What computers still can't do: a critique of artificial reason

776 Pages·1992·1.91 MB·English

by  Hubert L. Dreyfus

What Computers Still Can't Do : A Critique
title:
of Artificial Reason
author: Dreyfus, Hubert L.
publisher: MIT Press
isbn10 | asin: 0262540673
print isbn13: 9780262540674
ebook isbn13: 9780585330136
language: English
subject Artificial intelligence.
publication date: 1992
lcc: Q335.D74 1992eb
ddc: 006.3
subject: Artificial intelligence.
Page iii
What Computers Still Can't Do
A Critique of Artificial Reason
Hubert L. Dreyfus
Page iv
To my parents
Sixth printing, 1999
©1972, 1979, 1992 Hubert L. Dreyfus
All rights reserved. No part of this book may be reproduced in any
form by any electronic or mechanical means (including photocopying,
recording, or information storage and retrieval) without permission in
writing from the publisher.
Printed and bound in the United States of America.
Library of Congress Cataloging-in-Publication Data
Dreyfus, Hubert L.
What computers still can't do : a critique of artificial reason /
Hubert L. Dreyfus.
p. cm.
Rev. ed. of: What computers can't do, 1979.
Includes bibliographical references and index.
ISBN 0-262-04134-0. ISBN 0-262-54067-3 (pbk.)
1. Artificial intelligence. I. Title.
Q335.D74 1992
006.3dc20 92-27715
CIP
Page v
The difference between the mathematical mind (esprit de géométrie) and the
perceptive mind (esprit de finesse): the reason that mathematicians are not
perceptive is that they do not see what is before them, and that, accustomed to
the exact and plain principles of mathematics, and not reasoning till they have
well inspected and arranged their principles, they are lost in matters of
perception where the principles do not allow for such arrangement. . . . These
principles are so fine and so numerous that a very delicate and very clear
sense is needed to perceive them, and to judge rightly and justly when they are
perceived, without for the most part being able to demonstrate them in order
as in mathematics; because the principles are not known to us in the same
way, and because it would be an endless matter to undertake it. We must see
the matter at once, at one glance, and not by a process of reasoning, at least
to a certain degree. . . . Mathematicians wish to treat matters of perception
mathematically, and make themselves ridiculous . . . the mind . . . does it
tacitly, naturally, and without technical rules.
, Pensées
PASCAL
Page vii
CONTENTS
Introduction to the MIT Press Edition ix
Acknowledgments liii
Introduction to the Revised Edition (1979) 1
Introduction 67

Part I. Ten Years of Research in Artificial Intelligence
(1957 1967)
1. Phase I (1957 1962) Cognitive Simulation 91

I. Analysis of Work in Language Translation, Problem
Solving, and Pattern Recognition

II. The Underlying Significance of Failure to Achieve
Predicted Results
2. Phase II (1962 1967) Semantic Information Processing 130

I. Analysis of Semantic Information Processing
Programs

II. Significance of Current Difficulties
Conclusion 149

Part II. Assumptions Underlying Persistent Optimism
Introduction 155
3. The Biological Assumption 159
4. The Psychological Assumption 163

Page viii

I. Empirical Evidence for the Psychological
Assumption: Critique of the Scientific Methodology of
Cognitive Simulation

II. A Priori Arguments for the Psychological
Assumption
5. The Epistemological Assumption 189

I. A Mistaken Argument from the Success of Physics

II. A Mistaken Argument from the Success of Modern
Linguistics
6. The Ontological Assumption 206
Conclusion 225

Part III. Alternatives to the Traditional Assumptions
Introduction 231
7. The Role of the Body in Intelligent Behavior 235
8. The Situation: Orderly Behavior Without Recourse to 256
Rules
9. The Situation as a Function of Human Needs 272
Conclusion 281

Conclusion: The Scope and Limits of Artificial Reason
The Limits of Artificial Intelligence 285

The Future of Artificial Intelligence
Notes 307
Index 346

Page ix
Introduction to the MIT Press Edition
This edition of What Computers Can't Do marks not only a change of
publisher and a slight change of title; it also marks a change of status.
The book now offers not a controversial position in an ongoing debate
but a view of a bygone period of history. For now that the twentieth
century is drawing to a close, it is becoming clear that one of the great
dreams of the century is ending too. Almost half a century ago
computer pioneer Alan Turing suggested that a high-speed digital
computer, programmed with rules and facts, might exhibit intelligent
behavior. Thus was born the field later called artificial intelligence
(AI). After fifty years of effort, however, it is now clear to all but a
few diehards that this attempt to produce general intelligence has
failed. This failure does not mean that this sort of AI is impossible; no
one has been able to come up with such a negative proof. Rather, it
has turned out that, for the time being at least, the research program
based on the assumption that human beings produce intelligence using
facts and rules has reached a dead end, and there is no reason to think
it could ever succeed. Indeed, what John Haugeland has called Good
Old-Fashioned AI (GOFAI) is a paradigm case of what philosophers
of science call a degenerating research program.
A degenerating research program, as defined by Imre Lakatos, is a
scientific enterprise that starts out with great promise, offering a new
approach that leads to impressive results in a limited domain. Almost
inevitably researchers will want to try to apply the approach more
Page x
broadly, starting with problems that are in some way similar to the
original one. As long as it succeeds, the research program expands
and attracts followers. If, however, researchers start encountering
unexpected but important phenomena that consistently resist the new
techniques, the program will stagnate, and researchers will abandon it
as soon as a progressive alternative approach becomes available.
We can see this very pattern in the history of GOFAI. The program
began auspiciously with Allen Newell and Herbert Simon's work at
RAND. In the late 1950s Newell and Simon proved that computers
could do more than calculate. They demonstrated that a computer's
strings of bits could be made to stand for anything, including features
of the real world, and that its programs could be used as rules for
relating these features. The structure of an expression in the computer,
then, could represent a state of affairs in the world whose features had
the same structure, and the computer could serve as a physical symbol
system storing and manipulating such representations. In this way,
Newell and Simon claimed, computers could be used to simulate
important aspects of intelligence. Thus the information-processing
model of the mind was born.
Newell and Simon's early work was impressive, and by the late 1960s,
thanks to a series of micro-world successes such as Terry Winograd's
SHRDLU, a program that could respond to English-like commands by
moving simulated, idealized blocks (see pp. 12 13), AI had become a
flourishing research program. The field had its Ph.D. programs,
professional societies, international meetings, and even its gurus. It
looked like all one had to do was extend, combine, and render more
realistic the micro-worlds and one would soon have genuine artificial
intelligence. Marvin Minsky, head of the M.I.T. AI project,
announced: "Within a generation the problem of creating 'artificial
intelligence' will be substantially solved."