Artificial Intelligence I]
Artificial Intelligence A Philosophical Introduction Jack Copeland
Copyright B. 1. Copeland, 1993 The right of B. J. Copeland to be identified as author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988. First published 1993 Reprinted 1994, 1995, 1997 (twice), 1998 Blackwell Publishers Ltd 108 Cowley Road Oxford OX4 IJF, UK Blackwell Publishers Inc 350 Main Street Malden, Massachusetts 02148, USA All rights reserved. Except for the quotation of short passages for the purposes of criticism and review, 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 the publisher. Except in the United States of America, this book is sold subject to the condition that it shall not, by way of trade or otherwise, be lent, re-sold, hired out, or otherwise circulated without the publisher's prior consent in any form of binding or cover other than that in which it is published and without a similar condition including this condition being imposed on the subsequent purchaser. British Library Cataloguing in Publication Data A CIP catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data Copeland, Brian Jack Artificial intelligence: a philosophical introduction / B. 1. Copeland p. cm. Includes bibliographical references and index. ISBN 0-631-1 8385-X (pbk: alk. paper) I. Artificial intelligence - Philosophy. I. Title. Q335.C583 1993 92-44278 006.3-dc20 CIP Typeset in 10.5 on 12pt Ehrhardt by Graphicraft Typesetters Limited, Hong Kong Printed and bound in Great Britain by Athenreum Press Ltd, Gateshead, Tyne & Wear This book is printed on acid-free paper
For Jean and Reg
Contents List of figures x Acknowledgements Xl Introduction 1 In outline 2 1 The beginnings of Artificial Intelligence: a historical sketch 4-1.1 The arrival of the computer 4-1.2 The Logic Theorist 7 1.3 The Dartmouth Conference 8 1.4 Alan Turing and the philosophy of AI 9 2 Some dazzling exhibits 11 2.1 Inside the machine 11 2.2 Parry the paranoid program 12 2.3 Eliza the psychotherapist 13 2.4 Shrdlu the robot 15 2.5 Hacker the program-writing program 18 2.6 Programs that play games 21 2.7 The General Problem Solver 24-2.8 Sam and the Frump 26 2.9 Expert systems 30 3 Can a machine think? 33 3.1 Is consciousness necessary for thought? 33 3.2 The Turing Test 37 3.3 Has the Test been passed already? 39 3.4 Four objections to the Turing Test 44-3.5 Assessing the Test 50 3.6 Decision time S2
viii Contents 4 The symbol system hypothesis 58 4.1 Symbol manipulation 59 4.2 Binary symbols 60 4.3 Programs as symbols 66 4.4 A full-scale program 71 4.5 The definition of a computer 78 4.6 The hypothesis 79 4.7 Multiple realizability 81 5 A hard look at the facts 83 5.1 The evidence for the hypothesis 83 5.2 Getting the evidence into perspective 85 5.3 Hype 95 5.4 Programming common sense 98 5.5 Data versus know-how 100 5.6 The CYC project 102 5.7 The complexity barrier 120 6 The curious case of the Chinese room 121 6.1 The Chinese room argument 123 6.2 What's wrong with the argument? 125 6.3 Deciding about understanding 130 6.4 Turing machines and the biological objection to AI 132 7 Freedom 140 7.1 Turbo Sam makes a choice 141 7.2 Is freedom of the will an illusion? 141 7.3 Two kinds of freedom 145 7.4 Kleptomania and other compulsions 149 7.5 Libertarianism 151 7.6 Predictivism and chaos 154 7.7 The inevitable 157 8 Consciousness 163 8.1 Neglect and disa"ay 163 8.2 The fuzzy baseline 164 8.3 Consciousness as a type of internal monitoring 166 8.4 The ineffable 'feel' of it all 170 8.5 Into the heart of the mystery 172 8.6 What is it like to be a bat? 174 8.7 What Mary doesn't know 176 8.8 Drawing the threads together 179
Contents ix 9 Are we computers? 180 9.1 The strong symbol system hypothesis 180 9.2 Hardware versus wetware 182 9.3 Goodbye, von Neumann 192 9.4 Putting meaning into meat 195 9.5 Believing what you don't believe 198 9.6 Productivity and systematicity 199 9.7 Evaluating the arguments 200 9.8 The meaning of 'computer' 204 10 AI's fresh start: parallel distributed processing 207 10.1 The basic ideas 208 10.2 English lessons 214 10.3 Escape from a nightmare? 216 10.4 The contrast with computers 219 10.5 Comparisons with wetware 221 10.6 Searle's Chinese gym 225 10.7 The Church-Turing thesis 230 10.8 Are our cognitive processes algorithmically calculable? 233 10.9 Simulating networks by computer 241 10.10 The battle for the brain 245 10.11 Concluding remarks 247 Epilogue 249 Notes 250 Bibliography 283 Index 299
List of figures Figure 2.1 Shrdlu's world of coloured blocks. Figure 2.2 Simple block-moving tasks. Figure 2.3 The tower of contrition. Figure 3.1 Can you read this? Figure 3.2 The Turing Test. Figure 4.1 A register. Figure 4.2 Recursion. Figure 4.3 Think of each binary digit as pointing to a box. Figure 4.4 Each box contains a base 10 number. Figure 4.5 How the compiled program is stored in memory. Figure 5.1 Shadows are a problem for visual recognition software. Figure 6.1 } T ' h" Figure 6.2 unng mac me mput. Figure 6.3 } T' h' F 6 4 unng mac me output. 19ure. Figure 8.1 Each hemisphere 'sees' only one half of the screen. Figure 8.2 The two hemispheres pick different objects when the split-brain subject is asked which pictured items he associates with what he can see on the screen. Figure 9.1 Human neuron. Figure 10.1 The neuron fires a discharge along its output fibres when the weighted sum of its inputs exceeds its threshold. Figure 10.2 Connections between the layers of artificial neurons in a PDP network. Figure 10.3 Inverting an input pattern. Figure 10.4 A few of the variations on the letter 'A' to be found in the Letraset catalogue. Figure 10.5 The curse of context. Figure 10.6 The level of output depends on the amount of input the unit is receiving from other units. 16 20 25 34 38 60 64 65 66 70 85 138 139 267 267 168 169 184 209 211 213 217 218 220
Acknowledgements Many people have helped me with this book in many different ways, large and small. I thank you all. David Anderson, John Andreae, Derek Browne, Peter Carruthers, Stephan Chambers, John Cottingham, Hubert Dreyfus, Flip Ketch, Carmel Kokshoorn, Justin Leiber, David Lewis, Michael Lipton, Bill Lycan, Michael Maclean, Pamela McCorduck,James McGahey, Jack Messenger, Alison Mudditt, Gill Rhodes, Michael Smallman, Steve Smith, Kerry Stewart, Bob Stoothoff, Stephen Viles. Particular thanks to Jenny Arkell for commenting extensively on early drafts and suggesting many improvements; to Ann Witbrock for the computer-generated illustrations; and above all to Diane Proudfoot for support, criticism, help and encouragement. Figure 2.1 is based on figure 3 of Terry Winograd, Understanding Natural Language, with kind permission of Academic Press and Terry Winograd. Figure 3.1 is redrawn from figure 15 of Douglas Hofstadter, Codel, Escher, Bach: An Eternal Colden Braid, with kind permission of Basic Books Inc. (copyright 1979 by Basic Books, Inc.). Figure 5.1 is redrawn from figures 2.1 and 2.2 of P.H. Winston (ed.), The Psychology of Computer Vision, with kind permission of McGraw-Hill, Inc. Figure 8.2 is redrawn from figure 42 of M.S. Gazzaniga and ].E. LeDoux, The Integrated Mind, with kind permission of Plenum and Michael Gazzaniga. Figure 10.4 is a reproduction of parts of the Letraset catalogue and appears by the kind permission of Esselte Letraset Ltd.
Introduction Artificial Intelligence is the science of making machines do things that would require intelligence if done by men. Marvin Minsky, founder of the MIT Artificial Intelligence Laboratoryl Not long ago I watched a TV interview with Edward Fredkin, a specialist in electronic engineering and manager of the MIT Artificial Intelligence Laboratory. Fredkin is an earnest man with a serious, authoritative manner. What he had to say was startling. 2 There are three great events in history. One, the creation of the universe. Two, the appearance of life. The third one, which I think is equal in importance, is the appearance of artificial intelligence. This is a form of life that is very different, and that has possibilities for intellectual achievement that are hard for us to imagine. These machines will evolve: some intelligent computers will design others, and they'll get to be smarter and smarter. The question is, where will that leave us? It is fairly difficult to imagine how you can have a machine that's millions of times smarter than the smartest person and yet is really our slave, doing what we want. They may condescend to talk to us, they may play games that we like to play, and in some sense, they might keep us as pets. 3 Has Professor Fredkin missed his vocation as a science fiction writer? Or is this a realistic view of the future, a sober prediction from a man who is better placed than most to understand the implications of current developments in AI research? Are computers that think really a technological possibility? Indeed, does it even make sense to speak of a machine thinking - or is a thinking computer a conceptual absurdity, like a grinning electron or a married bachelor? For the philosophically curious, Fredkin's words raise a tumult of intriguing questions. Is thought a biological phenomenon, and thus as far beyond the compass of a silicon-and-metal machine as photosynthesis, lactation, and every other biology-dependent process? Or are thinking and perceiving more like flying - something that both living creatures and metallic artefacts can do?