consciousness is mysterious; cognition is not. Consciousness is ontologically novel; cognition is an Ontological free lunch. Cognition can be explained functionally; consciousness resists such explanation. Cognition is governed entirely by the laws of physics; conscious-ness is governed in part by independent Psychophysical laws.

While the focus on these distinctions has been necessary in order to come to grips with the many subtle metaphysical and explanatory issues surrounding conscious experience, it may encourage a misleading picture of the mind. In this picture, consciousness and cognition are utterly detached from each other, living independent lives. One might get the impression that a theory of consciousness and a theory of cognition will have little to do with one another.

This picture is misleading. Our mental life is not alienated from itself in the way that the picture suggests. There are deep and fundamental ties be-tween consciousness and cognition. On one side, the contents of our conscious experiences are closely related to the contents of our cognitive states. When-ever one has a green sensation, individuated phenomenally, one has a corres-ponding green perception, individuated psychologically. On the other side, much cognitive activity can be centered on conscious experience. We know about our experiences, and make judgments about them; as I write this, a great deal of my thought is being devoted to consciousness. These relations between consciousness and cognition are not arbitrary and capricious, but systematic.

The Paradox of Phenomenal Judgment 173

An analysis of this systematic relationship may provide much of the basic material for a theory of consciousness. In this way, we can see that the nature of cognition is not irrelevant to consciousness, but central to its explanation. Of course a theory of cognition cannot do all the explanatory work on its own, but it can nevertheless play a major role. After all, it is through cognition that we get a handle on consciousness in the first place. A thorough investiga-tion of the links between consciousness and cognition can provide the pur-chase we need to constrain a theory of consciousness in a significant way, perhaps ultimately leading to an account of consciousness that neither mysti-fies nor trivializes the phenomenon.

In this chapter, I lay the groundwork for a study of the relationship be-tween consciousness and cognition. The next section introduces some notions that are at the center of this relationship. The remainder of the chapter is largely defensive, addressing various problems that the relationship between consciousness and cognition might seem to pose for a nonreductive view. In the next chapter, I begin the task of building a positive theory that system-atizes the relationship between consciousness and cognition, with the goal of drawing them together into a unified picture of the mind.

Phenomenal judgments

The primary nexus of the relationship between consciousness and cognition lies in phenomenal judgments. Our conscious experience does not reside in an isolated phenomenal void. We are aware of our experience and its con-tents, we form judgments about it, and we are led to make claims about it. When I have a red sensation, I sometimes form a belief that I am having a red sensation, which can issue in a verbal report. At a more abstract level, when one stops to reflect on the mysteries that consciousness poses, as I have been doing throughout this book, one is making judgments about consciousness. At a more concrete level, we frequently form judgments about the objects of our conscious experience (in the environment, for example), as when we think, "There is something red." The various judgments in the vicinity of consciousness I call phenomenal judgments, not because they are phenomenal states themselves, but because they are concerned with phenomenology or with its objects.

Phenomenal judgments are often reflected in claims about consciousness: verbal expressions of those judgments. At various times, people make claims about consciousness ranging from "I have a throbbing pain now" through "LSD gives me bizarre color sensations" to "The problem of consciousness is utterly baffling." These claims and judgments are intimately related to our phenomenology, but they are ultimately part of our psychology. Verbal reports are behavioral acts, and are therefore susceptible to functional expla-

174 The Irreducibility of Consciousness

nation. In a similar way phenomenal judgments are themselves cognitive acts, and fall within the domain of psychology.

It is often taken that beliefs should be understood as functional states, characterized by their causal ties to behavior, the environment, and other beliefs, but this view is not universally accepted. Some hold that phenomenal experience can be partly constitutive of belief or of belief contents. For beliefs about consciousness, the functional view is likely to be particularly controversial: if any beliefs are dependent on conscious experience, beliefs about consciousness are the most likely candidates. I will therefore adopt the less loaded label "judgment" for the functional states in question, and will leave open the question of whether a judgment about consciousness is all there is to a belief about consciousness. We can think of a judgment as what is left of a belief after any associated phenomenal quality is subtracted. That there are purely psychological states that qualify as these judgments should not be a controversial matter. For a start, the disposition to make verbal reports of a certain form is a psychological state; at the very least, we can use the label "judgment" for this disposition. Moreover, whenever I form a belief about my conscious experience, there are all sorts of accompa-nying functional processes, just as there are with any belief. These processes underlie the disposition to make verbal reports, and all sorts of other disposi-tions. If one believes that LSD produces bizarre color sensations, the accom-panying processes may underlie a tendency to indulge in or to avoid LSD in future, and so on. We can use the term "judgment" as a coverall for the states or processes that play the causal role in question. At a first approxima-tion, a system judges that a proposition is true if it tends to respond affirma-tively when queried about the proposition, to behave in an appropriate manner given its other beliefs and desires, and so on.

Judgments can perhaps be understood as what I and my zombie twin have in common. My zombie twin does not have any conscious experience, but he claims that he does; at least, his detailed verbal reports sound the same as my own. As I am using the term, I think it is natural to say that my zombie twin judges that he has conscious experience, and that his judgments in this vicinity correspond one-to-one with mine.

At the end of this chapter, I argue that the semantic content of my phenom-enal beliefs is partly constituted in subtle ways by conscious experience itself (for example, red sensations may play a role in constituting the content of certain beliefs about red sensations). If this is so, then some of the zombie's judgments will have contents that are not as rich as my corresponding belief contents. Nevertheless, they will at least correspond one-to-one to mine, will have the same form, and will function in the same way in directing behavior as mine. So when I talk of a zombie's judgment that he is having a red sensation, I am talking about something interesting in his psychology: at the

The Paradox of Phenomenal Judgment 175

very least, my words can be taken to refer in a deflationary way to the judgment that he expresses using the words "I am having a red sensation" (or words with that sound!). I will talk about "claims" in a similar way, abstracting away from these subtle issues of content.

Strictly speaking, all descriptions of phenomenal claims and judgments in terms of their content (e.g., references to the judgment that one is having a red sensation) should be read in this deflationary way. The full content attributed will certainly be possessed by a subject's phenomenal beliefs, but the question of the content of a judgment is not so clear, precisely because it is not clear what role consciousness plays in constituting the content of a phenomenal belief. I will not make too much of this distinction for much of this chapter, as I will be trying to raise some problems that phenomenal judgments pose for my view in the most acute way possible. At the end of the chapter, I will consider these questions about content in more detail.

Three kinds of phenomenal judgment

Judgments related to conscious experience fall into at least three groups. There are what I will call first-order, second-order, and third-order phenome-nal judgments. I will usually drop the qualifier and speak of "first-order judgments," and so on, where it is understood that these are always phenome-nal judgments.

First-order judgments are the judgments that go along with conscious experiences, concerning not the experience itself but the object of the experi-ence. When I have a red sensation-upon looking at a red book, for in-stance-there is generally an explicit or implicit judgment, "There is some-thing red." When I have the experience of hearing a musical note, there is an accompanying psychological state concerning that musical note. It seems fair to say that any object that is consciously experienced is also Cognitively represented, although there is more to say about this. Alongside every con-scious experience there is a content-bearing cognitive state. This cognitive state is what I am calling a first-order judgment. (One might argue that this state is unlike a belief or judgment in certain ways, as for example it need not be endorsed on reflection. I discuss this at more length in the next chapter, but for now I will speak of "judgments" at least as a first approximation.)

We may think of the contents of these first-order judgments as making up the contents of awareness, where awareness is the psychological counter-Part of consciousness mentioned in Chapter 1: information of which we are aware is roughly information that is accessible to the cognitive system, available for verbal report, and so on. These judgments are not strictly about consciousness. Rather, they are parallel to consciousness, and generally about objects and properties in the environment, or even in the head. In fact, it is

176 The Irreducibility of Consciousness

reasonable to say that a first-order judgment is about what the corresponding experience is about. Where I have an experience of a red book, there is a corresponding first-order judgment about the red book. In a certain sense, we can therefore say that experience and first-order judgments-and therefore consciousness and awareness-share their contents. I will give a more refined account of this relationship in the next chapter.

In this chapter, I will be most concerned with second-order judgments. These are more straightforwardly judgments about conscious experiences. When I have a red sensation, I sometimes notice that I am having a red sensation. I judge that I have a pain, that I experience certain emotional qualities, and so on. In general, it seems that for any conscious experience, if one possesses the relevant conceptual resources, then one at least has the capacity to judge that one is having that experience.

One can also make more detailed judgments about conscious experiences. One can note that one is experiencing a particularly vivid shade of purple, or that a pain has an all-consuming quality, or even that a green after-image is the third such after-image one has had today. Apart from judgments about specific conscious experiences, second-order judgments also include judgments about particular kinds of conscious experiences, as when one notes that some drug produces particularly intense sensations, or that the tingle one gets before a sneeze is particularly pleasurable.

What I will call third-order judgments are judgments about conscious experience as a type. These go beyond judgments about particular experi-ences. We make third-order judgments when we reflect on the fact that we have conscious experiences in the first place, and when we reflect on their nature. I have been making third-order judgments throughout this work. A typical third-order judgment might be, "Consciousness is baffling; I don't see how it could be reductively explained." Others include "Con-scious experience is ineffable," and even "Conscious experience does not exist."

Third-order judgments are particularly common among philosophers, and among those with a tendency to speculate on the mysteries of existence. It is possible that many people go through life without making any third-order judgments. Still, such judgments occur in a significant class of people. The very fact that people make such judgments is something that needs explanation.

To help keep the distinctions in mind, the various kinds of judgments related to consciousness can be represented by the following:

 First-order judgment: That's red!

 Second-order judgment: I'm having a red sensation now.

 Third-order judgment: Sensations are mysterious.

The Paradox of Phenomenal Judgment 177

2. The Paradox of Phenomenal Judgment

The existence of phenomenal judgments reveals a central tension within a nonreductive theory of consciousness. The problem is this. We have seen that consciousness itself cannot be reductively explained. But phenomenal judgments lie in the domain of psychology and in principle should be reduc-tively explainable by the usual methods of cognitive science. There should be a physical or functional explanation of why we are disposed to make the claims about consciousness that we do, for instance, and of how we make the judgments we do about conscious experience. It then follows that our claims and judgments about consciousness can be explained in terms quite independent of consciousness. More strongly, it seems that consciousness is explanatorily irrelevant to our claims and judgments about consciousness. This result I call the paradox of phenomenal judgment.

The paradox of phenomenal judgment does not seem to have received a great deal of attention, but it is put forward vividly by physicist Avshalom Elitzur (1989) as an argument against views that take consciousness to be "passive"; he argues instead for an interactionist dualism.1 The paradox is also expressed by psychologist Roger Shepard (1993), who suggests that it is something we should become reconciled to:

In short, we still seem to be left with a dilemma: No analysis of the purely physical processes in a brain (or in a computer) seems capable of capturing the particular quality of the subjective experience corresponding to those processes. Yet, some such analysis should surely be able to give a causal account of how an individual comes to type a sentence such as the preceding. Perhaps we shall have to reconcile ourselves to accepting that although both the existence of conscious experiences and the similarity relations among their qualia have physi-cal embodiments with physical causes and effects, the conscious experiences or qualia themselves are neither characterizable as physical events nor communica-ble between physical systems. (p. 242)

As we saw in the last chapter, the question of whether consciousness is causally irrelevant in the production of behavior is a complex metaphysical issue that is best left open. But the explanatory irrelevance of consciousness is clearer, and raises many of the same difficulties that would be raised by causal irrelevance. However the metaphysics of causation turns out, it seems relatively straightforward that a physical explanation of behavior can be given that neither appeals to nor implies the existence of consciousness.

When I say in conversation, "Consciousness is the most mysterious thing there is," that is a behavioral act. When I wrote in an earlier chapter "Con-sciousness cannot be reductively explained," that was a behavioral act. When I comment on some particularly intense purple qualia that I am experiencing, that is a behavioral act. Like all behavioral acts, these are in principle ex-

178 The Irreducibility of Consciousness

plainable in terms of the internal causal organization of my cognitive system. There is some story about firing patterns in neurons that will explain why these acts occurred; at a higher level, there is probably a story about cognitive representations and their high-level relations that will do the relevant explan-atory work. We certainly do not know the details of the explanation now, but if the physical domain is causally closed, then there will be some reductive explanation in physical or functional terms.

In giving this explanation of my claims in physical or functional terms, we will never have to invoke the existence of conscious experience itself. The physical or functional explanation will be given independently, applying equally well to a zombie as to an honest-to-goodness conscious experiencer. It therefore seems that conscious experience is irrelevant to the explanations of phenomenal claims and irrelevant in a similar way to the explanation of phenomenal judgments, even though these claims and judgments are cen-trally concerned with conscious experience!

One way to resist this claim would be to argue that the full content of my phenomenal claims and beliefs cannot be reductively explained, because consciousness plays a role in constituting that content. One might argue that a zombie's claims and beliefs are different claims and beliefs, for example (although they look and sound just the same!), because a zombie would not have the full concept of consciousness. But at the very least it is still puzzling that consciousness should be irrelevant to the sounds we make when talking about consciousness, to the finger movements I am making now, and so on; so this response does not remove the full sense of bafflement. So I will set aside this way of thinking about things for now, and will continue to think about claims and judgments in the "deflationary" way that allows that they can be reductively explained.

Another way to resist the point would be to argue that for any high-level property that might be thought relevant in explanation, there will be a low-level explanation that does not invoke the existence of that property. One could argue that a psychological property such as memory is explanatorily irrelevant, as one can give neurophysiological explanations of actions that never once mention memory; one could even argue that temperature is explanatorily irrelevant in physics, as explanatory appeals to temperature can in principle be replaced by a molecular account. (Kim [1989] calls this the problem of explanatory exclusion.) This might suggest that consciousness is no worse off than any other high-level property when it comes to explana-tory irrelevance. If consciousness is on a par with memory or temperature, this is not bad company to be in.

We have seen, however, that high-level properties such as temperature and memory are all logically supervenient on the physical. It follows that when one gives an explanation of some action in neurophysiological terms, this does not make memory explanatorily irrelevant. Memory can inherit

The Paradox of Phenomenal Judgment 179

explanatory relevance by virtue of its logically supervenient status. When we explain a man's desire for female companionship in terms of the fact that he is male and unmarried, this does not make the fact that he is a bachelor explanatorily irrelevant! The general principle here is that when two sets of properties are conceptually related, the existence of an explana-tion in terms of one set does not render the other set explanatorily irrelevant. In a sense, one of the explanations can be a retelling of the other, due to the conceptual relation between the terms involved.

When we tell a story about the interaction of memories, there is a sense in which we are retelling the physical story at a higher level of abstraction. This higher level will omit many details from the physical story, and will therefore often make for a much more satisfying explanation (all those details may have been irrelevant clutter), but it is nevertheless logically related to the lower-level story. The same goes for temperature. These high-level properties are no more rendered explanatorily irrelevant by the exis-tence of a low-level explanation than the velocity of a billiard ball is rendered explanatorily irrelevant by the existence of molecular processes within the ball. In general, the high-level properties in question will constitute a more parsimonious redescription of what a low-level explanation describes. One might say that even a low-level description will often implicitly involve high-level properties, by virtue of their logically supervenient status, even if it does not invoke them explicitly. Where there is logical Supervenience, there is no problem of explanatory irrelevance.

The problems with consciousness are much more serious. Consciousness is not logically supervenient on the physical, so we cannot claim that a physical or functional explanation implicitly involves consciousness, or that consciousness inherits explanatory relevance by logically superven-ing on the properties involved in such an explanation. A physical or func-tional explanation of behavior is independent of consciousness in a much stronger sense. It can be given in terms that do not even imply the exis-tence of conscious experience. Consciousness is conceptually independent of what goes into the explanation of our claims and judgments about con-sciousness.

This is not to say that one can never appeal to conscious experience in the explanation of behavior. It is perfectly reasonable to explain the fact of someone's withdrawal from a flame by noting that they experienced pain. After all, even on the nonreductive view there are lawful regularities between experience and subsequent behavior. Such regularities ultimately depend on regularities at the physical level, however. For any explanation of behavior that appeals to a pain sensation, there is a more fundamental explanation in purely physical/functional terms-perhaps in terms of psychological pain or pain perception-that do not invoke or imply any properties of expe-rience. Experience gains a sort of indirect explanatory relevance in virtue

180 The Irreducibility of Consciousness

of its nomic connection to these physical and functional processes, but it nevertheless remains superfluous to the basic explanation.

To see the problem in a particularly vivid way, think of my zombie twin in the universe next door. He talks about conscious experience all the time- in fact, he seems obsessed by it. He spends ridiculous amounts of time hunched over a computer, writing chapter after chapter on the mysteries of consciousness. He often comments on the pleasure he gets from certain sensory qualia, professing a particular love for deep greens and purples. He frequently gets into arguments with zombie materialists, arguing that their position cannot do justice to the realities of conscious experience.

And yet he has no conscious experience at all! In his universe, the mate-rialists are right and he is wrong. Most of his claims about conscious experi-ence are utterly false. But there is certainly a physical or functional explana-tion of why he makes the claims he makes. After all, his universe is fully law-governed, and no events therein are miraculous, so there must be some explanation of his claims. But such an explanation must ultimately be in terms of physical processes and laws, for these are the only processes and laws in his universe.

(As before, one might plausibly argue that a zombie does not refer to consciousness in the full sense with his word "consciousness." For now, talk of a zombie's claims and judgments about consciousness should be read in the deflationary way discussed earlier. But even if he does not have the full concept, there is no doubt that he judges that he has some property over and above his structural and functional properties-a prop-erty that he calls "consciousness"-and the problem arises as strongly in this form.)

Now my zombie twin is only a logical possibility, not an empirical one, and we should not get too worried about odd things that happen in logically possible worlds. Still, there is room to be perturbed by what is going on. After all, any explanation of my twin's behavior will equally count as an explanation of my behavior, as the processes inside his body are precisely mirrored by those inside mine. The explanation of his claims obviously does not depend on the existence of consciousness, as there is no consciousness in his world. It follows that the explanation of my claims is also independent of the existence of consciousness.

To strengthen the sense of paradox, note that my zombie twin is himself engaging in reasoning just like this. He has been known to lament the fate of his zombie twin, who spends all his time worrying about consciousness despite the fact that he has none. He worries about what that must say about the explanatory irrelevance of consciousness in his own universe. Still, he remains utterly confident that consciousness exists and cannot be reductively explained. But all this, for him, is a monumental delusion. There is no consciousness in his universe-in his world, the eliminativists have been right

The Paradox of Phenomenal Judgment 181

all along. Despite the fact that his cognitive mechanisms function in the same way as mine, his judgments about consciousness are quite deluded.

This paradoxical situation is at once delightful and disturbing. It is not obviously fatal to the nonreductive position, but it is at least something that we need to come to grips with. It is certainly the greatest tension that a nonreductive theory is faced with, and any such theory that does not at least face up to the problem cannot be fully satisfactory. We have to carefully examine the consequences of the situation and separate what is merely counterintuitive from what threatens the viability of a nonreductive view of consciousness.

Nietzsche said, "What does not kill us, makes us stronger." If we can cope with this paradox, we may be led to valuable insights about the relationship between consciousness and cognition. I devote the remainder of this chapter to facing up to the paradox, and related issues about the connection between consciousness and cognition will recur throughout the next few chapters. In this way a theory of consciousness can be set onto much firmer ground.

(One might think one could evade the paradox by embracing what I have called a type-B position, in which consciousness supervenes with metaphysi-cal necessity but not with conceptual necessity, or a type-C positions, in which phenomenal properties constitute the intrinsic nature of the physical. But the paradox arises almost as strongly for these views. Even if these views salvage a sort of causal relevance for consciousness, they still lead to explanatory irrelevance, as explanatory relevance must be supported by conceptual connections. Even on these views, one can give a reductive expla-nation of phenomenal judgments but not of consciousness itself, making consciousness explanatorily irrelevant to the judgments. There will be a processing explanation of the judgments that does not invoke or imply the existence of experience at any stage; the presence of any further "metaphysi-cally necessary" connection or intrinsic phenomenal properties will be con-ceptually quite independent of anything that goes into the explanation of behavior.

Another way to see this: on these views, zombies are still conceivable, and there will be a perfectly good explanation of the zombie's behavior. Because this explanation applies to a zombie, the existence of consciousness will play no essential role in the explanation. But what is going on within the zombies is also going on within us, so the same explanation will apply equally to us. So even on these views there will be an explanation of our phenomenal judgments to which consciousness is quite superfluous.)

Facing up to the paradox

When it comes to the explanation of most of our behavior, the fact that consciousness is explanatorily irrelevant may be counterintuitive, but it is

182 The Irreducibility of Consciousness

not too paradoxical. To explain my reaching for the book in front of me, we need not invoke my phenomenal sensation of the book; it is enough to invoke my perception instead. When a concert-goer sighs at a particularly exquisite movement, one might have thought that the experienced quality of auditory sensations might be central to an explanation of this behavior, but it turns out that an explanation can be given entirely in terms of auditory perception and functional responses. Even in explaining why I withdraw my hand from a flame, a functional explanation in terms of the psychological notion of pain will suffice.

In general, it turns out that where one might think that one would need to invoke phenomenal properties in the explanation of behavior, one can usually invoke psychological properties instead. We saw in Chapter 1 that there is a psychological state underlying every phenomenal state. Where one might have invoked a sensation, one invokes a perceptual registration; where one might have invoked the phenomenal quality of an emotion, one invokes a corresponding functional state; where one might have invoked an occurrent thought, one need only invoke the content of that thought. It is this correspon-dence between phenomenal and psychological properties that makes the explanatory irrelevance of phenomenal properties not too serious a problem in general. It is counterintuitive at first, but it is only counterintuitive. At least for behavior that is not directly concerned with conscious experience, there does not seem to be a pressing need to invoke phenomenal properties in explanation.

It is with our claims and judgments about consciousness that the explana-tory irrelevance of conscious experience becomes troubling. True, it may not be especially worrying that consciousness is explanatorily irrelevant to our first-order phenomenal judgments, such as "That is a red thing," It is reasonable that these should be explained purely in terms of perception and other psychological processes; after all, the judgments in question are not directly concerned with conscious experience, but with the state of the world. For second- and third-order phenomenal judgments, however, explanatory irrelevance seems to raise real problems. It is these judgments that are about conscious experience, and that are responsible for our talking about our sensations and for philosophers' worries about the mysteries of conscious-ness. It is one thing to accept that consciousness is irrelevant to explaining how I walk around the room; it is another to accept that it is irrelevant to explaining why I talk about consciousness. One would surely be inclined to think that the fact that I am conscious will be part of the explanation of why I say that I am conscious, or why I judge that I am conscious; and yet it seems that this is not so.

After all, part of the explanation of why we claim and judge that there is water will involve the fact that there is indeed water. In a similar way, it seems that the existence of stars and planets is almost certainly explanatorily

The Paradox of Phenomenal Judgment 183

relevant to our judging that there are stars and planets. As a rule, when we judge truly and reliably that P, the fact that P is true generally plays a central role in the explanation of the judgment. There are some judgments for which the objects of those judgments are explanatorily irrelevant to the judgments themselves. Think of religious beliefs, for instance, or beliefs about UFOs, which can arguably be explained without invoking any gods or UFOs. But these are all quite possibly false beliefs, and not obviously instances of knowledge. By contrast, we know that we are conscious.

Here we are faced with a difficult situation: how can knowledge of con-sciousness be reconciled with the fact that consciousness is explanatorily irrelevant to phenomenal judgments? If phenomenal judgments arise for reasons independent of consciousness itself, does this not mean that they are unjustified? This, above all, is the central difficulty posed by the paradox of phenomenal judgment, and I will address it at length later in this chapter.

The paradox is a consequence of the facts that (1) the physical domain is causally closed; (2) judgments about consciousness are logically supervenient on the physical; (3) consciousness is not logically supervenient on the physical; and (4) we know we are conscious. From premises (1) and (2) it follows that judgments about consciousness can be reductively explained. In combination with premise (3), this implies that consciousness is explanatorily irrelevant to our judgments, which lies in tension with premise (4). Thus we have the paradox. One might try to escape the paradox by denying any one of these premises. I will consider each of these escape routes briefly.

Some dualists will deny premise (1). Traditionally, a Cartesian interaction-ist dualism has been motivated by the thought that only this can give con-sciousness the relevance to action that it deserves. Indeed, Elitzur (1989) argues directly from the existence of claims about consciousness to the conclusion that the laws of physics cannot be complete, and that conscious-ness plays an active role in directing physical processes (he suggests that the second law of thermodynamics might be false). But I have already argued that interactionist dualism is of little help in avoiding the problem of explana-tory irrelevance.

Some might be tempted to deny premise (2), but recall that we have defined judgments so that they are functional states, logically supervenient on the physical. Now, some might argue that there is no such functional state that remotely resembles what we think of as a judgment; but even so, we can simply retreat to claims about consciousness, which are behavioral acts and so more straightforwardly logically supervenient, and which raise the difficulties almost as strongly. Even if someone argued that behavioral acts are not purely physical (they might argue that conscious experience is required for something to qualify as a claim rather than a noise, or as a claim about consciousness), it is still surprising that consciousness is explanatorily irrelevant to the sounds we produce, and to the marks we write, all of

184 The Irreducibility of Consciousness

which can be systematically interpreted as concerning consciousness. So some analogous problems will arise no matter how we define the relevant states. Still, this sort of consideration can play at least a subsidiary role in dealing with the paradox, as it is plausibly beliefs rather than claims that are most closely connected to knowledge, and some sort of phenomenal belief content may be constituted by experience itself. I return to this matter later in the chapter.

Reductionists and eliminativists will of course deny premise (3) or (4). I have argued exhaustively for (3) already, so I will not repeat the arguments here. Similarly, the denial of premise (4) leads to eliminativism, an option I have already rejected. Still, I will examine a way that a reductionist might exploit the paradox of phenomenal judgment shortly.

It seems to me that the most reasonable attitude to take is to recognize that all the premises are probably true; and to see how they can be reconciled with one other. We know there is conscious experience; the physical domain is almost certainly causally closed; and we have established earlier that consciousness is not logically supervenient on the physical. The trick is to learn to live with the combination.

3. On Explaining Phenomenal Judgments

Given what has gone before, explaining why we say the things we do about consciousness emerges as a reasonable and interesting project for cognitive science. These claims are behavioral acts, and should be as susceptible to explanation as any other behavioral act. Indeed, there should be rich pickings for any cognitive scientist who takes this path. Explaining our claims and judgments about consciousness may be difficult, but it will not be as difficult as explaining consciousness itself. This explanation will not automatically yield an explanation of consciousness, of course, but it may well point us in

the right direction.

We can do more than accept the possibility of such an explanation as an intellectual conclusion, derived from the causal closure of physics and the logical Supervenience of behavior. There are independent reasons for think-ing that phenomenal judgments will be natural concomitants of certain kinds of cognitive processes, and that on reflection one should expect such judg-ments from an intelligent system with a certain design. If so, then the explana-tion of the claims and judgments may not be as difficult as one might think; they might fall out of some basic principles about cognitive design.

Here, I will provide just a very brief sketch of why one might think this; I go into this matter in more detail in Chapter 8. To get some feel for the sit-uation, imagine that we have created computational intelligence in the form of an autonomous agent that perceives its environment and has the capacity

The Paradox of Phenomenal Judgment 185

to reflect rationally on what it perceives. What would such a system be like? Would it have any concept of consciousness, or any related notions?

To see that it might, note that on the most natural design such a system would surely have some concept of self-for instance, it would have the ability to distinguish itself from the rest of the world, and from other entities resembling it. It also seems reasonable that such a system would be able to access its own cognitive contents much more directly than it could those of others. If it had the capacity to reflect, it would presumably have a certain direct awareness of its own thought contents, and could reason about that fact. Furthermore, such a system would most naturally have direct access to perceptual information, much as our own cognitive system does.

When we asked the system what perception was like, what would it say? Would it say, "It's not like anything"? Might it say, "Well, I know there is a red tricycle over there, but I have no idea how I know. The information just appeared in my database"? Perhaps, but it seems unlikely. A system designed this way would be quite inefficient and unnatural; its access to its own perceptual contents would be curiously indirect. It seems much more likely that it would say, "I know there is a red tricycle because I see it there." When we ask it in turn how it knows that it is seeing the tricycle, the answer would very likely be something along the lines of "I just see it."

It would be an odd system that replied, "I know I see it because sensors 78-84 are activated in such-and-such a way." As Hofstadter (1979) points out, there is no need to give a system such detailed access to its low-level parts. Even Winograd's program SHRDLU (1972) did not have knowledge about the code it was written in, despite the fact that it could perceive a virtual world, make inferences about that world, and even justify its knowledge to a limited degree. Such extra knowledge would seem to be quite unnecessary, and would only complicate the processes of awareness and inference.

Instead, it seems likely that such a system would have the same kind of attitude toward its perceptual contents as we do toward ours, with its knowl-edge of them being direct and unmediated, at least as far as the system is concerned. When we ask how it knows that it sees the red tricycle, an efficiently designed system would say, "I just see it!" When we ask how it knows that the tricycle is red, it would say the same sort of thing that we do: "It just looks red." If such a system were reflective, it might start wonder-ing about how it is that things look red, and about why it is that red just is a particular way, and blue another. From the system's point of view it is just a brute fact that red looks one way, and blue another. Of course from our vantage point we know that this is just because red throws the system into one state, and blue throws it into another; but from the machine's point of view this does not help.

As it reflected, it might start to wonder about the very fact that it seems to have some access to what it is thinking, and that it has a sense of self. A

186 The Irreducibility of Consciousness

reflective machine that was designed to have direct access to the contents of its perception and thought might very soon start wondering about the mysteries of consciousness (Hofstadter 1985a gives a rich discussion of this idea): "Why is it that heat feels this way?"; "Why am I me, and not someone else?"; "I know my processes are just electronic circuits, but how does this explain my experience of thought and perception?"

Of course, the speculation I have engaged in here is not to be taken too seriously, but it helps to bring out the naturalness of the fact that we judge and claim that we are conscious, given a reasonable design. It would be a strange kind of cognitive system that had no idea what we were talking about when we asked what it was like to be it. The fact that we think and talk about consciousness may be a consequence of very natural features of our design, just as it is with these systems. And certainly, in the explanation of why these systems think and talk as they do, we will never need to invoke full-fledged consciousness. Perhaps these systems are really conscious and perhaps they are not, but the explanation works independently of this fact. Any explanation of how these systems function can be given solely in compu-tational terms. In such a case it is obvious that there is no room for a ghost in the machine to play an explanatory role.

All this is to say (expanding on a claim in Chapter 1) that consciousness is surprising, but claims about consciousness are not. Although consciousness is a feature of the world that we would not predict from the physical facts, the things we say about consciousness are a garden-variety cognitive phenom-enon. Somebody who knew enough about cognitive structure would immedi-ately be able to predict the likelihood of utterances such as "I feel conscious, in a way that no physical object could be," or even Descartes's "Cogito ergo sum." In principle, some reductive explanation in terms of internal processes should render claims about consciousness no more deeply surprising than any other aspect of behavior. I have gestured toward such an explanation above, and will consider the matter in more detail in a later chapter.

We will see later that the details of an appropriate explanation can be very useful in getting a theory of consciousness off the ground. The relationship between an explanation of phenomenal judgments and an explanation of consciousness is a subtle one, however. Before proceeding, I will consider a less subtle response to the situation we are placed in.

Is explaining the judgments enough?

At this point a natural thought has probably occurred to many readers, especially those of a reductionist bent: If one has explained why we say we are consciousness, and why we judge that we are conscious, haven't we explained all that there is to be explained? Why not simply give up on the quest for a theory of consciousness, declaring consciousness itself a chimera?

The Paradox of Phenomenal Judgment 187

Even better, why not declare one's theory of why we judge that we are conscious to be a theory of consciousness in its own right? It might well be suggested that a theory of our judgments is all the theory of consciousness that we need.

This position gets some support from considerations about judgments in other domains. It might be thought that the widespread belief in gods, found in all sorts of diverse cultures, provides an excellent reason to believe that gods exist. But there is an alternative explanation of this belief, in terms of social and psychological forces. Atheists might appeal to people's psychologi-cal insecurity in the face of the cosmos, to the need for a common outlet for spiritual or emotional expression, and to the intrinsically self-propagating nature of certain idea systems, to explain why it is all but inevitable that religious beliefs should be widespread, given our nature and circumstances. One can even point to the existence of certain highly plausible but faulty arguments for the existence of a god, such as the argument from design and the cosmological arguments. Although these arguments are faulty, they are not obviously faulty (in particular, the argument from design could reason-ably have been seen as compelling before the time of Darwin), and it is not hard to see why they should generally contribute toward the naturalness of religious belief.

The observation that widespread religious belief might be explained in this way, without appeal to the existence of any gods, is generally taken to provide further evidence that no gods in fact exist. On this interpretation, the atheistic hypothesis can not only explain the complex structure of nature as well as the theistic hypothesis; it can even explain why the theistic hypothe-sis is so popular. This is a powerful way to cut the support from underneath an opposing view. In the case of religious belief, the argument seems very strong. It makes a tempting argument in the case of consciousness, too.

This is surely the single most powerful argument for a reductive or elimina-tivist view of consciousness. But it is not enough. The analogy fails. Explaining our judgments about consciousness does not come close to removing the mysteries of consciousness. Why? Because consciousness is itself an expla-nandum. The existence of God was arguably hypothesized largely in order to explain all sorts of evident facts about the world, such as its orderliness and its apparent design. When it turns out that an alternate hypothesis can explain the evidence just as well, then there is no need for the hypothesis of God. There is no separate phenomenon God that we can point to and say: that needs explaining. At best, there is indirect evidence.2 Similarly, the existence of UFOs is often postulated to explain strange events in the sky, markings in the ground, disappearances in the Bermuda Triangle, the claims of UFO "survivors," and so on. If it turns out that this evidence can be explained without postulating the existence of UFOs, then our reason for believing in UFOs disappears.

188 The Irreducibility of Consciousness

But consciousness is not an explanatory construct, postulated to help explain behavior or events in the world. Rather, it is a brute explanandum, a phenomenon in its own right that is in need of explanation. It therefore does not matter if it turns out that consciousness is not required to do any work in explaining other phenomena. Our evidence for consciousness never lay with these other phenomena in the first place. Even if our judgments about consciousness are explained reductively, all this shows is that our judgments can be explained reductively. The mind-body problem is not that of explaining our judgments about consciousness. If it were, it would be a relatively trivial problem. Rather, the mind-body problem is that of ex-plaining consciousness itself. If the judgments can be explained without explaining consciousness, then that is interesting and perhaps surprising, but it does not remove the mind-body problem.

To take the line that explaining our judgments about consciousness is enough (just as explaining our judgments about God is enough) is most naturally understood as an eliminativist position about consciousness (as one analogously takes an eliminativist position about God). As such it suffers from all the problems that eliminativism naturally faces. In particular, it denies the evidence of our own experience. This is the sort of thing that can only be done by a philosopher-or by someone else tying themselves in intellectual knots. Our experiences of red do not go away upon making such a denial. It is still like something to be us, and that is still something that needs explanation. To throw out consciousness itself as a result of the para-dox of phenomenal judgment would be to throw out the baby with the bathwater.

There is a certain intellectual appeal to the position that explaining phe-nomenal judgments is enough. It has the feel of a bold stroke that cleanly dissolves all the problems, leaving our confusion lying on the ground in front of us exposed for all to see. Yet it is the kind of "solution" that is satisfying only for about half a minute. When we stop to reflect, we realize that all we have done is to explain certain aspects of our behavior. We have explained why we talk in certain ways, and why we are disposed to do so, but we have not remotely come to grips with the central problem, namely conscious experience itself. When thirty seconds are up, we find ourselves looking at a red rose, inhaling its fragrance, and wondering: "Why do I experience it like this?" And we realize that this explanation has nothing to say about the matter.

If this position is not taken as a kind of eliminativism, it can perhaps be taken as a kind of functionalist position, in which the notion of consciousness is construed as "the thing responsible for judgments about consciousness." But this is as inadequate as any other functional definition of consciousness. Whether or not consciousness is in fact responsible for judgments about consciousness, this does not seem to be a conceptual truth. After all, it is at

The Paradox of Phenomenal Judgment 189

least logically possible that one could explain the judgments without ex-plaining consciousness, whether or not it is plausible; and that is enough to show that this construal of consciousness is a false one.

There are other variations on this line of argument. For instance, one could argue that there is a purely reductive explanation of why I think that consciousness cannot be reductively explained, or of why I think conscious-ness is not logically supervenient, or of why I think it cannot be functionally defined. We might even reductively explain why I think conscious experience is an explanandum. This might be thought to undercut my arguments in earlier sections entirely, opening the way for a reductive view of conscious-ness. But again this view can be satisfying only as a kind of intellectual cut and thrust. At the end of the day, we still need to explain why it is like this to be a conscious agent. An explanation of behavior or of some causal role is simply explaining the wrong thing. This might seem to be mule-headed stubbornness, but it is grounded in a simple principle: our theories must explain what cries out for explanation.

This line of argument is perhaps the most interesting that a reductionist or eliminativist can take-if I were a reductionist, I would be this sort of reductionist-but at the end of the day it suffers from the problem that all such positions face: it does not explain what needs to be explained. Tempting as this position is, it ends up failing to take the problem seriously. The puzzle of consciousness cannot be removed by such simple means.3

Dennett on phenomenal judgments

One advocate of the position that our judgments about consciousness are all we need to explain is Daniel Dennett. In a 1979 paper he writes:

I am left defending the view that such judgments exhaust our immediate con-sciousness, that our individual stream of consciousness consists of nothing but such propositional episodes, or better: that such streams of consciousness, com-posed exclusively of such propositional episodes, are the reality that inspires the variety of misdescriptions that pass for theories of consciousness, both homegrown and academic: My view, put bluntly, is that there is no phenome-nological manifold in any such relation to our reports. There are the public reports we issue, and then there are the episodes of our propositional awareness, our judgments, and then there is-so far as introspection is concerned-dark-ness. (1979, p. 95)

To this, all I can say is that Dennett's introspection is very different from mine. When I introspect, I find sensations, experiences of pain and emotion, and all sorts of other accoutrements that, although accompanied by judg-ments, are not only judgments-unless one redefines the notion of judgment, or of "episodes of our propositional awareness," to include such experiences. If we redefine the terms in this way, then Dennett's position is reasonable,

190 The Irreducibility of Consciousness

but there is no longer any reason to suppose that our judgments can be reductively explained. If judgments are instead construed as functionally individuated states such as dispositions to report, as I think Dennett intends, then his thesis becomes implausible. It simply consists in a denial of the data that a theory of consciousness must explain.

What might be going on when someone claims that introspection reveals only judgments? Perhaps Dennett is a zombie.4 Perhaps he means something unusual by "judgment." Most likely, however, he has taken something else for introspection: what we might call extrospection, the process of observing one's own cognitive mechanisms "from the outside," as it were, and reflecting on what is going on. Observing one's mechanisms, it is easy to come to the conclusion that it is judgments that are doing all the work. All that is going on in the relevant cognitive processes is a lot of categorization, distinction, and reaction. The processes involved with my perception of a yellow object can plausibly be fully explained in terms of certain retinal sensitivities, trans-formations into internal representations, and categorization and labeling of these representations. But this does not explain the contents of introspection; it explains only the processes involved. Extrospection is not introspection, although it is easy to see how a philosopher inclined to speculate on his own internal mechanisms could take one for the other. Conscious experience remains untouched by this explanatory method. (Perhaps the descriptions just given might provide an excellent account of the phenomenology of blindsight (described in Chapter 6), if not of ordinary consciousness!)

A similar move is made by Dennett in what is perhaps the central argument of Consciousness Explained (1991). Having presented his theory of report-ability, Dennett needs to argue that it explains everything that needs to be explained, and in particular that it explains experience insofar as experience needs to be explained. After much preliminary skirmishing, he makes the crucial argument (pp. 363-64) that a theory of experience needs to explain why things seem the way they do to us. And he argues that his theory can explain why things seem the way they do to us. Hence, he concludes, his theory explains everything that needs to be explained.

This is an elegant argument, with a ring of plausibility that many reduction-ist arguments about consciousness lack. But its elegance derives from the way it exploits a subtle ambiguity in the notion of "seeming," which balances on the knife-edge between the phenomenal and psychological realms. There is a phenomenal sense of "seem," in which for things to seem a certain way is just for them to be experienced a certain way. And there is a psychological sense of "seem," in which for things to seem a certain way is for us to be disposed to judge that they are that way. It is in the first sense that a theory of experience must explain the way things seem. But it is in the second sense that Dennett's theory explains it.5

The Paradox of Phenomenal Judgment 191

Once this subtle equivocation is noted, the argument loses most of its force. When Dennett says that his theory explains the way things seem to us, what this ultimately comes to is that it explains why we say that things are that way, and why we behave correspondingly in other fashions. (As Dennett himself notes, his theory of consciousness is grounded in his quasi-behaviorist theory of content.) But that sort of explanation falls far short of what a theory of consciousness needs to explain. At the end of the day, call-ing a theory of this sort a theory of consciousness begs all the important questions.

In general, when one starts from phenomenal judgments as the explananda of one's theory of consciousness, one will inevitably be led to a reductive view. But the ultimate explananda are not the judgments but experiences themselves. No mere explanation of dispositions to behave will explain why there is something it is like to be a conscious agent.

4. Arguments Against Explanatory Irrelevance

We have seen that the paradox of phenomenal judgment leads to counterintu-itive consequences. But so far this is all that we have seen. Some people will think that the consequences are not just counterintuitive but fatal. To estab-lish this, these objectors need an argument. Such an argument would show us why the explanatory irrelevance of consciousness simply cannot be true.

Such arguments are surprisingly hard to come by, but they can be made. The general idea is to argue that explanatory irrelevance is inconsistent with some well-established fact about ourselves. I can see three ways this might go. It might be argued that explanatory irrelevance is inconsistent with the fact that we know about our conscious experiences; or that it is inconsistent with the fact that we remember our conscious experiences; or that it is inconsistent with the fact that we refer to our conscious experiences. I do not think that any of these arguments are compelling, but they all raise interesting issues and all need to be expressed.

Some of these arguments are most naturally framed in terms of causal irrelevance rather than explanatory irrelevance. In order to give these argu-ments their full power, I will temporarily concede the causal irrelevance of experience, in order to see whether the arguments succeed. It is possible that similar arguments could be made wholly in terms of explanatory irrele-vance, but they would be more complicated. In any case, I have at least allowed that it might turn out that experience is causally irrelevant, and it is interesting to see whether this would have fatal consequences.

In order to allow an opponent's objections their full force, I will also occasionally speak of "beliefs" rather than "judgments" in what follows. As

192 The Irreducibility of Consciousness

I noted earlier, my main line of defense will not turn on the distinction between beliefs and judgments, so I will not make too much of it here. That issue might still have a supporting role to play, though. In what follows, it should at least be kept in the back of one's mind that (1) when talking about a zombie's beliefs and judgments, a deflationary notion is being stipulated, and (2) my own phenomenal beliefs, in the full sense, may be partly consti-tuted by conscious experience.

5. The Argument from Self-Knowledge*

The most difficult problem posed by explanatory irrelevance is the one I have already discussed: our knowledge of our own conscious experiences. On the face of it, we do not just judge that we have conscious experiences; we know that we have conscious experiences. But if a nonreductive view is right, then experience is explanatorily irrelevant to the formation of the judgment; the same judgment would have been formed even if experience were absent. It may therefore seem hard to see how that judgment can qualify as knowledge.

This might simply be phrased as a challenge: If experience is explanatorily irrelevant, how can we know about experience? As such, it is an important challenge, and one of the central questions about conscious experience. There are already many such difficult questions, however, and we may not be able to answer them before we develop a detailed theory of conscious-ness. It can also be phrased more strongly as an argument: If experience is ex-planatorily irrelevant, then we could not know that we have experiences. It is arguments of this sort that I am concerned to answer here. I will also make some suggestions in answer to the challenge, but that is a project that will recur.

I can see two related ways that such an argument might go. First, it might proceed directly from the possibility of my zombie twin. My zombie twin makes the same phenomenal judgments that I do. Where I judge that I am conscious, he judges that he is conscious. Further, his judgments are produced by the same mechanisms as my judgments. If justification accrues to judg-ments solely in virtue of the mechanisms by which they are formed, as is often supposed, then the zombie's judgments will be as justified as mine. But surely his judgments are not justified at all. After all, they are utterly and systematically false. It seems to follow that my judgments cannot be justified, either. They are produced by the same mechanisms that are respon-sible for deluded judgments in a zombie, and so they surely cannot qualify as knowledge.

If my phenomenal judgments are no more justified than a zombie's, then the ground is cut out from under the nonreductive position. The very starting

The Paradox of Phenomenal Judgment 193

point of the nonreductive position, our knowledge of the fact of experience, would be destroyed. It follows that this point functions as both a challenge and an argument. As a challenge: How can my judgments be any more justified than a zombie's, given that they are formed by the same mechanisms? As an argument: If my judgments are formed by the same mechanisms as a zombie's, they cannot be any more justified.

The second argument appeals to a causal theory of knowledge. It is often held that the crucial factor in justifying a belief about an entity is an appro-priate causal connection between the belief and the entity it is about. My beliefs about the table I am looking at, for example, are justified at least in part by the fact that the table is causally responsible for the beliefs. Propo-nents of a causal theory hold that a judgment about some object or state of affairs must bear a causal relation to that object or state of affairs if it is to count as knowledge (perhaps with exceptions in a priori domains such as conceptual or mathematical knowledge). Certainly, it seems that if my belief that John is in the pool bears no causal relation to John or the pool, then I do not know that John is in the pool.

But experience is causally irrelevant, or so I am conceding for now. A conscious experience plays no causal role in the formation of a judgment about that experience. If a causal theory of knowledge is correct, it follows that we cannot know anything about our experiences. Again, there is a challenge and an argument. The challenge: How can I know about experi-ence, given that experience does not cause my judgments? The argument: If experience plays no causal role in the formation of my judgments, then they cannot count as knowledge.

Shoemaker (1975a) uses arguments like these to argue for materialism about consciousness, and in fact to argue for reductive functionalism. Shoe-maker explicitly assumes a causal theory of knowledge, arguing that if we are to know about experience, then it must cause our introspective beliefs about experience. He also uses a version of the zombie argument to support reductive functionalism. If zombies or their functional equivalents are logi-cally possible, then experience is inaccessible to introspection: zombies have the same introspective mechanisms that we do, so those mechanisms do not allow us to determine whether or not we are zombies. Shoemaker concludes that zombies and their functional equivalents must be logically impossible.

The response to all of these arguments is fairly clear, I think. A property dualist should argue that a causal theory of knowledge is not appropriate for our knowledge of consciousness, and that the justification of our judg-ments about consciousness does not lie with the mechanisms by which those judgments are formed. Knowledge of conscious experience is in many im-portant respects quite different from knowledge in other domains. Our knowledge of conscious experience does not consist in a causal relationship to experience, but in another sort of relationship entirely.

194 The Irreducibility of Consciousness

This conclusion can be supported on independent grounds. One way to come to this independent support is to first consider another way that a property dualist might try to respond: through a reliabilist theory of knowl-edge. This might seem a promising response at first, but I think that one can see that a reliabilist theory is inappropriate for dealing with our knowledge of consciousness. It turns out that a causal theory is inappropriate for the same reason.

On a reliabilist theory, beliefs about a subject matter are justified if they are formed by a reliable process; that is, if they are formed by a process that tends to produce true beliefs. Perceptual beliefs, for example, are justified if they come about via optical stimulation from objects in the environment, a process that generally produces true beliefs; they are not justified if they are produced by hallucination, which is a very unreliable mechanism. It is entirely compatible with a nonreductive theory of experience that in the actual world, our phenomenal judgments are reliable: at least as a matter of nomic correlation, it seems likely that when one judges that one is having a visual experience, one is having a visual experience. The phenomenal judgments of my zombie twin, by contrast, are entirely unreliable; his judg-ments are generally false.

It might therefore seem that a reliabilist theory is the answer to our difficulties: it implies that our judgments about experience might be justified even in the absence of a direct causal connection, and it has the resources to explain the fact that my judgments are justified while my zombie twin's are not. But many will find that the appeal to a reliabilist theory is unsatisfying all the same; it has the feeling of a slippery maneuver that cannot meet the burden it is being asked to carry. The knowledge that a reliabilist theory grants us seems too weak to count as the kind of knowledge that we have of our conscious experience. On reflection, it is not hard to see why.

The trouble is that if our beliefs about consciousness were justified only by a reliable connection, then we could not be certain that we are conscious. The mere existence of a reliable connection cannot deliver certainty, for we have no way to rule out the possibility that the reliable connection is absent and that there is no consciousness at the other end. The only way to be sure here would be to have some further access to the other end of the connection; but that would be to say that we have some further basis to our knowledge of consciousness. This situation is often deemed acceptable for our knowl-edge of the external world: we do not need to be certain that chairs exist in order to know (in an everyday sense) that chairs exist, so it is not a problem that we are not certain that there is a reliable connection between chairs and our judgments about chairs. But we are certain that we are conscious; at least, this certainty is at the foundation of the position I have advocated. Perhaps the knowledge that we are conscious can be doubted in various "philosophical" ways, but not in the very direct way-analogous to doubting

The Paradox of Phenomenal Judgment 195

our knowledge of the external world-that would be granted if our beliefs were justified only by a reliable connection.

Beliefs justified only by a reliable connection are always compatible with the existence of skeptical hypotheses. These concern scenarios where things seem exactly the same to a subject but in which the beliefs are false, because the reliable connection does not hold. In the case of perceptual knowledge, for example, one can construct a case in which the reliable connection is absent-a case where the subject is a brain in the vat, say-and everything will still seem the same to the subject. Nothing about a subject's core episte-mic situation rules this scenario out. But in the case of consciousness, one cannot construct these skeptical hypotheses. Our core epistemic situation already includes our conscious experience. There is no situation in which everything seems just the same to us but in which we are not conscious, as our conscious experience is (at least partly) constitutive of the way things seem.

It is notable that in constructing skeptical scenarios relevant to other sorts of knowledge, such as our knowledge of the external world, philosophers are always careful to stipulate that a skeptical scenario is experientially identi-cal to the original scenario. As Descartes noted, skepticism goes only so far. If a skeptical scenario involves a vastly different set of experiences at its center-a host of bright flashing yellow and green experiences with a deafen-ing noise, say-then it is ruled out automatically. We know (in a much stronger sense than before) that such a situation is not our situation.

It follows that a reliabilist account of knowledge cannot deliver knowledge that is strong enough to have the character of our knowledge of conscious experience, and is therefore inappropriate in this case. But everything I have said about a reliabilist account of knowledge also applies to a causal account of knowledge. Where there is causation, there is contingency: a causal con-nection that holds might not have held. If the sole source of justification for a belief about X is a causal connection to X, then a subject cannot know for certain that the causal connection exists. The only way they might know this for certain would be if they had some independent access to X or to the causal chain, but this would imply knowledge grounded in something more than the causal chain itself. There will always be a skeptical scenario in which everything seems just the same to the subject, but in which the causal connection is absent and in which X does not exist; so the subject cannot know for certain about X. But we do know for certain that we are conscious; so a causal account of this knowledge is inappropriate.

Of course, an opponent might simply deny that our knowledge of con-sciousness is certain, and assert that there are skeptical scenarios that we cannot rule out-a zombie scenario, for example. But anyone who takes this view will likely be an eliminativist (or a reductive functionalist) about consciousness from the start. If one accepts that our immediate evidence does not rule out the possibility that we are zombies, then one should embrace

196 The Irreducibility of Consciousness

the conclusion that we are zombies: it leads to a much simpler view of the world, for a start. But the reason there is a problem about consciousness is that our immediate evidence does rule out that possibility. To take conscious-ness seriously is to accept that we have immediate evidence that rules out its nonexistence. Of course, all this is open to argument in the usual way; but the point is that there is no special reason to start disputing this at this point in the argument. Eliminativists and reductive functionalists have departed long ago. If one takes consciousness seriously, then one has good reason to believe that a causal or reliabilist account of our phenomenal knowledge is inappropriate.

What justifies phenomenal judgments?

The basic problem with the accounts above is that they make our access to consciousness mediated, in the way that our access to objects in the environ-ment is mediated, by some sort of causal chain or reliable mechanism. This sort of mediation is appropriate when there is a gap between our core epistemic situation and the phenomena in question, as in the case of the external world: we are connected to objects in the environment from a distance. But intuitively, our access to consciousness is not mediated at all. Conscious experience lies at the center of our epistemic universe; we have access to it directly.

This raises a question. What is it that justifies our beliefs about our experi-ences, if it is not a causal link to those experiences, and if it is not the mechanisms by which the beliefs are formed? I think the answer to this is clear: it is having the experiences that justifies the beliefs. For example, the very fact that I have a red experience now provides justification for my belief that I am having a red experience. Change the red experience to a different sort of experience, or remove it altogether, and the chief source of justifica-tion for my belief is removed. When I believe that I am experiencing a loud noise, my warrant for that belief stems chiefly from my experience of a loud noise. Indeed, one might ask, from where else could it stem?

We can put the point by noting, as before, that experience is part of our core epistemic situation. Replace my bright red experiences by dull green experiences, and you change my evidence for some of my beliefs, including my belief that I am having a bright red experience. This is mirrored in the fact that there is no way to construct a skeptical scenario in which I am in a qualitatively equivalent epistemic position, but in which my experiences are radically different. My experiences are part of my epistemic situation, and simply having them gives me evidence for some of my beliefs.

All this is to say that there is something intrinsically epistemic about experience. To have an experience is automatically to stand in some sort

The Paradox of Phenomenal Judgment 197

of intimate epistemic relation to the experience-a relation that we might call "acquaintance." There is not even a conceptual possibility that a sub-ject could have a red experience like this one without having any epi-stemic contact with it: to have the experience is to be related to it in this way.

Note that I do not say that to have an experience is automatically to know about it, in the sense in which knowledge requires belief. I think that thesis would be false: we have many experiences that we do not have beliefs about, and so do not know about. Further, one might have an experience without conceptualizing the experience in any way. To have an experience, and consequently to be acquainted with the experience, is to stand in a relation-ship to it more primitive than belief: it provides evidence for our beliefs, but it does not in itself constitute belief.

Indeed, nothing I have said implies that all beliefs about experiences are incorrigible, in that every such belief is automatically fully justified. Because beliefs about experiences lie at a distance from experiences, they can be formed for all sorts of reasons, and sometimes unjustified beliefs will be formed. If one is distracted, for example, one may make judgments about one's experiences that are quite false. The claim is not that having an experi-ence is the only factor that may be relevant to the justification or lack of justification of a belief about experience. The claim is simply that it is a factor-perhaps the primary factor-and provides a potential source of jus-tification that is not present when the experience is absent.

Some might find all this an ad hoc construction to save a troubled theory, but I do not think it is ad hoc. We have very good reason, quite independent of any considerations about explanatory irrelevance, to believe that the epistemology of experience is special, and very different in kind from episte-mology in other domains. Many have spoken of our "direct knowledge" of or "acquaintance" with experience, without being forced into the position as a defensive maneuver. Many have even claimed that knowledge of experi-ence is the foundation of all knowledge, precisely because we stand in such a direct epistemic relationship to it. The claim that all knowledge derives from knowledge of experience may have been overblown, but the general point that there is something special about our knowledge of experience has never been overturned.6

Similarly, the claim that experiences themselves justify our beliefs about experience is easy to motivate on independent grounds. For example, in his careful discussion of our knowledge of our own minds, Siewert (1994)-who takes consciousness seriously, but who shows no signs of sympathy with the view that consciousness is explanatorily irrelevant-gives an in-depth defense of the view that we have a "first-person warrant" for our beliefs about our experiences, a warrant that is grounded at least partly in our having the

198 The Irreducibility of Consciousness

experiences. So it is not going out on a limb to see experience as providing a direct source of justification.

Answering the arguments

Given all this, the answer to the arguments against explanatory irrelevance is straightforward. In response to the argument from the causal theory of knowledge, we note that there is independent reason to believe that the causal theory is inappropriate to explicate our knowledge of experience: our knowledge of experience is grounded in a more immediate relation. And in response to the argument from my zombie twin, we note that the justification of my beliefs about experience involves more than the mechanisms by which the beliefs are formed: it crucially involves experiences themselves. Because my zombie twin lacks experiences, he is in a very different epistemic situation from me, and his judgments lack the corresponding justification.

It may be tempting to object that if my belief lies in the physical realm, its justification must lie in the physical realm; but this is a nonsequitur. From the fact that there is no justification in the physical realm, one might conclude that the physical portion of me (my brain, say) is not justified in its belief. But the question is whether I am justified in the belief, not whether my brain is justified in the belief, and if property dualism is correct then there is more to me than my brain. I am constituted by both physical and nonphysical properties, and the full story about me cannot be told by focusing on only one half. In particular, the justification of my belief accrues not just in virtue of my physical features but in virtue of some of my nonphysical features-namely the experiences themselves.

It might still be objected, "But the belief would still have been formed even if the experience had been absent!" To this, the answer is, "So what?" In this case, I have evidence for my belief, namely my immediate acquain-tance with experience. In a different case, that evidence is absent. To note that in a different case the belief might have been formed in the absence of the evidence is not to say that the evidence does not justify the belief in this case.7 I know I am conscious, and the knowledge is based solely on my immediate experience. To say that the experience makes no difference to my psychological functioning is not to say that the experience makes no difference to me.

Finally, there is a persistent refrain that comes up in these situations: "But your zombie twin would say the same thing!" If I say I know I am conscious, it is noted that my zombie twin says the same. If I say my belief is justified by my immediate acquaintance with experience, it is noted that my zombie twin says the same. To this, the answer is again, "So what?" At most this shows that from the third-person point of view, my zombie twin and I are

The Paradox of Phenomenal Judgment 199

identical, so that you cannot be certain that I am conscious; but we knew this all along. But it does nothing to imply that from the first-person view, I cannot know I am conscious. From the first-person point of view, my zombie twin and I are very different: I have experiences, and he does not. Because of this, I have evidence for my belief where he does not. Despite the fact that he says the same things I do, I know that I am not him (though you might not be sure) because of my direct first-person acquaintance with my experiences. This may sound somewhat paradoxical at first, but really it is simply saying the obvious: our experience of consciousness enables us to know that we are conscious.

Even when it is objected that my zombie twin would believe the same things that I would, this does nothing to make plausible the first-person skeptical hypothesis that I might be a zombie. Underlying this sort of objec-tion may be the implicit assumption that the beliefs themselves are the primary determinants of my epistemic situation; so if there is a situation in which I believe exactly the same things that I do now, it is a situation that is evidentially equivalent to my current one. But of course this is false. The evidence for my beliefs about experiences is much more primitive than the beliefs themselves. It is experience itself that is primary; the beliefs are largely a secondary phenomenon.

It should also be remembered that we are stipulating a deflationary (i.e., functional) notion of belief, so to say that my zombie twin believes the same things as me is still only to make a claim about our commonalities from the third-person point of view: he is disposed to make the same sorts of claims, the same sort of inferences, and so on. This says nothing about how things are from the inside. The feeling that "a zombie would have the same beliefs" provides an objection here may stem from assuming an inflationary notion of belief, in which belief is at least in part an experiential phenomenon. Only in that sense might it be the case that identity in beliefs would make the situations indistinguishable from the first-person point of view; but of course in that sense there is no reason to accept that a zombie has the same beliefs in the first place.

The upshot of all this is that arguments about self-knowledge provide no reason to reject the view I advocate. If one takes consciousness seriously, one will already have good reason to embrace an epistemology of conscious-ness that renders these arguments toothless. Although there are many tempt-ing arguments that can be made, none of them appear to stand up to scrutiny.

Very much remains to be done in clarifying the first-person epistemology of consciousness, of course. At best I have sketched the bare outline of a framework for thinking about these things; many issues remain to be dealt with. In particular, one would like an analysis of just how an experience justifies a belief; of what other factors are relevant in justifying beliefs about

200 The Irreducibility of Consciousness

experiences; of under just what circumstances a belief about experience is fully justified; and so on. All of these are important questions that deserve to be taken up at length in a study of the epistemology of consciousness. But all of these are part of the challenge, and on the face of it there is no reason to believe that the challenge cannot be met. What is important is that the arguments that self-knowledge might appear to provide against a nonreductive view of experience do not succeed.

6. The Argument from Memory*

The second objection to the causal or explanatory irrelevance of experience is that it is incompatible with the fact that we remember our experiences. It certainly seems that I often remember my old experiences, as when I recall the tangy odor of mothballs in a closet when I was a child, or when I recollect a particularly vivid experience of orange while I was watching the sun set last night. But to remember something, it is often held, is to stand in an appropriate causal relation to it; this is sometimes known as the causal theory of memory. If experiences are causally irrelevant to my psychological functioning, however, it seems that my old experiences are not causally related to any of my current states. If so, then we could not remember our experiences at all.

The causal theory of memory is not written in stone, however. It comes from an analysis of what seems the appropriate thing to say about various cases. As with the case of knowledge, it may be that a causal theory is appropriate in many domains without being appropriate across the board. In particular, it is not obvious that it is appropriate in the domain of experience. Causal theories might not be as inappropriate in the case of memory as they are in the case of knowledge, as there is no doubt that our relation to a remembered experience is mediated, and it is plausible that much of that mediation involves a causal chain. But this is not to say that the causal chain tells the whole story.

In the case of remembered experiences, there will certainly be a causal connection at the level of psychology: the underlying cognitive state at the time of the original experience will be causally connected to the cognitive state at the time of the memory. And it seems plausible that an appropriate causal connection of this sort is all that is required for memory of experience. For example, there may be a causal connection between a phenomenal belief at the earlier time and beliefs at the later time; and if what I have said in the previous section is correct, this original belief may count as knowledge, being justified by an acquaintance with the experience itself. This sort of causal connection between a belief justified by acquaintance and a later

The Paradox of Phenomenal Judgment 201

belief seems quite sufficient for the later belief to count as an instance of memory. So there seems to be good reason to believe that a causal connection to an experience is not required to remember that experience.

Of course the question of just what counts as a "memory" and what counts as merely a justified true belief about the past is largely a semantic decision in these cases. What is important is that a nonreductive theory can save the appearances by giving a mechanism by which true beliefs about one's past experiences are formed. As long as a nonreductive theory can do this, then any argument that memory provides against such a theory is defanged. If someone insists that a causal connection to an object is required for memory, then we can simply say that we "pseudo-remember" experiences instead, or some such, and nothing important will be lost. But in any case it seems to me that a causal connection to a relevant original psychological state is quite enough for these beliefs to qualify as memories.

7. The Argument from Reference*

The third argument against the causal or explanatory irrelevance of con-sciousness is that it is incompatible with our ability to refer to our conscious experiences. Certainly, it seems that we can think about our conscious experi-ences, and talk about them-I have been doing that throughout this book. But it is sometimes held that reference to an entity requires a causal connection to that entity; this is known as the causal theory of reference. If so, then it would be impossible to refer to causally irrelevant experiences.

There seems to be no principled reason why reference to an entity requires a causal connection to that entity, however. Reference frequently involves a causal connection, but it is not clear that things have to be that way. In referring to an entity, all that is required is that our concepts have intensions (in particular, primary intensions) that the entity might satisfy. For example, my concept "the largest star in the universe" has a primary intension, picking out a referent in any given centered world. In the actual world, this intension picks out a certain star S-whether or not I am causally connected to S-so S qualifies as the referent of the concept. Given that there is a primary intension that an entity in the actual world might satisfy, we have the basic ingredients needed for reference.

It happens that for many of our concepts, primary intensions are character-ized causally: at a given centered world, they pick out an appropriate entity that is causally connected to the center. This is the insight of the so-called causal theory of reference. But there is no reason why a primary intension has to work this way. There are many other functions that pick out, in a hypothetical centered world, an entity that has no causal connection to the

202 The Irreducibility of Consciousness

center. Such functions might make perfectly good primary intensions, with a perfectly good referent.

Further, the existence of a primary intension-even in cases where a pri-mary intension is characterized causally-does not depend in any way on a causal connection to the referent. The primary intension is independent of those actual-world goings on. A causal connection may often play a role in evaluating the primary intension at a world, but this is very different from playing a role in determining the primary intension itself. Indeed, some of our concepts (e.g., "Santa Claus") have no referent at all, but they still have a primary intension-an intension that could have picked out a referent if the world had turned out the right way.

It will often be the case that causal connection to a referent plays a role in acquiring a concept, and thus in forming a primary intension. One might argue that even in the case of "the largest star in the universe," causal connections to the world play a role in acquiring the basic concepts from which this compositional concept is formed. But again, there seems to be no principled reason why the existence of a primary intension requires a causal connection to relevant subject matter. Even a brain in a vat might have concepts with primary intensions, despite its causal isolation (though most of them would be intensions that nothing in its world satisfies). Again, the constitution of a primary intension is independent of such causal con-nections.

There is a natural reason why causation is central to so many of our concepts: it is because we generally refer to what we know about, and the things we know about are generally things we are causally connected to. But we have already seen that there is good reason to reject the causal model of knowledge at least in the case of consciousness: in that case we have knowledge of a more immediate variety. So to refer to consciousness, we do not need to refer via an intension that picks out something that the center is causally connected to; instead, we can refer via an intension that picks out something that the center is immediately acquainted with.

In any case, what is important is that (1) my concept of "consciousness" can have a primary intension, whether or not there is a causal connection to the referent (for the existence of a primary intension never depends on such a causal connection); and (2) the primary intension can pick out a referent whether or not there is a causal connection to the referent (for there is no reason why a primary intension must pick out its referent in virtue of a causal connection). The intension specifies a perfectly good func-tion from centered worlds to features of those worlds; in this world, there is something that satisfies the intension, so my concept has a referent. As we have seen, consciousness is something of a primitive concept (like space and time, perhaps), so there is no hope of characterizing the intension in