Download How the Body Shapes the Way We Think

How the Body Shapes the Way We Think
A New View of Intelligence
Rolf Pfeifer and Josh Bongard
with a contribution by Simon Grand
Foreword by Rodney Brooks
Illustrations by Shun Iwasawa
A Bradford Book
The MIT Press
Cambridge, Massachusetts
London, England
© 2007 Massachusetts Institute of Technology
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.
MIT Press books may be purchased at special quantity discounts for business or sales
promotional use. For information, please email [email protected] or write to
Special Sales Department, The MIT Press, 55 Hayward Street, Cambridge, MA 02142.
This book was set in Syntax and Times Roman by SNP Best-set Typesetter Ltd., Hong
Kong. Printed and bound in the United States of America.
Library of Congress Cataloging-in-Publication Data
Pfeifer, Rolf, 1947–
How the body shapes the way we think : a new view of intelligence / by Rolf Pfeifer
and Josh Bongard ; with a contribution by Simon Grand ; foreword by Rodney Brooks ;
illustrations by Shun Iwasawa.
p. cm.
Includes bibliographical references (p. ).
ISBN-13: 978-0-262-16239-5 (alk. paper)
ISBN-10: 0-262-16239-3 (alk. paper)
1. Artificial intelligence. 2. Cognitive science. I. Bongard, Josh. II. Grand, Simon.
III. Title.
Q335.P445 2006
006.3—dc22
2006044919
10 9 8 7 6 5 4 3 2 1
To my friends in Japan (R. P.)
To Toby, Carol, and Ralph (J. B.)
Contents
Foreword by Rodney Brooks
Preface xvii
I
1
Intelligence, Artificial Intelligence, Embodiment, and What the Book Is
About 1
Intelligence, Thinking, and Artificial Intelligence
1.1
1.2
1.3
1.4
1.5
1.6
2
xiii
Artificial Intelligence: The Landscape
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
5
Thinking, Cognition, and Intelligence 7
The Mystery of Intelligence 11
Defining Intelligence 14
Artificial Intelligence 17
Embodiment and Its Implications 18
Summary 22
25
Successes of the Classical Approach 27
Problems of the Classical Approach 30
The Embodied Turn 34
The Role of Neuroscience 37
Diversification 39
Biorobotics 41
Developmental Robotics 44
Ubiquitous Computing and Interfacing Technology
Artificial Life and Multiagent Systems 49
Evolutionary Robotics 53
Summary 54
47
viii
II
3
Contents
Toward a Theory of Intelligence
Prerequisites for a Theory of Intelligence
3.1
3.2
3.3
3.4
3.5
3.6
3.7
4
57
61
Level of Generality and Form of Theory
Diversity-Compliance 67
Frame of Reference 72
The Synthetic Methodology 77
Time Perspectives 82
Emergence 85
Summary 88
Intelligent Systems: Properties and Principles
62
89
4.1
4.2
4.3
Real Worlds and Virtual Worlds 90
Properties of Complete Agents 95
Agent Design Principle 1: The Three-Constituents
Principle 100
4.4 Agent Design Principle 2: The Complete-Agent
Principle 104
4.5 Agent Design Principle 3: Cheap Design 107
4.6 Agent Design Principle 4: Redundancy 113
4.7 Agent Design Principle 5: Sensory-Motor
Coordination 117
4.8 Agent Design Principle 6: Ecological Balance 123
4.9 Agent Design Principle 7: Parallel, Loosely Coupled
Processes 134
4.10 Agent Design Principle 8: Value 137
4.11 Summary and Conclusions 140
5
Development: From Locomotion to Cognition
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
5.9
5.10
141
Motivation 143
Toward Developmental Robot Design 145
From Locomotion to Cognition: A Case Study 149
From Gait Patterns to Body Image to Cognition 153
The Symbol Grounding Problem 159
Matching Brain and Body Dynamics 161
Broadening the Scope: Other Aspects of Development 164
Learning in Embodied Systems 168
Social Interaction 170
Development: Where Are We and Where Do We Go from
Here? 173
Contents
ix
5.11 Summary: Design Principles for Developmental
Systems 175
6
Evolution: Cognition from Scratch
177
6.1
6.2
6.3
6.4
Motivation 181
The Basics of Evolutionary Computation 184
The Origins of Evolutionary Computation 187
Artificial Evolution in the Real World: On Pipes, Antennas,
and Electronic Circuits 189
6.5 Evolutionary Robotics 192
6.6 Evolving Morphology and Control 194
6.7 Genetic Regulatory Networks and Developmental
Plasticity 196
6.8 Self-Organization: The Powerful Ally of Mutation and
Selection 204
6.9 Artificial Evolution: Where Are We and Where Do We Go
from Here? 206
6.10 Summary: Design Principles for Evolutionary Systems 208
7
Collective Intelligence: Cognition from Interaction
213
7.1
7.2
7.3
7.4
7.5
7.6
7.7
Motivation 215
Agent-Based Modeling 217
Simulation versus Real Robots 221
Groups of Robots 222
A Note on Cooperation 226
Modular Robots 228
Scalability, Self-Assembly, Self-Repair, Homogeneity, and
Heterogeneity 232
7.8 Self-Reproducing Machines 235
7.9 Collective Intelligence: Where Are We and Where Do We
Go from Here? 238
7.10 Summary: Design Principles for Collective Systems 241
III
8
Applications and Case Studies
245
Ubiquitous Computing and Interfacing Technology
8.1
8.2
8.3
249
Ubiquitous Technology as Scaffolding 251
Ubiquitous Technology: Properties and Principles
Interacting with Ubiquitous Technology 263
253
x
Contents
8.4
8.5
9
Building Intelligent Companies
9.1
9.2
9.3
9.4
9.5
9.6
10
Cyborgs 264
Summary and Conclusions
270
271
Management and Entrepreneurship: Decision and Action
under Uncertainty 272
Companies as Embodied Systems 274
A Synthetic Approach to Management 279
Design Principles for Building Intelligent Companies 282
Corroborating the Speculations 293
Summary and Conclusions 294
Where Is Human Memory?
295
10.1
10.2
10.3
10.4
Introduction 298
The Storehouse Metaphor and Its Problems 300
Concepts of Memory 302
The Frame-of-Reference Problem in Memory Research:
Ashby’s Proposal 304
10.5 The Embodied View of Memory: Applying the Design
Principles for Intelligent Systems 307
10.6 Implications for Memory Research: Summary and
Speculations 318
11
Robotic Technology in Everyday Life
323
11.1 Introduction: Everyday Robots 324
11.2 Vacuum Cleaners: Roomba, Trilobite, and Similar
Species 327
11.3 Entertainment Robots 328
11.4 Therapeutic, Medical, and Rescue Robots 333
11.5 Humanoid Companion Robots 335
11.6 Robots Capable of Social Communication 341
11.7 Robots Capable of Facial and Bodily Expression 344
11.8 A Theoretical Note 346
11.9 Summary and Conclusions 348
IV
12
Principles and Insights
351
How the Body Shapes the Way We Think
353
12.1 Steps Toward a Theory of Intelligence
12.2 Selected Highlights 358
354
Contents
12.3 Seeing Things Differently
12.4 Epilogue 370
Notes 373
References 375
Index 389
xi
367
Foreword
The great revolutions in science come about when what was formerly
thought to be true and unassailable is both assailed and shown to not be
true after all. Sometimes the assaults are brutal and front on, and sometimes they are gentle over a long period of time, gradually creeping up
on the soon to be discredited truth.
This book is a gentle assault on some of the collateral tenets of modern
rationalism; not an assault on rationalism itself, but an assault on many
of the things that are commonly assumed by rationalists. Rolf Pfeifer and
Josh Bongard question whether our nervous systems compute, whether
they are separate control systems for our bodies, and even whether there
can truly be disembodied reasoning. These three ideas are so ingrained
in our computational metaphors that they usually go unquestioned—
they make no sense within our normal frameworks of thinking in the
fields of computer science and artificial intelligence, and even neuroscience. Beyond the mere technical these questions challenge the intellectual father of rationalism Rene Descartes and his “Je pense, donc je
suis” (I think, therefore I am) from his Discourse on Method (written in
French, not Latin, in 1637).
While such questions can be seen as a challenge to the very underpinnings of the scientific world view, they really are not. Pfeifer and
Bongard are not suggesting throwing out the scientific method and
replacing it, as some might fear, with postmodern relativism. Rather they
are assaulting certain metaphors that have perhaps gone haywire in their
influence on how we approach the study of intelligence, the study of us.
In modern times there have been two important and perhaps underestimated influences on our view of intelligence.
1. As Alan Turing described in his 1950 paper “Computing Machinery
and Intelligence,” his earlier and today still dominant model of
xiv
Foreword
computation came from considering the externally observable behavior
of a human computer, a person who carried out computations with pen
and paper, and “is supposed to be following fixed rules.” It is worth
noting here that Turing modeled what a person does, not what a person
thinks.
2. Ever since the human brain has come to be considered as the seat of
our thought, desires, and dreams, it has been compared to the most
advanced technology possessed by mankind. In my own lifetime I have
seen popular “complexity” metaphors for the brain evolve. When I was
a young child the brain was likened to an electromagnetic telephone
switching network. Then it became an electronic digital computer. Then
a massively parallel digital computer. And delightfully, in April 2002,
someone in a lecture audience asked me whether the brain could be “just
like the world wide web.” Even otherwise serious scientists have become
enamored of their own complexity metaphors declaring for instance that
quantum phenomena and the brain are both so complex that they must
be about the same thing.
Turing’s metaphor has become the very definition of computation, and
he points out in his 1950 paper, using Babbage’s unrealized mechanical
engine as the exemplar, that such computation is independent of the
medium in which it is expressed. The metaphors for the brain (except for
the quantum speculations) have entrenched it as the equivalent of
Turing’s form of computation, and thus rationalism largely assumes that
the human brain is a Turing machine, carrying out Turing computation,
and controlling its periphery, the human body.
But when we consider the evolutionary history of nervous systems we
are faced with a dilemma not unlike one that is so often used to challenge evolution itself. How could evolution have incrementally produced
the components of an eye—the lens, the pupil, the retina—when all are
necessary, fully formed, to enable the other to carry out its function
within the ensemble? When we turn that skepticism on its head we are
left to ask what roles earlier versions of nervous systems played, before
they became fully functional control systems, like Turing’s “control” component which he talked about along with the “executive” and the “store.”
Metaphors are useful in science as a way of understanding systems we
wouldn’t otherwise understand—metaphors can suggest appropriate
questions to ask about a system, they can provide intuitive models about
how things might work, and they can bridge gaps as a more explicit
theory is being formed. But they can also lead to ways of thinking about