Exploring Intelligence - The Eye · 2019-05-30 · 2 Scientific American Presents Intelligence EXPLORING A Search in the Human, Animal, Machine and Extraterrestrial Domains 6 12 Twin

PR

ESE

NT

SIntelligenceEXPLORING

QUARTERLY $5.95

SC

IEN

TIF

IC A

ME

RIC

AN

PR

ESE

NT

SE

XPL

OR

ING

INT

EL

LIG

EN

CE

Quarterly

Volu

me 9

, Num

ber 4

A Search in the

Human, Animal, Machine

andExtraterrestrial

Domains

IQ: What Your ScoreReally Means

Multiple Intelligences

Giftedness and Savants

Smart Drugs in Your Kitchen

Animal Thinking and Empathy

Game-PlayingComputers

SETI Today

Copyright 1998 Scientific American, Inc.

2 Scientific American Presents

IntelligenceEXPLORING

A Search in the Human, Animal,Machine and Extraterrestrial Domains

6 12

Twin savantspage 32

18

24

30

32

P R E S E N T SWinter 1998 Volume 9 Number 4

Scientific American Presents (ISSN 1048-0943),Volume 9,Number 4,Winter1998,published quarterly by Scientific American,Inc.,415 Madison Avenue,New York, NY 10017-1111. Copyright © 1998 by Scientific American, Inc. Allrights reserved.No part of this issue may be reproduced by any mechanical,photographic or electronic process,or in the form of a phonographic record-ing, nor may it be stored in a retrieval system, transmitted or otherwisecopied for public or private use without written permission of the publisher.Periodicals Publication Rate.Postage paid at New York,N.Y.,and at additionalmailing offices.Canadian BN No.127387652RT; QST No.Q1015332537. Sub-scription rates: one year $19.80 (outside U.S.$23.80).To purchase additionalquantities: 1 to 9 copies: U.S.$5.95 each plus $2.00 per copy for postage andhandling (outside U.S.$5.00 P & H); 10 to 49 copies: $5.35 each,postpaid; 50copies or more: $4.75 each, postpaid. Send payment to Scientific American,Dept.SAQ,415 Madison Avenue,New York,NY 10017-1111.Postmaster:Sendaddress changes to Scientific American Presents,Box 5063,Harlan,IA 51593.Subscription inquiries:U.S.and Canada (800) 333-1199;other (515) 247-7631.

INTRODUCTION

Intelligence Consideredby Philip Yam, issue editor

Most people can identify intelligent signals, be they from a person, animalor machine. But can brainpower be measured, quantified and changed? Ishuman reasoning similar to how an animal might obtain a hidden treat orhow a machine decides to trade a rookfor a bishop? A definition is trickierthan it might appear.

HUMAN INTELLIGENCE

How Intelligent Is Intelligence Testing?by Robert J. Sternberg

SATs and IQ tests don’t tell everything about a person’s chances of suc-cess in college or at a job, the author claims. Creativity and practicalintelligence (“street smarts”) are critical components as well, and testscan be devised that accurately assess these abilities. Unfortunately, theyare overlooked in the big business of standardized testing.

A Multiplicity of Intelligencesby Howard Gardner

According to the theory of multiple intelligences, there are eight, possiblynine, different kinds of intelligence, including musical, athletic andpersonal. The originator of the theory discusses these ideas and arguesthat they are just as important as the intelligence measured by paper-and-pencil tests.

The General Intelligence Factorby Linda S. Gottfredson

Also known as g, the general intelligence factor is what IQ tests are allabout. Despite the political controversy surrounding it, the test scoresand their differences, the author argues, are meaningful indicators notonly of academic performance but also of future life outcomes, such asemployment, divorce and poverty.

For Whom Did the Bell Curve Toll?by Tim Beardsley, staff writer

The most controversial book on intelligence in the past decade createdmuch political and media upheaval. But its conclusions as they relateto social policy are poorly grounded, and little has actually come inthe way of policy changes.

Uncommon Talents: Gifted Children, Prodigies and Savantsby Ellen Winner

Often assumed to be well adjusted and easy to teach, gifted childrenand prodigies are generally out of step with their peers and can devel-op feelings of isolation that prevent them from achieving as adults.Most extreme are savants, who have a phenomenal capacity for calcu-lation or memory despite being autistic.


Exploring Intelligence 3

38

44

52

66

78

84

90

96

MACHINE INTELLIGENCE

On Computational Wings: Rethinking the Goals of Artificial Intelligenceby Kenneth M. Ford and Patrick J. Hayes

The “gold standard” of traditional artificial intelligence—passing the so-called Turing test and thereby appearing tobe human—has led expectations about AI astray. Drawingan analogy to flying—modern aircraft do it quite wellwithout mimicking birds—the authors argue that AI hasmade substantial achievements and, in fact, pervadeseveryday life.

Computers, Games and the Real Worldby Matthew L. Ginsberg

Deep Blue may have deep-sixed the world chess cham-pion last year, and machines are tops in checkers andOthello, but games such as bridge, Go and poker stillelude competent computer play. The issue, though, isn’tsimply pitting humans against machines. Games enableprogrammers to explore the algorithms and to decidewhich are best for particular problems.

Wearable Intelligenceby Alex P. Pentland

Soon you may no longer fumble through your memory for dates, figures or the location of your favorite restau-rant. Researchers are miniaturizing computer machineryso that the devices can be worn unobtrusively as cloth-ing, eyeglasses and shoes. They can provide travel direc-tions, Internet access, electric power and foreign-lan-guage translation.

THE SEARCH FOREXTRATERRESTRIAL INTELLIGENCE

Is There Intelligent Life Out There?by Guillermo A. Lemarchand

The odds say we aren’t alone, but radio telescopes have yet to pick up a definite intelligent signal beyondEarth. Improving the chance of first contact may dependon searches around supernovae and even sending outour own greetings to likely candidate star systems.

Table of Major SETI Projects

K 6 2

A

K

7

8 4 2

Q 6 3

6

North

A 8 5

K

A

K

7 3

J 8 4

8 5

South

J 9 7

Q

9

A

10 5

5

J 9 4 2

WestQ 10 4 3

J

10

Q

9 6

7 2

10 3

East

Seeking “Smart” Drugsby Marguerite Holloway, staff writer

Research on stemming the ravages of Alzheimer’s diseaseand other dementia conditions is paving the way for drugsthat might enhance the memory capacity of healthy indi-viduals. Pharmaceutical firms are racing to develop thesecognitive enhancers, but the most effective smart drugsmay already be in your kitchen.

The Emergence of Intelligenceby William H. Calvin

From evolution’s perspective, why did intelligence arise?The ability to anticipate and plan may have come aboutas a result of the need to organize ballistic movements,such as throwing, and language may have enabledhumans to develop an ability to conceptualize.

ANIMAL INTELLIGENCE

Reasoning in Animalsby James L. Gould and Carol Grant Gould

Mounting evidence indicates that many species can inferconcepts, formulate plans and employ simple logic tosolve problems. Much of what they learn, however, is dictated by instinct and limited by an inability to learnfrom observation.

Talking with Alex: Logic and Speech in Parrotsby Irene M. Pepperberg

Mimicry is the mainstay of a parrot’sspeech, but Alex the Grey parrot seems to understand what he says. He can count,identify the odd-man-out from a groupand determine what’s the same and what’sdifferent. The author, who has workedwith Alex for more than 20 years, describesthe teaching approach that permits theexploration of Alex’s cognitive abilities.

Animal Self-Awareness: A Debate

Can Animals Empathize?Yes. Animals that learn to recognize themselves in mirrors—chimpanzees, orangutans and humans—are self-aware andtherefore can infer the states of mind and emotions ofother individuals.by Gordon Gallup, Jr.

Maybe not. Chimps will beg for food from a blind-folded person as often as from a sighted one. Such tests suggest they cannot conceive of others’—and perhaps even their own—mental states.

by Daniel J. Povinelli

60

COVER ILLUSTRATION BY WATERS DESIGN ASSOCIATES, INC.

100

WIL

LIA

M M

UÑ

OZ


Exploring Intelligence is published bythe staff of Scientific American, withproject management by:

John Rennie, editor in chiefPhilip M. Yam, issue editor

Mark Alpert, assistant editorMichelle Press, managing editor

Timothy M. Beardsley, Marguerite Holloway,staff writers

ArtJana Brenning, art director

Bridget Gerety, photography editor

CopyMaria-Christina Keller, copy chief

Molly K. Frances; Daniel C. Schlenoff; Katherine A. Wong; Stephanie J. Arthur;

Eugene Raikhel; Myles McDonnell

ContributorsMeg Crane, design director

Meghan Gerety, Anna Armentrout, production editors

Michele Moise, William Stahl, researchers

AdministrationRob Gaines, editorial administrator

ProductionRichard Sasso, associate publisher/

vice president, productionWilliam Sherman, director, productionJanet Cermak, manufacturing manager

Silvia Di Placido, prepress and quality managerGeorgina Franco, print production manager

Norma Jones, assistant project managerMadelyn Keyes, custom publishing manager

Carl Cherebin, ad traffic

CirculationLorraine Leib Terlecki, associate publisher/

circulation directorKatherine Robold, circulation manager

Joanne Guralnick, circulation promotion manager

Rosa Davis, fulfillment manager

Business AdministrationMarie M. Beaumonte, general manager

Alyson M. Lane, business manager Constance Holmes, manager, advertising

accounting and coordination

Electronic PublishingMartin O. K. Paul, director

Ancillary ProductsDiane McGarvey, director

Chairman and Chief Executive OfficerJohn J. Hanley

Co-ChairmanRolf Grisebach

PresidentJoachim P. Rosler

Vice PresidentFrances Newburg

Scientific American, Inc. 415 Madison Avenue • New York, NY 10017-1111

(212) 754-0550

PRINTED IN U.S.A.


PRESENTS

®

OTHER EDITIONS OF SCIENTIFIC AMERICAN

Le ScienzePiazza della Repubblica, 8

20121 Milano, ITALYtel: +39-2-29001753

[email protected]

Spektrum der Wissenschaft Verlagsgesellschaft mbH

Vangerowstrasse 2069115 Heidelberg, GERMANY

tel: [email protected]

Investigacion y CienciaPrensa Científica, S.A.

Muntaner, 339 pral. 1.a

08021 Barcelona, SPAINtel: [email protected]

Pour la Science Éditions Belin8, rue Férou

75006 Paris, FRANCEtel: +33-1-55-42-84-00

Majallat Al-Oloom Kuwait Foundation for

the Advancement of SciencesP.O. Box 20856

Safat 13069, KUWAITtel: +965-2428186

Swiat NaukiProszynski i Ska S.A.

ul. Garazowa 7 02-651 Warszawa, POLAND


Nikkei Science, Inc. 1-9-5 Otemachi, Chiyoda-ku

Tokyo 100-8066, JAPANtel: +813-5255-2821

Svit NaukyLviv State Medical University

69 Pekarska Street290010, Lviv, UKRAINE


Ke XueInstitute of Scientific and

Technical Information of ChinaP.O. Box 2104

Chongqing, Sichuan PEOPLE’S REPUBLIC OF CHINA

tel: +86-236-3863170

NEW YORKKate Dobson, PUBLISHER

tel: 212-451-8522, [email protected] Madison Avenue

New York, NY 10017fax: 212-754-1138

Thomas Potratz, ADVERTISING DIRECTOR

tel: 212-451-8561, [email protected]

Kevin Gentzeltel: 212-451-8820, [email protected]

Randy Jamestel: 212-451-8528, [email protected]

Stuart M. Keatingtel: 212-451-8525, [email protected]

Wanda R. Knoxtel: 212-451-8530, [email protected]

Laura Salant, MARKETING DIRECTOR

tel: 212-451-8590, [email protected] Schube, PROMOTION MANAGER

tel: 212-451-8592, [email protected] Spirakis, RESEARCH MANAGER

tel: 212-451-8529, [email protected] Mongelli, PROMOTION DESIGN MANAGER

tel: 212-451-8532, [email protected]

ASSISTANTS: May Jung, Beth O’Keeffe

DETROITEdward A. Bartley, MIDWEST MANAGER

3000 Town Center, Suite 1435Southfield, MI 48075

tel: 248-353-4411, fax: [email protected]

OFFICE MANAGER: Kathy McDonald

CHICAGORandy James, CHICAGO REGIONAL MANAGER


LOS ANGELESLisa K. Carden, WEST COAST MANAGER

1554 South Sepulveda Blvd., Suite 212Los Angeles, CA 90025


ASSISTANT: Stacy Slossy

SAN FRANCISCODebra Silver, SAN FRANCISCO MANAGER

225 Bush Street, Suite 1453San Francisco, CA 94104


ASSISTANT: Rosemary Nocera

DALLASThe Griffith Group

16990 Dallas Parkway, Suite 201Dallas, TX 75248


International Advertising Contacts

CANADAFenn Company, Inc.

2130 King Road, Box 1060King City, OntarioL7B 1B1 Canada


EUROPERoy Edwards, INTERNATIONAL

ADVERTISING DIRECTOR

Thavies Inn House, 3/4, Holborn CircusLondon EC1N 2HB, England

tel: +44 171 842-4343, fax: +44 171 [email protected]

BENELUXReginald Hoe Europa S.A.

Rue des Confédérés 291040 Bruxelles, Belgium

tel: +32-2/735-2150, fax: +32-2/735-7310

MIDDLE EASTPeter Smith Media & Marketing

Moor Orchard, Payhembury, HonitonDevon EX14 OJU, England

tel: +44 140 484-1321, fax: +44 140 484-1320

JAPAN Tsuneo Kai

Nikkei International Ltd.1-6-6 Uchikanda, Chiyoda-ku

Tokyo 101-0047, Japantel: +813-5259-2690, fax: +813-5259-2679

KOREAJo, Young Sang

Biscom, Inc.Kwangwhamun, P.O. Box 1916

Seoul, Koreatel: +822 739-7840, fax: +822 732-3662

HONG KONGStephen Hutton

Hutton Media LimitedSuite 2102, Fook Lee

Commercial Centre Town Place33 Lockhart Road, Wanchai, Hong Kong

tel: +852 2528 9135, fax: +852 2528 9281

Advertising and Marketing Contacts


Introduction6 Scientific American Presents

Intelligence Considered by Philip Yam, issue editor

THE

ESTA

TE O

F KE

ITH

HA

RIN

G

Untitled, Keith Haring


Exploring Intelligence 7Intelligence Considered

or the past several years, the Sunday newspaper supple-ment Parade has featured a column called “Ask Marilyn.”People are invited to query Marilyn vos Savant, who at age 10had tested at a mental level of someone about 23 years old;that gave her an intelligence quotient of 228—the highestscore ever recorded. IQ tests ask you to complete verbal andvisual analogies, to envision paper after it has been foldedand cut, and to deduce numerical sequences, among othersimilar tasks. So it is a bit perplexing when vos Savant fieldssuch queries from the average Joe (whose IQ is 100) as, What’sthe difference between love and infatuation? Or what is thenature of luck and coincidence? It’s not obvious how thecapacity to visualize objects and to figure out numerical patternssuits one to answer questions that have eluded some of thebest poets and philosophers.

Clearly, intelligence encompasses more than a score on atest. Just what does it mean to be smart? How much of intelli-gence can be specified, and how much can we learn about itfrom neurobiology, genetics, ethology, computer science andother fields?

The defining term of intelligence in humans still seems tobe the IQ score, even though IQ tests are not given as often asthey used to be. The test comes primarily in two forms: theStanford-Binet Intelligence Scale and the Wechsler IntelligenceScales (both come in adult and children’s versions). Generallycosting several hundred dollars, they are usually given onlyby psychologists, although variations of them populate book-stores and the World Wide Web. (Superhigh scores like vosSavant’s are no longer possible, because scoring is now basedon a statistical population distribution among age peers,rather than simply dividing the mental age by the chronolog-ical age and multiplying by 100.) Other standardized tests,such as the Scholastic Assessment Test (SAT) and the GraduateRecord Exam (GRE), capture the main aspects of IQ tests.

Such standardized tests may not assess all the importantelements necessary to succeed in school and in life, arguesRobert J. Sternberg. In his article “How Intelligent Is IntelligenceTesting?”, Sternberg notes that traditional tests best assessanalytical and verbal skills but fail to measure creativity andpractical knowledge, components also critical to problemsolving and life success. Moreover, IQ tests do not necessarilypredict so well once populations or situations change. Researchhas found that IQ predicted leadership skills when the testswere given under low-stress conditions, but under high-stressconditions, IQ was negatively correlated with leadership—thatis, it predicted the opposite. Anyone who has toiled throughcollege entrance exams will testify that test-taking skill alsomatters, whether it’s knowing when to guess or what ques-tions to skip.

Sternberg has developed tests to measure the creative andpractical sides of the mind. Some schools and businesses usethem, and Sternberg has published work showing their predic-

tive value in subsequent tasks, but they have yet to gain muchacceptance in the mainstream testing business.

Still, conventional standardized testing has leveled thefield for most people—whatever their shortcomings, the examsprovide some standard by which universities can select stu-dents. Contrast this with the time before World War II, whenfamily background and attendance at elite prep schools werekey requirements for selective colleges.

That tests cannot capture all of a person’s skills in a neatnumber is an important crux of the article by Howard Gardner.In “A Multiplicity of Intelligences,” he espouses his view,developed in part after working with artists and musicianswho had suffered strokes, that human intelligence is bestthought of as consisting of several components, perhaps asmany as nine. Components such as spatial and bodily-kines-thetic, embodied by, say, architect Frank Lloyd Wright andhockey player Wayne Gretzky, elude test measures. Gardner’sclassifications are not arbitrary; he draws from evolution,brain function, developmental biology and other disciplines.

Gardner has been quite influential in education circles,where his theory is often required study for teachers-to-be. Hefeels, however, that some of his ideas are being misinterpreted.He mentions Daniel Goleman’s best-seller, Emotional Intelligence,the central concept of which is based on multiple-intelli-gences theory. Gardner maintains that the theory should notbe used to create a value system, as suggested in Goleman’sbook. People with high emotional quotients aren’t necessarilywell adjusted and kind to others—think Hannibal Lecter.

In Defense of IQ

In sharp contrast to Sternberg and Gardner is Linda S.Gottfredson. In “The General Intelligence Factor,” she makesthe case for the psychologist’s g—that is, a single factor forbrains. Other elements, such as linguistic ability and mathe-matical skill, fall below g in the hierarchy of human skills. Sheargues that IQ scores are important predictors for both acade-mic and life success and draws on biology to bolster her ideas.

The concept of g has a long and stormy history. First pro-posed in the early part of this century, it has waxed and wanedin popularity. Among the public and the media, the concepttook a hard hit in 1981, when Stephen Jay Gould publishedhis now classic The Mismeasure of Man. In it, he argues thatearly researchers (perhaps unconsciously) biased their mea-surements of intelligence based on race and points to short-comings of those trying to substantiate g. For instance, hetakes to task Catherine M. Cox’s 1926 publication of deducedIQ scores of past historical figures. Gould notes that Cox drewher assumptions based on written biographical accounts of aperson’s deeds. Unfortunately, the existence of such biogra-phies correlated with the prominence of the family—poorerfamilies were less likely to have documentation of their chil-

F

What does it mean to havebrainpower? A search for a definition of intelligence


dren’s accomplishments. Hence, pioneering British physicistMichael Faraday, from a modest background, gets a surprisinglylow childhood IQ score of 105.

Psychometricans (psychologists who apply statistics tomeasure intelligence) have a hostile view of Gould. Accordingto critics, many of whom recently have written new reviewsfor the rerelease of Mismeasure, Gould does not grasp factoranalysis—the statistical technique used to extract g. In a 1995review published in the journal Intelligence, John B. Carroll ofthe University of North Carolina at Chapel Hill writes that “itis indeed odd that Gould continues to place the burden of hiscritique on factor analysis, the nature and purpose of which, Ibelieve, he still fails to understand.” This is one of the milder

criticisms leveled at Gould by psychometricians.The stormy debate about g stems from its political, racial

and eugenics overtones. Historically, the idea of IQ has beenused to justify excluding certain immigrant groups, to maintainstatus quo policies and even to sterilize some people. Scientistswho hold views that intelligence is strongly hereditary areoften vilified by the general population, sometimes rightlyand sometimes wrongly. One researcher who has a bad publicimage that is not on par with the opinion of professionalpeers is Arthur R. Jensen of the University of California atBerkeley: even those working psychologists who disagree with

him consider his investigations to be solid research.Modern genetic studies threaten to inflame the racial con-

troversy even more. For example, this past May, Robert Plominof the Institute of Psychiatry in London and several collabora-tors reported the discovery of a gene variation that is statisti-cally linked with high intelligence. The variation lies in chro-mosome 6, within a gene that encodes for a receptor for aninsulinlike growth factor (specifically, IGF-2), which mightaffect the brain’s metabolic rate.

In some respects, the discovery is not truly surprising.Obviously, some people are born smarter than others. Butnote who Plomin and his colleagues used as subjects: 50 stu-dents with high SAT scores. Strictly speaking, the researchersfound a gene for performance on the SAT. True, SATs correlatewith IQ scores, which in turn reflect g—which not everyoneagrees is the sole indicator of smarts. Complicating the analysesis the fact that average SAT scores have been variable; theydipped in the 1980s but are now swinging back up. Thatcould be the result of better schooling, because the SAT mea-sures achievement more than inherent learning capacities (forwhich IQ tests are designed). But even IQ scores have not beenas stable as was once thought. James R. Flynn of the Universityof Otago in New Zealand discovered that worldwide, IQ scoreshave been rising by about three points per decade—by a fullstandard deviation (15 points) in the past 50 years.

Are we truly smarter than our grandparents? Researchersaren’t sure just what hascaused the rise. (Flynn him-self, who is profiled in theJanuary 1999 issue of ScientificAmerican, doesn’t think therise is real.) Genetics clearlycannot operate on such ashort time scale. Ulric Neisserof Cornell University thinksit may have to do with theincreasing visual complexityof modern life. Images ontelevision, billboards andcomputers have enriched thevisual experience, makingpeople more capable in han-dling the spatial aspects of theIQ tests. So even though genesmight play a substantial rolein individual differences in

IQ, the environment dictates how those genes are expressed.In part to probe the genetic-environment mechanisms, the

American Psychological Association (APA) convened a task forceof mainstream psychologists. They published a 1995 report,Intelligence: Knowns and Unknowns, which concluded thatalmost nothing can be said about the reason for the 15-pointIQ difference between black and white Americans: “There is cer-tainly no such support for a genetic interpretation. At thistime, no one knows what is responsible for the differential.”

The APA report was sparked by the publication of The BellCurve, by Charles Murray and Richard J. Herrnstein. The report


Sir Francis Galton 200Johann Wolfgang von Goethe 185Francois-Marie Arouet Voltaire 170Alfred Lord Tennyson 155William Wordsworth 150Sir Walter Scott 150Lord Byron 150Abraham Lincoln 125George Washington 125Nicolaus Copernicus 105Michael Faraday 105

BRAIN ACTIVITY recorded by James B. Brewer andhis colleagues at Stanford University is revealed byfunctional magnetic resonance imaging. It showspart of the neural areas that operate during recall of a visual scene (above). Such imaging techniquesare enabling neurobiologists to pinpoint functionswithin the brain.

ESTIMATED IQ SCORES of eminent historical figures were pub-lished in 1926 by Catherine M. Cox in The Early Mental Traitsof Three Hundred Geniuses. Although such lists generateinterest, poor assumptions often underlie the analyses, renderingthe results highly questionable and largely irrelevant.

JAM

ES B

.BRE

WER


actually does not disagree with the data presented in the bookabout IQ scores and the notion of g. The interpretation of thedata, however, is a different story. To many scholars, The BellCurve played on psychometric data to advance a politically con-servative agenda—arguing, for instance, that g is largely inher-ited and that thus enrichment programs for disadvantagedyouth are doomed to failure. As staff writer Tim Beardsleypoints out in “For Whom Did the Bell Curve Toll?”, severalinterpretations are possible, and other studies have producedresults that run counter to the dreary conclusions offered byMurray and Herrnstein. Although it engendered heated debate,the book ultimately had little impact on government policy.

Function and Form

Even those who fall on the right end of the bell curve,however, do not necessarily have it easy. In “UncommonTalents: Gifted Children, Prodigies and Savants,” Ellen Winnerexplores the nature of children who are so mentally advancedthat schools often do not know how to educate them. Thesewhiz kids are expected to achieve on their own even thoughthey often are misunderstood, ridiculed and neglected. Manyare unevenly gifted, excelling in one field but doing averagein others. The most extreme cases are the so-called savants(formerly called idiot savants), who can perform astoundingfeats of calculation and memory despite having autism orautismlike symptoms. Studies of such people offer valuableinsights into how the human brain works.

Observations of brain-damaged patients have done muchto identify the discrete functional areas of the brain [see pastSCIENTIFIC AMERICAN articles, such as “The Split Brain Revisited,”by Michael S. Gazzaniga, July 1998; “Emotion, Memory andthe Brain,” by Joseph LeDoux, June 1994; and the specialissue Mind and Brain, September 1992]. Modern imaging tech-nology, such as positron-emission tomography (PET) andfunctional magnetic resonance imaging (fMRI), have helpedinvestigators to map cognitive function with structure [see“Visualizing the Mind,” by Marcus E. Raichle; SCIENTIFIC

AMERICAN, April 1994]. With such imaging, researchers can seehow the brain “lights up” when certain cognitive tasks are per-formed, such as reciting numbers or recalling a visual scene.

Structure and function are of particular interest to neuro-biologists trying to boost the brainpower of the common per-son. Several researchers in fact have ties to pharmaceuticalcompanies hoping to capitalize on what would seem to be ahuge market in cognitive enhancers. In “Seeking ‘Smart’Drugs,” staff writer Marguerite Holloway reviews the diverseapproaches. If you’re a sea slug or a fruit fly, scientists can dowonders for your memory. Humans have somewhat limitedchoices at the moment; the vast majority of compounds nowsold have no solid clinical basis. For instance, package labelsof the popular herb gingko biloba overstate its efficacy: astudy has shown that it has some modest benefits inAlzheimer’s patients, but no study has indicated that gingkodefinitely helps healthy individuals. Prospective compounds,including modified estrogen and nerve growth factors, seempromising, but the best smart drug may already be in yourkitchen: sugar, the energy source of neurons.

The exploration of human intelligence naturally raises thequestion of how humans got to be intelligent in the first place.In “The Emergence of Intelligence” (updated since its appear-ance in the October 1994 issue of Scientific American), WilliamH. Calvin puts forth a kind of 2001: A Space Odyssey hypothe-

sis: that ballistic movement, whether it’s pitching a baseball orthrowing sticks and stones at black monoliths, is the key tointelligence, because a degree of foresight and planning isrequired to hit the target. And these ingredients may have per-mitted language, music and creativity to emerge, differentiat-ing us from the rest of the world’s fauna.

Do Animals Think?

That’s not to say that animals aren’t intelligent. In“Reasoning in Animals,” James L. Gould and Carol GrantGould make a persuasive case that animals have some abilityto solve problems. The examples they cite and the studies theydescribe make it unlikely that strict behaviorism—that animals’

actions are dictated by conditioned responses—can explain itall. Of course, not everything an animal does is an act of cog-nition: many of the actions of animals are accomplished andrestricted by instinct and genes.

Language plays a role in the development of cognitiveabilities, too, as suggested by Irene M. Pepperberg’s article,“Talking with Alex: Logic and Speech in Parrots.” Alex is thefamous Grey parrot that can make requests and provideanswers in a seemingly reasoned way. Alex is unique in partbecause he’s a bird: other communicating animals have beenprimates, such as the chimpanzees Washoe and Kanzi and thegorilla Koko. Rigorously speaking, these animals are communi-cating through learned symbols and sounds; whether they aretruly engaging in language, which permits planning andabstraction, remains to be proved.

Besides language, another hallmark of intelligence may beself-awareness. Many investigators have grappled with humanconsciousness from a scientific perspective [see “The Puzzle ofConscious Experience,” by David J. Chalmers; SCIENTIFIC

AMERICAN, December 1995; and “The Problem of Conscious-ness,” by Francis Crick and Christof Koch; SCIENTIFIC AMERICAN,September 1992]. But how can you tell if an animal is self-aware? In the late 1960s Gordon G. Gallup, Jr., devised a nowclassic test using mirrors. Gallup painted a red dot on thefaces of anesthetized animals and then observed them when


NEURON TRANSISTOR, using a leech ganglion,unites carbon with silicon. The nerve cell (green),about 80 microns wide, fires depending on the sig-nals sent to the transistor. The fuzzy object piercingthe nerve cell is a micromanipulator.

PETE

R FR

OM

HER

Z M

ax P

lanc

k In

stitu

te o

f Bio

chem

istr

y


they awoke and noticed themselves in the mirror. An animalthat would start poking at the red spot on its face seeminglyindicated an awareness that it was seeing itself in the mirror,not another creature. Of all the animals tested in this way,only humans, chimpanzees and orangutans pass.

With self-awareness comes the ability to take into accountanother creature’s feelings—at least, that’s the way it works inhumans. Taking the pro side of the debate, “Can AnimalsEmpathize?”, Gallup reasons that chimps and orangutans havea sense of self, which they might use to model other creature’smental states.

Daniel J. Povinelli, however, remains skeptical (in the besttraditions of scientific open-mindedness, he adopts the “maybenot” view). He tells how he tested chimpanzees under a varietyof clever conditions to see if they understand that anothercreature cannot see them. It turns out that chimps will beg forfood from a blindfolded person (who does not see the chimps)as well as from a sighted individual. Such results suggest thatchimps do not reason about another animal’s state of mind—or even their own. That they pass the mirror test suggests toPovinelli that they are not necessarily self-aware. Instead theylearn that the mirror images are the same as themselves.

I, Robot

If our closest relatives aren’t self-aware, is there any chancethat a computer can be? In seeking to make a machine thatcan pass the so-called Turing test—that is, produce responsesthat would be indistinguishable from those of humans—

artificial intelligence has proved to be a substantial disappoint-ment. Yet passing the Turing test may be an unfair measure ofAI progress. In “On Computational Wings: Rethinking theGoals of Artificial Intelligence,” Kenneth M. Ford and Patrick J.Hayes maintain that the obsession with the Turing test has ledAI researchers down the wrong road. They draw an analogywith artificial flight: engineers for centuries tried to produceflying machines by mimicking the way birds soar. But modernaircraft obviously do not fly like birds, and fortunately so.From this argument, Ford and Hayes note that AI is effectivelyall around us—in instrumentation, in data-recognition tasks,

in “expert” systems such as medical-diagnosticprograms and in search software, such as intelli-gent agents, which roam cyberspace to retrieveinformation [see “Intelligent Software,” by PattieMaes; SCIENTIFIC AMERICAN, September 1995].

Several more formal AI projects exist. One isthat of Douglas B. Lenat of Cycorp in Austin, Tex.,who for more than a decade has been workingon CYC, a project that aims to create a machinethat can share and manage information that wehumans might consider common sense [see“Artificial Intelligence,” by Douglas B. Lenat;SCIENTIFIC AMERICAN, September 1995]. Anotheris that of Rodney Brooks and Lynn Andrea Steinof the Massachusetts Institute of Technology,whose team has produced Cog, a humanoidrobot that its makers hope to endow with abili-ties of a conscious human, without its necessari-ly being conscious.

A realm of AI that sparks intense, thoughperhaps unjustified, feelings of anxiety andhuman pride is game-playing machines. In“Computers, Games and the Real World,”Matthew L. Ginsberg summarizes the main con-tests that machines are playing and how theyfare against human competitors. Garry Kasparov’sloss in a six-game match against IBM’s Deep Bluelast year may have inspired some soul searching.The point of game-playing computers, however,is not so much to best their makers as to explorewhich types of calculation are best suited to thearchitecture of the silicon chip. As Ginsbergreminds us, computers are designed not toreplace us humans but to help us.

Indeed, life without computers is now hard to imagine.And the machines will get more ubiquitous. In “WearableIntelligence,” Alex P. Pentland explains how devices such askeyboards, monitor screens, wireless transmitters and receiversare getting so small that we can physically wear them. Imaginereading e-mail on special eyeglasses as you walk down the street,generating power in your shoes that is converted to electricitythat powers your personal-area network for cellular communi-cations. Two M.I.T. students, Thad Starner and Steve Mann,have spent time in such cyborg existences—Starner has beendoing it since 1992. They look like less slick versions of thefuturistic Borg creatures seen on the Star Trek series.

A true melding of mind and machine is still far away,although the appeal apparently is irresistible. British Telecom-munications has a project called Soul Catcher; the goal is todevelop a computer that can be slipped into the brain to aug-ment memory and other cognitive functions. Hans Moravecof Carnegie Mellon University and others have argued, some-


HUMANOID ROBOT KISMET of the Massachusetts Institute of Tech-nology interacts socially with humans with emotive expressions. Itbelongs to the Cog project, which seeks in part to develop a robotthat behaves as if it were conscious without necessarily being so.

SAM

OG

DEN


what disturbingly, that it should be possible to remove thebrain and download its contents into a computer—and withit, one hopes, personality and consciousness.

Connecting neurons to silicon is only in its infancy. PeterFromherz and his colleagues at theMax Planck Institute of Biochemistryin Martinsried-München, Germany,have managed to connect the two andcaused the neuron to fire wheninstructed by the computer chip.Granted, the neuron used in theexperiment came from a leech. But inprinciple “there are no show-stoppers”to neural chips, says computer scien-tist Chris Diorio of the University ofWashington, adding that “the elec-tronics part is the easy part.” Thedifficulty is the interface.

Diorio was one of the organizersof a weeklong meeting this past Augustsponsored by Microsoft Research andthe University of Washington thatexplored how biology might help cre-ate intelligent computer systems.Expert systems, notes co-organizer EricHorvitz of Microsoft Research, doquite well in their rather singular tasksbut cannot match an invertebrate inbehavioral flexibility. “A leechbecomes more risk taking when hun-gry,” he notes. “How do you build acircuit that takes risk?” The hydrocar-bon basis of neurons might also meanthat the brain is more efficient with itsconstituent materials than a computer is with its silicon. “If weknew what a synapse was doing, we could mimic it,” Dioriosays, but “we don’t have the mathematical foundation yet.”

Beyond Earth

While we have much to learn from the neurons on Earth,we stand to gain even more if we could find neurons fromother planets. In “Is There Intelligent Life Out There?”,Guillermo A. Lemarchand reviews the history of the search forextraterrestrial intelligence, or SETI. The odds say that othertechnological civilizations are out there, so why haven’t wemade contact yet, government conspiracies notwithstanding?The answer is simple: astronomers have looked at only a tinyfraction of the sky—some 10-16 of it. Almost all SETI fundshave come from private sources, and time on radio telescopesis limited.

One ingenious attempt to enlist help from amateurs isSETI@home. Interested parties would download a specialscreen saver for personal computers that, when running,would sift through data gathered from the Arecibo RadioObservatory in Puerto Rico (specifically, from Project SERENDIP).In other words, as you take a break from work, your PC wouldlook for artificial signals from space. Organizers estimate that50,000 machines running the screen saver would rival all cur-rent SETI projects. At press time, investigators were still com-pleting the software and looking for sponsorship: they need atleast $200,000 to proceed to the final phases. Check it out athttp://setiathome.ssL.berkeley.edu/ on the World Wide Web.

Of course, there’s the chance that we have alreadyreceived alien greetings but haven’t recognized them as such.In Lemarchand’s view, sending salutations of our own may bethe best way to make first contact. He proposes relying on a

supernova, on the assumption that other civilizations wouldalso turn their sights onto such relatively rare stellar explo-sions. Radio telescopes on Earth could send signals to nearbystar systems that have good views of both Earth and thesupernova.

Defining Intelligence

In the end, most of us would feel rather confident inidentifying intelligent signals, be they from space, a machine,an animal or other people. An exact definition of intelligenceis probably impossible, but the data at hand suggest at leastone: an ability to handle complexity and solve problems insome useful context—whether it is finding the solution to thequadratic equation or obtaining just-out-of-arm’s-reachbananas. The other issues surrounding intelligence—its neuraland computational basis, its ultimate origins, itsquantification—remain incomplete, controversial and, ofcourse, political.

No one would argue that it doesn’t pay to be smart. Therole that intelligence plays in modern society depends not onthe amount of knowledge gained about it but on the valuesthat a society chooses to emphasize—for the U.S., thatincludes fairness, equal opportunity, basic rights and toler-ance. That intelligence studies could pervert these values is,ultimately, the root of anxiety about such research. Vigilanceis critical and so is the need for a solid base of information bywhich to make informed judgments—a base to which, I hope,this issue has contributed.


LUNCH INVITATION? A few researchers worried that calling attention to ourselves, such as with the gold plaque on the Pioneer spacecraft, might bringextraterrestrial aliens intent on consuming humans. SETI scientists disagree,and some advocate sending more greetings from Earth.

NA

SA

SA


Human Intelligence12 Scientific American Presents

by Robert J. Sternberg

A typical American adolescent spendsmore than 5,000 hours in high school andseveral thousand more hours studying inthe library and at home. But for those stu-dents who wish to go on to college, muchof their fate is determined in the three orso hours it takes to complete the ScholasticAssessment Test (SAT) or the AmericanCollege Test (ACT). Four years later theymay find themselves in a similar positionwhen they apply to graduate, medical, lawor business school.

The stakes are high. In their 1994 bookThe Bell Curve, Richard J. Herrnstein andCharles Murray pointed out a correlationbetween scores on such tests and a varietyof measures of success, such as occupation-al attainment. They suggested that the U.S.is developing a “cognitive elite”—consistingof high-ability people in prestigious, lucra-tive jobs—and a larger population of low-ability people in dead-end, low-wage posi-tions. They suggested an invisible hand ofnature at work.

But to a large extent, the hand is neitherinvisible nor natural. We have decided as asociety that people who score well on thesehigh-stakes tests will be granted admissionto the best schools and, by extension, to thebest access routes to success. People haveused other criteria, of course: caste at birth,membership in governmental party, religiousaffiliation. A society can use whatever itwishes—even height, so that very soon peo-ple in prestigious occupations would be tall.(Oddly enough, to some extent Americansand many people in other societies alreadyuse this criterion.) Why have the U.S. andother countries chosen to use ability testsas a basis to open and close the access gates?

Are they really the measures that should beused? The answers lie in how intelligencetesting began.

A Brief History of Testing

Sir Francis Galton, a cousin of CharlesDarwin, made the first scientific attempt tomeasure intelligence. Between 1884 and1890 Galton ran a service at the South Ken-sington Museum in London, where, for asmall fee, people could have their intelli-gence checked. The only problem was thatGalton’s tests were ill chosen. For example,he contrived a whistle that would tell himthe highest pitch a person could perceive.Another test used several cases of gun car-tridges filled with layers of either shot, woolor wadding. The cases were identical inappearance and differed only in weight. Thetest was to pick up the cartridges and thento discriminate the lighter from the heavier.Yet another test was of sensitivity to thesmell of roses.

James McKeen Cattell, a psychologist atColumbia University, was so impressed withGalton’s work that in 1890 he devised simi-lar tests to be used in the U.S. Unfortun-ately for him, a student of his, Clark Wissler,decided to see whether scores on such testswere actually meaningful. In particular, hewanted to know if the scores were relatedeither to one another or to college grades.The answer to both questions proved to beno—so if the tests didn’t predict school per-formance or even each other, of what usewere they? Understandably, interest inGalton’s and Cattell’s tests waned.

A Frenchman, Alfred Binet, got off to a better start. Commissioned to devise a

How Intelligent Is

INTELLIGENCE

HU

MA

N



Conventional measures, such asSATs and IQ tests, miss criticalabilities essential to academicand professional success

Intelligence Testing?

How Intelligent Is Intelligence Testing?

means to predict school performance, he cast aroundfor test items. Together with his colleague TheodoreSimon, he developed a test of intelligence, publishedin 1905, that measured things such as vocabulary(“What does misanthrope mean?”), comprehension(“Why do people sometimes borrow money?”) andverbal relations (“What do an orange, an apple and apear have in common?”). Binet’s tests of judgmentwere so successful at predicting school performancethat a variant of them, called the Stanford-BinetIntelligence Scale (fourth edition), is still in usetoday. (Louis Terman of Stanford University popular-ized the test in the U.S.—hence the name.) A com-peting test series, the Wechsler Intelligence Scales,measures similar kinds of skills.

It is critical to keep in mind that Binet’s missionwas linked to school performance and, especially, todistinguishing children who were genuinely mentallyretarded from those who had behavior problems butwho were able to think just fine. The result was thatthe tests were designed, and continue to be designed,in ways that at their best predict school performance.

During World War I, intelligence testing reallytook off: psychologists were asked to develop amethod to screen soldiers. That led to the ArmyAlpha (a verbal test) and Beta (a performance testwith pantomimed directions instead of words),which were administered in groups. (Psychologistscan now choose between group or individuallyadministered tests, although the individual tests gen-erally give more reliable scores.) In 1926 a new testwas introduced, the forerunner to today’s SAT.Devised by Carl C. Brigham of Princeton University,the test provided verbal and mathematical scores.

Shortly thereafter, a series of tests evolved, whichtoday are used to measure various kinds of achieve-ments and abilities, including IQ (intelligence quo-tient), “scholastic aptitude,” “academic aptitude” andrelated constructs. Although the names of these testsvary, scores on all of them tend to correlate highly

CH

ARL

ES G

UPT

ON

The

Sto

ck M

arke

t


with one another, so for the purposes ofthis article I will refer to them loosely asconventional tests of intelligence.

What Tests Predict

Typically, conventional intelligencetests correlate about 0.4 to 0.6 (on a 0 to1 scale) with school grades, which statis-tically speaking is a respectable level ofcorrelation. A test that predicts perfor-mance with a correlation of 0.5, however,

accounts for only about 25 percent of thevariation in individual performances,leaving 75 percent of the variation unex-plained. (In statistics, the variation is thesquare of the correlation, so in this case,0.52 = 0.25.) Thus, there has to be muchmore to school performance than IQ.

The predictive validity of the testsdeclines when they are used to forecastoutcomes in later life, such as job per-formance, salary or even obtaining a jobin the first place. Generally, the correla-

tions are only a bit over 0.3, meaningthat the tests account for roughly 10percent of variation in people’s perfor-mance. That means 90 percent of thevariation is unexplained. Moreover, IQprediction becomes less effective oncepopulations, situations or tasks change.For instance, Fred Fiedler of the Univer-sity of Washington found that IQ posi-tively predicts leadership success underconditions of low stress. But in high-stress situations, the tests negatively pre-dict success. Some intelligence tests,including both the Stanford-Binet andWechsler, can yield multiple scores. Butcan prediction be improved?

Curiously, whereas many kinds oftechnologies, such as computers andcommunications, have moved forwardin leaps and bounds in the U.S. andaround the world, intelligence testingremains almost a lone exception. Thecontent of intelligence tests differs littlefrom that used at the turn of the century.Edwin E. Ghiselli, an American industrialpsychologist, wrote an article in 1966bemoaning how little the predictive valueof intelligence tests had improved in 40years. More than 30 years later the situa-tion remains unchanged.

Improving Prediction

We can do better. In research withMichael Ferrari of the University ofPittsburgh, Pamela R. Clinkenbeard of theUniversity of Wisconsin–Whitewater andElena L. Grigorenko of Yale University, Ishowed that a test that measured notonly the conventional memory and ana-lytical abilities but also creative and prac-


4 5

27

1

1

22

3 4

6 106 5

3

3 4

2

2

2 3

2

1 1

A B C

1

A B C D

SIR FRANCIS GALTON made the firstscientific attempt to measure intelligence.His tests included determining the pitch ofwhistles and the weight of gun cartridges.They were not particularly useful.

ALFRED BINET developed the examina-tion that is the forerunner of the modernIQ test. He devised questions that probedvocabulary, comprehension and verbalabilities to predict school performance.

1. The same mathematical rules apply within each row to produce thenumbers in the circles. The upper row, for instance, might meanmultiplication, whereas the lower row means subtraction. Deducethe rules for the items below and write the answer in the circle.

2. Two of the shapes represent mirrorimages of the same shape.Underline that pair.

CO

RBIS

-BET

TMA

NN

CO

RBIS

-BET

TMA

NN

A.

B.

Example:

A.

B.

QUESTIONS REPRESENTATIVE OF IQ and other standardized tests include mathe-matical deduction and computation, spatial visualization and verbal analogies.

Courtesy of Self-Scoring IQ Tests, by Victor Serebriakoff and Barnes & Noble and Robinson Publishing

Answers: 1A. 5; 1B. 3; 2A. A and C; 2B. B and D


tical thinking abilities could improve pre-diction of course grades for high schoolstudents in an introductory psychologycourse. (A direct comparison of correla-tions between this test and conventionaltests is not possible because of the restrict-ed sample, which consisted of high-abil-ity students selected by their schools.)

In these broader tests, individualshad to solve mathematical problems withnewly defined operators (for example, Xglick Y = X + Y if X < Y, and X – Y if X ≥Y), which require a more flexible kind ofthinking. And they were asked to planroutes on maps and to solve problemsrelated to personal predicaments, whichrequire a more everyday, practical kindof thinking. Here is one example:

The following question gives youinformation about the situation involv-ing a high school student. Read the ques-tion carefully. Choose the answer thatprovides the best solution, given thespecific situation and desired outcomes.

John’s family moved to Iowa fromArizona during his junior year in highschool. He enrolled as a new student inthe local high school two months ago butstill has not made friends and feels boredand lonely. One of his favorite activitiesis writing stories. What is likely to be themost effective solution to this problem?

A. Volunteer to work on the schoolnewspaper staff

B. Spend more time at home writing

columns for the school newsletterC. Try to convince his parents to

move back to ArizonaD. Invite a friend from Arizona to

visit during Christmas break

Best answer: A

Creativity can similarly be measured.For example, in another study, ToddLubart, now at René Descartes University-Paris V, and I asked individuals to per-form several creative tasks. They had towrite short stories based on bizarre titlessuch as The Octopus’s Sneakers or 3853,draw pictures of topics such as the earthseen from an insect’s point of view orthe end of time, come up with excitingadvertisements for bow ties, doorknobs

or other mundane products, and solvequasiscientific problems, such as howsomeone might find among us extrater-restrial aliens seeking to escape detec-tion. The research found that creativeintelligence was relatively domain-specific—that is, people who are creativein one area are not necessarily creativein another—and that creative perfor-mance is only weakly to moderately cor-related with the scores of conventionalmeasures of IQ.

The implications for such testingextend to teaching. The achievement ofstudents taught in a way that allowedthem to make the most of their distinc-tive pattern of abilities was significantlyhigher than that of students who weretaught in the conventional way, empha-

Exploring Intelligence 15How Intelligent Is Intelligence Testing?

A

A

B BB

Bis to as is to

is to

is to

as

as

is to

is to

A

A B C D E

B C

J

J L LL LLL

3. Underline the analogous shape.

INTELLIGENCE TESTING by Galtontook place between 1884 and 1890 at theSouth Kensington Museum in London.

AR

CH

IVES

OF

THE

HIS

TORY

OF

AM

ERIC

AN

PSY

CH

OLO

GY,

UN

IVER

SITY

OF

AKR

ON

Example:

A.

B.

4. Complete each analogy by underliningtwo words from those in the parentheses.

A. dog is to puppy as (pig, cat, kitten)B. circle is to globe as

(triangle, square, solid, cube)

5. Underline the two words whose mean-ings do not belong with the others.

A. shark, sea lion, cod, whale, flounderB. baize, paper, felt, cloth, tinfoilC. sword, arrow, dagger, bullet, club

Answers: 3A. B; 3B. E; 4A. Cat, kitten; 4B. Square, cube; 5A. Sea lion, whale (others are fish); 5B. Cloth, tinfoil (others aremade of compressed fibers); 5C. Arrow, bullet (others are used by the hand)



sizing memory. Indeed, furtherresearch done by Bruce Torff ofHofstra University, Grigorenko andme has shown that the achieve-ments of all students improve, onaverage, when they are taught tothink analytically, creatively andpractically about the material theylearn, even if they are tested onlyfor memory performance.

Interestingly, whereas individ-uals higher in conventional (mem-ory and analytical) abilities tendedto be primarily white, middle- toupper-middle-class and in “better”schools, students higher in creativeand practical abilities tended to beracially, socioeconomically andeducationally more diverse, andgroup differences were not signif-icant. Group differences in conven-tional test scores—which are com-mon and tend to favor white stu-dents—therefore may be in part afunction of the narrow range ofabilities that standard tests favor.

Tests can also be designed toimprove prediction of job perfor-mance. Richard K. Wagner of Flor-ida State University and I haveshown that tests of practical intel-ligence in the workplace can pre-dict job performance as well as orbetter than IQ tests do, even thoughthese tests do not correlate withIQ. In such a test, managers mightbe told that they have a numberof tasks to get done in the next threeweeks but do not have time to do themall and so must set priorities. We havedevised similar tests for salespeople, stu-dents and, most recently, military leaders(in a collaborative effort with psycholo-gists at the U.S. Military Academy atWest Point). Such tests do not replaceconventional intelligence tests, whichalso predict job performance, but rathersupplement them.

A Question of Culture

Cultural prerogatives also affect scoreson conventional tests. Grigorenko and I,in collaboration with Kate Nokes andRuth Prince of the University of Oxford,Wenzel Geissler of the Danish BilharziasisLaboratory in Copenhagen, FrederickOkatcha of Kenyatta University in Nai-robi and Don Bundy of the Universityof Cambridge, designed a test of indige-nous intelligence for Kenyan children ina rural village. The test required them toperform a task that is adaptive for them:

recognizing how to use natural herbalmedicines to fight illnesses. Children inthe village knew the names of manysuch medicines and in fact treated them-selves once a week on average. (Westernchildren, of course, would know none ofthem.) The children also took conven-tional IQ tests.

Scores on the indigenous intelligencetest correlated significantly but negativelywith vocabulary scores on the Westerntests. In other words, children who didbetter on the indigenous tests actuallydid worse on the Western tests, and viceversa. The reason may be that parentstend to value indigenous education orWesternized education but not both, andthey convey those particular values totheir children.

People from different cultures mayalso interpret the test items differently.In 1971 Michael Cole, now at the Uni-versity of California at San Diego, andhis colleagues studied the Kpelle, wholive in western Africa. Cole’s team foundthat what the Kpelle considered to be a

smart answer to a sorting problem,Westerners considered to be stupid,and vice versa. For instance, giventhe names of categories such asfruits and vegetables, the Kpellewould sort functionally (forinstance, “apple” with “eat”),whereas Westerners would sort cat-egorically (“apple” with “orange,”nested under the word “fruit”).

Westerners do it the way theylearn in school, but the Kpelle do itthe way they (and Westerners) aremore likely to do it in everydaylife. People are more likely to thinkabout eating an apple than aboutsorting an apple into abstract taxo-nomic categories.

Right now conventionalWestern tests appear in translatedform throughout the world. Butthe research results necessarilyraise the question of whether sim-ply translating Western tests forother cultures makes much sense.

Toward a Better Test

If we can do better in testingthan we currently do, then, gettingback to the original question posedat the beginning of the article,how have we gotten to where weare? Several factors have conspiredto lead us as a society to weighconventional test scores heavily:1. The appearance of precision. Test

scores look so precise that institutionsand the people in them often accordthem more weight then they probablydeserve.

2. The similarity factor. A fundamen-tal principle of interpersonal attractionis that people tend to be attracted tothose who are similar to them. This prin-ciple applies not only in intimate rela-tionships but in work relationships aswell. People in positions of power lookfor others like themselves; because theyneeded high test scores to get where theyare, they tend to seek others who havehigh test scores.

3. The publication factor. Ratings ofinstitutions, such as those publishedannually in U.S. News and World Report,create intense competition among col-leges and universities to rank near thetop. The institutions cannot control allthe factors that go into the ranking. Buttest scores are relatively easier to controlthan, say, scholarly publications of fac-ulty, so institutions start to weigh testscores more heavily to prop up their rat-

KPELLE OF WESTERN AFRICA illustrate the short-coming of translating Western IQ tests for differentcultures. The Kpelle would sort items based on func-tionality—such as “apple” with “eat”—whereasstandard tests seek to sort based on category—“apple” with “orange.”

JAC

QU

ES J

AN

GO

UX

Ton

y St

one

Imag

es


ings. Publication of mastery-test scoresby states also increases the pressure onthe public schools to teach to the tests.

4. Confirmation bias. Once peoplebelieve in the validity of the tests, theytend to set up situations that confirmtheir beliefs. If admissions officialsbelieve, for example, that students withtest scores below a certain point cannot

successfully do the work in their institu-tion, they may not admit students withscores below that point. The result is thatthe institutions never get a chance to seeif others could successfully do the work.

Given the shortcomings of conven-tional tests, there are those who wouldlike to get rid of standardized testingaltogether. I believe this course of action

would be a mistake. Without test scores,we are likely to fall into the trap of over-weighting factors that should matter lessor not at all, whether it is political pullor socioeconomic status or just plaingood looks. Societies started using teststo increase, not to decrease, equity for all.

Others would like to use only perfor-mance-based measures, such as havingchildren do actual science experiments.The problem with such measures is that,despite their intuitive appeal, they are noless culturally biased than conventionaltests and have serious problems of statis-tical reliability and validity that haveyet to be worked out.

A sensible plan would be to continueto use conventional tests but to supple-ment them with more innovative tests,some of which are already available andothers of which have to be invented.Unlike most kinds of companies involvedin technology, testing firms spend littleor nothing on basic research, and theirapplied work is often self-serving. Giventhe monopoly a few companies have inthe testing industry and the importanceof tests, we might think as a society ofstrongly encouraging or even requiringthe testing companies to modify theirapproach. Or the public could fundresearch on its own. The innovationsshould be not just in the vehicles fortesting (such as computerized testing)but in the very content of the tests. Thetime has come to move testing beyondthe horse and buggy. We have the means;we just need the will.

Exploring Intelligence 17How Intelligent Is Intelligence Testing?

PREDICTING JOB PERFORMANCE can be accomplished with tests of prac-tical intelligence, which require solvingreal-world problems. Such tests do notcorrelate with IQ, however.

SA

The following task represents a work-related situation, followed by a series ofitems that are relevant to handling the situation. Briefly scan all the items and thenrate the quality of each item on the 1 to 7 scale provided.

An employee who reports to one of your subordinates has asked to talk with youabout waste, poor management practices and possible violations of both company policyand the law on the part of your subordinate. You have been in your present position onlya year, but in that time you have had no indications of trouble about the subordinatein question. Neither you nor your company has an “open door” policy, so it is expectedthat employees should take their concerns to their immediate supervisors before bring-ing a matter to the attention of anyone else. The employee who wishes to meet withyou has not discussed this matter with her supervisors because of its delicate nature.

1—————2—————3—————4—————5—————6—————7extremely neither good extremely

bad nor bad good

1. Refuse to meet with the employee unless the individual first discusses thematter with your subordinate.

2. Meet with the employee but only with your subordinate present.3. Schedule a meeting with the employee and then with your subordinate to

get both sides of the story.4. Meet with the employee and then investigate the allegations if an investiga-

tion appears warranted before talking with your subordinate.5. Find out more about the employee, if you can, before making any decisions.6. Refuse to meet with the employee and inform your subordinate that the

employee has attempted to sidestep the chain of command.7. Meet with your subordinate first before deciding whether to meet with the

employee.8. Reprimand the employee for ignoring the chain of command.9. Ask a senior colleague whom you respect for advice about what to do in this

situation.10. Turn the matter over to an assistant.

ROB

ERT

J.ST

ERN

BER

G

ROBERT J. STERNBERG wrote his own version of an intelli-gence test when he was just 13. “I wish I still had a copy,” he says.“It’s probably as good as anything I’ve published since.” Then,as now, Sternberg believed that the standard tests were not goodmeasures of intelligence. But his research was canceled by theschool psychologist. “For some reason, the guy didn’t like theidea of a 13-year-old giving IQ tests to his classmates,” he recalls.“But some people still don’t like my ideas, so nothing reallychanges in life.” Now Sternberg is professor of psychology andeducation at Yale University, where he had been an undergraduate.In addition to intelligence, Sternberg also studies love, creativity,conflict resolution and other psychology issues. “I’m a dabbler,”he admits. But a dabbler with a mission. “I want to have peopleview intelligence more broadly,” Sternberg says. “If you can openpeople’s eyes and get them to question what they’ve been doingor how they’ve been thinking about things, it’s really rewarding.”M

ICH

AEL

MA

RSLA

ND

Yal

e U

nive

rsit

y

About the Author



REPR

INTE

D W

ITH

PER

MIS

SIO

N F

ROM

ART

ISTS

RIG

HTS

SO

CIE

TY

A Man Can Conceal Another, by Max Ernst


As a psychologist, I was surprised by the hugepublic interest in The Bell Curve, the 1994 book onhuman intelligence by the late Harvard Universitypsychologist Richard J. Herrnstein and policy analystCharles Murray. Most of the ideas in the book werefamiliar not only to social scientists but also to thegeneral public. Indeed, educational psychologistArthur R. Jensen of the University of California atBerkeley as well as Herrnstein had written popularlyabout the very same ideas in the late 1960s and theearly 1970s. Perhaps, I reasoned, every quarter-cen-tury a new generation of Americans desires to beacquainted with “the psychologist’s orthodoxy”about intelligence—namely, that there is a single,general intelligence, often called g, which is reflectedby an individual’s intelligence quotient, or IQ.

This concept stands in contrast to my own viewdeveloped over the past decades: that human intel-ligence encompasses a far wider, more universal setof competences. Currently I count eight intelligences,and there may be more. They include what are tra-ditionally regarded as intelligences, such as linguis-tic and logical-mathematical abilities, but also somethat are not conventionally thought of in that way,such as musical and spatial capacities. These intelli-gences, which do not always reveal themselves inpaper-and-pencil tests, can serve as a basis for moreeffective educational methods.

Defining Brainpower

The orthodox view of a single intelligence,widely, if wrongly, accepted today in the minds ofthe general population, originated from the ener-gies and convictions of a few researchers, who by

the second decade of this century had put forth itsmajor precepts. In addition to its basic assumption,the orthodoxy also states that individuals are bornwith a certain intelligence or potential intelligence,that this intelligence is difficult to change and thatpsychologists can assess one’s IQ using short-answertests and, perhaps, other “purer” measures, such as thetime it takes to react to a sequence of flashing lights orthe presence of a particular pattern of brain waves.

Soon after this idea had been proposed—I liketo call it “hedgehog orthodoxy”—more “foxlike”critics arose. From outside psychology, commentatorssuch as American newspaper columnist Walter Lipp-mann challenged the criteria used to assess intelli-gence, contending that it was more complex andless fixed than the psychometricians had proposed.

From within psychology, scientists questionedthe notion of a single, overarching intelligence.According to their analyses, intelligence is betterthought of as a set of several factors. In the 1930sLouis L. Thurstone of the University of Chicagosaid it makes more sense to think of seven, largelyindependent “vectors of the mind.” In the 1960sJoy P. Guilford of the University of SouthernCalifornia enunciated 120 factors, later amended to150. Scottish investigator Godfrey Thomson of theUniversity of Edinburgh spoke around the 1940s ofa large number of loosely coupled faculties. And inour own day, Robert J. Sternberg of Yale Universityhas proposed a triarchic theory of intellect. Thesearches comprise a component that deals with stan-dard computational skill, a component that is sensi-tive to contextual factors and a component that isinvolved with novelty.

Somewhat surprisingly, all these commentators—

Exploring Intelligence 19A Multiplicity of Intelligences

A Multiplicity of Intelligences

Rather than having just an intelligence defined by IQ, humans are better thought of as having eight, maybe nine, kinds of intelligences, including musical, spatial and kinesthetic

by Howard Gardner

© 1

998

How

ard

Gar

dner



whether in favor of or opposed to thenotion of single intelligence—share oneconviction. They all believe that thenature of intelligence will be determinedby testing and analyzing the data thussecured. Perhaps, reason orthodoxdefenders like Herrnstein and Murray,performance on a variety of tests willyield a strong general factor of intelli-gence. And indeed, there is evidence forsuch a “positive manifold,” or high cor-relation, across tests. Perhaps, counterpluralists like Thurstone and Sternberg,the right set of tests will demonstratethat the mind consists of a number of rel-atively independent factors, withstrength in one area failing to predictstrength or weakness in other areas.

But where is it written that intelli-gence needs to be determined on thebasis of tests? Were we incapable ofmaking judgments about intellect beforeSir Francis Galton and Alfred Binet cob-bled together the first set of psychometricitems a century ago? If the dozens of IQtests in use around the world were sud-denly to disappear, would we no longerbe able to assess intellect?

Break from Orthodoxy

Nearly 20 years ago, posing thesevery questions, I embarked on quite adifferent path into the investigation ofintellect. I had been conducting researchprimarily with two groups: childrenwho were talented in one or more artform and adults who had suffered fromstrokes that compromised specific capac-ities while sparing others. Every day Isaw individuals with scattered profiles ofstrengths and weaknesses, and I was im-pressed by the fact that a strength or adeficit could cohabit comfortably withdistinctive profiles of abilities and dis-abilities across the variety of humankind.

On the basis of such data, I arrivedat a firm intuition: human beings arebetter thought of as possessing a numberof relatively independent faculties, ratherthan as having a certain amount of intel-lectual horsepower, or IQ , that can besimply channeled in one or anotherdirection. I decided to search for a betterformulation of human intelligence. Idefined an intelligence as “a psychobio-logical potential to solve problems or tofashion products that are valued in atleast one cultural context.” In my focuson fashioning products and cultural val-ues, I departed from orthodox psychome-tric approaches, such as those adopted byHerrnstein, Murray and their predecessors.

To proceed from an intuition to adefinition of a set of human intelligences,I developed criteria that each of the can-didate intelligences had to meet [see boxat left]. These criteria were drawn fromseveral sources:

• Psychology: The existence of a dis-tinct developmental history for a capaci-ty through which normal and giftedindividuals pass as they grow to adult-hood; the existence of correlations (orthe lack of correlations) between certaincapacities.

• Case studies of learners: Obser-vations of unusual humans, includingprodigies, savants or those sufferingfrom learning disabilities.

• Anthropology: Records of how dif-ferent abilities are developed, ignored orprized in different cultures.

• Cultural studies: The existence ofsymbol systems that encode certain kindsof meanings—language, arithmetic andmaps, for instance.

• Biological sciences: Evidence thata capacity has a distinct evolutionaryhistory and is represented in particularneural structures. For instance, variousparts of the left hemisphere dominatewhen it comes to motor control of thebody, calculation and linguistic ability;the right hemisphere houses spatial andmusical capacities, including the dis-crimination of pitch.

The Eight Intelligences

Armed with the criteria, I consid-ered many capacities, ranging fromthose based in the senses to those hav-ing to do with planning, humor andeven sexuality. To the extent that a can-didate ability met all or most of the cri-teria handily, it gained plausibility as anintelligence. In 1983 I concluded thatseven abilities met the criteria suffi-ciently well: linguistic, logical-mathe-matical, musical, spatial, bodily-kines-thetic (as exemplified by athletes, dancersand other physical performers), interper-sonal (the ability to read other people’smoods, motivations and other mentalstates), and intrapersonal (the ability toaccess one’s own feelings and to drawon them to guide behavior). The lasttwo can generally be considered togeth-er as the basis for emotional intelligence(although in my version, they focusmore on cognition and understandingthan on feelings). Most standard mea-sures of intelligence primarily probe lin-guistic and logical intelligence; somesurvey spatial intelligence. The other

Criteria for anIntelligence

1. Potential isolation bybrain damage. For example, lin-guistic abilities can be compro-mised or spared by strokes.

2. The existence of prodigies,savants and other exceptionalindividuals. Such individuals per-mit the intelligence to be observedin relative isolation.

3. An identifiable core opera-tion or set of operations. Musicalintelligence, for instance, consistsof a person’s sensitivity to melody,harmony, rhythm, timbre andmusical structure.

4. A distinctive developmen-tal history within an individual,along with a definable nature ofexpert performance. One examinesthe skills of, say, an expert athlete,salesperson or naturalist, as well asthe steps to attaining such expertise.

5. An evolutionary historyand evolutionary plausibility. Onecan examine forms of spatial intel-ligence in mammals or musicalintelligence in birds.

6. Support from tests inexperimental psychology.Researchers have devised tasks thatspecifically indicate which skills arerelated to one another and whichare discrete.

7. Support from psychometricfindings. Batteries of tests revealwhich tasks reflect the same under-lying factor and which do not.

8. Susceptibility to encodingin a symbol system. Codes such aslanguage, arithmetic, maps andlogical expression, among others,capture important components ofrespective intelligences.



four are almost entirely ignored. In1995, invoking new data that fit the crite-ria, I added an eighth intelligence—thatof the naturalist, which permits therecognition and categorization of natur-al objects. Examples are Charles Darwin,John James Audubon and RachelCarson. I am currently considering thepossibility of a ninth: existential intelli-gence, which captures the human pro-clivity to raise and ponder fundamentalquestions about existence, life, death,finitude. Religious and philosophicalthinkers such as the Dalai Lama andSøren A. Kierkegaard exemplify this kindof ability. Whether existential intelligencegets to join the inner sanctum dependson whether convincing evidence accruesabout the neural basis for it.

The theory of multiple intelligences(or MI theory, as it has come to becalled) makes two strong claims.The first is that all humans possessall these intelligences: indeed,they can collectively be considered adefinition of Homo sapiens, cognitivelyspeaking. The second claim is that justas we all look different and have uniquepersonalities and temperaments, we alsohave different profiles of intelligences.No two individuals, not even identicaltwins or clones, have exactly the sameamalgam of profiles, with the samestrengths and weaknesses. Even in thecase of identical genetic heritage, indi-viduals undergo different experiencesand seek to distinguish their profilesfrom one another.

Within psychology, the theory ofmultiple intelligences has generatedcontroversy. Many researchers are ner-vous about the movement away fromstandardized tests and the adoption of aset of criteria that are unfamiliar andless open to quantification. Many alsobalk at the use of the word “intelligence”to describe some of the abilities, prefer-ring to define musical or bodily-kines-thetic intelligences as talents. Such anarrow definition, however, devaluesthose capacities, so that orchestra con-ductors and dancers are talented but notsmart. In my view, it would be all rightto call those abilities talents, so long aslogical reasoning and linguistic facilityare then also termed talents.

Some have questioned whether MItheory is empirical. This criticism, how-ever, misses the mark. MI theory is basedcompletely on empirical evidence. Thenumber of intelligences, their delineation,their subcomponents are all subject toalteration in the light of new findings.

Indeed, the existence of the naturalistintelligence could be asserted only afterevidence had accrued that parts of thetemporal lobe are dedicated to the nam-ing and recognition of natural things,whereas others are attuned to human-made objects. (Good evidence for a neur-al foundation comes from clinical litera-ture, which reported instances in whichbrain-damaged individuals lost the capac-ity to identify living things but couldstill name inanimate objects. Experimentalfindings by Antonio R. Damasio of theUniversity of Iowa, Elizabeth Warring-ton of the Dementia Research Group atNational Hospital in London and othershave confirmed the phenomenon.)

Much of the evidence for the per-sonal intelligences has come fromresearch in the past decade on emotion-

al intelligence and on the developmentin children of a “theory of mind”—therealization that human beings have in-tentions and act on the basis of theseintentions. And the intriguing findingby Frances H. Rauscher of the Universityof Wisconsin–Oshkosh and her col-leagues of the “Mozart effect”—that earlymusical experiences may enhance spatialcapacities—raises the possibility thatmusical and spatial intelligences drawon common abilities.

It is also worth noting that themovement toward multiple intelligencesis quite consistent with trends in relatedsciences. Neuroscience recognizes themodular nature of the brain; evolution-ary psychology is based on the notionthat different capacities have evolved inspecific environments for specific purpos-es; and artificial intelligence increasinglyembraces expert systems rather thangeneral problem-solving mechanisms.Within science, the believers in a singleIQ or general intelligence are increasinglyisolated, their positions more likely tobe embraced by those, like Herrnsteinand Murray, who have an ideological axto grind.

If some psychologists expressedskepticism about the theory of multipleintelligences, educators around theworld have embraced it. MI theory notonly comports with their intuitions thatchildren are smart in different ways; italso holds out hope that more studentscan be reached more effectively if theirfavored ways of knowing are taken intoaccount in curriculum, instruction and

assessment. A virtual cottage industryhas arisen to create MI schools, class-rooms, curricula, texts, computer sys-tems and the like. Most of this work iswell intentioned, and some of it hasproved quite effective in motivating stu-dents and in giving them a sense ofinvolvement in intellectual life.

Various misconceptions, however,have arisen: for example, that every topicshould be taught in seven or eight waysor that the purpose of school is to identi-fy (and broadcast) students’ intelligences,possibly by administering an octet ofnew standardized tests. I have begun tospeak out against some of these lessadvisable beliefs and practices.

My conclusion is that MI theory isbest thought of as a tool rather than asan educational goal. Educators need to

determine, in conjunction with theircommunities, the goals that they areseeking. Once these goals have beenarticulated, then MI theory can providepowerful support. I believe schoolsshould strive to develop individuals of acertain sort—civic-minded, sensitive tothe arts, deeply rooted in the disci-plines. And schools should probe piv-otal topics with sufficient depth so thatstudents end up with a comprehensiveunderstanding of them. Curricular andassessment approaches founded on MItheory, such as Project Spectrum at theEliot-Pearson Preschool at TuftsUniversity, have demonstrated consider-able promise in helping schools toachieve these goals.

The Future of MI

Experts have debated various topicsin intelligence—including whether thereis one or more—for nearly a century, andit would take a brave seer to predict thatthese debates will disappear. (In fact, ifpast cycles repeat themselves, a latter-day Herrnstein and Murray will authortheir own Bell Curve around 2020.) Asthe person most closely associated withthe theory of multiple intelligences, Irecord three wishes for this line of work.

The first is a broader but not infinite-ly expanded view of intelligence. It ishigh time that intelligence be widenedto incorporate a range of human com-putational capacities, including thosethat deal with music, other persons andskill in deciphering the natural world.


All humans possess all these intelligences: indeed, they can collectivelybe considered a definition of Homo sapiens, cognitively speaking.



The examples of each intelligence are meant for illustrative purposes only and are not exclusive—one person can excel

in several categories. Note also that entire cultures might encourage the development of one or another intelligence;

for instance, the seafaring Puluwat of the Caroline Islands in the South Pacific cultivate spatial intelligence and excel at

navigation, and the Manus children of New Guinea learn the canoeing and swimming skills that elude the vast majori-

ty of seafaring Western children.

A Sampling of Intelligences

5. BODILY-KINESTHETICControlling and orches-trating body motions andhandling objects skillfully.

Dancers, athletes, actors:Marcel Marceau, MarthaGraham, Michael Jordan

1. LINGUISTIC A mastery and love oflanguage and words witha desire to explore them.

Poets, writers, linguists: T. S. Eliot, NoamChomsky, W. H. Auden

2. LOGICAL-MATHEMATICALConfronting and assessingobjects and abstractionsand discerning their rela-tions and underlyingprinciples.

Mathematicians, scien-tists, philosophers:Stanislaw Ulam, AlfredNorth Whitehead, HenriPoincaré, Albert Einstein,Marie Curie

3. MUSICALA competence not only incomposing and performingpieces with pitch, rhythmand timbre but also in lis-tening and discerning. Maybe related to other intelli-gences, such as linguistic,spatial or bodily-kinesthetic.

Composers, conductors,musicians, music critics:Ludwig van Beethoven,Leonard Bernstein,Midori, John Coltrane

4. SPATIALAn ability to perceive thevisual world accurately,transform and modify per-ceptions and re-createvisual experiences evenwithout physical stimuli.

Architects, artists,sculptors, mapmakers,navigators, chess players:Michelangelo, Frank LloydWright, Garry Kasparov,Louise Nevelson, HelenFrankenthaler

6. and 7. PERSONALINTELLIGENCESAccurately determiningmoods, feelings and othermental states in oneself(intrapersonal intelligence)and in others (interperson-al) and using the informa-tion as a guide for behavior.

Psychiatrists, politicians,religious leaders, anthro-pologists: Sigmund Freud,Mahatma Gandhi,Eleanor Roosevelt

8. NATURALISTRecognizing and catego-rizing natural objects.

Biologists, naturalists:Rachel Carson, JohnJames Audubon

9. EXISTENTIAL (possible intelligence):Capturing and ponderingthe fundamental questionsof existence. More evi-dence, however, is need-ed to determine whetherthis is an intelligence.

Spiritual leaders, philo-sophical thinkers: Jean-Paul Sartre, Søren A.Kierkegaard

Maya Angelou Paul Erdös Frida Kahlo

Alvin Ailey Margaret Mead Dalai LamaCharles Darwin

Joni Mitchell

EMIL

E W

AM

STEK

ER A

P Ph

oto

(An

gel

ou

);G

EOR

GE

CSI

CSE

RY,f

rom

th

e d

ocu

men

tary

film

N Is

a N

umbe

r:A

Por

trai

t of P

aul E

rdös

(Erd

ös)

;HEN

RY D

ILTZ

Cor

bis

(Mit

chel

l);A

RT

RESO

URC

E(S

elf-

Port

rait

with

a M

onke

y,19

40) (

Kah

lo);

NO

RMA

ND

MA

XON

Alv

in A

iley

Am

eric

an D

ance

The

ater

(Aile

y);A

PA/A

RCH

IVE

PHO

TOS

(Mea

d);

POPP

ERFO

TO/A

RCH

IVE

PHO

TOS

(Dar

win

);M

ICH

AEL

O’N

EIL

Out

line

(Dal

ai L

ama)


But it is important that intelligence notbe conflated with other virtues, such ascreativity, wisdom or morality.

I also contend that intelligenceshould not be so broadened that it cross-es the line from description to prescrip-tion. I endorse the notion of emotionalintelligence when it denotes the capaci-ty to compute information about one’sown or others’ emotional life. When theterm comes to encompass the kinds ofpersons we hope to develop, however,then we have crossed the line into avalue system—and that should not bepart of our conception of intelligence.Thus, when psychologist and New YorkTimes reporter Daniel Goleman empha-sizes in his recent best-seller, EmotionalIntelligence, the importance of empathyas part of emotional intelligence, I goalong with him. But he also urgesthat individuals care for oneanother. The possession of thecapacity to feel another’s suffer-ing is not the same as the decision tocome to her aid. Indeed, a sadistic indi-vidual might use her knowledge ofanother’s psyche to inflict pain.

My second wish is that society shiftaway from standardized, short-answerproxy instruments to real-life demonstra-tions or virtual simulations. During aparticular historical period, it was per-haps necessary to assess individuals byadministering items that were themselvesof little interest (for example, repeatingnumbers backward) but that werethought to correlate with skills or habitsof importance. Nowadays, however, giventhe advent of computers and virtual tech-nologies, it is possible to look directly atindividuals’ performances—to see howthey can argue, debate, look at data, cri-tique experiments, execute works of art,and so on. As much as possible, we

should train students directly in thesevalued activities, and we should assesshow they carry out valued performancesunder realistic conditions. The need forersatz instruments, whose relation toreal-world performance is often tenuousat best, should wane.

My third wish is that the multiple-intelligences idea be used for more effec-tive pedagogy and assessment. I have lit-tle sympathy with educational effortsthat seek simply to “train” the intelli-gences or to use them in trivial ways(such as singing the math times tablesor playing Bach in the background whileone is doing geometry). For me, the edu-cational power of multiple intelligencesis exhibited when these faculties aredrawn on to help students master conse-quential disciplinary materials.

I explain how such an approachmight work in my book, A Well-Disci-plined Mind, which will appear in thespring of 1999. I focus on three rich top-ics: the theory of evolution (as an exam-ple of scientific truth), the music ofMozart (as an example of artistic beau-ty), and the Holocaust (as an example ofimmorality in recent history). In eachcase, I show how the topic can be intro-duced to students through a variety ofentry points drawing on several intelli-gences, how the subject can be mademore familiar through the use of analo-gies and metaphors drawn from diversedomains, and how the core ideas of thetopic can be captured not merelythrough a single symbolic language butrather through a number of complemen-tary model languages or representations.

Pursuing this approach, the individ-

ual who understands evolutionary theory,for instance, can think of it in differentways: in terms of a historical narrative, alogical syllogism, a quantitative exami-nation of the size and dispersion of pop-ulations in different niches, a diagramof species delineation, a dramatic senseof the struggle among individuals (orgenes or populations), and so on. Theindividual who can think of evolutionin only one way—using only one modellanguage—actually has only a tenuouscommand of the principal concepts ofthe theory.

The issue of who owns intelligencehas been an important one in our soci-ety for some time—and it promises to bea crucial and controversial one for theforeseeable future. For too long, the restof society has been content to leave

intelligence in the hands of psychome-tricians. Often these test makers have anarrow, overly scholastic view of intel-lect. They rely on a set of instrumentsthat are destined to valorize certaincapacities while ignoring those that donot lend themselves to ready formula-tion and testing. And those with a polit-ical agenda often skirt close to the dan-gerous territory of eugenics.

MI theory represents at once aneffort to base the conception of intelli-gence on a much broader scientific basis,one that offers a set of tools to educatorsthat will allow more individuals to mas-ter substantive materials in an effectiveway. Applied appropriately, the theorycan also help each individual achievehis or her human potential at the work-place, in avocations and in the serviceof the wider world.


SA

HOWARD GARDNER is pure Harvard. Hestarted his career as a student there in 1961 andwent on to complete a Ph.D. and a postdoctoralfellowship at Harvard Medical School. NowGardner is a professor of education and co-director of Harvard’s Project Zero—an umbrellaproject that encompasses some two dozen dif-ferent studies related to cognition and creativi-ty. At one time a serious pianist, Gardner hasalways been involved in the arts. His interest inpsychology and the arts led him to do postdoc-toral work in neurology, studying how artistsand musicians are affected after a stroke. AtProject Zero, Gardner met his wife, Ellen Winner,

who was studying children’s understanding ofmetaphor. Gardner has four children, all ofwhom are somehow involved in the arts—oneplays piano, another plays bass, one is a photog-rapher and the oldest is an arts administrator.

Gardner has written several books on mul-tiple-intelligences theory and other topics, in-cluding Frames of Mind, The Mind’s New Scienceand The Unschooled Mind. Ironically, the popu-lar misinterpretation of his MI theory hasinspired Gardner to study ethics. “I’ve learnedthat when you develop ideas, you have to havea certain sense of responsibility for how they’reused,” he says.

About the Author


It is high time that the view of intelligence be widened to incorporatea range of human computational capacities.

CO

UR

TESY

OF

HO

WA

RD G

ARD

NER


No subject in psychology has pro-voked more intense public controversythan the study of human intelligence.From its beginning, research on howand why people differ in overall mentalability has fallen prey to political andsocial agendas that obscure or distorteven the most well-established scientificfindings. Journalists, too, often present aview of intelligence research that isexactly the opposite of what most intel-ligence experts believe. For these andother reasons, public understanding ofintelligence falls far short of public con-cern about it. The IQ experts discussingtheir work in the public arena can feelas though they have fallen down therabbit hole into Alice’s Wonderland.

The debate over intelligence andintelligence testing focuses on the ques-tion of whether it is useful or meaning-ful to evaluate people according to asingle major dimension of cognitivecompetence. Is there indeed a generalmental ability we commonly call “intel-ligence,” and is it important in the prac-tical affairs of life? The answer, based ondecades of intelligence research, is anunequivocal yes. No matter their formor content, tests of mental skills invari-ably point to the existence of a globalfactor that permeates all aspects of cog-nition. And this factor seems to haveconsiderable influence on a person’spractical quality of life. Intelligence asmeasured by IQ tests is the single mosteffective predictor known of individualperformance at school and on the job. Italso predicts many other aspects of well-being, including a person’s chances ofdivorcing, dropping out of high school,being unemployed or having illegitimatechildren.

By now the vast majority of intelli-gence researchers take these findings forgranted. Yet in the press and in publicdebate, the facts are typically dismissed,

downplayed or ignored. This misrepresen-tation reflects a clash between a deeplyfelt ideal and a stubborn reality. The ideal,implicit in many popular critiques ofintelligence research, is that all people areborn equally able and that social inequali-ty results only from the exercise of unjustprivilege. The reality is that MotherNature is no egalitarian. People are in factunequal in intellectual potential—andthey are born that way, just as they areborn with different potentials for height,physical attractiveness, artistic flair, ath-letic prowess and other traits. Althoughsubsequent experience shapes this poten-tial, no amount of social engineering canmake individuals with widely divergentmental aptitudes into intellectual equals.

Of course, there are many kinds oftalent, many kinds of mental ability andmany other aspects of personality andcharacter that influence a person’schances of happiness and success. Thefunctional importance of general mentalability in everyday life, however, meansthat without onerous restrictions onindividual liberty, differences in mentalcompetence are likely to result in socialinequality. This gulf between equalopportunity and equal outcomes is per-haps what pains Americans most aboutthe subject of intelligence. The publicintuitively knows what is at stake: whenasked to rank personal qualities in orderof desirability, people put intelligencesecond only to good health. But with amore realistic approach to the intellectualdifferences between people, society couldbetter accommodate these differencesand minimize the inequalities they create.

Extracting g

Early in the century-old study ofintelligence, researchers discovered thatall tests of mental ability ranked individ-uals in about the same way. Although

mental tests are often designed to mea-sure specific domains of cognition—ver-bal fluency, say, or mathematical skill,spatial visualization or memory—peoplewho do well on one kind of test tend todo well on the others, and people whodo poorly generally do so across theboard. This overlap, or intercorrelation,suggests that all such tests measuresome global element of intellectual abil-ity as well as specific cognitive skills. Inrecent decades, psychologists havedevoted much effort to isolating thatgeneral factor, which is abbreviated g,from the other aspects of cognitive abili-ty gauged in mental tests.

The statistical extraction of g is per-formed by a technique called factoranalysis. Introduced at the turn of thecentury by British psychologist CharlesSpearman, factor analysis determines theminimum number of underlying dimen-sions necessary to explain a pattern ofcorrelations among measurements. Ageneral factor suffusing all tests is not,as is sometimes argued, a necessary out-come of factor analysis. No general factorhas been found in the analysis of per-sonality tests, for example; instead themethod usually yields at least five dimen-sions (neuroticism, extraversion, consci-entiousness, agreeableness and opennessto ideas), each relating to different sub-sets of tests. But, as Spearman observed,a general factor does emerge from analy-sis of mental ability tests, and leadingpsychologists, such as Arthur R. Jensen ofthe University of California at Berkeleyand John B. Carroll of the University ofNorth Carolina at Chapel Hill, have con-firmed his findings in the decades since.Partly because of this research, most intel-ligence experts now use g as the workingdefinition of intelligence.

The general factor explains mostdifferences among individuals in perfor-mance on diverse mental tests. This is


The GeneralIntelligence Factor

Despite some popularassertions, a single factorfor intelligence, called g,can be measured with IQtests and does predict success in life

by Linda S. Gottfredson


true regardless of what specific ability atest is meant to assess, regardless of thetest’s manifest content (whether words,numbers or figures) and regardless of theway the test is administered (in writtenor oral form, to an individual or to agroup). Tests of specific mental abilitiesdo measure those abilities, but they allreflect g to varying degrees as well. Hence,the g factor can be extracted from scoreson any diverse battery of tests.

Conversely, because every mentaltest is “contaminated” by the effects ofspecific mental skills, no single test mea-sures only g. Even the scores from IQtests—which usually combine about adozen subtests of specific cognitiveskills—contain some “impurities” thatreflect those narrower skills. For mostpurposes, these impurities make no prac-tical difference, and g and IQ can be usedinterchangeably. But if they need to,

intelligence researchers can statisticallyseparate the g component of IQ. The abil-ity to isolate g has revolutionized researchon general intelligence, because it hasallowed investigators to show that thepredictive value of mental tests derivesalmost entirely from this global factorrather than from the more specific apti-tudes measured by intelligence tests.

In addition to quantifying individualdifferences, tests of mental abilities havealso offered insight into the meaning ofintelligence in everyday life. Some testsand test items are known to correlate bet-ter with g than others do. In these itemsthe “active ingredient” that demands theexercise of g seems to be complexity.More complex tasks require more mentalmanipulation, and this manipulation ofinformation—discerning similarities andinconsistencies, drawing inferences,grasping new concepts and so on—con-

stitutes intelligence in action. Indeed,intelligence can best be described as theability to deal with cognitive complexity.

This description coincides well withlay perceptions of intelligence. The g fac-tor is especially important in just thekind of behaviors that people usuallyassociate with “smarts”: reasoning, prob-lem solving, abstract thinking, quicklearning. And whereas g itself describesmental aptitude rather than accumulatedknowledge, a person’s store of knowledgetends to correspond with his or her glevel, probably because that accumulationrepresents a previous adeptness in learn-ing and in understanding new informa-tion. The g factor is also the one attribute

Exploring Intelligence 25The General Intelligence Factor

HIERARCHICAL MODEL of intelligenceis akin to a pyramid, with g at the apex;other aptitudes are arrayed at successivelylower levels according to their specificity.

BRI

DG

EMA

N A

RT L

IBR

ARY

Ad Parnassum, by Paul Klee


that best distinguishes among personsconsidered gifted, average or retarded.

Several decades of factor-analyticresearch on mental tests have confirmed ahierarchical model of mental abilities.The evidence, summarized most effec-tively in Carroll’s 1993 book, HumanCognitive Abilities, puts g at the apex inthis model, with more specific aptitudesarrayed at successively lower levels: theso-called group factors, such as verbalability, mathematical reasoning, spatialvisualization and memory, are just belowg, and below these are skills that aremore dependent on knowledge or experi-ence, such as the principles and practicesof a particular job or profession.

Some researchers use the term “mul-tiple intelligences” to label these sets ofnarrow capabilities and achievements.Psychologist Howard Gardner of HarvardUniversity, for example, has postulatedthat eight relatively autonomous “intelli-gences” are exhibited in differentdomains of achievement. He does notdispute the existence of g but treats it asa specific factor relevant chiefly to acade-mic achievement and to situations thatresemble those of school. Gardner doesnot believe that tests can fruitfully mea-sure his proposed intelligences; withouttests, no one can at present determinewhether the intelligences are indeed inde-

pendent of g (or each other). Further-more, it is not clear to what extentGardner’s intelligences tap personalitytraits or motor skills rather than mentalaptitudes.

Other forms of intelligence havebeen proposed; among them, emotionalintelligence and practical intelligence areperhaps the best known. They are proba-bly amalgams either of intellect and per-sonality or of intellect and informal expe-rience in specific job or life settings,respectively. Practical intelligence like“street smarts,” for example, seems toconsist of the localized knowledge andknow-how developed with untutoredexperience in particular everyday settingsand activities—the so-called school ofhard knocks. In contrast, general intelli-gence is not a form of achievement,whether local or renowned. Instead theg factor regulates the rate of learning: itgreatly affects the rate of return in knowl-edge to instruction and experience butcannot substitute for either.

The Biology of g

Some critics of intelligence researchmaintain that the notion of generalintelligence is illusory: that no suchglobal mental capacity exists and thatapparent “intelligence” is really just a

by-product of one’s opportunities tolearn skills and information valued in aparticular cultural context. True, theconcept of intelligence and the way inwhich individuals are ranked accordingto this criterion could be social artifacts.But the fact that g is not specific to anyparticular domain of knowledge or men-tal skill suggests that g is independent ofcultural content, including beliefs aboutwhat intelligence is. And tests of differ-ent social groups reveal the same con-tinuum of general intelligence. Thisobservation suggests either that culturesdo not construct g or that they constructthe same g. Both conclusions undercutthe social artifact theory of intelligence.

Moreover, research on the physiolo-gy and genetics of g has uncovered bio-logical correlates of this psychologicalphenomenon. In the past decade, stud-ies by teams of researchers in NorthAmerica and Europe have linked severalattributes of the brain to general intelli-gence. After taking into account genderand physical stature, brain size as deter-mined by magnetic resonance imagingis moderately correlated with IQ (about0.4 on a scale of 0 to 1). So is the speedof nerve conduction. The brains ofbright people also use less energy duringproblem solving than do those of theirless able peers. And various qualities of


Answers: 1. A; 2. D; 3. 10, 12; 4. 3, 6; 5. 3, 7; 6. 5, 25; 7. B; 8. D

SAMPLE IQ ITEMS resembling those on current tests requirethe test taker to fill in the empty spaces based on the pattern

in the images, numbers or words. Because they can vary incomplexity, such tasks are useful in assessing g level.

A B C D E

Number Series

2, 4, 6, 8, _, _

3,6,3,6, _,_

1,5,4,2,6,5, _, _

2,4,3,9,4,16, _,_

Analogies

brother: sister father:A. child B. mother C. cousin D. friend

joke: humor law:A. lawyer B. mercy C. courts D. justice

3.4.5.6.

7.

8.

A B C D E

Matrix Reasoning1. 2.

LIN

DA

S.G

OTT

FRED

SON



brain waves correlate strongly (about 0.5to 0.7) with IQ: the brain waves of indi-viduals with higher IQs, for example,respond more promptly and consistentlyto simple sensory stimuli such as audibleclicks. These observations have led someinvestigators to posit that differences ing result from differences in the speed andefficiency of neural processing. If thistheory is true, environmental conditionscould influence g by modifying brainphysiology in some manner.

Studies of so-called elementary cog-nitive tasks (ECTs), conducted by Jensenand others, are bridging the gap betweenthe psychological and the physiologicalaspects of g. These mental tasks have noobvious intellectual content and are sosimple that adults and most children cando them accurately in less than a second.In the most basic reaction-time tests, forexample, the subject must react when alight goes on by lifting her index fingeroff a home button and immediatelydepressing a response button. Two mea-surements are taken: the number of mil-liseconds between the illumination ofthe light and the subject’s release of thehome button, which is called decisiontime, and the number of millisecondsbetween the subject’s release of the homebutton and pressing of the response but-ton, which is called movement time.

In this task, movement time seemsindependent of intelligence, but the deci-sion times of higher-IQ subjects are slight-ly faster than those of people with lowerIQs. As the tasks are made more complex,correlations between average decisiontimes and IQ increase. These results fur-ther support the notion that intelligenceequips individuals to deal with com-plexity and that its influence is greaterin complex tasks than in simple ones.

The ECT-IQ correlations are compa-rable for all IQ levels, ages, genders andracial-ethnic groups tested. Moreover,studies by Philip A. Vernon of the Uni-versity of Western Ontario and othershave shown that the ECT-IQ overlapresults almost entirely from the commong factor in both measures. Reaction timesdo not reflect differences in motivationor strategy or the tendency of some indi-viduals to rush through tests and dailytasks—that penchant is a personalitytrait. They actually seem to measure thespeed with which the brain apprehends,integrates and evaluates information.Research on ECTs and brain physiologyhas not yet identified the biologicaldeterminants of this processing speed.These studies do suggest, however, that

g is as reliable and global a phenomenonat the neural level as it is at the level ofthe complex information processingrequired by IQ tests and everyday life.

The existence of biological corre-lates of intelligence does not necessarilymean that intelligence is dictated bygenes. Decades of genetics research haveshown, however, that people are bornwith different hereditary potentials forintelligence and that these geneticendowments are responsible for muchof the variation in mental ability amongindividuals. Last spring an internationalteam of scientists headed by RobertPlomin of the Institute of Psychiatry inLondon announced the discovery of thefirst gene linked to intelligence. Ofcourse, genes have their effects only ininteraction with environments, partlyby enhancing an individual’s exposureor sensitivity to formative experiences.Differences in general intelligence,whether measured as IQ or, more accu-rately, as g are both genetic and environ-mental in origin—just as are all otherpsychological traits and attitudes studiedso far, including personality, vocationalinterests and societal attitudes. This isold news among the experts. The expertshave, however, been startled by morerecent discoveries.

One is that the heritability of IQrises with age—that is to say, the extentto which genetics accounts for differ-ences in IQ among individuals increasesas people get older. Studies comparingidentical and fraternal twins, publishedin the past decade by a group led byThomas J. Bouchard, Jr., of the Universityof Minnesota and other scholars, showthat about 40 percent of IQ differencesamong preschoolers stems from geneticdifferences but that heritability rises to60 percent by adolescence and to 80percent by late adulthood. With age, dif-ferences among individuals in theirdeveloped intelligence come to mirrormore closely their genetic differences. Itappears that the effects of environmenton intelligence fade rather than growwith time. In hindsight, perhaps thisshould have come as no surprise. Youngchildren have the circumstances of theirlives imposed on them by parents,schools and other agents of society, butas people get older they become moreindependent and tend to seek out thelife niches that are most congenial totheir genetic proclivities.

A second big surprise for intelli-gence experts was the discovery thatenvironments shared by siblings have

little to do with IQ. Many people stillmistakenly believe that social, psycho-logical and economic differences amongfamilies create lasting and marked differ-ences in IQ. Behavioral geneticists referto such environmental effects as“shared” because they are common tosiblings who grow up together. Researchhas shown that although shared envi-ronments do have a modest influenceon IQ in childhood, their effects dissi-pate by adolescence. The IQs of adoptedchildren, for example, lose all resem-blance to those of their adoptive familymembers and become more like the IQsof the biological parents they have neverknown. Such findings suggest that sib-lings either do not share influentialaspects of the rearing environment or donot experience them in the same way.Much behavioral genetics research cur-rently focuses on the still mysteriousprocesses by which environments makemembers of a household less alike.

g on the Job

Although the evidence of geneticand physiological correlates of g arguespowerfully for the existence of globalintelligence, it has not quelled the crit-ics of intelligence testing. These skepticsargue that even if such a global entityexists, it has no intrinsic functionalvalue and becomes important only tothe extent that people treat it as such:for example, by using IQ scores to sort,label and assign students and employ-ees. Such concerns over the proper useof mental tests have prompted a greatdeal of research in recent decades. Thisresearch shows that although IQ testscan indeed be misused, they measure acapability that does in fact affect manykinds of performance and many life out-comes, independent of the tests’ inter-pretations or applications. Moreover, theresearch shows that intelligence testsmeasure the capability equally well forall native-born English-speaking groupsin the U.S.

If we consider that intelligencemanifests itself in everyday life as theability to deal with complexity, then itis easy to see why it has great functionalor practical importance. Children, forexample, are regularly exposed to com-plex tasks once they begin school.Schooling requires above all that stu-dents learn, solve problems and thinkabstractly. That IQ is quite a good pre-dictor of differences in educationalachievement is therefore not surprising.

The General Intelligence Factor


When scores on both IQ and standard-ized achievement tests in different sub-jects are averaged over several years, thetwo averages correlate as highly as dif-ferent IQ tests from the same individualdo. High-ability students also mastermaterial at many times the rate of theirlow-ability peers. Many investigationshave helped quantify this discrepancy.For example, a 1969 study done for theU.S. Army by the Human ResourcesResearch Office found that enlistees inthe bottom fifth of the ability distribu-tion required two to six times as manyteaching trials and prompts as did theirhigher-ability peers to attain minimalproficiency in rifle assembly, monitoringsignals, combat plotting and other basic

military tasks. Similarly, in school set-tings the ratio of learning rates between“fast” and “slow” students is typicallyfive to one.

The scholarly content of many IQtests and their strong correlations witheducational success can give the impres-sion that g is only a narrow academicability. But general mental ability alsopredicts job performance, and in morecomplex jobs it does so better than anyother single personal trait, includingeducation and experience. The army’sProject A, a seven-year study conductedin the 1980s to improve the recruitmentand training process, found that generalmental ability correlated strongly withboth technical proficiency and soldier-

ing in the nine specialties studied,among them infantry, military policeand medical specialist. Research in thecivilian sector has revealed the same pat-tern. Furthermore, although the additionof personality traits such as conscien-tiousness can help hone the predictionof job performance, the inclusion ofspecific mental aptitudes such as verbalfluency or mathematical skill rarely does.The predictive value of mental tests inthe work arena stems almost entirelyfrom their measurement of g, and thatvalue rises with the complexity andprestige level of the job.

Half a century of military and civil-ian research has converged to draw aportrait of occupational opportunityalong the IQ continuum. Individuals inthe top 5 percent of the adult IQ distrib-ution (above IQ 125) can essentiallytrain themselves, and few occupationsare beyond their reach mentally. Personsof average IQ (between 90 and 110) arenot competitive for most professionaland executive-level work but are easilytrained for the bulk of jobs in theAmerican economy. In contrast, adultsin the bottom 5 percent of the IQ distri-bution (below 75) are very difficult totrain and are not competitive for anyoccupation on the basis of ability.Serious problems in training low-IQ mil-itary recruits during World War II ledCongress to ban enlistment from thelowest 10 percent (below 80) of the pop-ulation, and no civilian occupation inmodern economies routinely recruits itsworkers from that range. Current mili-tary enlistment standards exclude anyindividual whose IQ is below about 85.

The importance of g in job perfor-mance, as in schooling, is related tocomplexity. Occupations differ consider-ably in the complexity of their demands,and as that complexity rises, higher glevels become a bigger asset and lower glevels a bigger handicap. Similarly, every-day tasks and environments also differsignificantly in their cognitive complex-ity. The degree to which a person’s g levelwill come to bear on daily life dependson how much novelty and ambiguitythat person’s everyday tasks and sur-roundings present and how much con-tinual learning, judgment and decision


70 80 90 100 110 120 130

HighRisk

UphillBattle

KeepingUp

OutAhead

Yoursto Lose

LifeChances

TrainingStyle

Slow, simple,supervised

CareerPotential

Out of laborforce more than 1 monthout of year (men)

Unemployedmore than 1 month out of year (men)

Divorced in5 years

Had illegitimatechildren (women)

Lives inpoverty

Everincarcerated(men)

Chronic welfarerecipient(mothers)

High schooldropout

IQ

Mastery learning,hands-on

Collegeformat

Very explicit,hands-on

Written materials,plus experience

Gathers, infersown information

Assembler, food service,nurse’s aide

Clerk, teller,police officer,

machinist, sales

Manager,teacher,

accountant

Attorney,chemist,executive

22 19 15 14 10

12 10 7 7 2

21 22 23 15 9

32 17 8 4 2

30 16 6 3 2

7 7 3 1 0

31 17 8 2 0

55 35 6 0.4 0

Population Percentages

Total population distribution

5 20 50 20 5

Adapted from Intelligence, Vol. 24, No. 1; January/February 1997

CORRELATION OF IQ SCORES withoccupational achievement suggests thatg reflects an ability to deal with cogni-tive complexity. Scores also correlate withsome social outcomes (the percentagesapply to young white adults in the U.S.).

JOH

NM

ENG

EL P

onzi

& W

eill


making they require. As gamblers,employers and bankers know, even mar-ginal differences in rates of return willyield big gains—or losses—over time.Hence, even small differences in g amongpeople can exert large, cumulative influ-ences across social and economic life.

In my own work, I have tried to syn-thesize the many lines of research thatdocument the influence of IQ on life out-comes. As the illustration on the oppositepage shows, the odds of various kinds ofachievement and social pathology changesystematically across the IQ continuum,from borderline mentally retarded(below 70) to intellectually gifted (above130). Even in comparisons of those ofsomewhat below average (between 76and 90) and somewhat above average(between 111 and 125) IQs, the odds foroutcomes having social consequence arestacked against the less able. Young mensomewhat below average in generalmental ability, for example, are morelikely to be unemployed than mensomewhat above average. The lower-IQwoman is four times more likely to bearillegitimate children than the higher-IQwoman; among mothers, she is eighttimes more likely to become a chronicwelfare recipient. People somewhatbelow average are 88 times more likelyto drop out of high school, seven timesmore likely to be jailed and five timesmore likely as adults to live in povertythan people of somewhat above-averageIQ. Below-average individuals are 50percent more likely to be divorced thanthose in the above-average category.

These odds diverge even moresharply for people with bigger gaps in IQ,and the mechanisms by which IQ createsthis divergence are not yet clearly under-stood. But no other single trait or circum-

stance yet studied is so deeply implicatedin the nexus of bad social outcomes—poverty, welfare, illegitimacy and educa-tional failure—that entraps many low-IQindividuals and families. Even the effectsof family background pale in comparisonwith the influence of IQ. As shown mostrecently by Charles Murray of the Amer-ican Enterprise Institute in Washington,D.C., the divergence in many outcomesassociated with IQ level is almost as wideamong siblings from the same householdas it is for strangers of comparable IQlevels. And siblings differ a lot in IQ—onaverage, by 12 points, compared with 17for random strangers.

An IQ of 75 is perhaps the mostimportant threshold in modern life. Atthat level, a person’s chances of master-ing the elementary school curriculumare only 50–50, and he or she will havea hard time functioning independentlywithout considerable social support.Individuals and families who are onlysomewhat below average in IQ face risksof social pathology that, while lower, arestill significant enough to jeopardizetheir well-being. High-IQ individualsmay lack the resolve, character or goodfortune to capitalize on their intellectualcapabilities, but socioeconomic successin the postindustrial information age istheirs to lose.

What Is versus What Could Be

The foregoing findings on g’s effectshave been drawn from studies conductedunder a limited range of circumstances—namely, the social, economic and politi-cal conditions prevailing now and inrecent decades in developed countriesthat allow considerable personal freedom.It is not clear whether these findings

apply to populations around the world,to the extremely advantaged and disad-vantaged in the developing world or, forthat matter, to people living underrestrictive political regimes. No oneknows what research under different cir-cumstances, in different eras or with dif-ferent populations might reveal.

But we do know that, wherever free-dom and technology advance, life is anuphill battle for people who are belowaverage in proficiency at learning, solv-ing problems and mastering complexity.We also know that the trajectories ofmental development are not easilydeflected. Individual IQ levels tend toremain unchanged from adolescenceonward, and despite strenuous effortsover the past half a century, attempts toraise g permanently through adoptionor educational means have failed. Ifthere is a reliable, ethical way to raise orequalize levels of g, no one has found it.

Some investigators have suggestedthat biological interventions, such asdietary supplements of vitamins, may bemore effective than educational ones inraising g levels. This approach is based inpart on the assumption that improvednutrition has caused the puzzling rise inaverage levels of both IQ and height inthe developed world during this century.Scientists are still hotly debating whetherthe gains in IQ actually reflect a rise in gor are caused instead by changes in lesscritical, specific mental skills. Whateverthe truth may be, the differences in men-tal ability among individuals remain,and the conflict between equal opportu-nity and equal outcome persists. Onlyby accepting these hard truths aboutintelligence will society find humanesolutions to the problems posed by thevariations in general mental ability.

Exploring Intelligence 29The General Intelligence Factor

LINDA S. GOTTFREDSON is professorof educational studies at the University ofDelaware, where she has been since 1986,and co-directs the Delaware–Johns HopkinsProject for the Study of Intelligence andSociety. She trained as a sociologist, andher earliest work focused on career devel-opment. “I wasn’t interested in intelli-gence per se,” Gottfredson says. “But itsuffused everything I was studying in myattempts to understand who was gettingahead.” This “discovery of the obvious,”as she puts it, became the focus of herresearch. In the mid-1980s, while at JohnsHopkins University, she published severalinfluential articles describing how intelli-

gence shapes vocational choice and self-perception. Gottfredson also organizedthe 1994 treatise “Mainstream Science onIntelligence,” an editorial with more than50 signatories that first appeared in theWall Street Journal in response to the con-troversy surrounding publication of TheBell Curve. Gottfredson is the mother ofidentical twins—a “mere coincidence,”she says, “that’s always made me thinkmore about the nature and nurture ofintelligence.” The girls, now 16, followGottfredson’s Peace Corps experience ofthe 1970s by joining her each summer forvolunteer construction work in the vil-lages of Nicaragua.

SA

CO

UR

TESY

OF

LIN

DA

S.G

OTT

FRED

SON

About the Author


That Richard J. Herrnstein and Charles Murray’s 1994 bookThe Bell Curve should become a commercial blockbuster wasperhaps unsurprising, given its user-friendly presentation andits incendiary subject matter. The 800-page volume argued thatAmerican society is increasingly dividing into a wealthy “cog-nitive elite” and a dull, growing underclass. Because the authorsbelieve that cognitive ability is largely inherited and that itstrongly predicts important social outcomes such as avoidanceof poverty and criminality, they foresaw the emergence of a“custodial state” in which the elite keep the underclass under-foot. African-Americans, in Herrnstein and Murray’s vision,seemed doomed to remain disproportionately in the under-class, because that group is cognitively disadvantaged for reasonsthat are “very likely” to be in part genetic.

Among the authors’ recommendations for adapting to theseinevitable trends were dismantling affirmative action and thewelfare safety net and shifting funds from educational programsfor disadvantaged children to programs for the gifted—changesthat some might argue would speed stratification. The bookhas so far sold more than 500,000 copies.

Whether The Bell Curve will have an influence on socialscience or real-world policy comparable to its popularity seemsdoubtful. Murray wrote in an afterword to the paperback edition(Herrnstein died before the book was published) that the rela-tionships between IQ and social behaviors presented in The BellCurve are “so powerful they will revolutionize sociology.” Butthoughtful critics who have now had a chance to reanalyzecrucial data say new findings weaken or contradict most of TheBell Curve’s more abrasive conclusions.

Observers of the education scene see little evidence, more-over, that the book has had any effect on policy decisions, al-though it may in some minds have legitimized the status quobetween the haves and have-nots. The U.S. Congress, whichmight have been expected to give the book a hearing, has paidlittle attention to education policy in recent years. The BellCurve’s discussion of racial genetics probably ensured that pol-iticians would avoid allying themselves with its message, sayseducational evaluation expert Ernest R. House of the Universityof Colorado. What is left, as the dust settles, are some innocu-ous facts about intelligence that, while perhaps news to some,are hardly revolutionary, in the judgment of Christopher Jencksof Harvard University, an editor (with Meredith Phillips) of anew book, The Black-White Test Score Gap.

Starting with what is relatively uncontroversial, most schol-ars accept that the quantity measured by IQ tests, known asgeneral intelligence, is a meaningful construct that can predictmental performance—even though there are substantial differ-ences of opinion over its precise theoretical status, and nobodyknows its material basis. Most agree, too, that in today’s societysome nontrivial proportion of the variation in IQ scoresbetween individuals can be ascribed to different inherited genes.That proportion is called heritability.

Researchers differ, however, in their estimates of IQ’s heri-tability and the implications of that effect. Herrnstein andMurray adopted a “middling value” of 60 percent, while main-taining that it might be as high as 80 percent. Others disagree.In a recent book that reanalyzes The Bell Curve’s major argu-ments, Intelligence, Genes and Success, statisticians and geneticistsMichael Daniels, Bernie Devlin and Kathryn Roeder argue thatthe figure is actually about 48 percent.

The difference arises because estimates of the heritabilityof IQ turn largely on the similarity in IQ of twins who are rearedapart. Most twin studies ignore the possibility that sharing auterus for nine months may account for some later similaritiesin IQ. In reality, that effect appears to be substantial, and a sta-tistical analysis that compensates for it (by comparing monozy-gotic and fraternal twins as well as other siblings) produces thelower estimate of the heritability of IQ.

But that is not all that Daniels and his co-authors find faultwith in The Bell Curve’s use of heritability. The book erred inusing a “broad” definition of heritability as a basis for specula-tion about genetically based cognitive stratification, they say.They argue that for this purpose a “narrow” definition of heri-tability is the mathematically correct one and estimate its valueat only 34 percent, a figure that makes the emergence of cogni-tive castes “almost impossible.” (The narrow definition, unlikethe broad one, excludes interactions among genes.)

Raising IQ with the Environment

More fundamentally, and contrary to The Bell Curve,scholars point out that even if individual heritability of IQ werevery large, it might nonetheless be susceptible to environmentalimprovements. “A heritability estimate does not in any way‘constrain’ the effects of a changed environment,” notes psy-chologist Douglas Wahlsten of the University of Alberta.

Wahlsten gives the example of the inherited disease phe-nylketonuria, which can cause brain damage. It is successfullytreated by avoiding the amino acid phenylalanine in the diet.Likewise, Wahlsten cites studies in France showing that infantsadopted from a family having low socioeconomic status intoone of high socioeconomic status had childhood IQ scoresthat were 12 to 16 points higher than others who remained inpoverty with their biological mothers. In contrast to The Bell


For Whom Did theBellCurveToll?The most controversial

social science book indecades shook up

readers. Researchersare less easily

impressed

by Tim Beardsley, staff writer

BO

OK

JAC

KET

OF

THE

BELL

CU

RVE

© 1

994

SIM

ON

& S

CH

UST

ER,I

NC

.,RE

PRO

DU

CED

WIT

H P

ERM

ISSI

ON

OF

THE

FREE

PRE

SS,A

DIV

ISIO

N O

F SI

MO

N &

SC

HU

STER


Exploring Intelligence 31For Whom Did the Bell Curve Toll?

Curve’s judgment that “changing cognitive ability throughenvironmental intervention has proved to be extraordinarilydifficult,” Wahlsten concludes that even modest environmen-tal improvements can have substantial effects on ability testscores and that lasting gains in a child’s environment canexert “quite a large” effect.

Some such effects have been documented by Craig T.Ramey of the University of Alabama at Birmingham. Ramey hasdemonstrated how a preschool educational intervention for thefirst five years of life significantly boosted IQ scores of at-riskchildren throughout school years and into adolescence, with anaverage increase of five points still apparent at age 15. The mostdisadvantaged children showed gains twice as large. Academicachievement (as distinct from IQ) scores of at-risk kids showeven clearer benefits of preschool that persist well into theteenage years. But The Bell Curve shrugs off these benefits.

The book’s pessimistic assessment of the prospects for edu-cational interventions is its fatal flaw, according to psychologistRichard E. Nisbett of the University of Michigan. The authors“are probably right that there are limits to how much you canchange IQ, but they may be far wider than implied in thebook,” Nisbett says. Christopher Winship of Harvard andSanders Korenman of the City University of New York find thatconventional education itself boosts IQ by perhaps two to fourpoints a year, an estimate they say argues in favor of the publicinvestment. The Bell Curve argued that education had little orno effect on IQ. Perhaps the best conclusion is that the factorsthat feed into a measured IQ score are not fully understood.

A major problem that psychologists note for The Bell Curve’sargument is that unstandardized intelligence scores have beenincreasing rapidly for several decades in industrial countries, aphenomenon known as the Flynn effect. Because some environ-mental influence must have caused the effect—it is too rapidfor genetic changes to account for—environmental improve-ments that boost mental abilities must be possible.

Not So Black-and-White

One of the most painful issues that Herrnstein and Murrayexplored was the lower measured average scores of African-Americans on IQ tests, as compared with Caucasians. The BellCurve’s half-acceptance of a genetic influence was surely onereason for its notoriety (the question is entirely different fromthat of heritability of IQ between individuals). Yet accordingto Nisbett, the evidence—which includes adoption studies andother types—“offers almost no support for genetic explanationsof the IQ differences between blacks and whites.”

The test-score gap could be eliminated through practicableimprovements in the educational systems, contend Jencks andPhillips in The Black-White Test Score Gap. They cite three princi-pal arguments.

First, when black or mixed-race children are raised in whiterather than black homes, their preadolescent test scores risedramatically. That shows that improvements are feasible. Thescores tend to fall again during adolescence, but the reasons maynot be irremediable. Second, the Flynn effect argues againstgenetically based IQ differences between races. Third, black-white differences in academic achievement have already nar-rowed by almost half during this century, now being closer to10 than to the usually cited 15 points.

The Bell Curve elaborates on its racial claims by suggestingthat black-white differences in earnings are no greater than ex-pected because of IQ differences, a key plank in the book’s

attack on affirmative action. But an analysis by Alexander L.Cavallo of the University of Chicago and others, which looksat the sexes separately, contests this conclusion. After allowingfor ability, it seems, black males earn substantially less thanwhite males (in females the gap is in the opposite direction).Much of the differential, Cavallo asserts, is “contributed byfactors that may be influenced by racial discrimination,” aconclusion that undercuts The Bell Curve’s argument.

Researchers of a different political stripe from Herrnsteinand Murray have also found important qualifications to severalmore of The Bell Curve’s slew of conclusions about the predictiveeffect of IQ on life chances. Economist John Cawley of the Uni-versity of Chicago and his co-authors of a chapter in Intelligence,

Genes and Success analyze the same data studied by Herrnsteinand Murray but conclude that they “dramatically overstate”how much of the variation in wages between individuals can beexplained by intelligence. Sociologist Lucinda A. Manolakes ofthe State University of New York at Stony Brook likewise judgesIQ to “be only one of many variables” that affect criminality.

The list goes on. Winship and Korenman confirm an in-fluence of IQ on adult social outcomes such as earnings andavoidance of poverty. But they also find that family backgroundturns out to have effects comparable with those of IQ, whenproper allowance is made for the confounding effect of educa-tion. IQ is “not the dominant determinant.”

Stephen Fienberg of Carnegie Mellon University, one of theeditors of Intelligence, Genes and Success, notes that “everyoneknows that smart people do better in life.” But academics saythat “IQ matters in a much more nuanced way” than Herrnsteinand Murray maintain, according to Fienberg. The nuances makeit harder to issue policy recommendations.

The publicity firestorm over Herrnstein and Murray’s claimsseems to have died down in the past year. Jencks and Nisbettboth allow that The Bell Curve focused attention on the impor-tance of thinking about intelligence in debates about publicpolicy. Many readers, though, are likely to have come to cruderconclusions, such as that science has shown attempts to helpat-risk youth to be a waste of time. Nothing could be furtherfrom the truth. SA

19561956

MEAN SCORE

19771977

NU

MB

ER O

F ST

UD

ENTS

NU

MB

ER O

F ST

UD

ENTS

LAU

RIE

GR

AC

E

FLYNN EFFECT: means of raw IQ scores have increased world-wide, including those of third-graders in Edmonton, Alberta.



One evening a few years ago, while I was attending a con-cert, a young boy in the audience caught my attention. As theorchestra played a Mozart concerto, this nine-year-old childsat with a thick, well-thumbed orchestral score opened on hislap. As he read, he hummed the music out loud, in perfecttune. During intermission, I cornered the boy’s father. Yes, hetold me, Stephen was really reading the music, not just lookingat it. And reading musical scores was one of his preferredactivities, vying only with reading college-level computer pro-gramming manuals. At an age when most children concentrateon fourth-grade arithmetic and the nuances of playground eti-quette, Stephen had already earned a prize in music theorythat is coveted by adults.

Gifted children like Stephen are fascinating but alsointimidating. They have been feared as “possessed,” they havebeen derided as oddballs, they have been ridiculed as nerds.The parents of such young people are often criticized forpushing their children rather than allowing them a normal,well-balanced childhood. These children are so different fromothers that schools usually do not know how to educatethem. Meanwhile society expects gifted children to becomecreative intellectuals and artists as adults and views them asfailures if they do not.

Psychologists have always been interested in those whodeviate from the norm, but just as they know more about psy-

chopathology than about leadership and courage, researchersalso know far more about retardation than about giftedness.Yet an understanding of the most talented minds will provideboth the key to educating gifted children and a preciousglimpse of how the human brain works.

The Nature of Giftedness

Everyone knows children who are smart, hard-workingachievers—youngsters in the top 10 to 15 percent of all stu-dents. But only the top 2 to 5 percent of children are gifted.Gifted children (or child prodigies, who are just extreme ver-sions of gifted children) differ from bright children in at leastthree ways:

• Gifted children are precocious. They master subjects earlierand learn more quickly than average children do.

• Gifted children march to their own drummer. They makediscoveries on their own and can often intuit the solution to aproblem without going through a series of logical, linear steps.

• Gifted children are driven by “a rage to master.” They havea powerful interest in the area, or domain, in which they havehigh ability—mathematics, say, or art—and they can readilyfocus so intently on work in this domain that they lose senseof the outside world.

These are children who seem to teach themselves to read

UncommonTalents:

Gifted Children,Prodigies

and SavantsPossessing abilities wellbeyond their years, gifted children inspire admiration,but they also suffer ridicule,neglect and misunderstanding

by Ellen Winner

WA

NG

SH

IQIA

NG


Exploring Intelligence 33Uncommon Talents: Gifted Children, Prodigies and Savants

as toddlers, who breeze through college mathematics in mid-dle school or who draw more skillfully as second-graders thanmost adults do. Their fortunate combination of obsessiveinterest and an ability to learn easily can lead to high achieve-ment in their chosen domain. But gifted children are moresusceptible to interfering social and emotional factors thanonce was thought.

The first comprehensive study of the gifted, carried outover a period of more than 70 years, was initiated at StanfordUniversity in the early part of this century by Lewis M. Terman,a psychologist with a rather rosy opinion of gifted children. Hisstudy tracked more than 1,500 high-IQ children over the courseof their lives. To qualify for the study, the “Termites” were firstnominated by their teachers and then had to score 135 orhigher on the Stanford-Binet IQ test (the average score is 100).These children were precocious: they typically spoke early,walked early and read before they entered school. Their parentsdescribed them as being insatiably curious and as havingsuperb memories.

Terman described his subjects glowingly, not only assuperior in intelligence to other children but also as superiorin health, social adjustment and moral attitude. This conclu-sion easily gave rise to the myth that gifted children are happyand well adjusted by nature, requiring little in the way of spe-

cial attention—a myth that still guides the way these childrenare educated today.

In retrospect, Terman’s study was probably flawed. Nochild entered the study unless nominated by a teacher as oneof the best and the brightest; teachers probably overlookedthose gifted children who were misfits, loners or problematicto teach. And the shining evaluations of social adjustment andpersonality in the gifted were performed by the same admir-ing teachers who had singled out the study subjects. Finally,almost a third of the sample came from professional, middle-class families. Thus, Terman confounded IQ with social class.

The myth of the well-adjusted, easy-to-teach gifted childpersists despite more recent evidence to the contrary. MihalyCsikszentmihalyi of the University of Chicago has shown thatchildren with exceptionally high abilities in any area—notjust in academics but in the visual arts, music, even athletics—are out of step with their peers socially. These children tend tobe highly driven, independent in their thinking and introvert-ed. They spend more than the usual amount of time alone,and although they derive energy and pleasure from their soli-

GIFTED CHILD ARTIST WANG YANI from China painted ata nearly adult skill level at the age of five, when she completedthis painting in 1980. As a child, she produced a prodigiousnumber of works, at one point finishing 4,000 paintings withinthe space of three years.

Pull Harder, Wang Yani



tary mental lives, they also report feeling lonely. The moreextreme the level of gift, the more isolated these children feel.

Contemporary researchers have estimated that about 20to 25 percent of profoundly gifted children have social andemotional problems, which is about twice the normal rate; incontrast, moderately gifted children do not exhibit a higherthan average rate. By middle childhood, gifted children oftentry to hide their abilities in the hopes of becoming more pop-ular. One group particularly at risk for such underachievementis academically gifted girls, who report more depression, lowerself-esteem and more psychosomatic symptoms than academi-cally gifted boys do.

The combination of precocious knowledge, social isolationand sheer boredom in many gifted children is a tough challengefor teachers who must educate them alongside their peers.Worse, certain gifted children can leap years ahead of theirpeers in one area yet fall behind in another. These children,the unevenly gifted, sometimes seem hopelessly out of sync.

The Unevenly Gifted

Terman was a proponent of the view that gifted childrenare globally gifted—evenly talented in all academic areas.Indeed, some special children have exceptional verbal skills aswell as strong spatial, numerical and logical skills that enablethem to excel in mathematics. The occasional child who com-

pletes college as an early teen—or even as apreteen—is likely to be globally gifted. Suchchildren are easy to spot: they are all-aroundhigh achievers. But many children exhibitgifts in one area of study and are unremark-able or even learning disabled in others.These may be creative children who aredifficult in school and who are not imme-diately recognized as gifted.

Unevenness in gifted children is quitecommon. A recent survey of more than1,000 highly academically gifted adolescentsrevealed that more than 95 percent show astrong disparity between mathematical andverbal interests. Extraordinarily strong math-ematical and spatial abilities often accom-pany average or even deficient verbal abili-ties. Julian Stanley of Johns HopkinsUniversity has found that many gifted chil-dren selected for special summer programsin advanced math have enormous discrep-ancies between their math and verbal skills.One such eight-year-old scored 760 out of aperfect score of 800 on the math part of theScholastic Assessment Test (SAT) but only290 out of 800 on the verbal part.

In a retrospective analysis of 20 world-class mathematicians, psychologist BenjaminS. Bloom, then at the University of Chicago,

reported that none of his subjects had learned to read beforeattending school (yet most academically gifted children do readbefore school) and that six had had trouble learning to read.And a retrospective study of inventors (who presumably exhibithigh mechanical and spatial aptitude) showed that as childrenthese individuals struggled with reading and writing.

Indeed, many childrenwho struggle with languagemay have strong spatial skills.Thomas Sowell of StanfordUniversity, an economist bytraining, conducted a study oflate-talking children after heraised a son who did notbegin to speak until almostage four. These children tendedto have high spatial abilities—they excelled at puzzles, forinstance—and most had rela-tives working in professionsthat require strong spatial

DRAWING SAVANT NADIA was a “low-functioning” autistic child, whose mentalage was three years and three months whenshe was six. But this sketch by Nadia, doneat age five and a half in 1973, exhibits acommand of line, foreshortening and motionreminiscent of adult Renaissance masters.

TYPICAL DRAWING by a five-year-old of average ability lacksdetail and is highly schematic. C

OU

RTE

SY O

F EL

LEN

WIN

NER

© 1

977

HA

RCO

UR

T B

RA

CE

AN

D C

OM

PAN

Y


Uncommon Talents: Gifted Children, Prodigies and Savants

skills. Perhaps the most striking finding was that 60 percent ofthese children had engineers as first- or second-degree relatives.

The association between verbal deficits and spatial giftsseems particularly strong among visual artists. Beth Casey ofBoston College and I have found that college art students makesignificantly more spelling errors than college students major-ing either in math or in verbal areas such as English or history.On average, the art students not only misspelled more thanhalf of a 20-word list but also made the kind of errors associatedwith poor reading skills—nonphonetic spellings such as “physi-cain” for “physician” (instead of the phonetic “fisician”).

The many children who possess a gift in one area and areweak or learning disabled in others present a conundrum. Ifschools educate them as globally gifted, these students willcontinually encounter frustration in their weak areas; if theyare held back because of their deficiencies, they will be boredand unhappy in their strong fields. Worst, the gifts that thesechildren do possess may go unnoticed entirely when frustrated,unevenly gifted children wind up as misfits or troublemakers.

Savants: Uneven in the Extreme

The most extreme cases of spatial or mathematical giftscoexisting with verbal deficits are found in savants. Savantsare retarded (with IQs between 40 and 70) and are eitherautistic or show autistic symptoms. “Ordinary” savants usual-ly possess one skill at a normal level, in contrast to their oth-

erwise severely limited abilities. But the rarer savants—fewerthan 100 are known—display one or more skills equal toprodigy level.

Savants typically excel in visual art, music or lightning-fast calculation. In their domain of expertise, they resemblechild prodigies, exhibiting precocious skills, independentlearning and a rage to master. For instance, the drawing savantnamed Nadia sketched more realistically at ages three and fourthan any known child prodigy of the same age. In addition,savants will often surpass gifted children in the accuracy oftheir memories.

Savants are like extreme versions of unevenly gifted chil-dren. Just as gifted children often have mathematical or artis-tic genius and language-based learning disabilities, savantstend to exhibit a highly developed visual-spatial ability along-side severe deficits in language. One of the most promisingbiological explanations for this syndrome posits atypical brainorganization, with deficits in the left hemisphere of the brain(which usually controls language) offset by strengths in theright hemisphere (which controls spatial and visual skills).

According to Darold A. Treffert, a psychiatrist now in pri-vate practice in Fond du Lac, Wis., the fact that many savantswere premature babies fits well with this notion of left-sidebrain damage and resultant right-side compensation. Late inpregnancy, the fetal brain undergoes a process called pruning,in which a large number of excess neurons die off [see “TheDeveloping Brain,” by Carla J. Shatz; SCIENTIFIC AMERICAN,


THOMAS ALVA EDISON exemplifies the unevenly giftedindividual. Edison was a prolific inventor, obtaining1,093 patents for innovations ranging from the phono-graph to the incandescent light. As a child, he wasobsessed with science and spent much time tinkering ina chemistry laboratory in his parents’ cellar. Edison hadsome difficulties learning, though, especially in the ver-bal areas; he may have had symptoms of dyslexia. Thecoexistence of strong spatial-logical skills with a weak-ness in language is common in the unevenly gifted.

WOLFGANG AMADEUS MOZART is among the best-known child prodigies. He began picking out tunes onthe piano at three years of age; by four he could tell if aviolin was a quarter tone out of tune, and by eight hecould play without hesitation a complex piece he hadnever seen before. Mozart began composing at the ageof five, when he wrote two minuets for the harpsichord.Even as a young child, he could play pieces perfectlyfrom memory, having heard them only once, andimprovise on a theme without ever repeating himself.

ET A

RCH

IVE,

LON

DO

N/S

UPE

RSTO

CK

THE

CO

LLEC

TIO

NS

OF

THE

EDIS

ON

NAT

ION

AL

HIS

TORI

C S

ITE,

NAT

ION

AL

PARK

SER

VIC

E


September 1992]. But the brains of babies born prematurelymay not have been pruned yet; if such brains experience trau-ma to the left hemisphere near the time of birth, numerousuncommitted neurons elsewhere in the brain might remain tocompensate for the loss, perhaps leading to a strong right-hemisphere ability.

Such trauma to a premature infant’s brain could arise manyways—from conditions during pregnancy, from lack of oxygenduring birth, from the administration of too much oxygenafterward. An excess of oxygen given to premature babies cancause blindness in addition to brain damage; many musicalsavants exhibit the triad of premature birth, blindness andstrong right-hemisphere skill.

Gifted children most likely possess atypical brain organi-zation to some extent as well. When average students are test-ed to see which part of their brain controls their verbal skills,the answer is generally the left hemisphere only. But whenmathematically talented children are tested the same way,both the left and right hemispheres are implicated in control-

ling language—the right side of their brains participates intasks ordinarily reserved for the left. These children also tendnot to be strongly right-handed, an indication that their lefthemisphere is not clearly dominant.

The late neurologist Norman Geschwind of HarvardMedical School was intrigued by the fact that individuals withpronounced right-hemisphere gifts (that is, in math, music,art) are disproportionately nonright-handed (left-handed orambidexterous) and have higher than average rates of left-hemisphere deficits such as delayed onset of speech, stutteringor dyslexia. Geschwind and his colleague Albert Galaburdatheorized that this association of gift with disorder, which theycalled the “pathology of superiority,” results from the effect ofthe hormone testosterone on the developing fetal brain.

Geschwind and Galaburda noted that elevated testos-terone can delay development of the left hemisphere of thefetal brain; this in turn might result in compensatory right-hemisphere growth. Such “testosterone poisoning” might alsoaccount for the larger number of males than females whoexhibit mathematical and spatial gifts, nonright-handednessand pathologies of language. The researchers also noted thatgifted children tend to suffer more than the usual frequencyof immune disorders such as allergies and asthma; excess testos-terone can interfere with the development of the thymus gland,which plays a role in the development of the immune system.

Testosterone exposure remains a controversial explanationfor uneven gifts, and to date only scant evidence from the studyof brain tissue exists to support the theory of damage andcompensation in savants. Nevertheless, it seems certain thatgifts are hardwired in the infant brain, as savants and gifted


CALENDRICAL CALCULATORS GEORGE AND CHARLES,identical twins, are the most famous of such savants.Each could instantly compute the day of the week onwhich any given date, past or future, would fall. Thetwins were born in 1939 three months premature andretarded; their IQs tested between 40 and 70. Such anextraordinary ability to calculate in an otherwiseextremely mentally disabled child mirrors the milderunevenness of gifts seen in children highly talented inmathematics but learning disabled in language.

BIL

LEP

PRID

GE

Life

Mag

azin

e ©

TIM

E,In

c.

Human Intelligence


Exploring Intelligence 37Uncommon Talents:Gifted Children, Prodigies and Savants

children exhibit extremely highabilities from a very young age—before they have spent muchtime working at their gift.

Emphasizing Gifts

Given that many profound-ly gifted children are unevenlytalented, socially isolated andbored with school, what is thebest way to educate them? Mostgifted programs today tend totarget children who have testedabove 130 or so on standard IQtests, pulling them out of theirregular classes for a few hourseach week of general instructionor interaction. Unfortunately,these programs fail the mosttalented students.

Generally, schools arefocusing what few resourcesthey have for gifted educationon the moderately academicallygifted. These children make upthe bulk of current “pull-out”programs: bright students withstrong but not extraordinaryabilities, who do not face the challenges of precocity and iso-lation to the same degree as the profoundly gifted. These chil-dren—and indeed most children—would be better served ifschools instead raised their standards across the board.

Other nations, including Japan and Hungary, set muchhigher academic expectations for their children than the U.S.does; their children, gifted or not, rise to the challenge by suc-ceeding at higher levels. The needs of moderately gifted chil-dren could be met by simply teaching them a more demandingstandard curriculum.

The use of IQ as a filter for gifted programs also tends totip these programs toward the relatively abundant, moderatelyacademically gifted while sometimes overlooking profoundlybut unevenly gifted children. Many of those children do poorlyon IQ tests, because their talent lies in either math or language,but not both. Students whose talent is musical, artistic or ath-

letic are regularly left out as well. It makes more sense to iden-tify the gifted by examining past achievement in specific areasrather than relying on plain-vanilla IQ tests.

Schools should then place profoundly gifted children inadvanced courses in their strong areas only. Subjects in whicha student is not exceptional can continue to be taught to thestudent in the regular classroom. Options for advanced classesinclude arranging courses especially for the gifted, placing gift-ed students alongside older students within their schools, reg-istering them in college courses or enrolling them in accelerat-ed summer programs that teach a year’s worth of material in afew weeks.

Profoundly gifted children crave challenging work in theirdomain of expertise and the companionship of individuals withsimilar skills. Given the proper stimulation and opportunity,the extraordinary minds of these children will flourish.

WHEN BRILLIANCE ISN’T ENOUGH:William James Sidis (1898–1944) wasprofoundly gifted as a child, readingand spelling at the age of two, invent-ing a new table of logarithms at eight,speaking six languages by 10. By age11 he was enrolled at Harvard Univer-sity, delivering lectures on mathe-matics to the faculty. But Sidis’s fatherhad driven him mercilessly as a child,denying him any youthful pleasuresand letting the media hound him. Hegrew deeply bitter and resentful of hisfather and lost all interest in mathe-matics after graduating from Harvardat 16. This talented young man spentthe rest of his life in mindless clericaljobs, and his interests became obses-sive and autisticlike: at 28 he wrote acomprehensive book on the classifica-tion of streetcar transfer slips. He died,alone, from a brain hemorrhage at 46.

BRO

WN

BRO

THER

S

SA

ELLEN WINNER was a student of literatureand painting before she decided to explore devel-opmental psychology. Her inspiration was HarvardUniversity’s Project Zero, which researched thepsychological aspects of the arts. Her graduatestudies allowed her to combine her interests in artand writing with an exploration of the mind. Shereceived her Ph.D. in psychology from Harvard in1978 and is currently professor of psychology atBoston College as well as senior research associatewith Project Zero.

One of Winner’s greatest pleasures is writingbooks; she has authored three, one on the psychol-ogy of the arts, another on children’s use of meta-

phor and irony and, most recently, GiftedChildren: Myths and Realities. “I usually have sev-eral quite different projects going at once, so Iam always juggling,” she remarks. She is especial-ly intrigued by unusual children—children whoare gifted, learning disabled, gifted and learningdisabled, nonright-handed or particularly cre-ative. “The goal is to understand cognitivedevelopment in its typical and atypical forms.”

When she has time to play, Winnerdevours novels and movies and chauffeurs her13-year-old son on snowboarding dates. She ismarried to the psychologist Howard Gardnerand has three grown stepchildren. C

OU

RTE

SY O

F EL

LEN

WIN

NER

About the Author


1942

;PH

OTO

GR

APH

BY

DA

VID

HEA

LD,©

TH

E SO

LOM

ON

R.G

UG

GEN

HEI

M F

OU

ND

ATIO

N,N

EW Y

ORK


Untitled (Pharmacy), by Joseph Cornell

Human Intelligence


The ancient bards didn’t need them. Their well-tonedmemories bespoke tomes: the Iliad and the Odyssey, the RgVeda and the Mahabharata, among thousands of hours ofother recited epics. But in our era, filled with more informationin more forms than we could ever productively use, we seem towant them. Just as we want beauty sculpted not by our geneticheritage or by our exertion but rather by the scalpel or by sili-cone, we desire brains that are artificially boosted: we wantdrugs that make us think more quickly, that enable us toremember more readily, that give us a competitive edge.

The pursuit of these “smart” drugs has been celebratedsince the early 1990s, when books and bars (many of them inCalifornia) offered recommendations for diets or formulas orherbs such as ginkgo biloba that could better one’s brain. Inthe intervening years, a huge market for these items hassprung up, facilitated by the ease of sales over the Internet. InJapan alone, for instance, there are now 20 or so such com-pounds available and at least $2 billion in sales every year.

“Ninety-nine percent of that is hype,” says James L.McGaugh, head of the Center for the Neurobiology of Learningand Memory at the University of California at Irvine. And, tohim, worrisome hype. “We don’t know how many of these drugswork and how they interact with other drugs, so there is thepurely biological danger,” McGaugh explains.

Nevertheless, the public obsession with smart drugs mirrorsa scientific one. And what McGaugh and neuroscientists theworld over are studying could one day lead to clinically testeddrugs to enhance memory. The first wave of these are beingdesigned to help older people who are losing their ability toremember or those suffering from dementia. The only two drugsapproved by the U.S. Food and Drug Administration to boostmemory, in fact, are Tacrine and Donepezil, both for Alz-heimer’s patients. Several new compounds for this disease arein the final stage of testing and may soon be on the market.Hundreds more are being investigated. And behind this firstwave—but well off in the future—is the tsunami of promisethat such compounds could work in anyone.

The cognitive enhancers under study work in many differ-ent ways because research on memory is as rich and varied asmemories themselves. Scientists have looked at short-term (or“working”) memory, long-term memory, emotional memoryand olfactory memory; they have examined the molecular andgenetic webs of memory, the role of hormones in memory, andthe regions of the brain that light up in tomographic scanswhen a person remembers a sound as opposed to words. Ineach of these areas, neuroscientists garnered great insights overthe past few decades, offering the possibility that some of thegears of memory could be oiled or recast.

In spite of the advances and the optimism engendered,though, many investigators note that memory is so complexand so intertwined with other mental activities that it isunlikely that one drug could be precise enough to just helpyou find your glasses or remember names at a cocktail party.“It really calls for a carefully balanced approach, recognizingthat many of the mechanisms that may be critical for memorymay also be critical for transmissions that are deleterious,”

Exploring Intelligence 39Seeking “Smart” Drugs

Seeking“Smart”Drugs

New treatments for Alzheimer’sdisease and other neural disorders are pointing to drugsthat could boost memory inyoung, healthy individuals

by Marguerite Holloway, staff writer


observes Ira B. Black of Robert Wood Johnson Medical School.Further, augmenting short-term memory, say, or increasing

attention span does not necessarily translate into greater intel-

ligence. “It doesn’t make you smart,” McGaugh cautions. “Ifyou attend to the wrong things in life, that makes you dumb.”Larry Cahill, a colleague of McGaugh’s at Irvine, adds his owncaveat, borrowed from philosopher and psychologist WilliamJames: “‘Selection is the very keel on which our mental ship isbuilt.’ In other words, if we remembered everything we would‘be as ill off as if we remembered nothing.’”

Transmitter Turn-ons

How we recall anything comes down to the basic currencyof the nervous system: the giving and taking of neurotransmit-ters. These chemical messengers are released from a nerve cell

into a tiny space called the synapse. On the far side of thisgap sit other nerve cells studded with receptors shaped toreceive specific neurotransmitters. Once these receptors have

caught the molecules wafting across the synapse,they trigger chemical changes that allow informa-tion—in electrical form—to travel down the receiv-ing neuron to its end, where, in turn, more neuro-

transmitters set sail across a synapse. Understanding whichtransactions control memory is a matter of figuring out whichof the brain’s 100 billion neurons—each making an average of10,000 connections to other neurons—and which of the 50 orso neurotransmitters are involved.

Researchers have known since the 1950s that the hippo-campus—part of the limbic system, which controls emotionand sits under the cerebral cortex on top of the brain stem—iscrucial for memory. And since the 1980s they have known thatthe neurotransmitter glutamate, which binds to so-calledNMDA receptors, underlies a form of learning in the hippo-campus. Called long-term potentiation, it is thought to bringabout memory by strengthening the path of communication


MEMORY FORMATION includes many areas of the brain, butcentral to this activity is the hippocampus. Nestled in theinnermost part of the brain, the hippocampus, along with theamygdala and other structures, makes up the limbic system—the center of emotional response. The amygdala and hip-

pocampus also sit next to the olfactory nerve, which explainswhy smells can conjure up strong emotions and memories.Stress hormones released by the hypothalamus, the pituitarygland and the adrenal glands, which sit atop the kidneys,orchestrate some forms of memory as well.

Human Intelligence

We could end up worshipping intelligence even more thanwe already do—but using it even less.

CA

RO

L D

ON

NER

BASAL FOREBRAIN(BASAL GANGLIA)

OLFACTORY BULBAND TRACT

AMYGDALA

HIPPOCAMPUS

PITUITARYGLAND

HYPOTHALAMUS


between neurons—just as walking the same route through aforest again and again etches a permanent trail.

Several efforts to develop cognitive enhancers center onNMDA receptors—in particular, making them more active and,hence, more likely to establish long-term potentiation. Gary S.Lynch of U.C. Irvine, for instance, is investigating drugs—namedampakines—that interact with a particular kind of NMDA re-ceptor called AMPA.

NMDA receptors may also respond to neurotropins, com-pounds crucial for the survival and differentiation of neurons.In a surprising finding a few years ago, Black and his co-workersdiscovered that brain-derived neurotrophic factor—the king ofthe nerve growth factors—increases synaptic strength betweenneurons in the hippocampus. “We sort of wandered into thearea [of cognitive enhancers] through the back door,” Blackexplains. It now appears the hippocampus is lousy with neuro-tropins and that—at least in petri dishes and in rats—brain-derived neurotrophic factor may act on NMDA receptors.

Jump-Starting Genes

For the moment, Black is just figuring out the fundamen-tals. Getting large compounds across the blood-brain barrier andinto the brain is very hard. So Black and others are studyinghow to coax genes to turn on and produce growth factor in theright place. For example, James W. Simpkins, Edwin M. Meyerand their colleagues at the University of Florida at Gainesvilleare using a viral infection as the shuttle to carry a nerve growthfactor gene into the brains of laboratory animals. They watchto see which neurons take up the gene and generate nervegrowth factor, which ones take it up but do not do anythingwith it and which ones ignore it altogether. “We’re asking fun-damental questions,” Simpkins says. “How can we get the geneto the central nervous system? How can we enhance the hit?”

Genes, of course, orchestrate every mnemonic—and everyother—physiological activity, whether it is the creation of nervegrowth factor or of more NMDA receptors. By documenting themolecular and genetic machinations of memory, researchers atseveral institutions are hoping to find other forms of memoryboosters. Work on marine snails done by Eric R. Kandel’s teamat Columbia University and on fruit flies by Timothy Tully’sgroup at Cold Spring Harbor Laboratory in New York State haspinpointed a gene, known as CREB, that appears to be centralto some kinds of memory formation. With it, total recall.Without it, none. The hope is that years from now, cognitiveenhancers could perhaps tickle silent CREBs into life and, conse-quently, improve memory.

CREB may prove to be just one way of manipulating thesame process. Researchers at the University of Toronto report-ed recently in Science that long-term potentiation in the hip-pocampus can be forestalled by blocking the action of a par-ticular enzyme dubbed Src. Src belongs to a class of enzymes


RECEPTORS FOR ESTROGEN and nerve growth factor (dark spots in topimage) have been found in mice on the same neurons in the basal forebrain,a region damaged in Alzheimer’s disease. This discovery suggests that estro-gen may keep this—and perhaps other—parts of the brain healthy. Indeed, inthe presence of small amounts of estrogen, nerve cells flourish (middle);higher amounts yield even healthier and more robust cells (bottom). Studiesare now being conducted to see whether estrogen can prevent Alzheimer’sdisease or can improve memory function in women with the disease.

PHO

TOG

RA

PHS

BY

C.D

OM

INIQ

UE

TOR

AN

-ALL

ERA

ND


Human Intelligence

that had been shown by Kandel and others to be important tolong-term potentiation. Now it appears that Src regulates—nosurprise—NMDA receptors.

Hormonal Clout

Still, not all roads lead to NMDA. Estrogen, one of thestrongest and most promising cognitive enhancers currentlybeing studied, seems to work in a different way. In the early1990s C. Dominique Toran-Allerand of Columbia Universitynoticed that many neurons in the basal forebrain—an areanot far from the limbic system and one that is devastated byAlzheimer’s disease—had receptors for both estrogen andnerve growth factor. She hypothesized that these acetyl-choline-producing (or cholinergic, as they are called) neuronsneeded estrogen and nerve growth factor to stay healthy. Hernext thought was to consider what a sudden shortage of estro-gen would do to these neurons. (Previous work had shownthat the neurotransmitter acetylcholine is pivotal to memory,although how it works remains mysterious. Research had alsoestablished that people with Alzheimer’s have damaged cho-linergic neurons and low levels of acetylcholine. The twodrugs mentioned earlier—Tacrine and Donepezil—work byattacking the enzymes that break down acetylcholine.)

Toran-Allerand’s findings fit nicely with those of a fewother researchers as well as with anecdotal reports that morewomen than men develop Alzheimer’s. The large clinical trialsneeded to examine rigorously the protective effects of estrogenhave just begun: the Women’s Health Initiative-Memory Studyenrolled 6,000 women and will have data in 2005, and a studyof 900 women whose relatives have Alzheimer’s is under wayat Johns Hopkins University, in conjunction with ColumbiaUniversity and the Mayo Clinic. But in the past few years, sev-eral small studies have found that estrogen replacement thera-py not only reduces the risk of developing Alzheimer’s butalso improves short-term memory in women with the disor-der and in normally functioning postmenopausal women.

Men have a source of estrogen as well: testosterone is con-verted to its female counterpart in the brain. Unlike women,however, men do not experience a precipitous hormonal declinein their later years. Nevertheless, work by Simpkins and hiscolleagues shows that estrogen enhances short- and long-termmemory in animals of both sexes and that it protects the brainfrom damage such as that caused by the loss of oxygen duringstroke. So Simpkins and his team are developing nonfeminizingestrogens that could be used in men and that could reduce theestrogen-associated risk of cancer in women. “Our approachhas been to discover estrogenlike compounds that are cognitiveenhancers but, and we believe more important, are neuropro-tective compounds,” Simpkins explains.

A host of other hormones play a critical role in memory aswell. Researchers studying stress responses, including McGaughand Benno Roozendaal of U.C. Irvine, know that stress hor-mones such as corticosteroids can lead to powerful memoryformation. Cahill is among the many neuroscientists lookingat how such arousal lays down memory, which hormones dowhat and how to exploit this system. He is currently settingup clinical trials to test a beta blocker, Inderal, that could dullunpleasant memories in people who suffer post-traumatic stressdisorder. (Cahill is quick to point out that Inderal was listed asa cognitive enhancer in Smart Drugs and Nutrients, the bookthat started the U.S. craze. “We have shown in humans that itis quite bad for memory,” he laughs.)

The Proustian Connection: Popping a MadeleineIt is no surprise that smell triggers what are sometimes

described as the most powerful memories: the olfactorynerve is just two synapses away from the amygdala, thecenter of human emotions, and just three from the hip-pocampus, headquarters of at least some forms of memo-ry. Many researchers have been intrigued by this proximi-ty, among them Rachel S. Herz of the Monell ChemicalSenses Center in Philadelphia.

Herz recently examined whether smell could serve as aform of cognitive enhancer in emotional situations. Shetested students who were about to take an exam andwho were, as a consequence, exceedingly anxious. Oneset of nervous students was given a list of words toremember at the same time that they were exposed to asmell; the other group saw the same words, but theirroom remained odorless. A week later Herz found thatthose reexposed to the smell had 50 percent better recallthan the control subjects did.

In another experiment, Herz tried to determinewhether memory evoked by smell was more accuratethan memory evoked by other cues, such as images. Shefound that odor did not increase accuracy but rather theemotional intensity of the recollection. So if your cogni-tive enhancer of choice proves to be perfume or whateverspice you have on the shelf, beware: your emotions mayget the better of you. —M.H.

Those reexposed to the smell had 50 percent better recall.

PEPPERMINT is among the scents used inexperiments to evoke emotional memory.


PATT

I MU

RRAY

Ear

th S

cene

s


As cognitive enhancers, stress hormones pose a bit of aparadox. They can be good for memory, but they take theirtoll on the body. The same holds true of a few of the otherlegal and widespread smart drugs, such as caffeine, whichenhances mental alertness but can cause gastrointestinal prob-lems. It seems that caffeine works as a stimulant because itblocks one of the receptors for adenosine—a neurotransmitterthat seems, among other actions, to dull concentration.

Taking Taboos

A cup of coffee with sugar in it would work even better.“Perhaps the smartest smart drug out there is glucose,” Cahillcomments—although excess can be unhealthy. The first cluethat sugar is a smart drug came from stress studies. Among thehormones released during stress is epinephrine, which causesblood levels of glucose to shoot up. Paul E. Gold and others atthe University of Virginia at Charlottesville found that rodentsand people of all ages show memory improvement when givensugar. Gold also reports that glucose enhances some forms ofcognition in people with Alzheimer’s disease or Down syn-drome. Again, all the studies so far have been small and are notyet conclusive.

According to Gold, it appears that glucose directly triggersthe production of acetylcholine. Several other researchers havefingered the hormone insulin, rather than sugar, as the crucialmemory enhancer. Regardless, it would appear that a hit of jellybeans could be salubrious.

And while you are at it, a nicotine patch could help.Various studies have found that nicotine improves people’sshort-term memory. Nicotine, like other pleasure-inducingdrugs, is an analogue of a naturally occurring neurotransmitter:

it resembles acetylcholine and binds to the acetylcholine nico-tinic receptor. Because of the importance of cholinergic neuronsin Alzheimer’s disease, some researchers have focused on under-standing the nicotinic receptors. Meyer is one of these scientists,and after eight years of study, he has synthesized a drug that—to his surprise—proved to be a potent cognitive enhancer in asmall group of young, healthy men. “We didn’t expect the drugto work in normal people, because you don’t see Alzheimer’sagents working in non-Alzheimer folks,” Meyer says. More evi-dence, he adds, that “no one really knows how memory works.”

Which is why the wait for the right brain boosters may bea long one. But they are on their way, and given society’s desirefor elixirs and quick fixes, it is worth thinking about what itwould mean to rely even more on drugs, who would be able toafford them and what memory—and perhaps intelligence—would mean to us if we had dominion over it.

The idea of helping people who have impairments get somerelief is exciting. The idea of raising a normal, healthy person’sIQ a few points for certain tasks seems fair enough. But fartherdown this slippery slope lies the possibility of dramaticallyaugmenting someone’s intelligence—or, at least, that of some-one who can afford it. That possibility seems less fair and muchmore likely to institutionalize fully the social and economicstratification that already exists.

Paradoxically, such smart drugs could even inculcateintellectual lassitude, much as the good feeling from a moodenhancer or antidepressant allows some people to avoid grap-pling with emotional problems. We could end up worshippingintelligence even more than we already do—but using it evenless. “We should take care not to make the intellect our god,”Albert Einstein wrote in Out of My Later Life. “It has, of course,powerful muscles, but no personality.”


SA

STORYTELLING is an ancient tradi-tion. The poets who recited the greatIndian and Greek epics needed no“smart” drugs to keep their memoriespowerful. James L. McGaugh of theUniversity of California at Irvine passeson this thought from a friend: “All oneneeds to strengthen memory is applica-tion—application of the seat of thepants to the seat of the chair.”

INST

ITU

TE O

F O

RIEN

TAL

STU

DIE

S,ST

.PET

ERSB

URG

,RU

SSIA

/GIR

AU

DO

N/S

UPE

RSTO

CK



To most observers, the essence of intelligence is cleverness,a versatility in solving novel problems. Foresight is also said tobe an essential aspect of intelligence—particularly after anencounter with one of those terminally clever people who areall tactics and no strategy. Other observers will add creativityto the list. Personally, I like the way neurobiologist HoraceBarlow of the University of Cambridge frames the issue. Hesays intelligence is all about making a guess that discoverssome new underlying order. This idea neatly covers a lot ofground: finding the solution to a problem or the logic of anargument, happening on an appropriate analogy, creating apleasing harmony or guessing what’s likely to happen next.Indeed, we all routinely predict what comes next, even whenpassively listening to a narrative or a melody. That’s why ajoke’s punch line or a P.D.Q. Bach musical parody brings youup short—you were subconsciously predicting something elseand were surprised by the mismatch.

We will never agree on a universal definition of intelligencebecause it is an open-ended word, like consciousness. Bothintelligence and consciousness concern the high end of ourmental life, but they are frequently confused with more ele-mentary mental processes, such as ones we use to recognize afriend or to tie a shoelace. Of course, such simple neuralmechanisms are probably the foundations from which ourabilities to handle logic and metaphor evolved. But how didthat occur? That is both an evolutionary question and a neu-rophysiological one. Both kinds of answers are needed tounderstand our own intelligence. They might even help explainhow an artificial or an exotic intelligence could evolve.

Did our intelligence arise from having more of what otheranimals have? The two-millimeter-thick cerebral cortex is thepart of the brain most involved with making novel associa-tions. Ours is extensively wrinkled, but were it flattened out, itwould occupy four sheets of typing paper. A chimpanzee’s cor-tex would fit on one sheet, a monkey’s on a postcard, a rat’son a stamp. But a purely quantitative explanation seemsincomplete. I will argue that our intelligence arose primarilythrough the refinement of some brain specialization, such asthat for language. This specialization allowed a quantum leap

in cleverness and foresight during the evolution of humansfrom apes. If, as I suspect, the specialization involved a corefacility common to language, the planning of hand move-ments, music and dance, it has even greater explanatory power.

A particularly intelligent person often seems “quick” andcapable of juggling many ideas at once. Indeed, the twostrongest influences on your IQ score are how many novelquestions you can answer in a fixed length of time and howgood you are at simultaneously manipulating half a dozenmental images—as in those analogy questions: A is to B as C isto (D, E or F).

Versatility is another characteristic of intelligence. Mostanimals are narrow specialists, especially in matters of diet: themountain gorilla consumes 23 kilograms (50 pounds) of greenleaves each and every day. In comparison, a chimpanzeeswitches around a lot—it will eat fruit, termites, leaves and evena small monkey or piglet if it is lucky enough to catch one.Omnivores have more basic moves in their general behaviorbecause their ancestors had to switch between many differentfood sources. They need more sensory templates, too—mentalsearch images of things such as foods and predators for whichthey are “on the lookout.” Their behavior emerges through thematching of these sensory templates to responsive movements.

Sometimes animals try out a new combination of searchimage and movement during play and find a use for it later.Many animals are playful only as juveniles; being an adult is aserious business (they have all those young mouths to feed).Having a long juvenile period, as apes and humans do, surelyaids intelligence. A long life further promotes versatility byaffording more opportunities to discover new behaviors.

A social life also gives individuals the chance to mimicthe useful discoveries of others. Researchers have seen a troop

The Emergence of Intelligenceby William H. Calvin


Exploring Intelligence 45The Emergence of Intelligence

of monkeys in Japan copy one inventive female’s techniquesfor washing sand off food. Moreover, a social life is full ofinterpersonal problems to solve, such as those created bypecking orders, that go well beyond the usual environmentalchallenges to survival and reproduction.

Yet versatility is not always a virtue, and more of it is notalways better. When the chimpanzees of Uganda arrive at agrove of fruit trees, they often discover that the efficient localmonkeys are already speedily stripping the trees of ediblefruit. The chimps can turn to termite fishing, or perhaps catcha monkey and eat it, but in practice their population isseverely limited by that competition, despite a brain twice thesize of their specialist rivals.

The Impact of Abrupt Climate Change

Versatility becomes advantageous, however, when theweather changes abruptly. The fourfold expansion of thehominid brain started 2.5 million years ago, when the ice agesbegan. Ice cores from Greenland show that warming and cool-ing episodes occurred every several thousand years, superim-posed on the slower advances and retreats of the northern icesheets. The vast rearrangements in ocean currents lasted for

centuries, with sudden transitions that took less than a decade.The abrupt coolings most likely devastated the ecosys-

tems on which our ancestors depended. Because of lower tem-peratures and less rainfall, the forests in Africa dried up andanimal populations began to crash. Lightning strikes ignitedgiant forest fires, denuding large areas even in the tropics.There was very little food after the fires. Once the grassesreemerged on the burnt landscape, however, the survivinggrazing animals had a boom time. Within several centuries, asuccession of forests came back in many places, featuringspecies more appropriate to the cooler climate.

Cool, crash and burn. The progenitors of modern humanslived through hundreds of such episodes, but each was a pop-ulation bottleneck that eliminated most of their relatives. Hadthe cooling taken a few centuries to happen, the forests couldhave gradually shifted, and our ancestors would not have been

Language, foresight, musical skills and other hallmarks of intelligence may all be linked to the human ability to create rapid movements such as throwing

THROWING A STONE requires a surprising amount of brain-power. The complex sequence of movements is shown in afamous series of photographs taken by Eadweard Muybridge inthe 1880s. The improvement of throwing abilities in earlyhominids may have enhanced the dexterity of their mouthmovements as well and led to the development of language.

DO

VER

PU

BLI

CAT

ION

S


treated so harshly. The higher-elevation plant species wouldhave slowly marched down the hillsides to occupy the valleyfloors. Hominid generations could have made their living inthe way their parents taught them, culturally adapting to thenew milieu. But when the cooling and drought were abrupt, itwas one unlucky generation that suddenly had to improviseamid crashing populations and burning ecosystems. We arethe improbable descendants of those who survived—probablybecause they had ways of coping with these episodes that theother great apes did not exploit.

Improvising meant learning to eat grass—or managingregularly to eat animals that eat grass. The trouble is that suchanimals are fast and wary, whether rabbit or antelope. Smallor big, they are best tackled by cooperative groups. But shar-ing a rabbit leaves everyone hungry, so the hunters wouldhave tried for the bigger animals that cluster in herds. Andthat had an interesting consequence. If a single hunter killeda big animal, it was too much to eat; best to give most of themeat away and count on reciprocity when someone else suc-ceeded. Sharing food also meant fewer fights and more timeavailable to seek out scarce food.

Each population bottleneck temporarily exaggerated theimportance of such traits as cooperation, altruism and hunting

abilities. Even if each episode changed the inborn predilectionsof the hominids by only a small amount, the hundreds of rep-etitions of this scenario may explain some of the differencesbetween human abilities and those of our closest relativesamong the great apes. It is tempting to say that the abrupt cool-ings pumped up brain size, but what makes for better survival issomething much more specific: hunting abilities and perhapsaltruism. What might they have to do with intelligence?

Syntax and Structured Thought

One of the improvements that occurred during the iceages was the capacity for human language. In most of us, thebrain area critical to language is located just above our left ear.Monkeys lack this left lateral language area: their vocalizations(and simple emotional utterances in humans) employ a moreprimitive language area near the corpus callosum, the band offibers connecting the cerebral hemispheres.

Language is the most defining feature of human intelli-gence: without syntax—the orderly arrangement of verbalideas—we would be little more clever than a chimpanzee. Fora glimpse of life without syntax, look to the case of Joseph, an11-year-old deaf boy. Because he could not hear spoken lan-

guage and had never been exposed to fluent signlanguage, Joseph did not have the opportunity tolearn syntax during the critical years of early child-hood. As neurologist Oliver Sacks described him:“Joseph saw, distinguished, categorized, used; he hadno problems with perceptual categorization or gen-eralization, but he could not, it seemed, go muchbeyond this, hold abstract ideas in mind, reflect,play, plan. He seemed completely literal—unable tojuggle images or hypotheses or possibilities, unableto enter an imaginative or figurative realm.... Heseemed, like an animal, or an infant, to be stuck inthe present, to be confined to literal and immediateperception, though made aware of this by a con-sciousness that no infant could have.”

To understand why humans are so intelligent, we need tounderstand how our ancestors remodeled the apes’ symbolicrepertoire and enhanced it by inventing syntax. Wild chim-panzees use about three dozen different vocalizations to conveyabout three dozen different meanings. They may repeat a soundto intensify its meaning, but they do not string together threesounds to add a new word to their vocabulary. Humans also useabout three dozen vocalizations, called phonemes. Yet onlytheir combinations have content: we string together meaning-less sounds to make meaningful words. Furthermore, humanlanguage uses strings of strings, such as the word phrases thatmake up this sentence.

Our closest animal cousins, the common chimpanzee andthe bonobo (pygmy chimpanzee), can achieve surprising lev-els of language comprehension when motivated by skilledteachers. Kanzi, the most accomplished bonobo, can interpretsentences he has never heard before, such as “Go to the officeand bring back the red ball,” about as well as a two-and-a-half-year-old child. Neither Kanzi nor the child constructssuch sentences independently, but they can demonstrate bytheir actions that they understand them.

With a year’s experience in comprehension, a child startsconstructing sentences that nest one word phrase inside anoth-


Relative Size of Cerebral Cortex



er. The rhyme about the house that Jack built (“This is thefarmer sowing the corn / That kept the cock that crowed inthe morn /…That lay in the house that Jack built”) is anexample of such a sentence. Syntax has treelike rules of refer-ence that enable us to communicate quickly—sometimes withfewer than 100 sounds strung together—who did what towhom, where, when, why and how. Even children of lowintelligence seem to acquire syntax effortlessly, although intel-ligent deaf children like Joseph may miss out.

Something very close to syntax also seems to contributeto another outstanding feature of human intelligence: theability to plan ahead. Aside from hormonally triggered prepa-rations for winter, animals exhibit surprisingly little evidenceof advance planning. For instance, some chimpanzees uselong twigs to pull termites from their nests. Yet as authorJacob Bronowski observed, none of the termite-fishing chimps“spends the evening going round and tearing off a nice tidysupply of a dozen probes for tomorrow.”

Human planning abilities may stem from our talent forbuilding narratives. We can borrow the mental structures forsyntax to judge combinations of possible actions. To someextent, we do this by talking silently to ourselves, making nar-ratives out of what might happen next and then applyingsyntaxlike rules of combination to rate a scenario as unlikely,possible or likely. Narratives are also a major foundation forethical choices: we imagine a course of action and its effectson others, then decide whether or not to do it. But our think-ing is not limited to languagelike constructs. Indeed, we mayshout “Eureka!” when feeling a set of mental relationshipsclick into place yet have trouble expressing them verbally.

Ballistic Movements and Their Relatives

Language and intelligence are so powerful that we mightthink evolution would naturally favor their increase. But asHarvard University evolutionary biologist Ernst Mayr oncesaid, most species are not intelligent, which suggests “thathigh intelligence is not at all favored by natural selection”—or

that it is very hard to achieve. So we must consider indirectways of achieving it, rather than general principles.

Evolution often follows indirect routes rather than “pro-gressing” via adaptations. To account for the breadth of ourhigher intellectual functions (syntax, planning, logic, gameswith rules, music), we need to look at improvements incommon-core facilities. Humans certainly have a passion forstringing things together: words into sentences, notes intomelodies, steps into dances, narratives into games with rulesof procedure. Might stringing things together be a core facilityof the brain?

As improbable as the idea initially seems, the brain’s plan-ning of ballistic movements may have once promoted lan-guage, music and intelligence. Such movements are extremelyrapid actions of the limbs that, once initiated, cannot bemodified. Striking a nail with a hammer is an example. Apeshave only elementary forms of the ballistic arm movements atwhich humans are expert—hammering, clubbing and throw-ing. Perhaps it is no coincidence that these movements areimportant to the manufacture and use of tools and huntingweapons: in a setting such as cool-crash-and-burn, huntingand toolmaking were important additions to hominids’ basicsurvival strategies.

Compared with most movements, ballistic ones require asurprising amount of planning. Slow movements leave timefor improvisation: when raising a cup to your lips, if the cupis lighter than you remembered, you can correct its trajectorybefore it hits your nose. Thus, a complete advance plan is notneeded. You start in the right general direction and then cor-rect your path. For sudden limb movements lasting less thanone fifth of a second, feedback corrections are largely ineffec-tive because reaction times are too long. The brain has to planevery detail of the movement. Hammering, for example,requires planning the exact sequence of activation for dozensof muscles.

The problem of throwing is compounded by the briefnessof the launch window—the range of time in which a projectilecan be released to hit a target. Because the human sense of

SPECIALIZED SEQUENCING REGION ofthe left cerebral cortex is involved both inlistening to spoken language and in pro-ducing oral-facial movements. The shad-ing in the illustration at the right, basedon the data of George A. Ojemann of theUniversity of Washington, reflects theamount of involvement in these activities.

CEREBRAL CORTEX is the deeply convolutedsurface region of the brain that is moststrongly linked to intelligence (below). Ahuman’s cerebral cortex, if flattened, wouldcover four pages of typing paper (left); achimpanzee’s would cover only one page; amonkey’s would cover a postcard; and a rat’swould cover a postage stamp.

DA

NA

BU

RNS-

PIZ

ER A

ND

JU

DIT

H G

LIC

K



timing is inevitably jittery, when the distance to a target dou-bles, the launch window becomes eight times narrower, andshrinking the timing jitter requires the activity of 64 times asmany neurons. These neurons function as independent tim-ing mechanisms working in concert, like a chorus of medievalsingers reciting a plainsong in unison.

If mouth movements rely on the same core facility forsequencing as ballistic hand movements do, then improve-ments in dexterity might improve language, and vice versa.Accurate throwing abilities, which would have helped earlyhominids survive the cool-crash-and-burn episodes in thetropics, would also open up the possibility of eating meat reg-ularly and of being able to survive winter in a temperate zone.The gift of speech would be an incidental benefit—a freelunch, as it were, because of the linkage.

There certainly seems to be a sequencer common to bothhand movements and language. Much of the brain’s coordina-tion of movement occurs at a subcortical level in the basalganglia or the cerebellum, but novel movements tend todepend on the premotor and prefrontal cortex. Two majorlines of evidence point to cortical specialization for sequenc-ing, and both of them suggest that the lateral language areahas much to do with it. Doreen Kimura of the University ofWestern Ontario has found that stroke patients with languageproblems (aphasia) resulting from damage to left lateral brainareas also have considerable difficulty executing novel sequencesof hand and arm movements (apraxia). By electrically stimu-lating the brains of patients being operated on for epilepsy,George A. Ojemann of the University of Washington has alsoshown that at the center of the left lateral areas specialized forlanguage lies a region involved in listening to sound sequences.This perisylvian region seems equally involved in producingoral-facial movement sequences—even nonlanguage ones.

These discoveries reveal that the “language cortex,” aspeople sometimes think of it, serves a far more generalizedfunction than had been suspected. It is concerned with novel

sequences of various kinds: both sensations and movements,for both the hands and the mouth. The big problem withfashioning new sequences and producing original behaviors issafety. Even simple reversals in order can be dangerous, as in“Look after you leap.” Our capacity to make analogies andmental models gives us a measure of protection, however.Humans can simulate future courses of action and weed outthe nonsense off-line; as philosopher Karl Popper said, this“permits our hypotheses to die in our stead.” Creativity—indeed, the entire high end of intelligence and consciousness—involves playing mental games that improve the quality ofour plans. What kind of mental machinery might it take to dosomething like that?

Natural Selection in the Brain

By 1874, just 15 years after Charles Darwin published Onthe Origin of Species by Means of Natural Selection, Americanpsychologist William James was talking about mental processesoperating in a Darwinian manner. In effect, he suggested, ideasmight somehow “compete” with one another in the brain,leaving only the best or “fittest.” Just as Darwinian evolutionshaped a better brain in two million years, a similar Darwinianprocess operating within the brain might shape intelligentsolutions to problems on the timescale of thought and action.

Researchers have demonstrated that a Darwinian processoperating on a timescale of days governs the immune system.Through a series of cellular generations spanning several weeks,the immune system produces defensive antibody moleculesthat are better and better “fits” against invaders. By abstract-ing the essential features of a Darwinian process from what isknown about species evolution and immune responses, wecan see that any “Darwin machine” must have six properties.

First, it must operate on patterns of some type; in genet-ics, they are strings of DNA bases, but the patterns of brainactivity associated with a thought might qualify. Second,

APPROXIMATE DEPTH OF ICE-CORE SAMPLE (METERS)

3,000 2,500 2,000 1,500 1,000

DEV

IATI

ON

IN O

XYG

EN 1

8C

ON

TEN

T(P

ERC

ENT)

–30

LAST ICE AGE(APPROXIMATE DATES)

WA

RMER

–34

–32

–36

–38

–40

–42

–44

TIME (THOUSANDS OF YEARS AGO)

25250 150 120 100 80 70 60 50 40 35 30 20 18 16 14 12 11 10 9 8 7 6

CURRENTINTERGLACIAL

PERIOD

7° C INCREASEWITHIN 50 YEARS

RAPID CLIMATE CHANGES may have promoted behavioralversatility in the ancestors of modern humans. Studies ofGreenland ice cores show that during the last ice age, averagetemperatures were subject to abrupt fluctuations. During one

climatic oscillation (red line), the average temperature rose 13degrees Fahrenheit (seven degrees Celsius) in the space of afew decades. This graph is based on work by Willi Dansgaardof the University of Copenhagen and his colleagues.

JOH

NN

Y JO

HN

SON


copies must somehow be made ofthese patterns. Third, patterns mustoccasionally vary, either throughmutations, copying errors or areshuffling of their parts. Fourth,variant patterns must compete tooccupy some limited space (as whenbluegrass and crabgrass compete formy backyard). Fifth, the relativereproductive success of the variantsmust be influenced by their environ-ment; this result is what Darwincalled natural selection. And finally,the makeup of the next generationof patterns must depend on whichvariants survive to be copied. Thepatterns of the next generation willbe variations based on the more suc-cessful patterns of the current gener-ation. Many of the new variants willbe less successful than their parents,but some may be more so.

Let us consider how these prin-ciples might apply to the evolutionof an intelligent guess inside thebrain. Thoughts are combinations ofsensations and memories—in a way, they are movements thathave not happened yet (and maybe never will). They take theform of cerebral codes, which are spatiotemporal activity pat-terns in the brain that each represent an object, an action oran abstraction. I estimate that a single code minimally involvesa few hundred cortical neurons within a millimeter of oneanother, either keeping quiet or firing in a musical pattern.

Evoking a memory is simply a matter of reconstitutingsuch an activity pattern, according to the cell-assemblyhypothesis of psychologist Donald O. Hebb [see “The Mindand Donald O. Hebb,” by Peter M. Milner; SCIENTIFIC AMERICAN,January 1993]. Long-term memories are frozen patterns wait-ing for signals of near resonance to reawaken them, like rutsin a washboarded road waiting for a passing car to re-create abouncing spatiotemporal pattern.

Some “cerebral ruts” are permanent, whereas others areshort-lived. Short-term memories are just temporary alterationsin the strengths of synaptic connections between neurons, leftbehind by the last spatiotemporal pattern to occupy a patchof cortex; they fade in a matter of minutes. The transitionfrom short- to long-term memory is not well understood, butit appears to involve structural alterations in which the synap-tic connections between neurons are made strong and perma-nent, hardwiring the pattern of neural activity into the brain.

A Darwinian model of mind suggests that an activatedmemory can compete with others for “workspace” in the cor-tex. Both the perceptions of the thinker’s current environmentand the memories of past environments may bias that compe-

tition and shape an emerging thought. An active cerebral codemoves from one part of the brain to another by making a copyof itself, much as a fax machine re-creates a pattern on a dis-tant sheet of paper. The cerebral cortex also has circuitry forcopying spatiotemporal patterns in an adjacent region less thana millimeter away, although present imaging techniques lackenough resolution to see it in progress. Repeated copying ofthe minimal pattern could colonize a region, rather the waythat a crystal grows or wallpaper repeats an elementary pattern.

The picture that emerges from these theoretical considera-tions is one of a quilt, some patches of which enlarge at theexpense of their neighbors as one code copies more successful-ly than another. As you try to decide whether to pick an appleor a banana from the fruit bowl, so my theory goes, the cere-bral code for “apple” may be having a cloning competitionwith the one for “banana.” When one code has enough activecopies to trip the action circuits, you might reach for theapple. But the banana codes need not vanish: they couldlinger in the background as subconscious thoughts. Our con-scious thought may be only the currently dominant patternin the copying competition, with many other variants com-peting for dominance, one of which will win a moment laterwhen your thoughts seem to shift focus.

It may be that Darwinian processes are only the frostingon the cognitive cake, that much of our thinking is routine orrule-bound. But we often deal with novel situations in cre-ative ways, as when you decide what to fix for dinner tonight:You survey what’s already in the refrigerator and on the kitchen


DU

OM

O

BALLISTIC ARM MOVEMENTS,such as those displayed by New YorkYankees pitcher David Wells, are sorapid that the brain must plan thesequence of muscle contractions inadvance. Some of the neural mecha-nisms that plan such movements mayalso facilitate other types of planning.



shelves. You think about a few alternatives, keeping track ofwhat else you might have to fetch from the grocery store. Allof this can flash through your mind within seconds—and that’sprobably a Darwinian process at work.

Bootstrapping Intelligence

In both its phylogeny and ontogeny, human intelligencefirst solves movement problems and only later graduates toponder more abstract ones. An artificial or extraterrestrialintelligence freed of the necessity of finding food and avoid-ing predators might not need to move—and so might lack the“what happens next” orientation of human intelligence. It isdifficult to estimate how often high intelligence might emerge,given how little we know about the demands of long-termspecies survival and the courses evolution can follow. We can,however, evaluate the prospects of a species by asking howmany elements of intelligence each has amassed. Chimps and

bonobos may be missing a few of the elements—the ability toconstruct nested sentences, for example—but they are doingbetter than the present generation of artificial-intelligenceprograms.

Why aren’t there more species with such complex mentalstates? There might be a hump to get over: a little intelligencecan be a dangerous thing. A beyond-the-apes intelligence mustconstantly navigate between the twin hazards of dangerousinnovation and a conservatism that ignores what the RedQueen explained to Alice in Through the Looking Glass: “Ittakes all the running you can do, to keep in the same place.”Foresight is our special form of running, essential for the intel-ligent stewardship that Stephen Jay Gould of Harvard warns isneeded for longer-term survival: “We have become, by thepower of a glorious evolutionary accident called intelligence,the stewards of life’s continuity on earth. We did not ask forthis role, but we cannot abjure it. We may not be suited to it,but here we are.”

KANZI, A BONOBO, has been reared at Georgia State University in a lan-guage-using environment. By pointing at symbols that represent words, Kanziconstructs requests much like those of a two-year-old child. His comprehen-sion is as good as that of a two-and-a-half-year-old. Language experimentswith bonobos investigate how much of syntax is uniquely human.

ACQUISITION OF LANGUAGE by chil-dren occurs quickly and naturally throughexposure to adults. By the age of threeyears, the great majority of children areable to construct simple sentences.

AGE (MONTHS)8 16 24 30 36

ALL

CH

ILD

REN

(PER

CEN

T)

100

50

0

SPEAKING IN SINGLE WORDS

SPEAKING IN TWO-WORD PHRASES

SPEAKING IN SENTENCES OF FIVE

OR MORE WORDS

abc

–

–

–

a b c

WILLIAM H. CALVIN’s career has taken a Darwinian course: his scientific interestshave evolved significantly over the past four decades. He studied physics as an under-graduate at Northwestern University but devoted his spare time to a research projectexploring how the brain processes color vision. This project led to graduate work in neu-roscience at the Massachusetts Institute of Technology and Harvard Medical School, thento a Ph.D. in physiology and biophysics from the University of Washington in 1966. Hisearly research focused on neuron-firing mechanisms. “I wiretapped neurons, trying tofigure out how they transformed information,” he says. But in the 1980s he took on abigger question—how the human brain evolved—and his interests broadened to includeanthropology, zoology and psychology. He has written several acclaimed books, includ-ing The Cerebral Code, How Brains Think and (with George A. Ojemann) Conversationswith Neil’s Brain. “The puzzles I’m trying to solve require information from many differ-ent fields,” he says. Calvin is currently a theoretical neurophysiologist on the faculty ofthe University of Washington School of Medicine.

SA

MIC

HA

EL N

ICH

OLS

JOH

NN

Y JO

HN

SON

DO

UG

VA

ND

ERH

OO

F M

oder

n M

edia

About the Author


The ability to think and plan is taken by many of usto be the hallmark of the human mind. Reason, whichmakes thinking possible, is often said to be uniquelyhuman and thus sets us apart from the beasts. In the pasttwo decades, however, this comfortable assumption ofintellectual superiority has come under increasingly skep-tical scrutiny. Most researchers now at least entertain theonce heretical possibility that some animals can indeedthink. At the same time, several of the apparent mentaltriumphs of our species—language, for instance—haveturned out to owe as much to innate programming as toraw cognitive power.

This reversal of fortune for the status of humanintellectual uniqueness follows nearly a century of academ-ic neglect. The most devastating and long-lasting blowto the idea of animal intelligence stemmed from the 1904incident of Clever Hans the horse. Oskar Pfungst, theresearcher who unraveled the mystery of an animal thatseemed as intelligent as many humans, described the sit-uation vividly: “At last the thing so long sought for wasapparently found: a horse that could solve arithmeticalproblems—an animal, which thanks to long training,mastered not merely rudiments, but seemingly arrived at

Animal Intelligence52 Scientific American Presents

A mounting body of evidence suggests that anumber of species can inferconcepts, formulate plansand employ simple logic insolving problems

by James L. Gould and Carol Grant Gould

INTELLIGENCE

AN

IMA

L

CO

URT

ESY

OF

THE

MU

SEU

MO

F M

OD

ERN

ART

,NEW

YORK

Reasoning inAnimals


Exploring Intelligence 53Reasoning in Animals

The Dream, by Henri Rousseau


a power of abstract thought which surpassed, by far, the high-est expectation of the greatest enthusiast.” Hans could alsoread and understand spoken German.

After expert groups had tested the horse (often in theabsence of his owner, Mr. von Osten) and agreed that notrickery could be involved, Pfungst undertook to study the ani-mal in detail. After many months, he discovered the truesource of Hans’s cleverness: the animal watched for slightinvoluntary cues that invariably arose from his audience as heapproached the correct number of taps of his hoof.

The consequence of a mere horse having “tricked” the phil-osophical establishment was a wholesale retreat from work onanimal thinking: before the incident, it had been common toattribute reason and thought to animals. British comparativepsychologist George J. Romanes, in his 1888 book AnimalIntelligence, set the bar so low that even shellfish could be said tobe rational, as when “we find, for instance, that an oyster profitsby individual experience, or is able to perceive new relations andsuitably to act upon the result of its perceptions.” In short,Romanes felt that if instinct is not at work, reason must be.

As a result of the Clever Hans incident, however, the behav-iorist school of psychology came to dominate experiments onanimal behavior in the English-speaking world. This reactionaryperspective denied the existence of instinct, consciousness,thought and free will not only in animals but in humans aswell. As the founder of behaviorism, American psychologistJohn B. Watson, put it in 1912 (in characteristically uncom-promising terms), “Consciousness is neither a definite nor ausable concept.... [B]elief in the existence of consciousnessgoes back to the ancient days of superstition and magic.”

For Watson, all human and animal behavior was the resultof conditioning—even breathing and the circulation of theblood. From his perspective, humans do not really think,although they may form “verbal habits”—highly rational ones—with proper training. In the absence of words, Watson believed,animals could not possibly think. Species-specific behaviors—nest building by birds, for instance—were a result of the particu-lar anatomy of a species, the habitat into which it was born andthe experiences individuals typically underwent as they grew up.

And contrary to Romanes’s view, the behaviorist felt thateven learning can be mindlessly automatic, requiring no com-prehension on the part of the “student.” Classical condition-ing simply associates an innate stimulus-response reflex with anovel stimulus. Thus, it is possible to teach a dog that a bellor a flashing light means food. The other form of learning inthe behaviorist worldview—operant conditioning—merelyrequires animals to discover by trial and error which of theirmovements are rewarded and to use these data to fashionnovel behavior patterns. No understanding is needed in eithercase. Even the subsequent discovery of species-specific learn-ing programs (learning that is initiated and controlled byinstinct) did little to alter this passive-learning-machine viewof animals, although it did deliver a fatal blow to behaviorismitself [see “Learning by Instinct,” by James L. Gould and PeterMarler; SCIENTIFIC AMERICAN, January 1987].

The pervasive behaviorist taboo against investigatingwhether animals can think, however, has persisted. Only thepublication of Donald R. Griffin’s highly provocative and con-troversial 1976 book, The Question of Animal Awareness, hasbegun to erode it. Like most of our colleagues trained in the“don’t ask, don’t tell” intellectual atmosphere of the first threequarters of the century, we were astonished at first that any-one would risk raising this academically dangerous issue—least

of all someone with Griffin’s distinguished scientific credentials.His 1984 Animal Thinking and 1992 Animal Minds have widenedthe scope of his assault on conventional wisdom, but theshocked outrage in academia seems to have subsided into acivilized mixture of skepticism and interest. Maybe, after all,some animals might sometimes formulate simple plans. Buthow can we know?

What Criteria for Thought?

The kinds of behavior that Romanes found convincing nolonger seem very persuasive. For instance, the highly complexnest-building routines in birds and insects are known to belargely or entirely innate: individuals reared in isolation willnonetheless select appropriate nest sites, gather suitable mate-rial and fashion it into the kind of nest that wild-reared indi-viduals of the species create. True, nest building often improveswith practice and site selection benefits from experience, butthe basic elements of the behavior are in place before the ani-mal sets to work. Indeed, it is possible that the kinds of com-plex adaptive behavior that so impress us are just the types ofbehavior that must be innate, simply because they would beimpossible to learn from scratch.

And as the behaviorists showed, learning can be automatic,too. In fact, conditioning seems to involve an innate and com-plex weighing of probabilities—computing the chance that aparticular stimulus predicts the prompt appearance of an innatecue versus the chance it does not. Although it is possible thata duckling imprinting on its parents understands what it isdoing and why, the behavior of a young bird following a toytrain that was presented during the animal’s critical perioddoes not suggest any necessary comprehension. Thus, learningto modify behavior in the face of experience is not by itselfclear evidence of thinking.

Similarly, the frequently cited cases of apparent insight inanimals, such as the outbreak of cream-robbing behavior amongblue tits in England in the 1930s, may not mean what theyseem to mean. When unhomogenized milk was delivered to thedoorstep early each morning, a layer of cream would rise to thetop of the glass bottles. Blue tits, British cousins of the chick-adee, would remove the foil cap and sample the cream beforethe bottles were taken in. The inviting idea that some cagey


MR. VON OSTEN AND CLEVER HANS, his horse, stunnedthe world in the 1900s with the claim that Hans could doarithmetic, spell and even understand German. Hans wasactually responding to subtle cues of the people observing him.

CO

UR

TESY

OF

THE

NEW

YORK

PUB

LIC

LIB

RA

RY



bird had figured out this ploy and taught it to its friends ignoresthe natural history of the species: tits make their living peelingbark off trees to find insect larvae. So compulsive is their needto peel that hand-reared tits often strip the wallpaper from theirowners’ rooms in a presumably unrewarded search for insects.Perhaps the first blue tit to harvest cream from a milk bottlewas outstandingly stupid rather than amazingly bright, havingmistaken a bottle for a tree trunk.

Another common example is termite fishing among chimp-anzees. Some adult chimps strip long twigs of leaves and insertthem into the holes in termite mounds. When they withdrawthe twig, they eat the termites that cling to it. Photographs fre-quently show a younger chimp appearing to study the behaviorbefore trying it. But observations of lab-born chimpanzees revealthat chimps in general are obsessed with putting long, thinobjects into holes—pencils into electrical outlets, for instance.As with the blue tits, the behavior seems to be innate, and onlyknowing the proper place to perform it need be conditioned.

Early Hints of Thinking

To infer that an animal can think, therefore, enough mustbe known about the natural history and innate behavioral pro-pensities of the species, as well as the individual history of theanimal in question, to be able to exclude both instinct andconditioning as the source of a novel behavior. Before the cur-rent rebirth of interest in animal thinking, a few controversialstudies suggested that animals might be able to plan actionsin advance. They guide much of the experimental thinkingthat continues today.

In 1914 German psychologist Wolfgang Köhler was work-ing at a primate research center on the Canary Islands. He pre-sented his captive chimps with novel problems; often the pat-tern of solution suggested insight rather than trial and error.For instance, when Köhler first hung a bunch of bananas outof reach, the chimpanzee being observed made a few uselessleaps, then went off to a corner and “sulked.” But in time helooked back at the bananas, then around the large outdoorenclosure at the various objects he had to play with, back to thebananas, back to one specific toy (a box), then ran directly tothe box, dragged it under the fruit, climbed on top, leaped upand grabbed the prize.

In other variations the bananas were mounted higher,and the same pattern of seemingly sudden insight appeared,whether it involved stacking boxes, joining sticks to make apole long enough to knock down the fruit or using a single stickfrom atop one box. Criticisms of Köhler’s work focused on twoimportant points: the prior experience of these wild-caughtanimals was unknown (so they might be remembering a solu-tion they had learned in the wild), and lab-reared chimps spon-taneously pile boxes (which they then climb and use as jump-ing platforms) and also fit sticks together to make poles.

Well-controlled planning tests that avoided these problemswere performed by Edward C. Tolman of the University ofCalifornia at Berkeley in the 1940s. He would allow a rat toexplore an experimental maze with no differential reinforce-ment—a T-maze, for example, with the same food reward atthe end of each arm—but something the animal did not needto learn, and had not been trained to learn, would be differentat each end. In one instance, the left arm ended in a dark, nar-row box, whereas the right arm terminated in a wide, whitebox. (Rats inherently prefer dark, narrow boxes.) On anotherday the rat was taken to a different room, placed in a dark,

narrow box and electroshocked. On a subsequent day the ratwas returned to the original maze. Conditioning theory predictsthat the rat, not having been trained to any behavior in themaze, would explore at random. Alternatively, it might havelearned the location of the innately preferred dark, narrow box.The rat, however, went directly to the right end of the mazeand its white box.

The rat, Tolman concluded, had used two independent andapparently unrelated experiences to form a plan on the thirdday. He called this plan a “cognitive map.” Applying this per-spective to Köhler’s chimps, even if they had previously playedwith boxes and sticks in another context, to move the boxunder the bananas without any apparent trial and error wouldrequire enough insight to link the knowledge of what couldbe done with boxes to information on the desirability of ba-nanas. In the absence of trial-and-error conditioning, the chimpshad to conceive and execute a simple plan. This, at its mostbasic level, is evidence of animal thought.

Skeptics, however, found elaborate explanations forTolman’s results. For instance, one critic countered that rats areafraid of mazes; hence, removal of the rat from the maze whenit reached one of the end boxes was actually a reward that trig-gered learning. When returned to the maze, the rat knew itsbest chance of a reward (escape) was to go to the box it hadbeen taken from before. By chance, that was the white box.

Cognitive Maps

Although there is good evidence for cognitive maps increatures as phylogenetically remote as honeybees and jump-ing spiders, work on animal thinking has centered on birds

PROBLEM SOLVING IN CHIMPS, in this case, stackingboxes to reach bananas, was first documented byWolfgang Köhler around the time of World War I.

CO

UR

TESY

OF

THE

NEW

YORK

PUB

LIC

LIB

RARY



and primates. One important line of evidence is the apparentability of some animals to form concepts. The remarkable abili-ties of Alex the parrot, studied by Irene M. Pepperberg of theUniversity of Arizona, provide one clear example (see her arti-cle on page 60). Another, involving pigeons, was pioneered bythe late Richard J. Herrnstein of Harvard University (perhapsnow better known as the co-author of The Bell Curve). Histechnique was to provide lab-reared pigeons with a carousel ofslides, half with some example of the class of target objects—trees, perhaps, or fish or oak leaves. The birds were thenrewarded with food for pecking at any slide that contained,say, a tree. Learning was slow until the birds appeared tofigure out what the rewarded slides had in common. Undersome conditions, the pigeons would resort to memorizing thefull rewarded set of slides, revealing an astonishing ability torecall hundreds of pictures. In most cases, however, theycaught on to the common feature, demonstrating their knowl-edge by responding correctly to an entirely new set of slides.

Because of the huge range of variation among possibleexamples, a concept such as “tree” is difficult to formulate.There is no list of necessary and sufficient features, because we(and pigeons) recognize trees both with and without leaves,with and without central trunks, with and without substantialside branching, close up and far away, isolated or in densestands, with standard green or ornamental reddish leaves, andso on. For humans (and presumably for birds), a conceptincludes a list of properties that have individually predictiveprobabilities: leaves are highly correlated, for instance, where-as long, thin extensions have a lower (but still positive) associ-ation value. A tree is any object that has a sufficient “score” ofindividual properties—one high enough to exclude telephonepoles and television antennas. Many philosophers used toaccept concept formation as proof of thought; with the data onpigeons in hand, some have backed away from that criterion.

Many birds formulate and use mental maps of their homearea. The ability to devise a novel route to get from one familiarlocation to another is often taken as a literal example of a cog-nitive map. Whereas many, perhaps most, birds have local-areamaps, few are as spectacular as those of the African honeyguide,which makes its living feeding on the larvae and wax of bees.

The honeyguide has formed a symbiotic relationship withboth honey badgers (powerful, intelligent animals that are veryfond of honey) and humans. The bird locates a potential hiveopener—badger or human—and attempts to “recruit” it throughhighly visible and audible displays. One of the two most com-mon signals it uses is an onomatopoeic call that resembles thesound of tearing bark. Having engaged the attention of a suit-able helper, the honeyguide makes short flights in the approxi-mate direction of its target and calls again; if the helper failsto follow, the bird returns and tries again, perhaps with a shorterflight and louder calls this time. Once at the nest (which is gen-erally a quarter- to a half-mile away), the honeyguide waitswhile the badger or human it has led there breaks open thehive. The bird moves in for the larvae and wax after the hiveopener has left.

Numerous experiments have shown that honeyguidesknow the location of several hives and usually guide theiraccomplices to the nearest one—generally, by as direct a routeas the landscape allows. In one of the most interesting experi-ments, the human “helper” followed the bird but insisted onwalking steadily past the tree containing the hive. The honey-guide would first attempt to draw the person back to the tree;next the bird would change tactics and try to lead the human

on to another hive in the approximate direction of travel. Theseemingly inescapable conclusion is that honeyguides knowthe location of many colonies over a fairly wide area.

Distracting Predators and Getting Food

Flexibility in the use of innate alternatives may also be evi-dence for simple thinking. Two groups of ground-nesting birds,killdeers and plovers, have a variety of distraction ruses thatare used to lure potential predators away from their eggs. Eachdisplay begins with the bird leaving the nest and movinginconspicuously to a location well away from its eggs. The setof possible performances ranges from simply calling from ahighly visible spot to the complex feigning of a broken wing.There is even a highly realistic rodent-imitation ploy in whichthe bird scoots through the underbrush rustling provocativelyand uttering mouselike squeaks. Each species also has a separate“startle” display designed to keep harmless animals such asdeer from stumbling into the nest.

Anecdotal reports have suggested that the decision toleave the nest to perform a display, as well as which display toemploy, are suited to the degree of predator threat. A fox head-ing directly toward the nest, for example, is more likely to getthe high-intensity broken-wing performance. Carolyn Ristauof Columbia University put this reported ability to gaugethreats to a test by having distinctively dressed humans walkin straight lines near plover nests. Some were told to scan the

CREAM-ROBBING BEHAVIOR of blue tits, an outbreak ofwhich occurred in 1930s Britain, was probably not as inge-nious as it first seemed: the birds naturally peel bark offtrees in search of insect larvae.

J.M

ARK

HA

M B

ruce

Col

eman

Inc.



ground carefully, apparently searching for nests, whereas otherswere instructed to pay no attention to the ground. As timewent on, the plovers began to discriminate between the poten-tial hunters and the seemingly harmless humans: they did noteven bother to leave the nest for the latter group but performedelaborate distraction displays for the former. Some degree ofunderstanding seems evident in this ability to judge whichinnate response to select.

Two well-studied cases of unusual foraging behavior in birdsalso suggest an ability to plan. One involves green herons, birdsthat capture fish using several (presumably inborn) approaches.In addition, occasional herons have been observed bait fishing.They toss a morsel of food or a small twig into the water, andwhen a curious fish rises to investigate, the bird grabs it. Baitfishing has been observed in a few widely scattered spots in theU.S. and in a park in Japan. It appears on its own, seems(except once in Japan) not to spread to other birds and thenvanishes. Given the high success rate of the technique, andyet the rarity of its use, it is improbable that bait fishing is gen-etically programmed; most likely the trick has been indepen-dently invented by many different herons.

A more controlled study on the ontogeny of novel foragingtechniques was performed by Bernd Heinrich of the Universityof Vermont. He maintained five hand-reared ravens in a flightenclosure and thus knew just what kinds of learning opportu-nities had been available to the birds. He tested them withpieces of meat hung by strings from perches. The strings werefar too long to allow a raven on the perch to simply reachdown and grab the meat. The birds attempted to capture thefood in midair by flying up to it, but it was secured too wellfor this approach to work. After repeated failed attempts, theravens, like Köhler’s chimps, ignored the food.

Six hours after the test began, one raven suddenly solvedthe problem: it reached down, pulled up as much string as itcould manage, trapped this length of string in a pile betweenits foot and the perch, reached down again, trapped the nextlength of string under its claws and so on until it had hauledthe meat up to the perch. Again, as with Köhler’s chimps, therewas no period of trial and error.

After several days, a second raven solved the problem. Eventhough it had had ample opportunity to observe the first bird’srepeated successes, this individual pulled up the string, thenstepped along the perch, arraying the string in a line ratherthan a pile. It trapped the string under a foot, reached downand moved over again and again until the string was stretched

out along the perch and the meat was within reach. Otherbirds’ solutions of the problem involved looping the stringonto the perch. One bird never discovered how to obtain themeat; interestingly, this was also the one individual that neverlearned that flying away with the tied-down meat led to anasty jerk when the food reached the end of its tether.

Learning and Play

It may seem surprising that the ravens did not seem tolearn from one another but instead appeared to solve theproblem independently. In fact, there is very little evidencefor observational learning outside of primates. Most (somewould say all) of what researchers have initially assumed wasobservational acquisition of a technique has turned out to be“local enhancement”: learning where other animals congregate,not how they harvest what they find there. Bottle opening inblue tits spread quickly because the birds learned from otherswhere cream was to be found and had been born with the tech-nique for peeling back the lid.

As longtime cat and dog enthusiasts, we were originally asdubious as many readers probably are of the idea that animalsrarely, if ever, learn to copy behavior. A cat, for instance, that

TERMITE FISHING by chimpanzees is accomplished byinserting a stripped twig into a termite mound and then eatingthe insects that cling to it. As with blue tits, however, thebehavior seems innate, and only knowing the proper place toperform it needs to be conditioned.

GER

RY E

LLIS

EN

P Im

ages

FORMING A FUTURE PLAN,or a “cognitive map,” wasdemonstrated in rats in the1940s. On the first day, ratswere allowed to explore amaze that terminated withfood in a narrow, dark boxand in a wide, light box (ratsprefer dark boxes). On the nextday, the rats were taken to adark box and electroshocked.On the third day, the rats nav-igated to the light box, indicat-ing that the rats used twounrelated experiences to forma plan for the third day.

DAY 1 DAY 2 DAY 3

LAU

RIE

GR

AC

E



reaches for a doorknob apparently to obtain help in getting outseems to be imitating part of the human behavior associatedwith door opening. In fact, however, vertical stretches are partof the innate solicitation behavior of felines. What the cat mayhave had the insight to do is perform the solicitation at the doorwhen it wants to go out. This example, like so many others ofostensible copying, is a clever application of an innate or con-ditioned behavior in a novel location via local enhancement.

About the only possible exception to the local-enhance-ment theory outside of primates comes from work on octopuses.In one test done in 1992, an untrained octopus on one side ofa glass partition was allowed to view a trained octopus on theother side choose between two objects that differed in color.Later, when provided with the same choice, the observingoctopus selected the same-colored object as the “teacher” ani-mal had about 10 times as often. In other, more demandingexperiments, the observer octopus watched while a trainedoctopus in the adjoining compartment opened a containerholding food. When tested, the “student” could open a similarcontainer (using the same technique) much sooner than anuntrained peer. These findings, however, are still preliminary;researchers have had difficulty in verifying them fully.

One behavior that intelligent animals appear to share isplay, defined as performing seemingly pointless but energeticbehaviors that human observers consider (introspectively)playful. Octopuses, for instance, jet at floating objects. Parrotswill swim (they prefer the backstroke) and make snowballs.Dolphins and whales leap high in the air for no obvious reasonand engage objects on the surface. Ravens will toss rocks toone another in midair—it looks to be a game of catch—andrepeatedly slide down snowbanks. And primates are famousfor their antics—hanging upside down from a limb over astream to splash water noisily or covering their eyes with broadleaves to play blindman’s buff high overhead in trees. Althoughwe cannot know what is going on in the minds of these crea-tures, even the most hardened observer must wonder if theseare not intelligent but bored animals injecting some excitementinto their lives.

Evidence for Primate Thinking

One important aspect of logic is the ability to recognizeand act on relations between objects and individuals. Perhapsthe first well-documented example of how much animals knowabout one another came from the work of Robert M. Seyfarth

and Dorothy L. Cheney, now at the University of Pennsylvania,on vervet monkeys in Africa [see “Meaning and Mind inMonkeys,” by Robert M. Seyfarth and Dorothy L. Cheney;SCIENTIFIC AMERICAN, December 1992]. Dominance interactionshad already suggested that each monkey understood the posi-tion of every other vervet in the troop hierarchy. Seyfarth andCheney performed numerous ingenious experiments to inves-tigate what individual vervets knew. They discovered that themonkeys also kept track of each infant’s mother and her socialstatus. When the researchers played recordings they had madeof various infants’ distress calls, the infant’s mother would lookin the direction of the hidden loudspeaker, while all the otherfemales would look at the mother.

The logical operations that support the behavior specificto individual vervets are essential to the social calculus of thegroup. Studies by Frans B. M. de Waal of Yerkes RegionalPrimate Research Center and Emory University on the intri-cate social maneuvering that goes on in captive chimpanzeetroops show how this kind of knowledge can be exploited.More dramatic, however, are his descriptions of the chimpan-zees’ use of deceit, which had been reported from the wild bythe naturalist Jane Goodall.

The master of dissimulation in one of de Waal’s troops wasa (then) low-ranking male named Dandy. Usually alpha malesdo not permit other males to mate with females. Dandy andhis special female friend would meet as if by chance behindrocks or brush. The simultaneous disappearance of a female anda low-ranking male usually provokes suspicion in alphas, butDandy and his date would choose cover that hid only theirlower bodies. They would mate while pretending to forage, andthe female would suppress the shrieks that accompany typicalchimpanzee intercourse.

Dandy also took advantage of distractions to mate orwould even create them himself, as when he once rushed tothe front of the enclosure and began screaming at the passinghumans. The alphas hurried to see what was going on, andDandy slipped away in the confusion. Another time Dandyobserved a low-ranking male courting his own special female.Instead of throwing a tantrum—the usual response to thiskind of affront—Dandy fetched the nearest alpha and allowedhim to deal decisively with the transgressor.

NOVEL FORAGING TECHNIQUE can develop in animals.Green herons have been seen to toss food or twigs into thewater as bait to attract fish to the surface (left). In a con-

PHO

TOG

RA

PHS

BY

ROB

ERT

F.SI

SSO

N N

atio

nal G

eogr

aphi

c So

ciet

y


These and many other instances of Dandy’s expert use ofsocial logic imply an ability to think and plan—even scheme—at an impressive level. Combined with other extensive obser-vations of chimpanzees in both the wild and more controlledseminatural enclosures, these examples strongly implicate anevolutionary continuity in the ability to analyze situations andimagine solutions. Without words, the mental operationsinvolved may be of necessity pictorial; language doubtless per-mits our species to contrive far more elaborate plans, not tomention fantasies and self-delusions.

Humans in Perspective

Language has played a dominant role in discussions aboutthinking and consciousness. Some philosophers go so far as toassert that language is uniquely and essentially human, a cre-ation of a conscious intellect, a tool necessary to planning andthought. The discovery of symbolic languages in animals rang-

ing from vervets to honeybees has been sobering; the almostoverwhelming evidence for nonverbal planning has furtherblunted the authority of such sweeping generalizations. Butnothing has been as deflating to the human self-image as thediscovery that consonant recognition, language processing andeven grammar are largely innate [see “The Perception of Speechin Early Infancy,” by Peter D. Eimas; SCIENTIFIC AMERICAN,January 1985; “Creole Languages,” by Derek Bickerton;SCIENTIFIC AMERICAN, July 1983; and “Specializations of theHuman Brain,” by Norman Geschwind; SCIENTIFIC AMERICAN,September 1979].

Our uniqueness as a species, it would seem, depends on agenetic specialization not obviously more elaborate than theone that confers the power of echolocation on bats. But lan-guage empowers what already appears to be a phylogeneticallywidespread ability to reason and plan with an evolutionarilynew capacity for elaboration, communication and coordina-tion that has catapulted our species into a position of aston-ishing intellectual potential. When we look at the fascinatingfauna with which we share the planet, we should recall thatbut for the fickle logic of evolution, our species would be justanother variety of conniving, inarticulate primates.


CLA

REH

.GO

ULD

About the Authors JAMES L. GOULD and CAROL GRANT GOULD have co-authored manyarticles and books on animal behavior. James Gould became interested in ani-mal behavior in the 1960s, when he was a young draftee at a U.S. militarybase in Germany. Spending his off-duty hours in the library, he one daypicked up a copy of King Solomon’s Ring, by Konrad Lorenz, the Austrian ethol-ogist perhaps best known for his studies on imprinting in birds. When Gouldreturned to college at the California Institute of Technology, he signed up fora course in animal behavior. He went on to get his Ph.D., studying learning inhoneybees with Donald R. Griffin at the Rockefeller University. Although heenjoys working on animal cognition, Gould doesn’t recommend cutting one’sacademic teeth on such a controversial problem. “Studying animal intelli-gence is no longer tantamount to professional suicide, but it used to be,” hesays. As a professor of ecology and evolutionary biology at PrincetonUniversity, Gould is now studying how females choose their mates—particu-larly among guppies, mollies and other tropical fish.

Carol Grant Gould got her Ph.D. in Victorian literature from New YorkUniversity and often joins Jim in the field, whether tracking birds in upstateNew York or watching whales in Argentina. “When you’re married to a fieldbiologist, you have to become a biologist by association if you ever want tosee your spouse,” she says. “Fortunately, learning biology, like studying litera-ture, opens fascinating new windows on life.” Carol is a science writer by pro-fession and teaches English at a private high school in Princeton, N.J. Besideswriting projects, they have also collaborated to produce two kids and enjoybiking, canoeing and hanging out in Princeton.

trolled study, ravens had to figure out how to retrieve a pieceof meat dangling on a string from the perch (right). Mostdeveloped unique ways to pull up the string.

SA

BER

ND

HEI

NRI

CH



TalkingwithAlex:

Parrots were once thought to be no more than excellent mimics, but research is showing that they understand what they say.Intellectually, they rival great apes and marine mammals

Logic by Irene M. Pepperberg


Bye. I’m gonna go eat dinner. I’ll see you tomorrow,” Ihear Alex say as I leave the laboratory each night. What makesthese comments remarkable is that Alex is not a graduate stu-dent but a 22-year-old Grey parrot.

Parrots are famous for their uncanny ability to mimichuman speech. Every schoolchild knows “Polly wanna cracker,”but the general belief is that such vocalizations lack meaning.Alex’s evening good-byes are probably simple mimicry. Still, Iwondered whether parrots were capable of more than mind-less repetition. By working with Alex over the past two decades,I have discovered that parrots can be taught to use and under-stand human speech. And if communication skills provide aglimpse into an animal’s intelligence, Alex has proved thatparrots are about as smart as apes and dolphins.

When I began my research in 1977, the cognitive capacityof these birds was unknown. No parrot had gone beyond thelevel of simple mimicry in terms of language acquisition. Atthe time, researchers were training chimps to communicatewith humans using sign language, computers and special boardsdecorated with magnet-backed plastic chips that representwords. I decided to take advantage of parrots’ ability to pro-duce human speech to probe avian intelligence.

My rationale was based on some similarities between par-rots and primates. While he was at the University ofCambridge, Nicholas Humphrey proposed that primates hadacquired advanced communication and cognitive skillsbecause they live and interact in complex social groups. Ithought the same might be true of Grey parrots (Psittacuserithacus). Greys inhabit dense forests and forest clearingsacross equatorial Africa, where vocal communication plays animportant role. The birds use whistles and calls that they mostlikely learn by listening to adult members of the flock.

Further, in the laboratory parrots demonstrate an abilityto learn symbolic and conceptual tasks often associated withcomplex cognitive and communication skills. During the1940s and 1950s, European researchers such as Otto D. W.Koehler and Paul Lögler of the Zoological Institute of theUniversity of Freiburg had found that when parrots are exposedto an array of stimuli, such as eight flashes of light, some ofthem could subsequently select a set containing the same num-ber of a different type of object, such as eight blobs of clay.Because the birds could match light flashes with clay blobs onthe basis of number alone means that they understood arepresentation of quantity—a demonstration of intelligence.

But other researchers, including Orval H. Mowrer, foundthat they were unable to teach these birds to engage in referen-tial communication—that is, attaching a word “tag” to a partic-ular object. In Mowrer’s studies at the University of Illinois, aparrot might learn to say “hello” to receive a food reward when

its trainer appeared. But the same bird would also say “hello”at inappropriate times in an attempt to receive another treat.Because the parrot was not rewarded for using the word incor-rectly, eventually it would stop saying “hello” altogether. Someof Mowrer’s parrots picked up a few mimicked phrases, butmost learned nothing at all.

Because parrots communicate effectively in the wild, itoccurred to me that the failure to teach birds referential speechmight stem from inappropriate training techniques ratherthan from an inherent lack of ability in the psittacine subjects.For whatever reason, parrots were not responding vocally tothe standard conditioning techniques used to train other speciesto perform nonverbal tasks. Interestingly, many of the chim-panzees that were being taught to communicate with humanswere not being trained with the standard paradigms; perhapsparrots would also respond to nontraditional training. To testthis premise, I designed a new method for teaching parrots tocommunicate.

Go Ask Alex

The technique we use most frequently involves two hu-mans who teach each other about the objects at hand whilethe bird watches. This so-called model/rival (M/R) protocol isbased, in part, on work done by Albert Bandura of StanfordUniversity. In the early 1970s Bandura showed that childrenlearned difficult tasks best when they were allowed to observeand then practice the relevant behavior. At about the same time,Dietmar Todt, then at the University of Freiburg, independentlydevised a similar technique for teaching parrots to replicatehuman speech.

In a typical training session, Alex watches the trainer pickup an object and ask the human student a question about it:for example, “What color?” If the student answers correctly,he or she receives praise and is allowed to play with the objectas a reward. If the student answers incorrectly, however, thetrainer scolds him or her and temporarily removes the objectfrom sight. The second human thus acts as a model for Alexand a rival for the trainer’s attention. The humans’ interac-tions also demonstrate the consequences of an error: themodel is told to try again or to talk more clearly.

We then repeat the training session with the roles of train-er and model reversed. As a result, Alex sees that communica-tion is a two-way street and that each vocalization is not specificto an individual. In Todt’s studies, birds were exposed only topairs of individuals who maintained their respective roles. Asa result, his birds did not respond to anyone other than thehuman who initially posed the questions. In contrast, Alexwill respond to, interact with and learn from just about anyone.


TIM

OTH

Y A

RCH

IBA

LD

and Speech in ParrotsCopyright 1998 Scientific American, Inc.

The fact that Alex works well with different trainers suggeststhat his responses are not being cued by any individual—oneof the criticisms often raised about our studies. How could anaive trainer possibly cue Alex to call an almond a “corknut”—his idiosyncratic label for that treat?

In addition to the basic M/R system, we also use supple-mental procedures to enhance Alex’s learning. For example,once Alex begins to produce a word describing a novel item,we talk to him about the object in full sentences: “Here’s thepaper” or “You’re chewing paper.” Framing “paper” within asentence allows us to repeat the new word frequently andwith consistent emphasis, without presenting it as a single,

repetitive utterance. Parents and teachers often usesuch vocal repetition and physical presentation ofobjects when teaching young children new words.We find that this technique has two benefits. First,Alex hears the new word in the way that it is usedin normal speech. Second, he learns to produce theterm without associating verbatim imitation of histrainers with a reward.

We also use another technique, called referentialmapping, to assign meaning to vocalizations thatAlex produces spontaneously. For example, afterlearning the word “gray,” Alex came up with theterms “grape,” “grate,” “grain,” “chain” and “cane.”Although he probably did not produce these specificnew words intentionally, trainers took advantage ofhis wordplay to teach him about these new itemsusing the modeling and sentence-framing proce-dures described earlier.

Finally, all our protocols differ from those usedby Mowrer and Todt in that we reward correctresponses with intrinsic reinforcers—the objects towhich the targeted questions refer. So if Alex cor-

rectly identifies a piece of wood, he receives a piece of woodto chew. Such a system ensures that at every interaction, thesubject associates the word or concept to be learned with theobject or task to which it refers. In contrast, Mowrer’s pro-grams relied on extrinsic reinforcers. Every correct answerwould be rewarded with a preferred food item—a nut, forexample. We think that such extrinsic rewards may delaylearning by causing the animal to confuse the food item withthe concept being learned.

Of course, not every item is equally appealing to a parrot.To keep Alex from refusing to answer any question that doesn’tinvolve a nut, we allow him to trade rewards once he has cor-

rectly answered a question. If Alex correctlyidentifies a key, he can receive a nut—a more desir-able item—by asking for it directly, with a simple “Iwanna nut.” Such a protocol provides some flexibili-ty but maintains referentiality of the reward.

What’s Different, What’s the Same

I began working with Alex when he was 13months old—a baby in a species in which individualslive up to 60 years in captivity. Through his years oftraining Alex has mastered tasks once thought to bebeyond the capacity of all but humans and certainnonhuman primates. Not only can he produce andunderstand labels describing 50 different objectsand foods but he also can categorize objects bycolor (rose, blue, green, yellow, orange, gray or pur-ple), material (wood, wool, paper, cork, chalk, hideor rock) and shape (objects having from two to sixcorners, where a two-cornered object is shaped likea football). Combining labels for attributes such ascolor, material and shape, Alex can identify, request


ALEX CAN IDENTIFY ITEMS on a tray by shape,color, substance or quantity.

Animal Intelligence

TARGET OBJECTS vary and include blocks, letters,numbers and small toys.PH

OTO

GR

APH

S B

Y W

ILLI

AM

MU

ÑO

Z


and describe more than 100 different objects with about 80percent accuracy.

In addition to understanding that colors and shapes repre-sent different types of categories and that items can be catego-rized accordingly, Alex also seems to realize that a single objectcan possess properties of more than one category—a green trian-gle, for example, is both green and three-cornered. When pre-sented with such an object Alex can correctly characterize eitherattribute in response to the vocal queries “What color?” or“What shape?” Because the same object is the subject of both

questions, Alex must change his basis for classification to answereach query appropriately. To researchers such as Keith J. Hayesand Catherine H. Nissen, who did related work with a chim-panzee at the Yerkes Regional Primate Research Center at EmoryUniversity, the ability to reclassify items indicates “abstract apti-tude.” On such tests, Alex’s accuracy averages about 80 percent.

Alex has also learned the abstract concepts of “same” and“different.” When shown two identical objects or two itemsthat vary in color, material or shape, Alex can name whichattributes are the same and which are different. If nothing

Exploring Intelligence 63Talking with Alex: Logic and Speech in Parrots

TRANSCRIPTS OF DIALOGUES indicatethat Alex can count objects on a tray.Dialogue 1, recorded in 1986, shows thatAlex can distinguish five objects of twodifferent types—in this case, plant stakesand keys. Dialogue 2, from 1997, revealsthat Alex has become more sophisticatedin his ability: presented with a more com-plex set of objects (photograph), Alexcan count the number of blue blocks andgreen wool balls without being distractedby the other items on the tray.

DIALOGUE 1

Alex is shown two plant stakes and three keys on a tray.

Trainer: How many key?Alex: Wood

Irene (with back to tray, to trainer): Are there any wood?Trainer (to Irene): Yes.

Irene: Try that.Trainer: Okay, tell me, how many wood?

Alex: Two.Irene: Two?

Trainer: Yes.

Alex is given one stake, which he chews apart. It is replaced, and the trayis presented again.

Trainer: Now, how many key?Alex: Key.

Trainer: That’s right, keys. How many?Alex: Two wood.

Trainer: There are two wood, but you tell me, how many key?Alex: Five.

Trainer: Okay, Alex, that’s the number of toys; you tell me, how many key?

Alex: Three.Irene: Three?

Trainer: Good boy! Here’s a key.

DIALOGUE 2

Irene: Okay, Alex, here’s your tray. Will you tell me howmany blue block?

Alex: Block.Irene: That’s right, block…how many blue block?Alex: Four.

Irene: That’s right. Do you want the block?Alex: Wanna nut.

Irene: Okay, here’s a nut. (Waits while Alex eats the nut.)Now, can you tell me how many green wool?

Alex: Sisss...Irene: Good boy!

WIL

LIA

M M

UÑ

OZ


about the objects is the same or different, he replies, “None.”He responds accurately even if he has not previously encoun-tered the objects, colors, materials or shapes.

Alex is indeed responding to specific questions and not justrandomly chattering about the physical attributes of the objects.When presented with a green, wooden triangle and a blue,wooden triangle, his accuracy was above chance on questionssuch as “What’s same?” If Alex were ignoring the question andresponding based on his prior training, he might have respond-ed with the label for the one anomalous attribute—“color”—rather than either of the correct answers—“matter” or “shape.”

Alex’s comprehension matches that of chimpanzees anddolphins. He can examine a tray holding seven differentobjects and respond accurately to questions such as “Whatcolor is object-X?” or “What object is color-Y and shape-Z?” Acorrect response indicates that Alex understood all parts of thequestion and used this understanding to guide his search for theone object in the collection that would provide the requestedinformation. His accuracy on such tests exceeds 80 percent.

We also used a similar test to examineAlex’s numerical skills. He currently usesthe terms “two,” “three,” “four,” “five” and“sih” (the final “x” in “six” is a difficultsound for a parrot to make) to describequantities of objects, including groupingsof novel or heterogeneous items. When weshow Alex a “confounded number set”—acollection of blue and red keys and toy cars,for example—he can correctly answer ques-tions about the number of items of a partic-ular color and form, such as “How manyblue key?” His accuracy in this test, 83.3percent, equals that of adult humans whoare given a very short time to quantify simi-larly a subset of items on a tray, accordingto work done by Lana Trick and ZenonPylyshyn of the University of WesternOntario.

Alex also comprehends at least one rela-tive concept: size. He responds accuratelyto questions asking which of two objects isthe bigger or smaller by stating the color ormaterial of the correct item. If the objectsare of equal size, he responds, “None.”Next, we will try to get Alex to tackle rela-tive spatial relations, such as over andunder. Such a proposition presents anadded challenge because an object’s posi-tion relative to a second object can change:what is “over” now could be “under” later.

One last bit of evidence reinforces ourbelief that Alex knows what he is talkingabout. If a trainer responds incorrectly tothe parrot’s requests—by substituting anunrequested item, for example—Alex gener-ally responds like any dissatisfied child: hesays, “Nuh” (his word for “no”), and

repeats his initial request. Taken together, these results stronglysuggest that Alex is not merely mimicking his trainers but hasacquired an impressive understanding of some aspects ofhuman speech.

Tricks of the Training

What is it about our technique that allows Alex to masterthese skills? To address that question, we enlisted a few yearsago the help of Alo, Kyaaro and Griffin—three other juvenileGrey parrots. Of the many different variations on our techniquewe tried with these parrots, none worked as well as the two-trainer interactive system.

We attempted to train Alo and Kyaaro using audiotaperecordings of Alex’s training sessions. The birds also watchedvideo versions of Alex’s sessions while they were in isolation(with an automated system providing rewards) or in the pres-ence of trainers who were slightly interactive. Griffin viewed


Wood

2

3

2

4

4

2

Rock

Wool

2

Keys

Wood

Rocks

Keys

Wood

Jacks

Yellow wool

Corks

Rocks

Wood

Object set Target items CorrectAlex's response

ALEX’S ACCURACY in identifying the num-ber of targeted items in 1986 was 70 percent(seven out of 10 questions); now Alex is cor-rect more than 80 percent of the time.

GEO

RGE

RETS

ECK


the same videos in the presence of a highly interactive humantrainer who rephrased material on the video and questionedthe bird directly. Although all three parrots occasionally mim-icked the targeted labels presented in the interactive video ses-sions, they failed to learn referential speech in any of thesesituations.

When we then trained these birds using the standard M/Rprotocol, their test scores improved dramatically. In the pasttwo years Griffin, for example, has acquired labels for sevenobjects and is beginning to learn his colors. The parrots’ fail-ure to learn from the alternative techniques suggests thatmodeling and social interactions are important for maintain-ing the birds’ attention during training and for highlightingwhich components of the environment should be noted, hownew terms refer to novel objects and what happens whenquestions are answered correctly or incorrectly. All these con-cepts are critical in training birds to acquire some level ofhuman-based communication.

The M/R technique and some variants have also provedvaluable in teaching other species referential communication.Diane Sherman of New Found Therapies in Monterey, Calif.,uses the M/R technique for teaching language skills to devel-opmentally delayed children. Even Kanzi, the bonobo (pygmychimpanzee) trained by Sue Savage-Rumbaugh and her col-leagues at Georgia State University, initially learned to com-municate with humans via computer by watching his motherbeing trained—a variant of our modeling technique. Kanzi’sabilities are probably the most impressive of all primates’trained to date. Chimpanzees have been taught human-basedcodes through a variety of techniques; however, apes thatwere trained using protocols similar to those developed byMowrer demonstrated communication skills that were far lessflexible and less “languagelike” than those of apes trainedusing systems that had more in common with our techniques.

Bird Brains

Alex continues to perform as well as apes and dolphins intests of intellectual acuity, even though the structure of theparrot brain differs considerably from that of terrestrial and

aquatic mammals. Unlike primates, parrots have little graymatter and thus not much of a cerebral cortex, the brainregion associated with cognitive processing in higher mam-mals. Other parts of Alex’s brain must power his cognitivefunction.

The parrot brain also differs somewhat from that of song-birds, which are known for their vocal versatility. Yet Alex hassurpassed songbirds in terms of the relative size of their“vocabularies.” In addition, he has learned to communicatewith members of a different species: humans. With each newutterance, Alex and his feathered friends strengthen the evi-dence indicating that parrots are capable of performing com-plex cognitive tasks. Their skills reflect the innate abilities ofparrots and suggest that we should remain open to discover-ing advanced forms of intelligence in other animals.


IRENE M. PEPPERBERG’s work is for the birds—or sothe funding agencies first thought. “My early grants cameback with pink sheets basically asking what I was smoking,”she jokes. Pepperberg actually trained as a theoreticalchemist: as a Ph.D. student at Harvard University, she gener-ated mathematical models to describe boron compounds.But an episode of Nova featuring “signing” chimps, singingwhales and squeaking dolphins drew her to her currentwork. “I was fascinated to see that people could study animalbehavior as a career,” she says. Now Pepperberg is an associ-ate professor at the University of Arizona at Tucson, a citythat brings tears to her eyes—literally. “I’m allergic to every-thing that grows in Tucson,” Pepperberg says of the trees,grasses, molds and weeds. In 1997 she used the funds from aJohn Simon Guggenheim Memorial Foundation fellowship towrite a book on parrot cognition and communication, InSearch of King Solomon’s Ring: Studies on the Communicativeand Cognitive Abilities of Grey Parrots (currently in press).

Alex also has a life in publishing—he is the title charac-ter in Alex and Friends, a children’s book about the animalsthat have learned to communicate with humans. Throughthe Internet, you can order a special copy—one thatPepperberg has signed and Alex has chewed. It is available atwww.azstarnet.com/nonprofit/alexfoundation/ on the WorldWide Web.

SA

MODEL/RIVAL PROTOCOL used to teach Alex has the train-er ask the student about objects; a correct response by the stu-dent earns praise and possession of the object. In this way, thestudent acts as a model for Alex and a rival for the trainer’sattention. Roles are often reversed to demonstrate that thesame person is not always the questioner.

PHO

TOG

RA

PHS

BY

WIL

LIA

MM

UÑ

OZ

About the Author

Talking with Alex: Logic and Speech in Parrots



I used to tell students that no one ever heard, saw,tasted or touched a mind. There is no way for me toexperience your experience, let alone that of a speciesother than my own. So although minds may exist, theyfall outside the realm of science.

I have since changed my mind. A number of yearsago I began to study whether primates could recognizethemselves in a mirror. Most animals react to theirimages as if confronted by another animal. But chim-panzees, orangutans and, of course, humans learn thatthe reflections are representations of themselves—thesecreatures are objects of their own attention and areaware of their own existence. In the past three decades,I and other researchers have used the mirror test in vari-ous ways to explore self-awareness in animals. I con-clude that not only are some animals aware of them-selves but that such self-awareness enables these animalsto infer the mental states of others. In other words,species that pass the mirror test are also able to sympa-thize, empathize and attribute intent and emotions inothers—abilities that some might consider the exclusivedomain of humans.

I began exploring self-awareness with mirrors in1969, when I was at Tulane University. I presented afull-length mirror to preadolescent chimpanzees at theuniversity’s Delta Regional Primate Research Center.Initially, they reacted as if they were seeing other chim-panzees, but after a few days they grew accustomed to

the mirror and began to use it to make faces, look at theinside of their mouths, and groom and inspect otherparts of their bodies that they had never seen before.

The Mirror Test

To determine whether they had learned to recognizetheir own reflections, I anesthetized each animal andapplied red dye to an eyebrow ridge and to the top halfof the opposite ear. Later, on awakening and seeing them-selves in the mirror, the chimpanzees reached up andtouched the red marks on their faces, following this insome instances with looking at and smelling their fingers.Chimpanzees that did not have the benefit of prior expe-rience with mirrors acted as if confronted by anotherchimpanzee and failed to locate the marks on their faces.These findings of self-recognition have now been repli-cated with chimpanzees more than 20 times by scientistsall over the world.

Many other animals, including a variety of primates,elephants, birds and even dolphins, have been tested forself-recognition. But only chimpanzees, orangutans andhumans have consistently passed this test. (Marc D. Hauserof Harvard University reported that cotton-top tamarinspass the mirror test when their white tufts of hair aremarked, but no one has been able to replicate these results.)

The failure to find self-recognition in other animalsis not for want of trying. Susan D. Suarez of the Sage

Animals that pass the mirrortest are self-aware and thuscan infer the states of mindof another individual

Yes

Can Animals

by Gordon Gallup, Jr.

Continued on page 68


Exploring Intelligence 67Can Animals Empathize?

Let me begin with a point on which Gordon Gallup,Jr., and I agree: the reactions of chimpanzees when theysee themselves in mirrors reveal that these animals pos-sess a self-concept. Furthermore, we agree that this self-concept appears to be restricted to the great apes andhumans. Beyond this point, however, our views diverge.Gallup speculates that the capacity for self-recognitionmay indicate that chimpanzees are aware of their owninternal, psychological states and understand that otherindividuals possess such states as well. I have come todoubt this high-level interpretation of the chimpanzees’reactions to seeing themselves in mirrors. More generally,I question whether chimpanzees possess the deep psy-chological understanding of behavior that seems socharacteristic of our species. In what follows, I describewhy I have come to this conclusion, and I offer an expla-nation of how humans and chimpanzees can behave sosimilarly and yet understand this behavior in radicallydifferent ways.

Knowing That Others See

Consider the simple act of seeing. When we witnessother people turning their eyes toward a particular object,we automatically interpret this behavior in terms of theirunderlying psychological states—what they are attendingto, what they are thinking about, what they know orwhat they intend to do next. These inferences are often

solely based on fairly subtle movements of their eyesand heads.

Do chimpanzees understand seeing in this manner?Gallup thinks they do, and at first glance it seems hard todeny it. For example, chimpanzees exhibit a strong inter-est in the eyes of their fellow apes. Frans B. M. de Waalof the Yerkes Regional Primate Research Center at EmoryUniversity has reported that chimpanzees do not appearto trust the reassurance gestures of their former opponentsunless such gestures are accompanied by a mutual gaze—that is, unless they stare directly into one another’s eyes.Research from our own laboratory has established thatchimpanzees follow the gaze of other apes—and ofhumans as well. If you stand face-to-face with a chimp,lock your gaze with hers and then suddenly look overher shoulder, the ape will reliably turn around, as if try-ing to determine what you are looking at.

In short, the spontaneous behavior of chimpanzeesseems to make a fairly persuasive case that they can rea-son about the visual perspectives of others. Does thisbehavior, then, provide confirmation of Gallup’s model?Maybe, but maybe not. The problem is that there areother equally plausible interpretations that do not assumethat chimpanzees are reasoning about one another’svisual experiences. The case of gaze following illustratesthe problem quite well. A chimpanzee who follows yourgaze leads you to assume that the animal is trying tofigure out what you are looking at. But what excludes

by Daniel J. Povinelli

Maybe notEven though chimpanzees pass the mirror test, they do not seem to conceive ofothers’—or even their own—mental states

DO

NN

A B

IERS

CH

WA

LE U

SL N

ew Ib

eria

Res

earc

h C

ente

r

Continued on page 72

Empathize?



Colleges and I gave a pair of rhesus mon-keys, reared together in the same cage,continuous exposure to themselves in afull-length mirror for 17 years (morethan 5,000 hours of mirror exposure ayear). Despite this extended opportunityto learn about the mirror, neither monkeyever showed any evidence of self-recogni-tion. On the other hand, when I wouldwalk into the room where they were keptand they saw my reflection in the mirror,they would immediately turn to confrontme directly. So it was not that they wereincapable of learning to interpret mirroredinformation about other objects correctly.

Experiments have also failed touncover compelling evidence of self-recog-nition in gorillas. After pondering those results, Suarez and Idecided to give gorillas the benefit of the doubt, reasoningthat maybe gorillas do not care about the superimposed marks.We tested this hypothesis at the Yerkes Regional Primate ResearchCenter at Emory University by applying marks to gorillas’ wristsas well as to their faces. We discovered that on recovery fromanesthesia all the gorillas touched and inspected the marks ontheir wrists. But despite extensive prior experience with mirrors,none of the gorillas were able to locate comparable marks ontheir faces that could be seen only in the mirror.

Gorillas naturally avoid making eye contact with oneanother, so a possible reason for their mirror-test failure is thatthey avoid eye contact with their reflection and hence never

learn to recognize themselves. Daniel J. Shillito and Benjamin B.Beck of the National Zoological Park in Washington, D.C., and Irecently tested this hypothesis, relying on a technique developedby James R. Anderson of the University of Stirling in England.It calls for a pair of mirrors placed together at an angle thatrenders it impossible to make eye contact with the reflection.But none of the gorillas showed evidence of self-recognition, noteven one that had more than four years of exposure to mirrors.

In other tests of learning, problem solving and cognitivefunctioning, differences in performance among species are typ-ically a matter of degree, not kind. What is to be made of suchdecisive differences in self-recognition? Maybe the reason mostspecies cannot process mirrored information about themselvesstems from an inability to conceive of themselves. Correctlyinferring the identity of the reflection presupposes an identityon the part of the organism making that inference.

That conclusion seems reasonable, considering the waymembers of Homo sapiens interpret mirror images. Humans donot begin to show compelling evidence of mirror-guided self-recognition until they reach 18 to 24 months of age—about thesame time at which the prefrontal cortex begins to mature in

structure and function. Younger infants react to themselves inmirrors as though they were seeing other children, just as mostspecies do. At about the time that children learn to recognizethemselves, they begin to show other evidence of self-concep-tion, such as using personal pronouns, smiling after masteringa task and engaging in self-conscious play.

Before about two years of age, no one has experiences thatcan be consciously recalled in later life. Consistent with myinterpretation, this period of “infant amnesia” stops at about

the same time that children begin toshow self-recognition. As would beexpected, the onset of an autobiographi-cal memory only begins with the emer-

gence of self-conception.That may terminate prematurely at the other end of the life

span if dementia sets in. Disturbances in self-awareness andimpaired structure and function of the prefrontal cortex oftenaccompany this condition. Thus, for some, human developmentmay be bounded at both ends by periods of unconsciousness.

Knowing Mental States

Some practical advantages are derived from being able toconceive of the self. I argue that self-awareness, consciousnessand mind are an expression of the same underlying process, sothat organisms aware of themselves are in a unique positionto use their experience as a means of modeling the experienceof others. When you see someone in a situation similar to oneyou have encountered, you automatically assume his or herexperience will be similar to yours. Although it is probablytrue that no two people experience the same event in exactlythe same way, as members of the same species we share thesame sensory and neurological mechanisms. So there is boundto be considerable overlap between your experience and mine.

We ought to be able to identify animals that can or cannot recognizethemselves in mirrors and their empathetic tendencies.

GALLUPYes

SELF-RECOGNITION is evident when chimpanzees touch thered dot painted on their faces. Once familiar with the mirrors,they will inspect themselves and make faces.

PHO

TOG

RA

PHS

BY

DO

NN

A B

IERS

CH

WA

LE

Continued from page 66



Moreover, given a knowledge of how external events influencemy mental states (and vice versa), I have a means of modelingthe mental states of others.

To see my point, imagine you have a dog that returns homeone day in obvious distress: it has porcupine quills in its nose.You could either have a veterinarian remove the quills, or youcould attempt to extract them yourself using a pair of pliers. Ifyou were to opt for the latter, it would be an excruciating ordealfor you. Not that you would experience any pain in the process,but as you pulled the quills from the dog’s nose and witnessedits reaction, it would prove virtually impossible not to empa-thize with the dog. That is, you would use your prior experi-ence with pain to model your dog’s ostensible experience.

But how do you think another unrelated dog witnessingthis transaction would respond? Pet owners may be surprisedto learn—and any veterinarian can tell you—that dogs areempathetically oblivious to pain and suffering in other, unre-lated dogs. I suspect that dogs experience pain in much thesame way that we do, but because they cannot conceive ofthemselves, dogs cannot use their experience with pain tomodel painful experiences in other creatures. (They might, ofcourse, react to the yelping.)

Another way to illustrate this incapacity involvespeople who have a condition called blindsight experi-ence. These patients have sustained extensive damageto the visual cortex and often act as if they were blind,even though their primary visual system remains intact.Lawrence Weiskrantz of the University of Oxford andhis colleagues discovered that such patients can show asurprising ability to “guess” the identity of objects andtheir location. In other words, vision in such patientshas been reduced to an unconscious sensation. Blind-sight patients can still respond to visual information,but they are not aware of it. As a consequence, they havebeen rendered mindless when it comes to vision. Iwould predict that individuals born with blindsight cangrow up using guessing strategies and hence act visuallynormal. Their condition would become apparent only ifthey were placed in a situation that required them tomake inferences about the visual experiences in otherpeople—understanding how high-beam headlights

affect oncoming drivers on a dark country road, for instance.So back to my main point: I maintain that knowledge of

mental states in others presupposes knowledge of mental statesin oneself and, therefore, that knowledge of self paves the wayfor an inferential knowledge of others. Most humans routinelymake inferences and attributions about what other people mayor may not know, want or plan to do. By the same token,species that fail to recognize themselves in mirrors should failto use introspectively based social strategies such as sympathy,empathy, attribution, intentional deception, grudging, gratitude,pretense, role playing or sorrow.

Evidence for Empathy

We ought to be able to identify animals that can or cannotrecognize themselves in mirrors and their empathetic tenden-cies in some fairly definitive ways. If you were to cover the eyesof an animal at some point, how would it later respond to acagemate wearing a blindfold? An animal that is self-awareought to be in a position to use its prior experience with blind-folds to take into account its cagemate’s inability to see. If youwere to teach an animal to vocalize for a food reward everytime you entered the room and then blocked its hearing withearplugs or headphones, how would it respond the next day ifyou entered the room wearing headphones? If self-aware, itshould vocalize more loudly to compensate for your impairedability to hear.

In these kinds of tests, monkeys that fail to show evidenceof self-recognition (as distinct from chimpanzees and orang-utans, which are great apes) seem completely incapable of tak-ing into account what other monkeys may or may not know.Dorothy L. Cheney and Robert M. Seyfarth of the University ofPennsylvania have found that vervet monkeys give alarm callson seeing a predator even if other monkeys have already seenit, too. Likewise, they found that Japanese monkey mothersdo not distinguish between offspring that know or do notknow about food or danger when it comes to alerting theirbabies to the presence of one or the other.

Monkeys that cannot recognize themselves in mirrorsapproximate what psychologists call radical behaviorists. Their


interactions with other monkeysseem to be based entirely on ananalysis driven by the external fea-tures of the other monkey and noton what it might be thinking or whatit might want to do. Chimpanzees,on the other hand, ought to repre-sent primitive, albeit imperfect, cog-nitive psychologists—they should beable to respond empathetically andmodify their behavior accordingly.

Initial experiments by Daniel J.Povinelli and Sarah T. Boysen ofOhio State University showed thatchimpanzees appear to distinguishbetween what humans may or maynot know. When two humans point-ed toward different cups, the chim-panzees learned to pick the cupimplicated by the human who hadwitnessed which cup had been baitedwith food. Although chimpanzeesseemed to recognize ignorance onthe part of human informants, rhesus monkeys did not.

Further evidence for cognitive empathy in chimpanzeescomes from a mutual problem-solving experiment in whichhumans and chimpanzees had to perform different tasks. Forinstance, a chimpanzee had to pull a handle to bring food cupswithin reach but could not see which cup had been baited,whereas the human who could not reach the cups had to pointto the baited cup. The chimpanzees were able to switch roleswith the humans with no decrement in performance. Rhesusmonkeys, however, failed to show any evidence of transferwhen the roles were reversed.

Arguing against self-awareness and empathy of chimpan-zees, Povinelli cites experiments that failed to find evidence inchimpanzees for an ability to take into account what anothercreature sees. He concludes that chimpanzees cannot evenconceive of their own mental states, let alone those of others.

There are some explanations for the negative results, how-ever. Povinelli’s experiments relied on chimpanzees that might

have been too young; the onset of self-recogni-tion in chimpanzees does not occur until ado-lescence. Still another possibility is that wehumans categorize our experiences (for example,by sight, hearing or smell). Lacking language,chimpanzees may not distinguish between visual,auditory and tactile experiences. Therefore,inferences they make about attention may bemore global.

Also, most studies have focused on whetherchimpanzees can take various informationalstates of mind into account (that is, whether theycan figure out what another individual knows).But the data on humans show that childrenattribute feelings and motivation before theyhave the ability to attribute informational statesof mind. Beginning at about the time or shortlyafter (but never before) they learn to recognize

themselves in mirrors, chil-dren start to make primitiveinferences about emotionalstates of mind in others;the more sophisticated abil-ity to infer informationalstates of mind does nothappen until a year or twolater. Autistic children, incontrast, have difficultytaking into account whatother people may know,want or feel. As expected,self-recognition in autisticchildren is often delayed oreven absent.

Because chimpanzeesand orangutans pass the




SA

mirror test, Povinelli hypothesizes that they possess a motorself-concept rather than a psychological one. That is, they donot really recognize themselves but simply learn an equivalencebetween their behavior and what they see in the mirror.

But matters of appearance have little to do with move-ment. So why should chimps and orangutans seem so intent onusing mirrors to look at and inspect parts of their bodies theyhave never seen before? Why should theybother to respond to strange but motoricallyinconsequential red marks on their own faces?Suzanne Calhoun and Robert Thompson ofHunter College describe the reaction of a chimpanzee that, onbeing reintroduced to a mirror a year after learning to recognizeherself, became very agitated when she opened her mouth andsaw several missing teeth. It is hard to see how this reactioncould be understood purely in motor terms.

Self-Awareness and the Brain

Other, more speculative clues about self-awareness lie in thephysical makeup of the brain: certain areas seem to be respon-sible for it. Donald T. Stuss of the Rotman Research Institutein Toronto and I have been collaborating on a long-term projectthat focuses on human patients who have damage to the frontalcortex, the part of the brain responsible for some of the mostcomplex activities of the mind. Preliminary data show thatsuch patients seem unable to model mental states in others.

Self-awareness may correlate with activity in the right pre-frontal cortex. Julian P. Keenan and Alvaro Pascual-Leone ofHarvard Medical School, along with my colleagues N. BruceMcCutcheon and Glenn S. Sanders and me, tested how fasthumans can recognize faces. When responding with the lefthand (controlled by the right hemisphere), subjects identifiedtheir own faces faster than the faces of friends or co-workers.In addition, subjects viewing their own faces displayed signifi-cant changes in electrical potentials in the right prefrontalcortex. Moreover, when we altered the electrical activity inthis brain area with magnetic fields, subjects changed theirresponse rates to their own faces but not to the other faces.

Given this evidence of functional lateralization of self-awareness in humans, it is interesting to note that comparedwith other great apes, gorilla brains are the least anatomically

lateralized. The absence of ahighly specialized right hemi-sphere might explain the gorilla’sweak and inconsistent perfor-mance in the mirror tests.Povinelli claims that the gorilla’sfailure here is a “crucial test” ofhis theory of the motor self-con-cept. In particular, he speculatesthat it arose as an adaptation tolife in the trees: unlike chim-panzees and orangutans, gorillasspend most of their time on theground. But at night, gorillas stillreturn to the trees to sleep, eventhough they are at a greater riskof falling because they are solarge. In fact, humans are theones that have much more com-pletely emancipated themselvesfrom the branches. Therefore,

humans and not gorillas ought to fail to recognize themselvesin mirrors.

Povinelli’s data notwithstanding, I think most peopleworking in this area would agree that the jury is still out onwhether great apes can attribute mental states to others. Muchof the research in this topic is consistent with the conclusionreached by Nicholas Humphrey of the University of Cambridge

that many species may have clever brains but blank minds:clever brains in the sense of learning, memory and problemsolving, but blank minds in the sense of being unable to usetheir experience to take into account the experience of others.As evidenced by the behavior of people who sleepwalk andthose who suffer from blindsight, you do not have to knowwhat you are doing in order to do it in an appropriate way.Humans and possibly a few species of great apes appear tohave entered a unique cognitive domain that sets us apartfrom other creatures.

This model of consciousness and mind based on self-aware-ness has brought me full circle. When I devised the initial test ofself-recognition almost 30 years ago, it is apparent that I wasusing my experience and imagination about how I wouldrespond to strange facial marks to anticipate how chimpanzeesmight respond to such marks if they could recognize themselvesin mirrors. Moreover, if this model or some modified version ofit eventually proves correct, it would mean that the ability toconceive of oneself in the first place is what makes conscious-ness and thinking possible. The famous quote from Descarteswould have to be rewritten as “I am, therefore I think.”

Many species may have clever brains but blank minds.

GORDON GALLUP, JR.’sresearch interests run thegamut: he has hypnotizedchickens to examine how theirimmobility serves as a defenseagainst predation, worked onopen-field behavior in rats,looked at depression andreproductive failure in people,theorized about the demise ofthe dinosaurs and monitoredrisk-taking behaviors in men-struating women. Each proj-ect—odd though it mightsound—adds to our under-standing of the evolutionaryforces that underlie behavior,both animal and human.Gallup devised the mirror testwhile he was a graduate student at Washington State University. The ideacame to him one day while he was shaving in front of the mirror.

After serving on the faculty at Tulane University, Gallup accepteda position as a professor and chair of psychology at the State Universityof New York at Albany. He lives on an old dairy farm, where he growshis own food—potatoes, tomatoes, beans and corn—and maintains asmall herd of beef cattle. In the fall, Gallup bales hay and chops woodbefore heading off to teach class and grade exams. And he enjoys it. “Ijust love being outside and doing physical work,” he says. “It helps mekeep one foot firmly planted in reality.”

JAM

ESLE

YNSE

SABA

PHO

TOG

RAPH

S BY

DO

NN

A B

IERS

CH

WA

LE

About the Author


the possibility that evolution has simply produced “mind-blind” mechanisms that lead social primates to look whereother animals look, without entertaining any ideas about theirvisual perspective?

To disentangle these issues, we need to study the behaviorof these animals in more revealing experimental situations. Onemethod occurred to us after watching our chimpanzees in theireveryday play. They frequently covered their heads with blan-kets, toy buckets or even their palms and then frolicked aroundtheir compound until they bumped into something—or some-one. Occasionally they would stop and lift the obstructionfrom their eyes—to peek, as it were—before continuing theirblind strolls. On more than one occasion I made the mistakeof imitating these behaviors while playing with the animals, amaneuver that left me vulnerable to a well-timed play attack!

Does this behavior mean that chimpanzees have a conceptof seeing? For example, when they play with someone else who

covers his or her head, do they know that this person cannotsee them, or do they simply learn that this person is unable torespond effectively?

To answer these questions, we examined one of our chim-panzees’ most common communicative gestures: begging.First, we allowed them to beg for food from an experimenterwho was sitting just out of their reach. When they did so,they were handed an apple or banana. Next, we confrontedthem with two familiar experimenters, one offering a piece offood and the other holding out an undesirable block of wood.As we expected, the chimps had no trouble: after glancing atthe two experimenters, they immediately gestured to the oneoffering the food.

This set the stage for our real objective, which was to pro-vide the apes with a choice between a person who could seethem and a person who could not. If the high-level model ofchimpanzee understanding were correct, the chimps wouldgesture only to the person who could see them. We achievedthe “seeing/not-seeing” contrast by having the two experiment-ers adopt different postures. In one test, one experimenter worea blindfold over her eyes while the other wore a blindfold overher mouth. In the other tests, one of the experimenters wore abucket over her head, placed her hands over her eyes or sat

with her back turned to the chimpanzee. All these postureswere modeled after the behaviors we had observed during thechimpanzees’ spontaneous play.

The results of this initial experiment were astonishing. Inthree of the four tests—the ones involving blindfolds, bucketsand hands over the eyes—the apes entered the lab and pausedbut then were just as likely togesture to the person who couldnot see them as to the personwho could. In several cases, theapes gestured to the person whocould not see them and then,when nothing happened, ges-tured again, as if puzzled by thefact that the experimenter didnot respond.

We were not prepared forsuch findings. Surely our apesunderstood that only one of the experimenters could see them.Indeed, the apes did perform excellently in one of the tests,where one experimenter sat with her back turned to the chim-panzees. But why only this one? At first we assumed that theback/front test was simply the most obvious or natural con-trast between seeing and not seeing. In this test the apes might

have been demonstrating their gen-uine understanding of seeing—anunderstanding that was obscured bythe arguably less natural postures inthe other tests.

Another idea, however, began tonag at us. Perhaps the apes’ excellentperformance on the back/front testhad nothing to do with their reason-ing about who could or could not seethem. Maybe they were just doingwhat we had taught them to do in thefirst part of the study—gesture to thefront of someone who was facingthem. Or perhaps the act of gesturingto the front of a social partner is sim-

ply a hardwired social inclination among chimpanzees, uncon-nected to a psychological concept of seeing or attention.

As a first attempt to distinguish among these possibilities,we conducted another test in which both experimenters satwith their backs to the chimpanzees, but one looked over hershoulder at them. This posture was quite familiar to the apes—in their daily interactions, they frequently looked over theirshoulders at one another. The high-level model of chimpanzeeunderstanding predicted that the animals would gesture onlyto the experimenter who could see them. The low-level modelpredicted that the apes would choose at random because theycould not see the front of either experimenter. Their perfor-mance turned out to be random—they were just as likely togesture to either experimenter.

I should point out that what I am describing are the apes’initial reactions to these situations. As you might guess, withenough experience of not being handed a banana after gesturingto someone whose face was not visible, our chimpanzeesquickly learned to choose the other option. But what exactlydid the apes learn? Did they finally realize what we were ask-ing them—“Oh, I get it! It’s about seeing!”—or had they sim-ply learned another rule that could work every time: “Gestureto the person whose face is visible.”


Maybe NotPOVINELLI

GAZE FOLLOWING is a common behavior among chimpanzees. When the experimenterlooks over the chimpanzee’s shoulder (left), the ape looks in the same direction (right).

PHO

TOG

RAPH

S BY

DO

NN

A B

IERS

CH

WA

LE

Continued from page 67


We examined this ques-tion in an extended series ofstudies, the results of whichwere consistent with the low-level model. For example,after the chimpanzees learnednot to gesture to an exper-imenter whose head wasobscured by a cardboard disk,we retested the animals using

the original conditions (buckets, blindfolds, hands over theeyes and looking over the shoulder). We realized that if theapes had genuinely understood the idea of seeing, they oughtto gesture only to the experimenters who could see them inall the other tests as well. But if the chimpanzees had simply

learned to gesture to a person whose face was visible, theywould still choose randomly in the blindfold test, because thefaces of the experimenters were equally visible (one had theblindfold over her eyes; the other had it over her mouth). Justas the low-level model predicted, the chimpanzees were morelikely to gesture to the experimenter who could see them inall the tests except one—the blindfold test.

These findings contrast sharply with the development ofthese abilities in human infants. John H. Flavell and his col-leagues at Stanford University have shown that children asyoung as two or three years seem to understand the conceptof seeing. And indeed, when we tested young children usingour seeing/not-seeing method, we found that even two-and-a-half-year-old children performed at levels suggesting that theyunderstood that only one of the experimenters could see them.

Growing Up Ape

Let me address one important criticism of our work raisedby Gallup concerning the age of our animals. The initial testswere conducted in 1993 and 1994, when the chimpanzees were

five to six years old. Although several of our apes were display-ing all the traditional evidence of recognizing themselves inmirrors, some of them were still on the cusp of developing thisability. Could it be that older chimpanzees might fare better inthe seeing/not-seeing tests?

One year after the initial research—and after our apes hadbeen engaged in many other studies—we assessed their reactionsto several of the original seeing/not-seeing tests. Much to oursurprise, the chimpanzees initially responded at random, even tothe test where one of the experimenters hid her head behind acardboard disk—a test the apes had learned extremely well a yearearlier. Our chimpanzees’ performance improved only gradually,after considerable trial and error. Furthermore, after another yearhad passed and our apes had become young adults, additionaltests revealed that they were still relying on rules about thefrontal posture, faces and eye movements of the experimenters—not about who could see them. Thus, despite the fact that manyof our chimpanzees had displayed evidence of self-recognitionfor more than four years, we had no evidence that they gen-uinely understood one of the most basic empathic aspects ofhuman intelligence: the understanding that others see.

The Meaning of Self-Recognition

If we knew nothing more about chimpanzees, we mightsimply conclude that they understand visual perception in avery different manner than we do. Other studies in our labora-tory, however, have suggested that chimpanzees may notunderstand any behavior in a psychological manner. For exam-ple, careful tests revealed that our apes do not comprehend

pointing gestures as referential actions,nor do they understand the differencebetween accidental and intentional behav-ior. Furthermore, recent tests conductedwith Daniela K. O’Neill of the Universityof Waterloo suggest that our original inter-pretations of our earlier studies on cooper-ation—which Gallup cites in support of histheory—may have been incorrect. Although

our chimpanzees easily learn to cooperate with one another,our new results cast doubt on whether they truly appreciatethe differing subjective mind-sets of their partners.

If chimpanzees do not genuinely reason about mentalstates in others, what can we say about their understanding ofself? Exactly what is revealed by their antics in front of mir-rors? And do such reactions to mirror images really indicate


CHIMPANZEE UNDERSTANDING of theconcept of seeing was tested in a series ofexperiments. Confronted with pairs of“seeing” and “not-seeing” experimenters(above), the chimps were equally likely togesture to either one. The apes performedbetter in the back/front test (right), buttheir performance was random in thelooking-over-the-shoulder test (below).

PHO

TOG

RAPH

S BY

DO

NN

A B

IERS

CH

WA

LE


the onset of autobiographical memory—in both apes andhumans—as Gallup suggests?

As a first attempt to answer these questions, we shifted ourattention to humans—specifically, two-, three- and four-year-oldchildren. In a series of studies, we individually videotaped thechildren as they played an unusual game with an experimenter.

During the game, the experimenter praised the child and usedthis opportunity to place a large, brightly colored stickersecretly on top of the child’s head. Three minutes later thechildren were shown either a live video image of themselvesor the recording we had made several minutes earlier, whichclearly depicted the experimenter placing the sticker on thechild’s head.

These tests revealed that the younger children—the two-and three-year-olds—responded very differently depending onwhether they observed the live or delayed images. When con-fronted with a live image, the vast majority of the two- andthree-year-olds reached up and removed thesticker from their heads. When confrontedwith three-minute-old images, however, onlyabout one third of the younger childrenreached up for the sticker. Did the others sim-ply not notice the sticker in the delayedvideo? Hardly. After experimenters drew theirattention to the sticker in the video and askedthem, “What is that?” the majority of thechildren responded, “It’s a sticker.” But thisacknowledgment did not cause them to reachup and remove the sticker.

In one sense, of course, the children clearly “recognizedthemselves” in the delayed video. When they were asked,“Who is that?” even the youngest children confidently replied,“Me!” or stated their proper names. This reaction, however,did not seem to go beyond a recognition of facial and bodilyfeatures. When asked, “Where is that sticker?” the children fre-quently referred to the “other” child: “It’s on her [or his]head.” It was as if the children were trying to say, “Yes, thatlooks like me, but that’s not me—she’s not doing what I’mdoing right now.” One three-year-old girl summarized thispsychological conflict quite succinctly: “It’s Jennifer,” she stated,only to hurriedly add, “but why is she wearing my shirt?”

So when do children come to think of themselves as hav-ing a past and a future? Our studies have revealed that by aboutfour years of age, a significant majority of the children beganto pass our delayed self-recognition test. Unlike their youngercounterparts, most four- and five-year-olds confidently reachedup to remove the sticker after they observed the delayed videoimages of themselves. They no longer referred to “him” or “her”or their proper names when talking about their images. Thisfinding fits nicely with the view of Katherine Nelson of the CityUniversity of New York and others, who believe that genuineautobiographical memory appears to emerge in children be-tween 3.5 and 4.5 years old—not at the two-year mark thatGallup favors. Of course, any parent knows that two-year-oldscan recall past events, but this is very different from under-standing that those memories constitute a genuine “past”—ahistory of the self leading up to the here and now.

Although it is still too early to rule out Gallup’s model alto-

gether, our research suggests that self-recognition in chim-panzees and human toddlers is based on a recognition of theself’s behavior, not the self’s psychological states. When chim-panzees and orangutans see themselves in a mirror, they forman equivalence relation between the actions they see in themirror and their own behavior. Every time they move, the

mirror image moves with them.They conclude that everythingthat is true for the mirror image isalso true for their own bodies, andvice versa. Thus, these apes can

pass the mirror test by correlating colored marks on the mirrorimage with marks on their own bodies. But the ape does notconclude, “That’s me!” Rather the animal concludes, “That’sthe same as me!”

Thus, although Gallup and I agree that passing the mirrortest reveals the presence of a kind of self-concept, we differ onthe nature and scope of that concept. Gallup believes thatchimpanzees possess a psychological understanding of them-selves. In contrast, I believe these apes possess an explicit men-tal representation of the position and movement of their ownbodies—what could be called a kinesthetic self-concept.

Ironically, this may be close to what Gallup himself hadin mind when he originally published his discovery nearly 30years ago. He noted that self-recognition appears to requirethe ability to project “kinesthetic feedback onto the reflectedvisual image so as to coordinate the appropriate visually guid-ed movements via the mirror.”

But why do humans, chimpanzees and orangutans possessthis kinesthetic self-concept, whereas other nonhuman pri-mates—such as monkeys—do not? One clue may be the largedifference in body size between the great apes and other pri-mates. Consider orangutans, which may represent the closestliving approximation to the common ancestor of the great apesand humans. Several years ago, John G. H. Cant of the Uni-versity of Puerto Rico and I spent months in the Sumatran


Self-recognition in chimpanzees and human toddlers is based on arecognition of the self’s behavior, not the self’s psychological states.

SELF-RECOGNITION TEST for chim-panzees (left) was modified for humanchildren with the help of video images(below). The experimenter secretly placed asticker on each child’s head. Young childrenreached for the sticker (right) only whenthe video image was live.


rain forest observing the orangutan’s chaotic blend of slow, care-fully planned movements and sudden, breathtaking acrobatics.We concluded that the problems encountered by these 40- to80-kilogram (90- to 180-pound) animals in bridging the gapsbetween trees were qualitatively different from the problemsfaced by the much smaller monkeys and lesser apes. Wehypothesized that as the ancestors of the great apes evolved,quadrupling in body size over 10 to 20 million years, they mayhave needed to evolve a high-level self-representational systemdedicated to planning their movements in their arboreal envi-ronment. Ultimately, this unprecedented increase in body sizefor a tree-dwelling mammal may have left a psychologicalimprint on the great apes: an explicit kinesthetic self-concept.It was this self-concept that Gallup tapped millions of yearslater in his tests of chimpanzee self-recognition.

A crucial test for our theory is the gorilla, the largest non-

human primate. Although gorillas share the same commonancestor as humans, chimpanzees and orangutans, they havereadapted to spending most of their lives on the ground. Thesurprising absence of self-recognition in this species may reflectthe fact that gorillas no longer needed to execute the complexmovements that were necessary to transport their enormousbody weight across the gaps between trees. Their evolutionappears to have focused on aspects that were more relevant totheir new terrestrial way of life, including a more rapid physicalgrowth rate than is found in chimpanzees and orangutans. Thisprocess may have interfered with the development of a kines-thetic self-concept. Humans, in contrast, slowed down theirgrowth rate, allowing more years for cognitive development.

If self-recognition depends on a kinesthetic rather than apsychological self-concept, it would help explain some puzzlingfacts. Several studies have found no connection between theability of 18- to 24-month-old infants to pass the mirror test andtheir ability to understand that a mirror reflects any objectplaced in front of it. Our theory explains this result by postulat-ing that the infants do not see their mirror images as represen-tations of themselves. Rather they see their images as a specialclass of entities that share their behavior and appearance.

Our theory also explains why toddlers often fail the self-recognition test if there is even a minimal disruption of thevisual feedback—for example, a two-second delay in the videoimages of themselves. Although the children continue to recog-nize their facial and bodily features, the two-second disjunctionbetween their actions and the movements of their images leadsthem to conclude that the images are not equivalent to them-selves. Finally, our theory explains why both toddlers and

chimpanzees, after recognizing themselves in the mirror, maynonetheless persist in looking behind the mirror, as if search-ing for the “other” child or ape.

Understanding Minds: A Human Specialization

At this point we are still left with a troubling question:How can humans and chimpanzees share such sophisticatedsocial behaviors but understand them so differently? Why dohumans interpret these behaviors in terms of psychologicalstates, but apes do not?

My answer may become more obvious if we imagine ourplanet 60 million years ago, long before any of the modernprimates had evolved. Alison Jolly of Princeton University hasspeculated that as the solitary lifestyle of the small, early primatesgave way to existence in large groups, these animals wereforced to cope with increasingly complex social interactions. Asa result, Jolly argues, the primates became stunningly adept atreasoning about one another’s actions, slowly evolving the richarray of social behaviors now observed among the modernprimates: gaze following, deception, appeasement and so on.

But, in my view, none of these behaviors required theearly primates to reason about one another’s mental states.Our research suggests that only one primate lineage—thehuman one—evolved the unique cognitive specialization thatenables us to represent explicitly our own psychological statesand those of others. But in evolving this specialization, we didnot discard our array of basic primate behaviors. Our newawareness of the mental dimension of behavior was woveninto our existing neural circuitry, forever altering our under-standing of our own behavior and the behavior of thosearound us. Other species, including chimpanzees, may simplybe incapable of reasoning about mental states—no matter howmuch we insist on believing that they do.


DANIEL J. POVINELLIfirst became interested inchimpanzee behavior in 1979,when he was 15 years old.While doing research for ahigh school debate, Povinellicame across an article byGordon Gallup, Jr., inAmerican Scientist describinghis mirror tests on chim-panzees. “The elegance andingenuity of Gallup’s testsreally struck me,” he says. “Ithought that here might be aspecies profoundly similar toour own.” Inspired, Povinelli studied primate behavior as an under-graduate at the University of Massachusetts at Amherst and earnedhis Ph.D. in biological anthropology from Yale University in 1991. Hethen joined the University of Southwestern Louisiana’s New IberiaResearch Center, a 150-acre facility that is home to more than 300chimpanzees. Now an associate professor, Povinelli directs the center’sdivision of behavioral biology, which studies cognitive developmentin both chimpanzees and young children. Over the years Povinellihas become a friend and colleague of Gallup’s, but his view of thechimpanzee’s mental abilities has diverged from that of his mentor.“It took a lot of patience on the part of the chimpanzees,” he says,“but they’ve finally taught me that they’re not hairy human children.”

About the Author

Can Animals Empathize?

SA

PHO

TOG

RAPH

S BY

DO

NN

A B

IERS

CH

WA

LE


Machine Intelligence78 Scientific American Presents

On

ComputationalRethinking the Goals of Artificial Intelligence

The greatest value ofartificial intelligence may

lie not in imitating humanthinking but in extending it

into new realms

by Kenneth M. Ford and Patrick J. Hayes

INTELLIGENCE

MA

CH

INE

INTELLIGENT SYSTEMS are cropping up everywhere.Current and future applications include (counterclock-wise from the top) computer programs that composemusic, advanced software on the Deep Space 1 probe,derivative pricing at the Chicago Board of Trade, anautonomous rover for exploring Mars, navigation sys-tems for trucks and emulsion monitoring in steel mills.

ILLU

STRA

TIO

N B

Y LA

URI

E G

RA

CE.

PHO

TOG

RAPH

S C

OU

RTES

Y O

F SP

ECTR

UM

AST

RO (s

pace

pro

be),

JEFF

ERY

TITC

OM

B L

iais

on In

tern

atio

nal(

trad

ing

floor

),N

ASA

AM

ES R

ESEA

RCH

CEN

TER

(Mar

sro

ver)

,MIT

CH

KEZ

AR

Tony

Sto

ne Im

ages

(tru

ck) A

ND

JO

NAT

HA

N K

IRN

Lia

ison

Inte

rnat

iona

l(st

eel m

ill).

MU

SIC

ILLU

STR

ATIO

N C

OU

RTE

SY O

F D

AVID

CO

PE U

nive

rsit

y of

Cal

iforn

ia a

t San

ta C

ruz.


Many philosophers and humanist thinkers are convincedthat the quest for artificial intelligence (AI) has turned out tobe a failure. Eminent critics have argued that a truly intelligentmachine cannot be constructed and have even offered mathe-matical proofs of its impossibility. And yet the field of artificialintelligence is flourishing. “Smart” machinery is part of the in-formation-processing fabric of society, and thinking of the brainas a “biological computer” has become the standard view inmuch of psychology and neuroscience.

While contemplating this mismatch between the criticalopinions of some observers and the significant accomplishmentsin the field, we have noticed a parallel with an earlier endeav-or that also sought an ambitious goal and for centuries wasattacked as a symbol of humankind’s excessive hubris: artificialflight. The analogy between artificial intelligence and artificialflight is illuminating. For one thing, it suggests that the tradi-tional view of the goal of AI—to create a machine that cansuccessfully imitate human behavior—is wrong.

For millennia, flying was one of humanity’s fondest dreams.The prehistory of aeronautics, both popular and scholarly,dwelled on the idea of imitating bird flight, usually by somehowattaching flapping wings to a human body or to a frameworkworn by a single person. It was frustratingly clear that birdsfound flying easy, so it must have seemed natural to try to cap-ture their secret. Some observers suggested that bird featherssimply possessed an inherent “lightness.” Advocates of the pos-sibility of flight argued that humans and birds were fundamen-tally similar, whereas opponents argued that such comparisonswere demeaning, immoral or wrongheaded. But both groupsgenerally assumed that flying meant imitating a bird. Even rel-atively sophisticated designs for flying machines often includedsome birdlike features, such as the beak on English artistThomas Walker’s 1810 design for a wooden glider.

This view of flying as bird imitation was persistent. An arti-cle in English Mechanic in 1900 insisted that “the true flyingmachine will be to all intents and purposes an artificial bird.”A patent application for a “flying suit” covered with featherswas made late in the 19th century, and wing-flapping meth-ods were discussed in technical surveys of aviation publishedearly in this century.

The Turing Test

Intelligence is more abstract than flight, but the long-termambition of AI has also traditionally been characterized as theimitation of a biological exemplar. When British mathematicianAlan M. Turing first wrote of the possibility of artificial intelli-gence in 1950, he suggested that AI research might focus onwhat was probably the best test for human intelligence avail-

able at the time: a competitive interview. Turing suggested thata suitable test for success in AI would be an “imitation game”in which a human judge would hold a three-way conversationwith a computer and another human and try to tell them apart.The judge would be free to turn the conversation to any topic,and the successful machine would be able to chat about it asconvincingly as the human. This would require the machineparticipant in the game to understand language and conversa-tional conventions and to have a general ability to reason. If thejudge could not tell the difference after some reasonable amountof time, the machine would pass the test: it would be able toseem human to a human.

There is some debate about the exact rules of Turing’s imi-tation game, and he may not have intended it to be taken soseriously. But some kind of “Turing test” has become widely per-ceived, both inside and outside the field, as the ultimate goalof artificial intelligence, and the test is still cited in most text-books. Just as with early thinking about flight, success is definedas the imitation of a natural model: for flight, a bird; for intel-ligence, a human.

The Turing test has received much analysis and criticism,but we believe that it is worse than often realized. The test hasled to a widespread misimpression of the proper ambitions ofour field. It is a poorly designed experiment (depending toomuch on the subjectivity of the judge), has a questionable tech-nological objective (we already have lots of human intelligence)and is hopelessly culture-bound (a conversation that is passableto a British judge might fail according to a Japanese or Mexicanjudge). As Turing himself noted, one could fail the test by beingtoo intelligent—for example, by doing mental arithmetic ex-tremely fast. According to media reports, some judges at thefirst Loebner competition in 1991—a kind of Turing test contestheld at the Computer Museum in Boston—rated a human as amachine on the grounds that she produced extended, well-written paragraphs of informative text. (Apparently, this is nowconsidered an inhuman ability in parts of our culture.) Withthe benefit of hindsight, it is now evident that the central defectof the test is its species-centeredness: it assumes that humanthought is the final, highest pinnacle of thinking againstwhich all others must be judged. The Turing test does not admitof weaker, different or even stronger forms of intelligence thanthose deemed human.

Most contemporary AI researchers explicitly reject the goalof the Turing test. Instead they are concerned with exploringthe computational machinery of intelligence itself, whether inhumans, dogs, computers or aliens. The scientific aim of AIresearch is to understand intelligence as computation, and itsengineering aim is to build machines that surpass or extendhuman mental abilities in some useful way. Trying to imitate ahuman conversation (however “intellectual” it may be) con-tributes little to either ambition.

In fact, hardly any AI research is devoted to trying to passthe Turing test. It is more concerned with issues such as howmachine learning and vision might be improved or how todesign an autonomous spacecraft that can plan its own actions.Progress in AI is not measured by checking fidelity to a humanconversationalist. And yet many critics complain of a lack ofprogress toward this old ambition. We think the Turing testshould be relegated to the history of science, in the same waythat the aim of imitating a bird was eventually abandoned bythe pioneers of flight. Beginning a textbook on AI with theTuring test (as many still do) seems akin to starting a primer

Exploring Intelligence 79On Computational Wings:Rethinking the Goals of Artificial Intelligence

Wings:


on aeronautical engineering with an explanation that the goalof the field is to make machines that fly so exactly like pigeonsthat they can even fool other pigeons.

Imitation versus Understanding

Researchers in the field of artificial intelligence may take auseful cue from the history of artificial flight. The developmentof aircraft succeeded only when people stopped trying to imi-tate birds and instead approached the problem in new ways,thinking about airflow and pressure, for example. Watchinghovering gulls inspired the Wright brothers to use wing warp-ing—turning an aircraft by twisting its wings—but they did notset out to imitate the gull’s wing. Starting with a box kite, theyfirst worked on achieving sufficient lift, then on longitudinaland lateral stability, then on steering and finally on propulsionand engine design, carefully solving each problem in turn. Afterthat, no airplane could be confused with a bird either in itsoverall shape or in its flying abilities. In some ways, aircraft maynever match the elegant precision of birds, but in other ways,

they outperform them dramatically. Aircraft do not land intrees, scoop fish from the ocean or use the natural breeze tohover motionless above the countryside. But no bird can fly at45,000 feet or faster than sound.

Rather than limiting the scope of AI to the study of how tomimic human behavior, we can more usefully construe it as thestudy of how computational systems must be organized in orderto behave intelligently. AI programs are often components oflarger systems that are not themselves labeled “intelligent.”There are hundreds of such applications in use today, includingthose that make investment recommendations, perform med-ical diagnoses, plan troop and supply movements in warfare,schedule the refurbishment of the space shuttle and detectfraudulent use of credit cards. These systems make expert deci-sions, find meaningful patterns in complex data and improvetheir performances by learning. All these actions, if done by ahuman, would be taken to display sound judgment, expertiseor responsibility. Many of these tasks, however, could not bedone by humans, who are too slow, too easily distracted or notsufficiently reliable. Our intelligent machines already surpassus in many ways. The most useful computer applications,including AI applications, are valuable exactly by virtue of theirlack of humanity. A truly humanlike program would be just asuseless as a truly pigeonlike aircraft.

Waiting for the Science

The analogy with flight provides another insight: techno-logical advances often precede advances in scientific knowledge.The designers of early aircraft could not learn the principles ofaerodynamics by studying the anatomy of birds. Evolution isa sloppy engineer, and living systems tend to be rich with adhoc pieces of machinery with multiple uses or mechanismsjury-rigged from structures that evolved earlier for a differentreason. As a result, it is often very difficult to discover basicprinciples by imitating natural mechanisms.

Experimental aerodynamics became possible only in theearly part of this century, when artificial wings could be testedsystematically in wind tunnels. It did not come from studyingnatural exemplars of flight. That a gull’s wing is an airfoil isnow strikingly obvious, yet the airfoil was not discovered byexamining the anatomy of birds. Even the Wright brothersnever really understood why their Flyer flew. The aerodynamicprinciples of the airfoil emerged from experiments done in 1909by French engineer Alexandre-Gustave Eiffel, who used a windtunnel and densely instrumented artificial wings. The first air-craft with “modern” airfoils—which were made thicker afterengineers demonstrated that thicker airfoils improved lift with-out increasing drag—did not appear until late in World War I.As is true for many other disciplines, a firm theoretical under-standing was possible only when controlled experiments couldbe done on isolated aspects of the system. Aerodynamics wasdiscovered in the laboratory.

The same reasoning applies to the study of human intelli-gence. It may be impossible to discover the computationalprinciples of intelligent thought by examining the intricaciesof human thinking, just as it was impossible to discover theprinciples of aerodynamics by examining bird wings. TheWright brothers’ success was largely attributed to their percep-tion of flight in terms of lift, control and power; similarly, ascience of intelligence must isolate particular aspects ofthought, such as memory, search and adaptation, and allow usto experiment on these one at a time using artificial systems.


JUDGE

HUMAN PARTICIPANT

COMPUTER PARTICIPANT

TURING TEST for artificial intelligence was proposed in 1950by British mathematician Alan M. Turing (photograph). Inthe test, a human judge would hold a three-way conversationwith a computer and another human. If the judge could notdistinguish between the responses of the human and those ofthe computer, the machine would pass the test.

GEO

RGE

RETS

ECK

CO

UR

TESY

OF

KEN

NET

H M

.FO

RD A

ND

PAT

RIC

K J.

HAY

ES


Exploring Intelligence 81On Computational Wings:Rethinking the Goals of Artificial Intelligence

By systematically varying functional parameters ofthought, we can determine the ways in which variouskinds of mental processes can interact and support oneanother to produce intelligent behavior.

Several areas of AI research have been transformed inthe past decade by an acceptance of the fact thatprogress must be measurable, so that different techniquescan be objectively compared. For example, large-scaleempirical investigations must be conducted to evaluatethe efficiency of different search techniques or reasoningmethods. In this kind of AI research, computers are pro-viding the first wind tunnels for thought.

A Science of Intelligence

Rejecting the Turing test may seem like a retreatfrom the grand old ambition of creating a “humanlike”mechanical intelligence. But we believe that the properaim of AI is much larger than simply mimicking humanbehavior. It is to create a computational science of intelli-gence itself, whether human, animal or machine. This isnot a new claim; it has been made before by AI pioneersAllen Newell and Herbert A. Simon, cognitive psychologistZenon Pylyshyn and philosopher Daniel C. Dennett,among others. But it was not until we noted the analogywith artificial flight that we appreciated the extent towhich the Turing test, with its focus on imitating humanperformance, is so directly at odds with the proper objec-tives of AI. Some of our colleagues say their ultimate goalis indeed the imitation of human intelligence. Even withthis limited aim, however, we believe that the perspectivesketched here provides a more promising way to achievethat ambition than does the method outlined by Turing.

Consider again the analogy with flight. Just as theprinciples of aerodynamics apply equally to any wing,natural or artificial, the computational view of intelligence—or,more broadly, of mentality—applies just as well to naturalthinkers as to artificial thinkers. If cognitive psychology andpsycholinguistics are like the study of bird flight in all its com-plexity, then applied AI is like aeronautical engineering.Computer science supplies the principles that guide the engi-neering, and computation itself is the air that supports thewings of thought.

The study of artificial intelligence, like a large part of com-puter science, is essentially empirical. To run a program is oftento perform an experiment on a large, complex apparatus (madepartly of metal and silicon and partly of symbols) to discoverthe laws that relate its behavior to its structure. Like artificialwings, these AI systems can be designed and instrumented toisolate particular aspects of this relation. Unlike the researchmethodology of psychology, which employs careful statisticalanalysis to discern relevant aspects of behavior in the tangledcomplexity of nature, the workings of AI systems are open todirect inspection. Using computers, we can discover and exper-iment directly with what Newell and Simon have called the“laws of qualitative structure.”

This picture of AI defines the field in a more useful andmature way than Turing could provide. In this view, AI is theengineering of cognitive artifacts based on the computationalunderstanding that runs through and informs current cogni-tive science. Turing correctly insisted that his test was notmeant to define intelligence. Nevertheless, in giving us thistouchstone of success, he chose human intelligence—in fact,

the arguing skill of an educated, English middle-class man play-ing a kind of party game—as our goal. But the very sciencethat Turing directed us toward provides a perspective fromwhich a much broader and more satisfying account of intelli-gence is emerging.

Scholastic Critics

Artificial intelligence and artificial flight are similar even inthe criticisms they attract. The eminent American astronomerSimon Newcomb argued passionately in the early 1900s againstthe idea of heavier-than-air flight. Newcomb’s fulminationsseem amusing now, but his arguments were quite impressiveand reflected the view of the informed intelligentsia of his day.Like British mathematical physicist Roger Penrose, who usesGödel’s theorem to “prove” that AI is impossible, Newcombemployed mathematical arguments. He pointed out that as birdsget bigger, their wing area increases in proportion to the squareof their size, but their body weight increases in proportion tothe cube, so a bird the size of a man could not fly. He was stillusing this argument against the possibility of manned flightseveral years after the Wright brothers’ success at Kitty Hawk,N.C., when aircraft were regularly making trips lasting severalhours. It is, in fact, quite a good argument—aircraft takeoffweights are indeed roughly proportional to the cube of theirwingspan—but Newcomb had no idea how sharply the liftfrom an airfoil increases in proportion to its airspeed. Hethought of a wing as simply a flat, planar surface.

COMPARISON OF SKELETONS of a human and a bird—heretaken from a 16th-century manuscript by French naturalistPierre Belon—examined similarites in anatomy in an attempt tounderstand how birds can fly.

CU

LVER

PIC

TURE

S


Newcomb also used a combination of thought experimentand rhetoric to make his point—the same tactic that philoso-pher John R. Searle has employed in his famous “ChineseRoom” argument against AI [see “Is the Brain’s Mind a Com-puter Program?” by John R. Searle; SCIENTIFIC AMERICAN, Jan-uary 1990]. Newcomb stated scornfully, “Imagine the proudpossessor of the aeroplane darting through the air at a speedof several hundred feet per second! It is the speed alone thatsustains him. How is he ever going to stop?” Newcomb’s argu-ments, with their wonderful combination of energy, passion,cogency and utter wrongheadedness, are so similar to contem-porary arguments against artificialintelligence that for several years wehave offered the annual SimonNewcomb Award for the silliest newargument attacking AI. We welcomenominations.

A common response to our analo-gy between artificial intelligence andartificial flight is to ask what will bethe Kitty Hawk of AI and when will ithappen. Our reply follows that ofHerbert Simon: it has already hap-pened. Computers regularly performintelligent tasks and have done so formany years. Artificial intelligence isflying all around us, but many simplyrefuse to see it. Among the thousandsof applications in use today, here arejust a few examples: AI systems nowplay chess, checkers, bridge andbackgammon at world-class levels,compose music, prove mathematicaltheorems, explore active volcanoes,synthesize stock-option and derivativeprices on Wall Street, make decisionsabout credit applications, diagnosemotor pumps, monitor emulsions in a

steel mill, translate technical ser-vice manuals, and act as remedialreading tutors for elementaryschool children. In the nearfuture, AI applications will guidedeep-space missions, exploreother planets and drive trucksalong freeways.

But should all this reallycount as “intelligent”? The perfor-mance of AI systems, like thespeed or altitude of aircraft, is notopen to dispute, but whether ornot one chooses to call it “intelli-gent” is determined more bysocial attitude than by anythingobjective. When any particularability is mechanized, it is oftenno longer considered to be ahallmark of mental prowess. It iseasy now to forget that when

Turing was writing, a “computer” was a human being who didarithmetic for a living, and it was obvious to everyone thatcomputing required intelligence. The meaning of the word hasnow changed to mean a machine, and performing fast, accu-rate arithmetic is no longer considered a hallmark of mentalability, just as the meaning of “flying” has changed to coverthe case, once inconceivable, of dozing quietly in an airplaneseat while traveling at hundreds of miles an hour far above theclouds. Newcomb—who was famous as one of the finest com-puters of his time—went to his deathbed refusing to concedethat what early aircraft did should be called “flying.”

Turing suggested his test as a wayto avoid useless disputes aboutwhether a particular task counted astruly intelligent. With considerableprescience, he anticipated that manypeople would never accept that theaction of a machine could ever belabeled as “intelligent,” that mosthuman of labels. But just as there wasno doubt that the early flyers movedthrough the air at certain altitudes andspeeds, there is no doubt that electron-ic computers actually get arithmeticdone, make plans, produce explana-tions and play chess. The labels areless important than the reality.

The arbitrariness of the sociallabeling can be illustrated by a thoughtexperiment in which the machine isreplaced by something mysterious butnatural. Whereas a dog will never passthe Turing test, no one but a philoso-pher would argue that a dog does notdisplay some degree of intelligence—certainly no one who has owned a dogwould make such an argument. It isoften claimed that Deep Blue, the com-


SIMON NEWCOMB, American astronomerand mathematician, argued passionatelyagainst the possibility of artificial flight—even after the Wright brothers’ successfultests of their aircraft in 1903.

WOODEN GLIDER designed in1810 by English artist ThomasWalker included a birdlike beak.

CU

LVER

PIC

TURE

S

MA

RY L

EA S

HA

NE

ARC

HIV

ES,L

ICK

OB

SER

VATO

RY


puter that defeated chess championGarry Kasparov, is not really intelli-gent, but imagine a dog that playedchess. A chess-playing dog that couldbeat Kasparov would surely beacclaimed a remarkably smart dog.

The idea that natural intelligenceis a complex form of computationcan only be a hypothesis at present.We see no clear reason, however,why any mental phenomenon can-not be accounted for in this way.Some have argued that the computa-tionalist view cannot account for thephenomenology of consciousness. Ifone surveys the current theories ofthe nature of consciousness, howev-er, it seems to us that a computation-alist account offers the most promise.Alternative views consider conscious-ness to be some mysterious physical property, perhaps arisingfrom quantum effects influenced by the brain’s gravity or evensomething so enigmatic as to be forever beyond the reach ofscience. None of these views seems likely to explain how aphysical entity, such as a brain in a body, can come to be awareof the world and itself. But the AI view of mental life as theproduct of computation provides a detailed account of howinternal symbols can have meaning for the machine and howthis meaning can influence and be influenced by the causalrelations between the machine and its surroundings.

The scientific goal of AI is to provide a computationalaccount of intelligence or, more broadly, of mental ability

itself—not merely an explanation of human mentality. Thisvery understanding, if successful, must deny the uniquenessof human thought and thereby enable us to extend and amplifyit. Turing’s ultimate aim, which we can happily share, was notto describe the difference between thinking people and unthink-ing machines but to remove it. This is not to disparage or reducehumanity and still less to threaten it. If anything, understand-ing the intricacies of airflow increases our respect for how extra-ordinarily well birds fly. Perhaps it seems less magical, but itscomplexity and subtlety are awesome. We suspect that the samewill be true for human intelligence. If our brains are indeedbiological computers, what remarkable computers they are.


SA

On Computational Wings:Rethinking the Goals of Artificial Intelligence

THE WRIGHT FLYER, shown atKitty Hawk, N.C., with OrvilleWright piloting, proved that aircraftneed not imitate birds.

AR

CH

IVE

PHO

TOS

KENNETH M. FORD (left) is associate director at the National Aero-nautics and Space Administration Ames Research Center and directorof the NASA Center of Excellence for Information Technology. He is onleave from the Institute for Human and Machine Cognition at theUniversity of West Florida, where he is the director. Ford entered com-

puter science and artificial intelligence through the back door of phi-losophy. After studying epistemology as an undergraduate, he joinedthe navy and wound up fixing computers. “I had no interest in com-puters,” he says. “I didn’t even know how to program them.” ButFord learned quickly—he received his Ph.D. in computer science fromTulane University in 1987. He still finds a compelling connectionbetween computers and philosophy. “I’m convinced,” he says, “thatwere they reborn into a modern university, Plato, Aristotle and Leibnizwould most suitably take up appointments in the computer sciencedepartment.” Over the years he has developed a love of stalking waryredfish and speckled trout on the shallow flats of the Gulf of Mexico—quiet hours that are often the source of fresh ideas.

PATRICK J. HAYES is John C. Pace Eminent Scholar at theInstitute for Human and Machine Cognition at the University of WestFlorida and is a past president of the American Association forArtificial Intelligence. His path toward AI started at age 12, when heconstructed a robot that could wander around a tabletop withoutfalling off. He went on to study mathematics at the University ofCambridge and machine intelligence at the University of Edinburgh.Visiting Stanford University to work with AI pioneer John McCarthysparked his interest in the use of logic in artificial intelligence. Later,at the University of Rochester, Hayes headed one of the first multidis-ciplinary programs in cognitive science. His current research focuseson the representation of knowledge, the underpinnings for anartificial consciousness and the philosophical questions raised by AI.Hayes also has a more temporal interest: his wife, Jacqueline, and hecollect and restore antique mechanical clocks.

About the Authors

INST

ITU

TE F

OR

HU

MA

N A

ND

MA

CH

INE

CO

GN

ITIO

N



Computers, Games and the Real World

More than just competing with people, game-playingmachines complement human thinking by offeringalternative methods to solving problems

by Matthew L. Ginsberg

LOU

ISE

AN

D W

ALT

ER A

NN

ENB

ERG

CO

LLEC

TIO

N,P

HIL

AD

ELPH

IA M

USE

UM

OF

ART

The world watched with considerable amaze-ment in May 1997 as IBM’s chess computer, DeepBlue, beat Garry Kasparov, the world champion, ina six-game match. With a machine’s victory in thismost cerebral of games, it seemed that a line hadbeen crossed, that our measurements of ourselvesmight need tailoring.

The truth of who ultimately won and who lost,of course, is not so black-and-white. Kasparovplayed poorly, resigning a game that would haveled to a draw early in the match and making acompletely uncharacteristic error in the last game.And while chess-playing computers have been gain-ing an edge on their human competitors for sometime, in many other games, such as Go and bridge,computer players remain relatively weak. Still, incheckers and Othello, machines have been theworld’s strongest players for years. Backgammon,like chess, is currently too close to call, whereasmachines have a slight but definite edge in Scrabble.

The board and card games that researchersbuild programs to play provide an environmentwith specific rules and objective outcomes, a closedsystem allowing theories to be tested and achieve-ments to be tracked. As an element of artificialintelligence, game-playing software highlights thekey differences between the brute-force calculationof machines and the often intuitive, pattern-matching abilities of humans.

Considering Moves

People have been designing machines and pro-grams to play games almost as long as computercode has been written. In 1946 British mathemati-cian Alan M. Turing began designing a chess playeron a souped-up code-breaking machine used byBritain during World War II. The computer became

the first machine to play a full game of chess, albeitat an extremely slow beginner level. Turing and hiscolleagues at the University of Manchester laterwent on to tackle the more basic programmingchallenges of tic-tac-toe and checkers.

Since then, many people have designed differ-ent types of programs and computers with varyingdegrees of success. The most recent and best gameprograms [see box on pages 86 and 87] are interestingbecause they generally play their games quite well.But perhaps more interesting is that they achievethis level of performance by using techniques verydifferent from those used by their human counter-parts. In games such as chess, a human player willconsider some tens (or perhaps hundreds) of posi-tions when selecting a move. A machine, on theother hand, will consider billions, searchingthrough many possible lines of play in the processof selecting an action.

Considering the far greater computationalcapacity of computers, how is it that humans canwin at all? The answer is that although we look at arelative handful of successor positions, we look atthe right handful. Emanuel Lasker, world chesschampion from 1894 to 1920, was once asked howmany moves he considered when analyzing a chessposition. “Only one,” he replied. “But it’s alwaysthe best move.”

We identify the best positions to consider by aprocess known as pattern matching. It is how weimmediately identify the four-legged platform in aroom as a table or the images in a police mug shotas those of the same person despite the differentorientations of the front and profile views. Anexpert chess player compares a given chess positionto the tremendous number of such positions thathe has seen over the course of his career, usinglessons learned from analyzing those other posi-


Exploring Intelligence 85Computers, Games and the Real World

Portrait of Chess Players, by Marcel Duchamp



tions to identify good moves in the current game.Pattern matching is a parallel process; if you had n comput-

ers, you could do it n times faster. For example, suppose you aretrying to compare a particular chess position to the 100,000positions that you have seen previously. Each comparison is insome sense independent: given 100,000 computers, you coulddo the overall comparison 100,000 times more quickly.

Searching is not an inherently parallel process; problems

cannot be split up, because what you want to do nextdepends on what you just did. For Deep Blue, there is no pre-existing list of positions that it evaluates. Rather it can exam-ine a position only after it has constructed its predecessor.

It is important to note that when I say that patternmatching is parallel, whereas brute-force searching is not, I amreferring not to the problem being solved but only to themethod being used to solve it. I have not said that playing agood game of chess is a parallel problem or that it is not. Infact, the evidence in some sense is that chess is not inherentlyparallel or serial, because parallel techniques (human patternmatching) and serial ones (Deep Blue’s brute-force searching)can be applied equally well to the game.

Function Follows Form

There is a good reason that Kasparov and other humanplayers use a parallel technique to play chess: the human brainappears to be a parallel machine, consisting of about 100 bil-lion neurons, each capable of operating 1,000 times a second.

BACKGAMMON• 2 players• 15 pieces each• Goal: Move all pieces off the

board• Rules:

Dice roll determines numberof movesPlayers move in oppositedirectionsPiece cannot land on a pointoccupied by 2 or more ofopponent’s piecesSingle piece can be “hit” iflanded on by opponent; hitpiece must start anew

• Program: TD-Gammon*• Web site:

www.research.ibm.com/massdist/tdl.html

• Advantage: Too close to call

BRIDGE• 4 players in 2 teams• 13 cards dealt to each player• Goal: Make 2 “game con-

tracts,” or a “rubber”• Rules:

The bid: Each player predictshow many times his or hercard will be the highest (atrick)The play: Put down 1 cardat a time and compare it withothers; this occurs 13 timesThe scoring: Points scored ifbid is made or exceeded;otherwise points go to theopposing team

• Program: GIB*• Web site: www.gibware.com• Advantage: Human

CHECKERS• 2 players• 12 pieces each• Goal: Avoid being the player

who can no longer move (usu-ally when a player has nopieces left)

• Rules:Move forward on dark diago-nal, 1 square at a timeOpponent’s piece capturedwhen jumped to emptysquare diagonally behindopponent’s pieceCreation of a “king,” a piecethat can move backward andforward, occurs when piece ismoved to opponent’s last row

• Program: Chinook• Web site: www.cs.ualberta.ca/

~chinook• Advantage: Machine

K 6 2

A

K

7

8 4 2

Q 6 3

6

North

A 8 5

K

A

K

7 3

J 8 4

8 5

South

J 9 7

Q

9

A

10 5

5

J 9 4 2

WestQ 10 4 3

J

10

Q

9 6

7 2

10 3

East

Games Computers PlayJO

NAT

HA

N E

LDER

FIEL

D G

amm

a Li

aiso

n

CHESS• 2 players• 16 pieces each (1 king,

1 queen, 2 rooks, 2 bishops, 2knights, 8 pawns)

• Goal: Capture opponent’s king(checkmate)

• Rules:Pieces are captured whenlanded on by opponent’s pieceType of piece dictates move-ment options

• Program: Deep Blue• Web site: www.chess.ibm.com/

meet/html/d.3.html• Advantage: Too close to call

*Indicates commercial software that runs on personal computers

MAN AGAINST MACHINE: Although the IBM computer DeepBlue officially beat world champion Garry Kasparov (two wins,one loss, three ties), it is not entirely clear whether the computerwould have won had Kasparov played in his usual top form.


Exploring Intelligence 87Computers, Games and the Real World

About 30 billion neurons are laid out in six layers of the cor-tex, the gray matter (“thinking” neurons) making up the outerfolds of the human brain. An additional 70 billion constitutethe white matter (“connecting” neurons). This massively paral-lel configuration is good at recognizing patterns but is chal-lenged by serial calculations, such as searching.

In contrast, Deep Blue contains only 480 chess-specificprocessors, each of which is capable of examining about twomillion chess positions a second. This setup enables it tosearch many positions in very little time. Although this search-intensive approach provides a certain advantage, it has limita-tions. Even with the most powerful computer imaginable—say,one performing perhaps 1017 operations per second (one oper-ation in the amount of time it takes light to traverse the spaceof a hydrogen atom)—you still would not be able to make adent in solving a game such as Go, for which there are some10170 possible positions.

When humans play a game that depends on a brute-forcesearch through an enormous number of positions and strate-gies, such as Othello, we cannot use methods that depend onthe possibility of executing a billion instructions a second; theneurons in our brain fire at a millionth of that rate. Becauseour brains operate so slowly, serial methods to solve problemsare generally ineffective for us.

Of course, 1,000 operations per second may be relatively

slow, but humans can still use the serial method in a pinch.We use serial methods when we multiply large numbers, forinstance. We also commonly use them to solve puzzles, suchas Rubik’s cube, and brainteasers, such as the old problem offiguring out how to get three missionaries and three cannibalsacross a river using a single rowboat that can hold two peo-ple, subject to the condition that the cannibals cannot out-number the missionaries on either bank because they will eatthem. (In the more modern version of the problem, the mis-sionaries cannot outnumber the cannibals because they willconvert them.) Although we are capable of applying reason-ing to solve brainteasers, the reasoning itself seems veryunnatural because we typically give up on parallel methodsand instead search through the possible combinations. Forartificial intelligence, this method is referred to as “puzzlemode” reasoning. Machines are good at it, because their hard-ware is designed for it, but humans are not.

And unlike the human brain, which is stuck where it is,computers can improve their game-solving ability with fasterhardware and more efficient programs. In most game-playingprograms, a computer is programmed to analyze only thosemoves close to the current position, to avoid a massively deepsearch. Its move options can be assigned values using a proce-dure known as minimaxing. An additional technique, calledalpha-beta pruning, allows the computer to compute these

SCRABBLE• 2 to 4 players• 100 tiled letters• Goal: Accumulate most points

by creating high-scoring words• Rules:

Each player draws 7 lettersEach letter has a valueSquares on the board havevaluesWords created must join anarray

• Program: Maven* (used inScrabble CD-ROM)

• Web site: www.hasbroscrabble.com/cd/cd.html

• Advantage: Machine, by aslight margin

GO• 2 players• Black-and-white stones• Grid size of board can vary:

typical game is on 19-by-19grid points

• Goal: Conquer a larger part ofthe board (conquered partencompasses stones placedon board plus stones thatcould be added safely—thatis, within the player’s walls)

• Rules:Both sides alternate in plac-ing stones on the boardStones surrounded by anopponent’s stones are cap-tured and removed from theboard

• Program: Handtalk*• Web site: www.webwind.com/go• Advantage: Human, by a huge

margin

S1

I1

L1

I1

C3 A1 R1

A1 L1 K5 Z10 T1 B3 D2

B2 O1 N1

O1

N1

K10 7J 3

POKER (Texas Hold ’Em)• 3 to 20 players• 2 cards dealt to each player;

5 cards placed in center of table• Goal: Obtain the best hand and

win the “pot”• Rules:

5 center (community) cards startfacedownFirst round of betting ensues; 3 community cards are turnedoverSubsequent rounds of bettingensue; 4th and 5th communitycards turned overPlayers select best 5 from thecommunity cards and theirhands to obtain identical kindsof cards (pairs, 3- and 4-of-a-kind), flushes (all same suit),straights (sequential) or theircombinationsFinal round of betting ensues

• Program: LOKI• Web site: www.cs.ualberta.ca/

~games/poker• Advantage: Human, by a huge

margin

OTHELLO• 2 players• Black-and-white disks• Goal: Have most disks on the

board at the end of the game• Rules:

Players alternate placingdisks on unoccupied boardspacesIf opponent’s disks aretrapped between other player’sdisks, opponent’s disks areflipped to the other player’scolor

• Program: Logistello• Web site:

www.neci.nj.nec.com/homepages/mic/log.html

• Advantage: Machine

*Indicates commercial software that runs on personal computers JOH

N M

ENG

EL P

onzi

& W

eill



values without examining every imaginable possibility. Thisstrategy enables the computer to perform a more effectivesearch given its fixed computational resources [see illustrationabove].

There have been attempts to produce chess computersthat play the game in a more humanlike way. But the perfor-mance of these systems has been modest at best. Nobel laure-ate Herbert A. Simon of Carnegie Mellon University believesthat the basic differences in architecture may prevent efficienthumanlike reasoning in computers: “It could be that because ofthe radical differences between electronic devices and brains,programs designed to be efficient [intelligent programs] wouldbe totally different in architecture and process from systemsdesigned to simulate human thinking.” Thus, the hardware maymake what we consider efficient reasoning not so efficient whenrun on a silicon-based system.

Prowess Follows Process

In summary, some problems are best solved using patternmatching, whereas others are best solved using serial search.Go seems to be a fundamentally parallel challenge; Othelloseems to be serial. Other games, such as chess, seem equallyamenable to both methods.

This division also exists for games of imperfect informa-tion, where the players do not know what cards or other assetsare held by their opponents. (A game of perfect informationprovides all players with complete data about the state of thegame; backgammon, chess, checkers and Othello are exam-ples.) Imperfect-information games have historically eluded

competent computer play. In these games, such as poker andbridge, humans often rely on experience and intuition to playwell. Computers are at a disadvantage; there is just too muchinformation to process. But recently massive computationalresources have brought games of imperfect information closerto being solved. Daphne Koller and Avi Pfeffer of StanfordUniversity suggest that it will be possible to “solve poker.”They base this conclusion on successful algorithms theydesigned to play perfectly using an eight-card deck. The ques-tion now is if enough computational power will ever be avail-able to apply the same approach to a 52-card deck.

In bridge, too, the power of exhaustive methods is begin-ning to be felt. Earlier programs modeled human thinking,but a modern approach works as follows. The opponents arerepeatedly given specific random hands, and the result is thenanalyzed assuming that the locations of all the cards areknown. This approach identifies the best play or plays in oneparticular situation, and the best play overall is the one that isbest in as many of the random situations as possible. Onceagain, this style of analysis is possible primarily because ofincreasing hardware power. Although the best humans stilloutplay the best programs, the gap is narrowing. As in chess,programs that attempt to model human thinking are nomatch for their search-based competitors.

In all these cases, Simon’s speculation on the innateimportance of architecture has proved correct. The strongestmachine players do indeed bear little resemblance to theirhuman counterparts. The only game that seems to be anexception is backgammon. In this case, Gerald J. Tesauro of IBMcreated a program called TD-Gammon that appears, at least

Player 1 (maximizer): computer

Player 2 (minimizer): human

a

b c d

e f g

h i j

l m

k

n o

rqp

1

-1

1

11

-1

-1

-1 -11

Alpha-beta pruningeliminates parts of thetree that are poor pathchoices relative toother pathways. Forexample, once thecomputer recognizesthat it can win bymoving to node c, itno longer needs toanalyze any of its otheroptions from node a.

How Game-Playing Computers Think

Minimaxing assigns one player apositive value and the other anegative value. (In this case, thecomputer is player 1, the maxi-mizer, and the human is player2, the minimizer.) Once estab-lished, the values assigned eachnode can be judged as benefitingthe maximizer (the “positive”player) or the minimizer (the“negative” player). Each noderepresents the positions of allthe pieces after a move has beenmade—more generally, the stateof the game at the moment.

MEG

CR

AN

E


Computers, Games and the Real World

on the surface, to have a humanlike architecture. Specifically,it relies on an artificial neural network, which is softwaredesigned to mimic the function and structure of neurons.

Games People (and Computers) Play

The nature of the game and the type of solution process-ing most conducive to success often dictate whether humanor machine plays better. Machines excel at checkers, a gamethat involves deep-search methods to evaluate possible combi-nations. Chinook, currently the best checkers program, knowsimmediately which player (if either) will win once there areeight or fewer checkers on the board. It determines the out-come by accessing its enormous database of endgames. Thistask is not a pattern-matching problem, because there are noreliable patterns that characterize all the positions; insteadChinook simply stores a huge table of the positions and val-ues and looks in that table when a result is needed.

Machines also have a tremendous advantage in Othellobecause it is a difficult game for humans to get a feel for; it isalmost impossible to tell at a glance who is winning at any par-ticular point. The markers flip back and forth so frequently thathaving a preponderance of markers with your color in the mid-dle of the game does not necessarily indicate that you are win-ning. In other words, Othello is a game that is difficult to playusing pattern matching but is perfectly suited to a machine’sbrute-force search methods. Therefore, machines excel.

Go, however, is the computer’s Achilles’ heel. It is a pat-tern-matching game in which human experts can generallyrecognize large configurations of stones that are dead (sur-rounded and fated to eventual capture), alive (permanentlysafe from capture) or whose fate is currently undetermined.Playing Go using brute-force search, however, is extremelydifficult, because at any given point there will be some 250legal moves. (In comparison, there are only approximately 30legal moves in chess and seven in Othello.) Building an end-game database such as Chinook’s is completely out of thequestion because the number of possible ending positions in aGo game is beyond a computer’s computational power.

And other games? The best in backgammon is a close callbetween human and computer and is very likely to remain so:because of the relative simplicity of the game, both people andmachines play backgammon nearly perfectly. Scrabble is alsoclose but not very likely to remain so, as machines become bet-ter at the strategic parts of the game. (Both humans andmachines have no difficulty playing the highest scoring moveat any point, but high-level Scrabble is also a matter of keepinga good balance of tiles in your rack and minimizing your oppo-nent’s opportunities.) Humans won a two-game match in 1997but lost an 11-game match in 1998. Bridge is likely to becomeclose in about five years, as hardware becomes faster and algo-rithms improve. Humans narrowly won a two-hour match thispast July, and at the 1998 World Bridge Championships, held inAugust in Lille, France, the bridge program I wrote placed 12thin a field of 34 of the top human bridge players.

Playing Nicely Together

As artificial intelligence has developed, success has comemost often through the application of good serial algorithms.These successes are not limited to game playing: serial algo-rithms are the most effective known solutions for selectingthe order in which to assemble the component parts of a

Boeing 747. These algorithms lead to production schedulessome 10 to 20 percent shorter than the best schedules pro-duced by humans. Pattern matching and other parallel tech-niques are not terribly well suited to scheduling complextasks, because a schedule that looks good may in fact not be.

At least for the foreseeable future, humans will be betterthan machines at solving parallel problems, and machines willbe better at solving serial ones. There should be nothing particu-larly surprising or threatening here; we have designed machinesto achieve other results that we ourselves are not capable ofachieving, such as airplanes to fly us or forklifts to raise objectsour muscles cannot move. And we do not feel compelled to pithumans against forklifts in Olympic weight lifting.

The lesson is that intelligent machines are not our com-petitors but our collaborators. The complementary skills ofhumans and machines enable problems to be solved that nei-ther could have figured out alone. And that is exceptionallygood news for us all, carbon and silicon alike.


MATTHEW L. GINSBERG is a senior research associate at theUniversity of Oregon and founder of the university’s ComputationalIntelligence Research Laboratory (CIRL). He received his doctorate inmathematics from the University of Oxford in 1980, where he alsocaptained the bridge team. Although he currently plays little competi-tive bridge, he has remained in the game via his alter ego, an expert-level bridge-playing program called GIB (Ginsberg’s Intelligent Bridgeplayer). “On a personal level, GIB has been a blast,” Ginsberg says. “Ican interact with top players, but no one is angry when the computerbeats them.” Pending software adjustments, Ginsberg predicts that GIBwill become the world’s top bridge player in five years. When not con-ducting research, supervising graduate work at CIRL or tinkering withGIB, Ginsberg likes to relax with a few loops and rolls in his kit-builtstunt plane. Lately he has cut down on that thrill to once a week,choosing instead to spend time with his wife, his infant daughter andhis four-year-old son, who has just learned to play chess.

About the Author

SA

ROB

ERT

KA

ISER


Research on intelligence is mostly aboutinvestigating how brains work or buildingintelligent machines or creating “smart”environments such as a house that canidentify and track its occupants. But whatabout making people smarter? To accom-plish this goal, one can consider biochem-istry or bioimplants, but the easiest way toimprove intelligence is by augmenting theitems we wear all the time—glasses, wrist-watches, clothes and shoes—with miniaturecomputers, video displays, cameras andmicrophones. These high-tech “wearables,”which are being developed at theMassachusetts Institute of Technology MediaLaboratory, can extend one’s senses, improvememory, aid the wearer’s social life and evenhelp him or her stay calm and collected.

The idea of increasing intelligence withwearable devices is very old. English physi-cist Robert Hooke wrote in 1665 (in thepreface to Micrographia): “The next care tobe taken, in respect of the Senses, is a sup-plying of their infirmities with Instruments,and as it were, the adding of artificial Organsto the natural.... And as Glasses have highlypromoted our seeing ... there may be foundmany mechanical inventions to improveour other senses of hearing, smelling, tastingand touching.”

One must draw a distinction betweenwearable devices and those that are merely

Miniature computers built into clothes, shoes and eyeglasses may become the“smartest” new fashion accessories

Machine Intelligence

by Alex P. Pentland

Wearable Intelligence

INTELLIGENT CLOTHES were recently dis-played at a fashion show at the Massachu-setts Institute of Technology Media Lab. Atthe left, a model wears a television reporter’soutfit, equipped with a video camera in herglove and a heads-up display in her glasses.

SAM

OG

DEN


Exploring Intelligence 91Wearable Intelligence

portable, the classic example being thepocket watch and wristwatch. The dif-ference is simple: you have to pull outthe pocket watch and open it to see thetime, whereas the wristwatch enablesyou to see the time instantly, even whileworking with both hands. Although thismay seem like a minor difference, itgreatly affects how you use the deviceand how completely it is integrated intoyour life. Watches, eyeglasses and radioshave all evolved from handheldportable versions to wearable items, andmany of today’s portables are destinedto become tomorrow’s wearables.

Electronic devices are beginning tomake the transition from portable towearable. There are now wristwatchesthat contain medical monitors andpagers; eyeglasses with embedded com-puter displays that only the user cansee; vests, belts and watches with com-puters inside them; and cell phones andpagers that come with Internet connec-tions and tiny teleconferencing cameras.

Equipped with wearable computersand other devices, people can conve-niently check messages or finish a pre-sentation while sitting on the subway orwaiting in line at a bank. Even moreimportant, they can also ignore thesemachines while attending to otheraffairs. Operating portable devices, incontrast, often requires your full atten-tion and both hands. You have to stopeverything you are doing and concen-trate on the device. To appreciate howinconvenient this situation is, imaginehaving human aides (instead of elec-tronic aids) who grabbed your handsand shouted in your face every timethey had something to say.

Wearable devices can be much lessdisruptive, and people relate to themdifferently than they do to other tools.Something that’s with you all the timecan change the sense of who you areand what you can do. As we adapt towearable devices and shape our personalhabits around them, over time the cul-ture as a whole will shift to incorporatethem.

Hardware That’s Made to Wear

Psychological studies show thevalidity of the phrase “the clothes makethe man.” Our self-perception and self-confidence can indeed change with ourclothes. The same is true for any con-stantly available device—and not alwaysfor the better. Those of us who are con-tinually “on call” via a pager know how

fundamentally these tools can alterone’s life. The personal effects of manyinformation and communications tech-nologies recall Marshall McLuhan’s dic-tum, “the medium is the message’’—that is, the way in which a new technol-ogy changes our way of life can be moreimportant than the information it con-veys. But wearables are more personalthan traditional communications toolsbecause they are a constant part of one’sphysical presence: they are not onlypart of what you wear but also part ofwho you are.

In the near future, the trend-settingprofessional may wear several smalldevices, perhaps literally built into theirclothes. A person “dressed for success”in this manner may appear to have afantastic memory, to be amazinglyknowledgeable and to have powers ofdetection and deduction second only toSherlock Holmes. These wearable intelli-gence devices can enhance one’s “mem-ory” by providing instant access tobooks, digitized maps, calendars andvarious databases; providing wirelessconnections to the Internet and e-mail;and boosting one’s awareness withvarious sensors.

The hardware technology forthis scenario has already beendeveloped at research universitiessuch as Carnegie Mellon, theGeorgia Institute of Technologyand the M.I.T. Media Laboratory,at large companies such as IBM,Toshiba and Motorola, and atstart-up companies such asMicroOptical in Boston and theFlexible PC Company inNorthfield, Minn. In my “wear-ables closet’’ I have glasses with aprivate, full-resolution computerdisplay; a health monitor in awatch that records my tempera-ture, heart rate and blood pres-sure; a computer-in-a-belt with awireless Internet connection; alapel pin that doubles as a cameraand microphone; and a touchpador keyboard literally sewn into ajacket. Soon there may be no needfor batteries because sufficientpower can be generated by har-vesting the excess energy in nor-mal walking (for example, byputting piezoelectric materials inshoes that generate electricitywhen compressed). Connectingwires might also become unneces-sary by making use of conductivefibers woven into clothes or by

transmitting small amounts of powervia radio signals and skin conduction—all of which have been demonstrated atthe Media Lab.

It is too early to tell whichapproaches to wearable design will provepopular. Some people, for example, maybe comfortable with head-mountedvideo displays; others may find themunwieldy. Some may like the feel ofheadsets similar to those used by tele-phone operators; others may prefer lessconspicuous audio and speech interfaces.The devices can be built in many ways,and it will take a fashion and style battleto determine what people really want tobuy. The Media Lab, however, is not tak-ing a passive attitude toward this issue.On the contrary, several years ago I ini-tiated a collaboration with some of theworld’s most famous design schools tosee what future wearable fashion wouldlook like. Some tantalizing visions havealready emerged from this effort.

As with most new technologies,wearables will probably make theirbiggest inroads in specialized tasks beforebecoming widely adopted by the gener-

SAM

OG

DEN

VEST WORN by a model is designed totranslate a person’s speech into anotherlanguage. Microphones on the front of thevest record the voice, and speakers on theshoulders broadcast the translation.



al public. Wall Street traders could relyon the devices for information neededto make quick decisions on the tradingfloor. Doctors could use them to storemedical records and take pictures andnotes. Industrial inspectors and scientistsworking in the field could jot down theirobservations while walking around.Repair workers could obtain assistancein the midst of a complicated job.

Thousands of Federal Express deliv-ery people are now equipped with wear-ables so that the company can bettercoordinate its efforts around the world.Another company, Symbols Technologiesin Holtsville, N.Y., makes a wearabledevice in the form of a ring with a built-in Universal Product Code (UPC) readerand computer display. When pointed atany product that has a UPC bar code,the ring can automatically pull up con-sumer reviews, instruction manuals andother information about the product.

Tailoring for a Better Fit

Although their potential is vast,many of these devices suffer from a com-mon problem: they are mostly obliviousto you and your situation. They do notknow what information is relevant toyou personally or when it is sociallyappropriate to “chime in.” The goal insolving this problem is to make electronicaids that behave like a well-trained but-ler. They should be aware of the user’ssituation and preferences, so they know

what actions are appropriateand desirable—a task I call“situation awareness.” Theyshould also make relevantinformation available beforethe user asks for it and with-out forcing it on the user—atask I call “anticipation andavailability.”

To enhance situationawareness, the wearabledevice can employ sensors ofvarious types to determinewhere the user is and whathe or she is doing. The devicecan also monitor the user’schoices and build a model ofhis or her preferences. A per-son can actively train thecomputer by saying, “Yes,that was a good choice; showme more,” or “No, neversuggest country music tome.” The models can alsowork solely by statisticalmeans, gradually compiling

information about the user’s likes anddislikes. (Firefly Network, a companystarted by my Media Lab colleagues, takesthis approach, recommending books ormovies to people depending on how theirtastes match the profiles of millions ofother users.)

For anticipation and availability, thewearable device can take a few key factsabout the user’s situation to promptsearches through a digital database or theWorld Wide Web. The informationobtained in this manner is then presentedin an accessible, secondary display out-side the user’s main focus of attention.

A good example of memory aug-mentation devices that use these design

principles are electronic navigation aidsrelying on Global Positioning System(GPS) satellites. The typical navigationaid has a display that constantly showsone’s current position on a map andindicates with an arrow how to reachthe stated destination. There are hand-held versions for hikers and plug-ins forlaptops, but most are currently found inautomobiles.

The promise of never being lost is astrong selling point for these aids, butthey can do much more: for example,the devices can use your current loca-tion to call up information about near-by landmarks. You don’t have to type inqueries to find the local restaurants orgas stations; everything is retrieved onthe basis of your present position. Theease and utility of such automaticindexing is making navigation aids ahuge commercial success.

A wearable version of this device isnow being manufactured by Motorolawith advice from my students and me.The U.S. Army will soon be field-testing50 of these GPS navigation and commu-nications systems on its troops andplans to outfit tens of thousands of sol-diers with the technology in the nextfew years. Of course, the same wearableaids can be used just as easily by tourists.The system would include a GPS sensor,a wireless connection to a digital data-base (such as the Web), a microphoneand a digital camera. When you visit atourist attraction, the device could showyou the historical facts about the siteand give you directions to the next stopon your itinerary. You could then wan-der around at your leisure with thewearable equivalent of a personalizedtour guide.

RING WITH BAR-CODE READER is alreadybeing sold for use in warehouses and loadingdocks. The device allows workers to identifythe contents of a carton and to read shippingand handling instructions.

SAM

OG

DEN

DANCING SHOESdeveloped by JoeParadiso, a scientistat the M.I.T. MediaLab, convert dancesteps into music. Theshoes could alsoinform runners of akink in their stride.

SAM

OG

DEN



This type of sys-tem relies on database“filters’’ and “agents”that use informationabout the wearer’s sit-uation—for example,his or her location andthe time of day—tofetch pertinent dataand to label theimages, text andsounds that the wearermight find interesting.Although softwareagents will never beable to magicallyanticipate our desires,they can discernsomething about ourpatterns from statisticalanalyses. For instance,depending on theuser’s instructions,today’s tools can bringto one’s immediateattention e-mail froma spouse or boss and save the rest forfuture perusal.

Although it is impossible to sayexactly where wearable intelligencetechnology is heading, I suspect thetrend will be toward devices with greatersituation awareness, achieved in part byusing additional sensors such as camerasand microphones. My research, forinstance, focuses on building wearablesthat attempt to see what the user seesand hear what the user hears. My idea isthat you can’t know what people mightbe interested in unless you know whatthey are hearing and seeing.

In essence, I want to give thesoftware agents eyes and ears so thatthey can better understand—and thusbetter help—their human users. As thetools my students and I build becomemore reliable, and people becomemore comfortable with them, theagents could be allowed greater ini-tiative. I expect that eventually manyof the small tasks that complicate ourlives will be delegated to such agents.

To this end, my research grouphas built wearables with small cam-eras mounted on the user’s hat orglasses, which employ computervision techniques to recognize whatthe person is looking at, without theneed for bar codes or other tags.Once the computer knows what theperson is looking at, it automaticallygathers information about that object.When you meet someone, the com-

puter can run a face-recognition programon the camera image to remind you ofthe person’s name and other noteworthyfacts. Alternatively, this computer-camerawearable can function as an expertadviser, analyzing the layout of balls ona pool table, for example, and identify-ing the best available shot.

A wearable with a downward-look-ing camera provides another interestingsource of situation awareness, because itcan see what you’re doing and observeyour hand gestures. One version of thissystem reads American Sign Language

and converts it into audible English.Another version can distinguish betweenactivities such as handshakes, typing anddriving, so that it can tailor its informa-tion retrieval to fit the current activity.(Technology similar to that used forspeech recognition can be used to iden-tify patterns of motion rather than pat-terns of sound.)

Other kinds of situation awarenesscan be achieved with audio input. Wehave built wearables that know whenyou and another person are talking, sothat they don’t interrupt. Although thesystem is not 100 percent accurate, weplan to make it better by using a camerathat can determine whether sounds arecoming from you or someone you’retalking with, rather than from anybodyin the vicinity.

We have also built devices that rec-ognize what you are saying and thentranslate your phrases (albeit crudely)into another language. This system couldbe helpful for travelers and for peoplewith serious speech impediments.Currently the technology works only for


AFFECTIVE WEARABLE monitors thestress of Jennifer Healey, a graduate stu-dent at the M.I.T. Media Lab, by mea-suring skin conductivity and tempera-ture. The data can be downloaded to aPalmPilot, which can display the user’svital signs.

SAM

OG

DEN

SOCIAL WEARABLES canbe used as icebreakers atparties. The devices flashinfrared light to communi-cate the names of theirusers and any informationthey would like to share.

SAM

OG

DEN



a user who has “trained” the system.Simultaneous translation of a two-wayconversation, unfortunately, is still adistant dream.

Wearable items can offer importantsocial benefits, helping us recognize peo-ple we might otherwise ignore or pro-viding new ways for strangers to findcommon ground. In addition to assistingour memory, wearable intelligence aidscan also augment our other talents andabilities in such areas as music, danceand athletics. One device (called “danc-ing shoes”) converts dance steps intomusic; another can assist people withtheir baseball swings.

The Body Electric

Wearable medical monitors shouldbecome increasingly useful in the future.Today your doctor gets a “snapshot” ofyour physical condition about once ayear, which is far too infrequent to catchincipient diseases. The failure to do moreregular health monitoring is particularlyproblematic for the elderly, whose con-dition can change quite quickly. Evenmore troubling is the fact that current

medical specialists cannot explain howmost problems develop, because theyonly get to see people when somethinghas gone wrong.

Yet it is now possible to build smalldevices that continuously monitor a widerange of vital signs. My students and Iare working with the Center for FutureHealth at the University of Rochester todevelop such medically oriented wear-ables, including early-warning systemsfor people with high-risk medical prob-lems and “elder care’’ wearables that willhelp keep seniors out of nursing homes.Another simple but important applica-tion for medical wearables is to give peo-ple feedback about their alertness andstress level—an approach currentlyunder study at the Media Lab by RosalindW. Picard’s “affective wearables” group.

A system that constantly tracksone’s vital signs could yield helpfulinformation, but it could also overwhelmthe wearer with raw data, making itdifficult to reach any decision. Similarconcerns can be raised about wearableintelligence systems in general, whichcould potentially swamp users with toomuch data and leave them feeling

burned out from the lack of “downtime.” The devices might also encouragepeople to retreat further into themselvesand their machines, leading to greatersocial isolation.

I agree that poorly designed wear-ables could cause such problems. Butinformation overload and social disrup-tion are usually not caused by too muchinformation or connectivity per se. Afterall, business and government leadershave always dealt with huge amounts ofinformation and organizations withthousands of people. Instead it seems tome that most difficulties arise wheninformation and communications arenot properly integrated into our dailyroutine.

To avoid these problems, we needwearable devices that are organizedaround the pattern of our lives, ratherthan organizing our lives around them.I think we can accomplish this goal bymaking wearable devices that aresufficiently aware of their surroundingsand the likes and dislikes of their users.

It should be noted that when booksfirst became cheap and portable items,many feared that family life and popular

POOL-PLAYING WEARABLE includes a head-mounted camera thatrecords the position of the balls on the pool table. Specially designed soft-ware analyzes the possible shots and identifies the easiest one. A heads-up display (inset) helps graduate student Tony Jebara line up the shot.

SAM

OG

DEN

SAM

OG

DEN

SIGN-LANGUAGE TRANSLATOR is demonstratedby its developer, Thad Starner, one of the MediaLab’s original “cyborgs.” The video camera on hishat records his hand gestures, and motion-recogni-tion software converts the signs into English. In thephotograph Starner is signing the word “bicycle.”


culture would disintegrate, as people spent more time readingand less time talking. The impacts of eyeglasses and watcheswere also hotly debated in their time. But despite gloomy pre-dictions, books, watches and glasses are now an accepted partof our lives. We’ve grown accustomed to them, and I think weare considerably better off, even though we are different peo-ple because of them. Wearable intelligence aids will cause sim-ilar adjustments.

The great advantage of wearable devices is that they can bewith you constantly, serving as a mental aid that is part of yourbody and part of your everyday life. If we can endow thesetools with sufficient situation awareness to make them a helprather than a hindrance, they offer the promise of enhancinghuman intelligence in a seamless and enjoyable way.


ALEX P. PENTLAND is the academic head ofthe M.I.T. Media Laboratory, Toshiba Professor ofMedia Arts and Sciences at M.I.T. and external direc-tor of the Center for Future Health at the Universityof Rochester. He is a founder of the IEEE WearableComputing technical area and has published over200 papers in the fields of wearable computing,machine and human vision, human-machine inter-face, computer graphics and artificial intelligence.Newsweek magazine recently named him one of the100 Americans most likely to shape the next centu-ry. “I got into this field because I’ve always beenunhappy with traditional theories of intelligence,”he says. “There are different aspects of humanintelligence, each of which can be augmented.”

In addition to developing wearable devices,Pentland enjoys exploring their use in dance, fash-ion and other artistic endeavors. In recent years hehas organized wearables dance performances inHollywood, wearables fashion shows in Paris,Tokyo and Boston, and wearables performancepieces at various sites around the world. He wouldlike to thank the Media Lab’s former and currentgraduate students—particularly Thad Starner,Bradley Rhodes and Steve Mann—and colleaguesNeil Gershenfeld, Mike Hawley, Pattie Maes, JoeParadiso, Alice Pentland, Rosalind Picard andMitch Resnick.

FURTHER READING

THE ROMANCE OF AERONAUTICS: AN INTERESTING

ACCOUNT OF THE GROWTH AND ACHIEVEMENTS OF ALL

KINDS OF AERIAL CRAFT. Charles Cyril Turner. J. B.Lippincott Company, 1912.

THE PREHISTORY OF FLIGHT. Clive Hart. University ofCalifornia Press, 1985.

THE EMPEROR’S NEW MIND. Roger Penrose. OxfordUniversity Press, 1989.

TURING TEST CONSIDERED HARMFUL. P. J. Hayes and K.M. Ford in 14th International Joint Conference onArtificial Intelligence, pages 972–977. MorganKaufmann Publishers, 1995.

WHY GÖDEL’S THEOREM CANNOT REFUTE

COMPUTATIONALISM. G. LaForte, P. J. Hayes and K. M.Ford in Artificial Intelligence, Vol. 104, Nos. 1–2,pages 265–286; October 25, 1998.

THE AGE OF INTELLIGENT MACHINES. RaymondKurzweil. MIT Press, 1990.

ESSENTIALS OF ARTIFICIAL INTELLIGENCE. Matthew L.Ginsberg. Morgan Kaufmann Publishers, 1993.

DO COMPUTERS NEED COMMON SENSE? Matthew L.Ginsberg in Proceedings of the 5th InternationalConference on Principles of Knowledge Representationand Reasoning (Cambridge, Mass.). MorganKaufmann Publishers, 1996.

ONE JUMP AHEAD: CHALLENGING HUMAN SUPREMACY IN

CHECKERS. Jonathan Schaeffer. Springer-Verlag,1997.

DEEP BLUE VS. KASPAROV, ROUND TWO. On theScientific American Web site atwww.sciam.com/explorations/042197chess/

SMART ROOMS. Alex P. Pentland in Scientific American,Vol. 274, No. 4, pages 68–76; April 1996.

AFFECTIVE COMPUTING. Rosalind W. Picard. MIT Press,1997.

AUGMENTED REALITY THROUGH WEARABLE COMPUTING.Thad Starner et al. in Presence: Teleoperators andVirtual Environments, Vol. 6, No. 4, pages 386–398;August 1997.

FIRST INTERNATIONAL SYMPOSIUM ON WEARABLE

COMPUTERS. Digest of papers. IEEE ComputerSociety Press, 1997.

WEARABLE COMPUTING: A FIRST STEP TOWARD PERSONAL

IMAGING. Steve Mann in Computer, Vol. 30, No. 2,pages 25–32; 1997.

WEARING YOUR COMPUTER. Wendy M. Grossman inScientific American, Vol. 278, No. 1, page 46;January 1998.

Information on wearable devices is available atwww.media.mit.edu/wearables on the World Wide Web.

GAME-PLAYING MACHINES

RETHINKING ARTIFICIAL INTELLIGENCE

WEARABLE COMPUTING

MAC

HIN

EIN

TELL

IGEN

CE

SAM

OG

DEN


About the Author

SA


Making first contact

is equivalent to

finding a needle in

a haystack 35 times

the size of Earth.

Actively sending

announcements to

introduce ourselves

may be the best way

Is There Intelligentby Guillermo A. Lemarchand

INTELLIGENCETo consider Earth the only populatedworld in infinite space is as absurd asto assert that in an entire field sownwith millet, only one grain will grow.

—Metrodorus of Chios,4th century B.C.

EXTR

ATER

RES

TRIA

LTH

ESE

ARCH

FOR

The Search for Extraterrestrial Intelligence96 Scientific American Presents


Life Out There?The land lies sleeping under theenveloping mantle of night.Bright stars gleam like jewels fromthe velvet darkness. Beyond, indepths frightening in their sheerimmensity, the Milky Way trailsits tenuous gown of stardustacross the heavens, and wellbeyond that billions of stars,galaxies and planets dance in acosmic symphony.

From our earliest days, humanshave strongly sensed that this end-less majesty is too huge to be con-templated by a single intelligentspecies, and one thread that linksthe ancient Greek philosophers tomodern space scientists is thedesire to know whether otherinhabited worlds exist. Vast andold beyond understanding, theuniverse forces us to ponder theultimate significance of our tinybut exquisite life-bearing planetand to long for the knowledge thatsomewhere out there, someone likeus is gazing toward the heavensand having similar thoughts.

We have the means to test the

possibility that advanced extrater-restrial civilizations exist. This isthe search for extraterrestrialintelligence, known as SETI.

The Chances of Intelligent Life

Despite its roots in some ofour most profound questions, thegoal of SETI is not to fulfill a spiri-tual longing. Instead it is a realis-tic, practical response to the sta-tistical likelihood that the evolu-tion of life is a natural occurrenceeverywhere across the universe.

SETI operates under a two-pronged hypothesis. The firstassumption, known as the princi-ple of mediocrity, is that thedevelopment of life is an unexcep-tional consequence of physicalprocesses taking place in appro-priate environments—in this case,on Earth-like planets. Because ourgalaxy has hundreds of billions ofstars and the universe has billionsof galaxies, many habitable Earth-like planets should exist, and lifeshould be common.

The second assumption is thaton some planets that shelter livingcreatures, at least one species willdevelop intelligence and a techno-logical culture. They will have aninterest in communicating withother sentient beings elsewhere inthe cosmos and will beam signalsinto space with that goal. Assum-ing that these cultures would useelectromagnetic signals to com-municate and that their signalswould have an artificial signaturewe could recognize, it should bepossible to establish contact andexchange information.

Life as we know it could onlyexist on Earth-size planets becauseliquid water—apparently a prereq-uisite for organisms similar to ter-restrial ones—seems to occur onlyon planetary bodies that size.Recent astronomical observations indicate that planetary systemsare common. In just the past threeyears, researchers have detected 13planetary systems orbiting sunlikestars. Current detection methodsprevent us from knowing how sim-

Is There Intelligent Life Out There? Exploring Intelligence 97

MEG

CR

AN

E


ilar these new systems are to our ownsolar system. At best, we know the newplanets to be gas giants like Jupiter.

Based on these discoveries, SETI pio-neer Philip Morrison of the Massachu-setts Institute of Technology has esti-mated that the number of planetary sys-tems just in our own galaxy could rangefrom as few as 10 million to as many as100 million. From this estimate, I believewe can realistically speculate about thenumber of Earth-like planets.

The most important determining fac-tors are how frequently Earth-like planetsare formed when a planetary systemdevelops and how soon afterward lifewould appear. The best current estimatesof the number of planets close in mass toEarth, combined with the best currentestimates of the long-term stability ofoceans, suggest that one or two possibleworlds around every sunlike star haveenvironments suitable for life—essential-ly the profile of our own solar system.

Based on Earth’s history, life emergesrelatively quickly. When Earth formedsome 4.6 billion years ago, it was a life-less, inhospitable place. Only one billionyears later the whole planet was teemingwith one-celled organisms resemblingblue-green algae. The principle of medi-ocrity suggests a logical progression: theemergence of life will lead to the emer-

gence of intelligence, which will give riseto interstellar communications technol-ogy. Viewed from one angle, it may besaid that SETI is simply an attempt totest this theory.

There are no guarantees, though,that life everywhere will develop alongthe path followed on Earth and lead tointelligence. For example, Ernst Mayr ofthe Museum of Comparative Zoology atHarvard University notes that out ofsome 50 billion species that have arisenon Earth, only one achieved the kind ofintelligence needed to establish a civi-lization. Intelligent life on Earth occu-pies less than 0.025 percent of the totalhistory of life here. Mayr believes thatsuch high intelligence may simply notbe favored by natural selection: after all,every other species on Earth gets alongfine without it.

Another possibility is that highintelligence is extraordinarily difficult ordangerous to acquire. For example, twoor more competing intelligent speciescould destroy each other before eithercould give rise to a technological civi-lization. If this is so, we probably cannotexpect more than one intelligent speciesto exist on any planet.

Technological civilizations mustalso survive long enough to be discov-ered. The late Carl Sagan referred to our

period as “technological adolescence,”when technology brings the civilization-ending threats of ecological catastrophe,exhaustion of natural resources andnuclear war. There may be 100 millionsuitable planets in our galaxy, and if evena small fraction of civilizations survivetechnological adolescence, then the pos-sible number of galactic civilizations maystill be very large. Sagan considered anestimate of one million of them in thegalaxy to be conservative.

Making Conversation

SETI researchers also assume that thephysical laws governing the universe arethe same everywhere. If so, then weshould be able to communicate throughour common principles of mathematics,physics, chemistry and so on.

Not all researchers, however, agree.Nicholas Rescher, a philosopher at theUniversity of Pittsburgh, argues thatextraterrestrials would be organisms withdifferent needs, senses and behaviors. Sodespite sharing universal laws with us,they are extremely unlikely to have anytype of science we would recognize.

But artificial-intelligence pioneerMarvin Minsky of M.I.T. argues thatintelligent extraterrestrials will think likeus, in spite of different origins, becauseall intelligent problem solvers are sub-ject to the same ultimate constraints:limitations on space, time and resources.According to Minsky, in order for intelli-gent life-forms to evolve powerful waysto deal with such constraints, they mustbe able to represent the situations theyface and to manipulate those representa-tions. To do this, every intelligence willinevitably discover the same basic prin-ciples. As a result, he says, aliens willhave evolved thought processes andcommunications strategies that willmatch our own to a degree that willenable us to comprehend them. SETIproponents largely concur.

As a tool that cuts across culturaland linguistic boundaries, mathematicswould seem to be a “cognitive univer-sal” that could be used to communicatewith extraterrestrial intelligences (ETIs).As early as 1896 Sir Francis Galton, acousin of Charles Darwin, published anessay describing a mathematical lan-guage he developed for extraterrestrialcommunication.

In 1960 Dutch mathematician HansFreudenthal created a language for a cos-mic dialogue, known as Lincos (LinguaCosmica), based on mathematical princi-


LEVE

L OF

COS

MIC

NOI

SE (

KELV

INS)

FREQUENCY (GIGAHERTZ)

GALACTIC BACKGROUND

2.76 K COSMICBACKGROUND

QUANTU

M NOIS

E LI

MIT

OF RAD

IO TEL

ESCO

PES

HYDROGENEMISSION

HYDROXYLEMISSION

0.1 1 10 100 1,0001

10

100

1,000

RADIO SPECTRUM OF SPACE has a “quiet” area between one and 100 gigahertz, inwhich noise comes only from the cosmic background radiation. The “magic frequen-cies” in this area include the frequency at which hydrogen emits energy, at 1,420 mega-hertz, and that of the hydroxyl radical (oxygen and hydrogen), at 1,667 megahertz. Thecombination of both substances yields water—possibly a universal signature of life.

AN

NA

ARM

ENTR

OU

T


ples for exchanging concepts of time,space, mass and motion. Recently Louis E.Narens of the University of California atIrvine noted that many kinds of cognitiveuniversals can be surmised by consideringpragmatic requirements—for example,what the extraterrestrials must know tobuild sending and receiving equipment.

Efforts in First Contact

Since the first formal SETI efforts,researchers have used the microwaveregion of the electromagnetic spectrumfor detecting interstellar communica-tions, because microwave signals requirelittle energy to exceed natural back-ground radiation and are not deflectedby galactic or stellar fields. Moreover,they are easy to generate, detect andbeam and are not absorbed by the inter-stellar medium or by planetary atmo-spheres. A quiet cosmic frequency win-dow exists in the microwave region,between one and 100 gigahertz.

In September 1959 GiuseppeCocconi and Morrison, both then atCornell University, proposed the first real-istic strategy for searching for ETIs. Itwould use radio-astronomy telescopes toscan the nearest sunlike stars for artificialsignals at or near the 21-centimeter wave-length (1,420 megahertz), which corre-sponds to the frequency of microwaveenergy that neutral hydrogen emits.(Hydrogen is the most abundant elementin the universe, so presumably radioastronomers in any technological civiliza-tion would scan at this wavelength tostudy the substance.) Independently,Frank D. Drake, an astronomer then atthe National Radio Astronomy Observa-tory (NRAO) in Green Bank, W.Va., wasplanning an actual search. On April 8,1960, he turned the 26-meter-wideHoward Tatel radio telescope toward thenearby solar-type stars Tau Ceti andEpsilon Eridani. Drake dubbed the searchProject Ozma, in reference to a princesswho appeared in sequels to L. FrankBaum’s book The Wonderful Wizard of Oz.

With $2,000 worth of parts, thelow-profile, low-budget Ozma systemhad only one channel with a spectralresolution of 100 hertz and a sensitivityof one one-hundred septillionth (10–22)of a watt per square meter: with currentreceiver technology, the same searchwould be thousands of times more sen-sitive. In the end, no signals were foundafter 150 hours of observation. Despiteits failure, however, Project Ozma firedthe imagination of the public.

In the Days before SETI

The first initiatives to communicate with extraterrestrial beings onthe moon or on Mars began more than 150 years ago. German mathe-matician Carl F. Gauss (1777–1855) suggested that there be erected inSiberia a giant figure of the diagram used in Euclid’s demonstration ofthe Pythagorean theorem. The hypothetical Selenites (moon dwellers),on seeing this figure through their telescopes, would recognize it ashaving been made by intelligent terrestrial beings and would respondaccordingly. In 1869 French intellectual prodigy Charles Cros(1846–1888) suggested that rays from electric lights could be focusedby parabolic mirrors so as to be visible to hypothetical inhabitants ofMars or Venus. He also presented a code using periodic flashes.

During the 1920s, an extensive debate about how to communi-cate with the hypothetical Martians began in the pages of ScientificAmerican. In those days, radio pioneers Nikola Tesla (1856–1943),Guglielmo Marconi (1874–1937) and David Todd (1855–1939) begantheir speculations about the use of radio waves for interplanetarycommunication. On January 27, 1920, the New York Times reportedthat Marconi occasionally detected with his radio equipment “veryqueer sounds and indications, which might come from somewhere out-side the Earth.” No less a scientific authority than Albert Einstein wasquoted as believing that Mars and other planets might be inhabitedbut that Marconi’s strange signals stemmed either from atmosphericdisturbances or experiments with other wireless systems.

In a 1920 Scientific American article, H. W. Nieman and C. WellsNieman proposed a system to encode messages to other planets,arguing that the key to communication was the timing—in the durationof the signal to produce dots and dashes and in the lack of a signal toproduce pauses. Their proposal was the basis for the encoding systemused in 1974 by Frank D. Drake and his colleagues in the first inter-stellar message sent from the Arecibo Observatory in Puerto Ricotoward the Great Cluster in the constellation Hercules. —G.A.L.

ENCODED MESSAGES can yield images for interstellar communications,

as proposed in the March 20, 1920,Scientific American.

SCIE

NTI

FIC

AM

ERIC

AN

Is There Intelligent Life Out There? Exploring Intelligence 99


In the early 1970s the NationalAeronautics and Space Administrationbegan to show interest in SETI. The lateBernard Oliver, vice president for devel-opment at Hewlett-Packard, and JohnBillingham, a NASA scientist, headedProject Cyclops, a summer school con-vened to design an array of 1,000 100-

meter-wide antennae to eavesdrop onthe television, radar and other “domes-tic” transmissions of hypothetical galac-tic neighbors 1,000 light-years away.Project Cyclops, however, was too ambi-tious for NASA funding and was neverbuilt. Instead, during the 1970s and1980s, NASA funding for SETI was limit-

ed to workshops and conferences. In1992 NASA launched a 10-year, $100-million SETI project, but Congress can-celed the program after a year.

Still, SETI researchers have managedwithout NASA funding. From 1973 to1997 the Ohio State University RadioObservatory, led by John D. Kraus and


MAIN SETI PROJECTS NOW UNDER WAY

Observatory

Starting observation date

Site

Antenna diameter (meters)

Range of telescope motion(declination, in degrees)

Channels (in millions)

Spectral resolution (hertz)

Operation frequency (gigahertz)

Instantaneous bandwidth (megahertz)

Total bandwidth coverage (megahertz)

Sensitivity (watts per square meter)

Approximate sky coverage (percent)

Types of signals

Funding sources

BETA

Oak Ridge

1995

Harvard, Mass.

26

–30 to +60

250 x 8

0.5

1.4–1.7

40

320

3 x 10–24

70

C, Slow CH

PlanetarySociety,

Bosack-Kruger and Shulsky Foundations

META II

IAR

1990

Buenos Aires,Argentina

30

–90 to –10

8.4

0.05

1.4, 1.6, 3.3

0.4–2

1.2– 6

8 x 10–24

50

C, CH

PlanetarySociety,

CONICET

SERENDIP IV

Arecibo

1997

305

–2 to +38

168

Down to 0.6

1.4

100

180

~10–24

30

C, CH, P

PlanetarySociety, Friends

of SERENDIP, SETI Institute

SOUTHERNSERENDIP

Parkes

1998

NSW,Australia

64

–90 to +26

4.2 x 2

0.6

1.4

2.4

2.4

2 x 10–25

75

C, CH, P

Universityof New

South Wales

ITALIANSERENDIP

Medicina

1998

Bologna,Italy

32

–30 to +90

4.2 x 2

1.2

0.4, 1.4, 1.6

5

5

~10–25

75

C, CH, P

ItalianResearchCouncil

PHOENIXPROJECT

Parkes

1995

NSW,Australia

22 and 64

–90 to +26

Arecibo

1998

305

–2 to +38

NRAO

1996–1998

Green Bank, W. Va.

30 and 43

–35 to +80

28.7 x 2

Down to 1

1–3

20

2,000

~10–25

Not applicable (targeted survey)

C, CH, P

SETIInstitute

Puerto Rico Puerto Rico

IAR ARECIBO GREEN BANK PARKES

CO

UR

TESY

GU

ILLE

RM

O A

.LEM

AR

CH

AN

D

SETH

SHO

STA

CK

SETI

Inst

itut

e

SETH

SHO

STA

CK

SETI

Inst

itut

e


Robert Dixon, conducted the longest full-time, dedicated SETI search, entirely on avolunteer basis. In 1985 the PlanetarySociety, an organization based in Pasa-dena, Calif., with more than 100,000members around the world, built an 8.4-million-channel analyzer known asMETA (Mega-channel Extraterrestrial

Assay). The society installed the device,designed by Paul Horowitz of HarvardUniversity, at the 26-meter antenna ofHarvard’s Oak Ridge Observatory. Fiveyears later the society installed a similarspectral analyzer, META II, in one of thetwo 30-meter antennae of the ArgentineInstitute of Radio Astronomy near

Buenos Aires. They were the first privatelyfunded, dedicated, full-sky SETI surveys.

During the past 40 years, more than90 different professional SETI projectshave been carried out at observatories inAustralia, Argentina, Canada, France,Germany, Italy, Russia, the Netherlandsand the U.S. Together they have accu-mulated more than 320,000 observinghours, mostly in the so-called magic fre-quencies where we believe ETIs wouldbroadcast. Unfortunately, none providedany conclusive evidence of the detectionof an intelligent extraterrestrial signal.

Today private financial support forSETI comes from individual donations tononprofit organizations, such as the SETIInstitute, the Planetary Society andFriends of SERENDIP. Logistical supportcomes from institutions such as Harvard,the University of California at Berkeley,the National Astronomy and IonosphericCenter at Arecibo in Puerto Rico, NRAO,the University of New South Wales inAustralia and the National ResearchCouncil of Argentina (CONICET). Thanksto the generosity of Hewlett-Packard’sOliver, the SETI Institute has a $20-mil-lion endowment that allows it to devel-op more ambitious projects. In particular,Project Phoenix is a mobile programthat uses a 56-million-channel system;Berkeley’s Project SERENDIP IV has 168million channels; and the PlanetarySociety–Harvard’s BETA has 250 million.These systems range in cost from severalhundred thousand to several milliondollars. The 64-meter antenna in Parkes,Australia, and the 43-meter antenna ofthe NRAO have already finished twoobservation campaigns and are nowplanning to extend their work using theobservatories at Arecibo and Jodrell Bankin England.

Over the past few years, Paul Shuchof the SETI League, a nonprofit organi-zation with members in more than 40countries, has been trying to coordinate5,000 small antenna dishes—built, main-tained and operated by private individu-als—in such a way that they will notmiss any likely sky positions. Prototypestations went into operation in 1996,and many hundred enthusiasts world-wide are taking part in this project.

Another initiative, called SETI@home,is trying to use the Internet to organize50,000 to 100,000 volunteers to performmassive parallel computation on desk-top computers. Participants could down-load a screen-saver program that willnot only provide the usual graphics butalso perform sophisticated analyses of


MEDICINA

OAK RIDGE

BETA, META and the three SERENDIPsmake full-sky surveys for ultra-nar-rowband signals, limited only by therange of motion of each radio tele-scope (defined by its declinationrange). Project Phoenix searches forsignals around nearby stars, usingthe Arecibo antenna and the 76-meter-wide antenna at Jodrell Bankin England as a means to confirmArecibo signals. The number of chan-nels, spectral resolution and instanta-neous and total bandwidth are char-acteristics of each spectral analyzer.The sensitivity, which varies with fre-quency, is proportional to eachantenna diameter and to the techni-cal characteristics of the receiversand spectrometer. Carrier signals (C)are continuous, modulated waves(radio and television signals ride oncarrier waves). Pulses (P) are brief,intermittent signals; chirps (CH) arepulses whose frequency changes.

SETH

SHO

STA

CK

SETI

Inst

itut

e

STEL

IO M

ON

TEB

UG

NO

LI

PLA

NET

AR

Y S

OC

IET

Y

Is There Intelligent Life Out There?



SETI data using the host computer. Thedata would be tapped from ProjectSERENDIP IV’s receiver at Arecibo.

SETI Search Methods

The practical requirements of SETImake the notion of “searching for a nee-dle in a haystack” pale in comparison,because an electromagnetic transmissionchannel always has several variablesthat must be set. These include a four-dimensional aspect: the location of theextraterrestrial civilization (three dimen-sions in space) and a temporal dimen-sion that coordinates transmission andreception (you can be looking to thecorrect place but at a moment whennobody is transmitting, or vice versa).Other factors include the frequency, thesignal intensity and the cryptographicvariables, such as polarization, modula-tion, information rate, code and seman-tics (which all must be overcome todecode any message). All these variablesmake up our “cosmic haystack.”

Leaving aside certain complicatingfactors, there are roughly 3 × 1029 places,or “cells,” in the sky to explore. Eachcell’s dimensions are 0.1 hertz wide, mul-tiplied by the number of beams that a300-meter Arecibo-type radio telescope(the world’s largest) would need to con-duct a full-sky survey. The calculationassumes a receiver sensitivity of 10–20 to10–30 watt per square meter—less thanthe energy we would receive from a 100-watt lightbulb shining on Pluto.

Assuming Sagan’s estimate of onemillion technological civilizations, thissearch is comparable to looking for anactual five-centimeter-long sewing nee-dle in a haystack 35 times the size ofEarth. So far only a small fraction of thewhole haystack has been explored, amere 10–16 to 10–15 of the total possiblenumber of cells.

The success of the search dependsnot only on the number of civilizationsin our galaxy but also on their transmis-sion strategies. Historically, researchersassumed that some “supercivilizations”

can make omnidirectional transmissionsstrong enough to be detected by full-skysurveys. Yet even if many civilizationsare communicating with one anotheracross the galaxy, only a vanishinglysmall probability exists that we onEarth, randomly observing differentdirections in the sky, would be able toeavesdrop on narrow-beam ETI signals.Full-sky surveys made by the Harvard,Arecibo, Ohio and Buenos Aires SETIprojects did not find any evidence ofomnidirectional supercivilization trans-missions at a distance of 22 megaparsecs(70 million light-years).

What kind of intentional signalmight we expect? It will most likely benarrowband, approximately one hertz orless in width and ideally a single wave-length (and thus obviously artificiallygenerated), because the senders wouldwant their signal to stand out as artificialagainst similar natural signals andbecause such a signal travels farthest fora given transmitting power. Most SETIprojects can distinguish only the pres-

Supernova

ETI

Earth

R1

R3R2

Supernova

ETI

Earth

R1

R3R2

signaldirection

signaldirection

ReceptionEllipsoid

(where we listenfor signals)

TransmissionHyperboloid

(from Earth)


(from ETIs)

signaldirection

signaldirection

ReceptionEllipsoid

(where we listenfor signals)


(of Earth)


(of ETI)

JOH

N M

ENG

EL P

onzi

& W

eill

STRATEGY TO USE SUPERNOVA AS A BEACON can narrowthe search for an extraterrestial intelligence (ETI), which isassumed to rely on the exploding star as a source of attention.Earth radio telescopes would look for signals among stars thatfall within an ellipsoidal region of space defined by Earth’s dis-tance to the supernova (R1) and to the possible ETI planet (R2),and by the distance from the supernova to the ETI (R3). (Therelation is R2 + R3 – R1 = a constant for each specific time afterthe supernova explosion. The ellipsoid enlarges over time, and

R1 and R3 can be several hundred thousand light-years.)Moreover, we could transmit greetings to stars that fall withina particular region, which is a “hyperboloid of transmission”—ashape defined by time and the supernova’s position in the sky.It basically corresponds to a zone that provides the best viewof the supernova and Earth, so that an ETI studying the super-nova would also see our signal coming to it from the samedirection. Likewise, Earth would fall in the transmissionhyperboloid of the ETI, if the ETI chose an identical strategy.


ence of such a signal among the broadband of cosmic noises but cannot ascer-tain the content of a message that mightbe coded in some unknown form.

Current SETI detection devices simul-taneously analyze several million or bil-lion spectral channels, whereas computerscheck for any strong narrowband signalsamong the cacophony of cosmic noise.Other instrumentation eliminates allhuman-made terrestrial and space radiointerference. After years of observationand the analyses of hundreds of billionsof signals, fewer than 100 signals havelooked like potential extraterrestrial sig-nals. Unfortunately, none of them couldbe detected in follow-up observations.

To cull the false alarms, James M.Cordes, Joseph W. Lazio and Sagan ofCornell University derived tests to analyzethe unexplained signals detected by theMETA and META II projects between 1986and 1995. Their analyses found signalsthat originated near the galactic planethat could not be ruled out as alien.

To check the origin of these andother unexplained signals, SETIresearchers instituted new observationstrategies. Horowitz’s Project BETA nowuses a billion-channel analyzer and threedifferent antenna beams in order toexclude any possible terrestrial interfer-ence. Project Phoenix uses what its mem-bers call a FUDD (Follow-Up DetectionDevice): a second antenna hundreds ofkilometers away that simultaneouslyanalyzes signals and can screen out falseones (as shown in the recent movieContact). Using these improvements,SETI researchers identified the METAand META II candidate signals and otherunexplained blips as different kinds ofterrestrial interference.

Is it possible that we have alreadyreceived an intelligent contact signaland that we missed it? I believe weprobably have. But our detectors werenot sensitive enough to distinguish itfrom the cosmic noise. Or it could bethat our antenna was not pointed to thecorrect place at the right moment, or thatwe are searching in the wrong frequencyor using the wrong observation strategy.Perhaps the signal faded because itpassed through charged plasma cloudsthat pervade interstellar space.

I believe that the first evidence ofan intelligent signal will probably comeaccidentally, when a traditional astrono-mer, unable to explain some anomalousobservation, will realize that his or herdata can be explained only as a conse-quence of some technological extrater-

restrial activity. He or she will be able todraw this conclusion by using what wehave learned through SETI.

This prediction reflects the distinctlyunglamorous character of the day-to-daywork of SETI. There are very few full-timeSETI researchers: much time is spentdesigning and testing computer pro-grams, and computers automatically domost of the work related to observation.The typical SETI researcher is alsoinvolved with learning and designingnew hardware and software, developingnew observational strategies and, mostimportant, interpreting observationaldata to discern in them any possibleintelligent signal patterns among thewaterfall of cosmic noise.

Supernova Beacons

The greatest difficulty in making thefirst contact stems from the requirementthat the incoming signal must arrive atthe same time that the target civilizationhas its receiver pointed toward theunknown transmitter. The need for thissynchronization is one of the weakestparts of our search strategy. But anextraterrestrial civilization using thiskind of “active” search method (broad-casting signals to be discovered, in con-trast to “passive” listening only) mightovercome this problem by using a natur-al astronomical phenomenon—probablya supernova—as a “beacon” that wouldattract the attention of other civiliza-tions. The sending civilization wouldtransmit its own message in the diametri-cally opposite, or antipodal, direction ofthe supernova as seen from the trans-mitting planet.

In 1976 and 1977 Tong B. Tang ofthe Cavendish Laboratory of the Univer-sity of Cambridge and P. V. Makovetskiiof the Leningrad Institute of AeronauticalInstrument Manufacture independentlyargued that we might improve the prob-ability of contact if we assumed that ETIstransmitting signals might use supernovabeacons. They calculated that we shouldobserve only those stars within an ellip-soidal volume, with Earth at one of thefoci of the ellipse and a supernova at theother [see illustration on opposite page].

In fact, my colleagues and I havesuggested that SETI researchers useexactly this strategy to attempt to contactETI civilizations, using as the beacon thesupernova detected in the Large Magel-lanic Cloud on February 23, 1987. Giventhat there are on average only one to foursupernova explosions in our galaxy every

100 years and that the supernova, dubbedSN1987A, was the brightest one in 383years, we can assume that most of thepossible galactic civilizations would bepaying close attention to it.

We would transmit our message inthe direction antipodal to the supernova,in a field defined by a hyperboloid(roughly speaking, it corresponds to anarea around Earth that provides the bestview of the supernova within the ellip-soid). There are only 33 nearby objectsinside this hyperboloid, which wouldfocus the effort even further: of these 33objects, 16 are solar-type stars that couldhave planets with other civilizations.

Are We Lunch?

The idea of using this kind of activesearch strategy concerns people in manyquarters of the SETI community aboutwhether to send such signals at all. Theheat surrounding this issue can be tracedto the first—and to date the only—attempt to send such a signal. OnNovember 16, 1974, the Arecibo Observa-tory transmitted an interstellar messagedescribing some characteristics of life onEarth toward the Great Cluster inHercules, M13, a group of about 300,000stars 25,000 light-years distant.

The action provoked some majorprotests. Former U.S. diplomat MichaelA. G. Michaud considered the attempt apolitical act. He suggested a public dis-cussion of the potential benefits and risksof ETI contact and urged that a decisionbe made openly, “with the involvementof public authorities.” Martin Ryle, aNobel laureate and Astronomer Royal ofEngland, wrote to leading astronomerssaying that he felt it was very hazardousto reveal our existence and location tothe galaxy. For all we know, he said,“any creatures out there are malevolentor hungry,” and once they knew of us“they might come to attack or eat us.”He strongly recommended that no suchmessages be sent again.

Frank Drake replied to Ryle in a let-ter stating: “It’s too late to worry aboutgiving ourselves away. The deed is done,and repeated daily with every televisiontransmission, every military radar signal,every spacecraft command....” Accordingto Drake, Ryle seemed satisfied with therejoinder.

Ben R. Finney, an anthropologist atthe University of Hawaii at Manoa, hasdescribed human responses to contactwith ETIs as “paranoid” (assuming thatextraterrestrials are malevolent) or “pro-

Exploring Intelligence 103Is There Intelligent Life Out There?


noid” (assuming that interaction withETIs would be extremely beneficial tohumanity). Ever since H. G. Wells intro-duced in 1898 the idea of the invasionof Earth by murderous aliens in The Warof the Worlds, the paranoid idea hasdominated not only science fiction butalso the thinking of some scientists. Forexample, in the early 1960s theBrookings Institution in Washington,D.C., prepared a report for NASA thatconcluded that “the discovery of life onother worlds could cause the Earth’s civ-ilization to collapse.”

Medical anthropologist MelvinKonner of Emory University has said,“Evolution predicts the existence of self-ishness, arrogance and violence on otherplanets even more surely than it predictsintelligence. If they could get to Earth,extraterrestrials would do to us what wehave done to lesser animals for centuries.”

“Any creature we contact will beevery bit as nasty as we are,” echoesMichael Archer, a biologist at theUniversity of New South Wales. Hethinks the gold-coated copper phono-graph records affixed to each Voyagerspacecraft—which contain, among otherindications of intelligent life, 118 photo-graphs of our planet, ourselves and ourcivilizations—are giant dinner invitationsto the cosmos.

The pronoid school of thought isreflected in the writings of William J.Newman and Sagan, who have suggest-ed that there may be universal impedi-ments against cosmic imperialism. They

have gone so far as to suggest that aCodex Galactica, produced by moremature civilizations, might exist to edu-cate younger societies on cosmic eti-quette. They have further argued thatadvanced civilizations with long histo-ries must have learned how to be benignand how to treat an adolescent societylike ours “delicately.”

“As our own species is in the processof proving, one cannot have superior sci-ence and inferior morals. The combina-tion is unstable and self-destroying,” thescience-fiction writer Arthur C. Clarke hassaid. This position was shared by the lateIsaac Asimov and by other SETI propo-nents, including myself. If other civiliza-tions agree, we might expect advancedsocieties to make only limited informa-tion available to emerging societies. Thisview is opposite to the contact scenariousually advocated by SETI pioneers, suchas Sagan, who expect vast amounts ofinformation or some kind of Encyclo-pedia Galactica.

Going beyond Adolescence

One final argument may be made infavor of active search strategies, whichmay imply strongly that they are essen-tial if we are to have any hope at all ofcontacting extraterrestrial intelligence.

Communication is a two-wayprocess. If all beings in the universe aretrying to detect signals from other beingswithout sending out any of their own,then no one will receive a signal. This

possibility conjures up the disturbingimage of a galaxy filled with technologi-cal civilizations eager to make contactwith one another but with all of themonly listening and thus forever con-signed to isolation.

Of course, such a “listeners-only”universe is unlikely, even if no inten-tional signals are being sent. As Drakenoted, humanity has already madeknown its existence and location to alarge part of our galaxy. Perhaps thisannouncement is typical for emergingadolescent civilizations, which may beno more cautious and quiet than adoles-cent humans.

Indeed, our position relative to theoutcome of SETI is very much like thatof an adolescent setting out on life’sjourney: the possibilities are infinite, thefuture is wide open, and we have grandplans, but much of the shape of thatfuture hangs not only on what we dobut also on luck—whether or not certaincritical “ifs” actually come to pass. SETImay yield the greatest discovery in thehistory of humankind if life is ubiquitousacross the cosmos; if life inevitably givesrise to intelligence and technology; iftechnological civilizations routinely sur-vive long enough to broadcast andreceive interstellar signals; if such civiliza-tions want to be found; if we are usingthe correct search strategies and are tunedto the right frequencies; and if we recog-nize the signal when it arrives. Untilthen, we must do what most adolescentsdo very poorly: we must wait.


EL LLAMADO DE LAS ESTRELLAS, G. A.Lemarchand. Ediciones Lugar,Buenos Aires, 1992.

WE ARE NOT ALONE. WalterSullivan. Plume, 1994.

THE BIOLOGICAL UNIVERSE. S. J. Dick.Cambridge University Press, 1996.

ASTRONOMICAL AND BIOCHEMICAL

ORIGINS AND THE SEARCH FOR LIFE IN

THE UNIVERSE. Edited by C. B.Cosmovici, S. Bowyer and D.Werthimer in Proceedings of the5th International Conference onBioastronomy, IAU ColloquiumNo. 161, Editrice Compositori,Bologna, Italy, 1997.

CARL SAGAN’S UNIVERSE. Edited by Y.Terzian and E. Bilson. CambridgeUniversity Press, 1997.

ORIGINS OF PLANETS AND LIFE. Editedby H. J. Melosh. Annual Reviews,Inc., Palo Alto, Calif., 1997.

GUILLERMO A. LEMARCHAND was five years old whenNeil Armstrong first set foot on the moon. “So I have alwaysbeen interested in space,” he says. “And I’ve always wonderedwhether life could have started on another planet in anotherplace in the universe.” But Lemarchand did more than wonder.As an undergraduate student in physics at the University ofBuenos Aires, Lemarchand organized an international meetingon intelligent life in the universe. Some 500 students attendedthe meeting and listened to presentations by eminent biologists,astronomers and radio astronomers. A few months laterLemarchand made the first SETI observations at the ArgentineInstitute of Radio Astronomy—a project that he helped to estab-lish and now coordinates.

When he’s not scanning the sky for signs of intelligent life,Lemarchand works with mathematical models trying to under-stand long-term dynamics of social and economic systems anddevotes his energies to promoting scientists’ sense of socialresponsibility. He established the Argentine branch of Pugwash and organized an international sympo-sium on scientists, peace and disarmament, where he proposed a Hippocratic oath for scientists, which isnow used in graduation ceremonies at the University of Buenos Aires. Lemarchand’s activism broughthim to the attention of Carl Sagan of Cornell University, where he subsequently spent a year as a visitingfellow before returning to the University of Buenos Aires.

SA

Further Reading

CO

UR

TESY

OF

GU

ILLE

RMO

A.L

EMA

RCH

AN

D

About the Author


Documents

Exploring Intelligence - The Eye · 2019-05-30 · 2 Scientific American Presents Intelligence EXPLORING A Search in the Human, Animal, Machine and Extraterrestrial Domains 6 12 Twin