Indian J Physiol Pharmacol 2000; 44 (4): 379-391

REVIEW ARTICLE

THE ORIGINS AND NATURE OF THE COGNITIVE PARADIGM:AN OVERVIEW

SUBRATA DASGUPTA

Institute of Cognitive Science,Center for Advanced Computer Studies, andDepartment of History,University of Louisiana at Lafayette,Lafayette, Louisiana 70504 - 3772 (U.S.A.)

(Received OD July 22, 2000)

Abstract: Although cognitive science, the 'science of mind' is generallyregarded as of relatively recent vintage, its origins can actually be tracedback to ideas from the 1930s. The purpose of this paper is to offer a viewof the essential nature of what has come to be called the cognitiveparadigm - the framework around which cognitive science has come to beconstructed - and about its origins and development. Of particular note isthe interdisciplinarity of the field, with its strong links to psycholog.y,linguistics, neurobiology, computer science and philosophy. As will bee:w:.plained below, this interdisciplinary character is rooted in the historicaldevelopment of the science itself, and can only be understood in such acontext. Thus, the present article is strongly historical in spirit andcontent.

Key words: cognitive science cognitionartificial intelligenceneuroscience linguistics

psychologysymbolic representation

information processing

INTRODUCTION

In 1977, a new journal called CognitiveScience was inaugurated; two years later,the Cognitive Science Society was institutedat a conference in La Jolla, California (1).

If the founding of a first journal and thecreation of a professional society are anyindicators that a field of study has come ofage, then these dates project a sense of howrecent has been the eme'rgence of cognitivescience as a distinct scientific arena.However, this is not to say that cognitive

science emerged dramatically, de novo, inthe mid-1970s; on the contrary. Thehistorical origins of cognitive science is acomplex, tangled web involving manydisciplines, some of which, like philosophyof mind, go back to antiquity while others,such as computer science, are almost asyoung as cognitive science itself. And whilea few histories of the subject have beenpublished (1, 2), a truly comprehensive,analytical study of the ways in whichcognitive science emerged from its variouscontributing remains to be written.

380 Dasgupta

The purpose of this article is vastly moremodest. Taking a cue from Kuhn that thehallmark of a genuine scientific field is thepresence of a paradigm that is, a networkof theories, laws, principles, models andexemplars (3), the aim of this paper is tooffer a view of the essential nature of thecognitive paradigm. This article is, thus,strongly historical is spirit and content.

What is the cognitive paradigm?

We all have a rough idea of whatconstitutes a revolution. As in the realm ofpolitics and societies so also in the realm ofideas, certain transforma:tions occur that areradical enough from what preceded them tobe described as 'revolutionary'. Moreprecisely, a revolution in the world of ideasinvolves at least two characteristics. One isthat it involves a radically new way oflooking at things: new kinds of questionsare suggested and posed about the fieldof interest, new tools are employed, anew vocabulary or even language mayari"e, a new world view emerges. Thiscomplex of entities is precisely what Kuhnmeant by the word 'paradigm'. Thus, onecharacteristic of a revolution in ideas is theemergence of a new paradigm.

The second characteristic is the presenceof a significant community of people whoare committed to the new paradiglTb It isnot so much the size of the community thatis important but the influence its memberscommand.

It is within this context that we can talkmeaningfully about the cognitive revolutionwhich, in essence, refers to the emergence,development and deployment of a particular

Indian J Physiol Pharroac:;ol 2000; 44(4)

paradigm-the cognitive paradigm-as a wayof understanding the nature and workingof mind.

What is the essence of this paradigm?Well, on the one hand, we have behavior:observable characteristics of humans andanimals, manifested as speech, interactionswith the external world, generation of ideas,and so on. On the other hand, we have thebrain: a body of physical matter, thevarious components and activities of which,by way of its interaction with the outsideworld, give rise to behavior.

The problem is that there is a hugeconceptual gap between overt behavior andbrain matter, which makes it rather difficultto try to understand behavior directly interms of neurophysiological events. Whatscientists very often do in order to bridge aconceptual gap between two types of eventsthat are known to be causally related andyet appear vastly removed from each otheris to propose one or more intermediate levelsof descriptions between the two extremes.Effectively, what is created is a hierarchyof description levels. The idea is that weare better off trying to bridge the narrowerconceptual gaps that are thus createdbetween the adjacent levels in such ahierarchy.

The cognitive paradigm (or the idea ofcognitivism) is based on this notion ofhierarchy. The essence of the cognitiveparadigm is the development of anintermediate layer (or possibly severallayers) of explanation, understa,~ding anddes;cription of ('mel~tal') phenomena lyingbetween the exteremities of overt behaviorand the physico-chemical activities of brainmatter.

Indian J Physiol Pharmacol 2000; 44(4)

The problem of understanding the mindcan thus be broken down into severalsmaller, more manageable subproblems: toexplain overt behavior-our utterances, thegeneration of ideas and thoughts, the waywe make sense of the external world, thefact that we learn, the manner in whichchildren grasp their first concepts ofnumbers-in terms of intermediate levelprocesses; and then explain the primitiveoperations from which the intermediatelevel processes are built i.n terms ofneurobiological processes. Thus, if overtbehavior pertains to the world at large-thesocial world-and if brain matter belongs tothe natural world of physics and chemistry,the 'cognitive level' mediates between thesocial and the physico-chemical.

Origins of the cognitive paradigm

The origins of the cognitive paradigmcan be traced back at the very least to the1930s. In America, the dominant paradigmin psychology between World Wars I and IIwas, of course, behaviorism, which more orless ignored mental phenomena altogether;rather, behaviorism owed allegiance only toovert behavior. Yet, even in America, therewere a handful of scientists who, whileadmitting to being behaviorists, were notaverse to talking about 'cognition' and'mental processes'. One such psychologistwas E.C. Tolman in whose work we findthe idea of mind as being some entityphysically located in then brain but which,in terms of its properties, is qistinct fromthe brain. In particular, Tolman regardedthe mind as a purposive entity; this conceptof purpose is, of course, entirely absent inthe realm of physics, chemistry andneurobiology (4).

The Origins and Nature of Cognitive Paradigm 381

Tolman's was a relatively rarecognitivistic voice in an America dominatedby behaviorism. Outside the United States,behaviorism was never quite so universallyaccepted and, indeed, the basic structure ofthe cognitive paradigm was far more inevidence. From Cambridge, England, F.e.Bartlett published his book Remembering(1932) in which he postulated the presence,in the head, of an "organization ofpast reactions or of past experiences" - inwords, a schema representing one's priorexperiences-which then participated ininterpreting and making sense of newexperiences 10 such mental acts asremembering (5). A decade later, KennethCraik suggested that thinking entails theconstructions, within the nervous system,of a model of the external situation one isthinking about; and that "thought modelsor parallels reality" (6, p. 57). Craik wenton to say that:

the essential feature (of thought) is ...symbolism, and .... this symbolism islargely of the same kind as thatwhich is familiar to us in mechanicaldevices which aid thought andcalculation. (6, p. 57).

Craik was writing at a time whencalculating devices were (a) mechanical orelectro-mechanical, and (b) analoguedevices. But he bad somehow grasped thenotion that thought entails the use ofsymbols which represent things andsituations in the real world. He alsoappealed to the analogy of the Gnlculatingmachine which itself, in some sense,models the phenomenon about which itcalculates.

382 Dallguptll

Craik's insight needs to be appreciated.It is not merely, he wrote, that "thoughtemploys symbols", but also "the whole ofthought (is) ... a process of symbolism" (6,p. 58). As for the nature of this 'process ofsymbolism', Craik went on to explain that:

If the organism carries a "small scalemodel" of external reality and of its ownpossible actions within its head, it is ableto tryout various alternatives, concludewhich is the best of them, react to futuresituations before they arise, utilise theknowledge of past events in dealing withthe present and future, and in every wayto react in a much fuller, safer and morecompetent manner to the emergencieswhich face it. (6, p. 61).

In other words, Craik conceived thinkingas involving symbolic representations(or models) in the head of aspects ofthe external world, and symbolicrepresentations of the human agent'sactions. Thought would then entail themanipulation of the symbolic models by therepresented actions - that is, by mentallysimulating actions and their effects onexternal reality.

The advent and impact of the digital computer

Craik's premature death in 1945precluded further development of hiS" ideasand, certainly, while well known inEngland, his work seemed to have hadvirtually no effect in America - which iswhere a number of quite remarkableinterdisciplinary sciences developed inthe years immediately following World WarII - including information theory, switchingcircuit theory based on Boolean logic, game

Indian J PhYliiol Pharmacol 2000; 44(4)

theory, general systems theory, andoperations research. In the tangled historyof the origins of cognitive science, thesewould all leave their imprint in one way oranother. But the most influentialdevelopment was that of the electronicdigital computer between 1946 and 1949.We find evidence of the immediate appealof, what was apparently, a purelytechnological development for thinkers onmind, brain and thought in theinterdisciplinary conference, sponsored bythe Hixon Foundatioo on "CerebralMechanisms in Behavior", held in Pasadena,California in 1948 (7). Here, John vonNeumann, mathematician and co-inventorof game theory, as well as co-inventor ofthe concept of the stored program digitalcomputer, compared certain features of thecomputer to the brain (8). As an example,he pointed out that functionally, neuronstransmit impulses which possess an "all-or­nothing" character, similar to the digitalelements in a computer. Functionally, the,neurons are analogous to digital electronicdevices, even though the matter of whichthey are made bear no resemblance to eachother. In fact, in analogizing between brainand computer, von Neumann was drawingupon earlier work by Pitts and McCullochwho had shown that the functional behaviorof a network of neurons could be modeledin terms of the principles of mathematicallogic, the resulting model being termed a'formal neural network' (9).

But the Hixon Symposium participantshad more on their minds than computersand digital circuit elements; they had mindon their minds. Behaviorism, the dogma thathad eschewed any discussion of mind,mentalism or cognition as being within the

I

~

Indian J PhYliol Pblrr:Dleol 2000; 44(4)

pUTvew of scientific psychology, came underserious attack by such scientists as thepsychologist Karl Lashley and themathematician-neurophysiologist WarrenMcCulloch. Thus, Lashley would argue thatthe nervous system, far from being a"quiescent OT static system- is organiz.ed and'actively excited', and that 'behavior is theresult of interact.ion of this background ofexcitat.ion with input from any designatedstimulus' (l0, p. 112). What one needed toknow, then, was the "general character ofthis background of information-; for onlythen can the effect of some input stimulusbe understandable. And McCulloch, forwhom 'mind' meant "ideas and purposes·,argued that the 'mind is in the head'precisely because:

t.here, and only there, are hosts ofpossible connections (between neurons)to be formed as time and circumstancedemand. Each new connection serves toset the stage for others yet. to come andbetter fitted to adapt us to the world.(I t, p. 56).

Two years after the Hixon Symposium,the logician nnd computational theorist AlanTuring suggested an experimental situationin which one might not be able todistinguish between the manifestation ofhuman intelligence and the behavior of acomputer (12). Turing's idea was thefollowing. Imagine an experimental setupin which a computer is in one room and aperson in another; in a third room there isa second person who serves as aninterrogator. The computer is programmedto receive and answer qu.estions from theinterrog3tor; and the latter, usingappropriate devices, puts questions to both

The OriCinl and Nature of Cornilive Paradigm 383

the computer and the other person, thoughhe does not know where either is located.Through appropriate devices, theinterrogator also receives answers to hisquestions from human and machine. Underthis set up, if the interrogator is unable todistinguish between the computer's answersfrom those of the person, then the computercan be said to think and be intelligent inthe sense that human beings are normallysaid to be intelligent.

Turing's essay remains one of thegenuine classics in the early history ofcognitive science; his testing criterion(known as the "Turing teat') set the ultimatestandard of what would count as 'artificialintelligence'.

AnDUI mill'abilil: 1956

Turing's essay, though provocative and,ultimately, influential, was fundamentallyspeculative. If we are to identify an annusmirabilis for the cognitive revolution, thatyear must be 1956, for in that year, a seriesof relatively independent developmentsbecame public, the collective impact ofwhich would be felt very quickly.

Of these developments, one was entirelylinked to the digital computer; and itsspecific nature was to utilize the computera8> a means for talking about mentalprocesses, and as a physical instrument withwhich experiments concerning mentalphenomena could be performed.

The work being alluded to was due toHerbert Simon and Allen Newell, one asocial scientist, the other a mathematician.At a symposium held in September 1956 at

384 Dasgupta

the Massachusetts Institute of Technology,Newell and Simon described a 'machine'­actually a computer program - called theLogic Theorist that was capable ofdiscovering proofs for theorems In

mathematical logic (13). Here, then, was amachine exhibiting behavior akin to thehighest form of human intelligence, able toprove the first few theorems stated in Russelland Whitehead's Principia Mathematica.Leaving aside the question of whether the'architecture of cognition' functioned in wayssimilar to the program, the Logic Theoristsuggested the possibility of artificialintelligence; and the possibility that,perhaps, not all minds have to be biologicalin origin. Logic Theorist also demonstratedrather formally and precisely what suchconcepts as symbol processing, symbolicrepresentation, and information processingmight mean in an empirical or operationalsense.·It is important to note that amongstthe sources of ideas that would directly orindirectly feed into the design of the LogicTheorist were Simon's earlier work on thepsychology of human decision making inorganizations (14), and Von Neumann andMorgensteru's seminal work on game theory(I5) - both, seemingly very unlikely sourceindeed (Dasgupta, unpublished)!

This paper was not the only contributiouthis particula.r conference made to thedevelopment of the cognitive paradigm. Ayoung linguist named Noam Chomskypresented a paper which would become theroot of yet another major movement in thecognitive field, one which is sometimes calledthe 'Chomskeyan revolution' in linguistics(16). Chomsky would argue in his variouspublications that the creativity we observein linguistic behavior, our ability to generate

Jndian J Physiol Pharmacol 2000; 44{4}

sentences we may not have ever utteredbefore, and the ability to understandsentences we may not have ever heardbefore, can only be explained by theexistence, in the mind, of a system of rulesand representations that encode universalprinciples of language that are common toall specific languages. Chomsky connectedlinguistics, the systematic study oflanguage structure and use, withcognitivism in the following terms:

The person who has acquired knowledgeof a language has internalized asystem of rules that relate sound andmeaning in a particular way. Thelinguist constructing a grammar of alanguage is in effect proposing ahypothesis concerning this internalizedsystem. The linguist's hypothesis, ifpresented with sufficient explicitnessand preCISion, will have certainempirical consequences with regard tothe form of utterances and theirinterpretations by the native speaker.07, p. 26).

For Chomsky, the 'internalized systemof rules' is knowledge of language.

There were also other seminalpublications in that same year that wouldcontribute significantly to the emergenceof the cognitive paradigm. The psychologistGeorge Miller suggested that there weresome basic limitations on humaninformation processing capacity (I8). IfNewell and Simon were influenced directlyby the workings of the computer, Millerwas influenced by the new mathematicaltheory of information which had beeninvented in the 1940s. Also in 1956, Jerome

Indian J Phyaiol Pharmacol 2000: 44(4)

Bruner, Jacqueline Goodnow and GeorgeAustin published a work in which thetask of categorization or concept formationwas explained as a constructive processinvolving the deployment of strategies (9).

The important lesson that emergesfrom these landmark studies, studiesthat led to the shaping of the cognitiveparadigm, is that the latter, unlike mostscientific paradigms, is fundamentallyinter~isciplinary. Indeed, one widelyaccepted view is that cognitive science liesat the intersection of five 'mainstream'disciplines, namely, psychology, philosophy,linguistics, artificial intelligence andneuroscience. Some would disagree thatthese constitute all the participatingdisciplines at the intersection. For example,Gardner includes anthropology as anothercontributing field (2), while this writer hascome to believe, on the basis of historicalcase studies of scientific creativity, that80me aspects of cognition cannot beexplained without appeal to the socialsciences.

However, the one shared belief thatis more or less unchallenged by thosephilosophers, psychologists, linguists,artificial intelligence researchers and(perhaps to a lesser extent) neuroscientistswho call themselves cognitive scientists, isthat cognition entails manipulation andprocessing of representations (also calledinformation).

Indeed, a large part of the researchprogram in cognitive science, whetherconducted by psychologists, linguists,computer scientists or philosophers, isconcerned with the question of what can be

The Origins and Nature of Cognitive Paradil"m 385

the nature of mental representation fordifferent kinds of cognitive tasks. Forinstance, researchers in the field ofcognitive development studying how andwhen infants first develop the concept ofnumbers debate on the form of the infant'srepresentation of numbers. One proposalsuggests that each distinct number isrepresented in the infant's mind by adistinct symbol; an alternative hypothesisis that each distinct object perceived by theinfant is represented by a distinct symbol(20). Psycholinguists have used certainformal structures called trees (that lookroughly like organization charts) to depicthow humans may represent, construct,parse and understand sentences (17, p. 29).Cognitive psychologists and artificialintelligence researchers interested in howhumans (or computers) learn concel-ts-forexample, the concept 'computer'-havesuggested representations called semanticnetworks that show relationships betweenthe concept of interest and other concepts,facts, objects, and properties, so that theconcept of interest (such as 'computer') ismeaningful by virtue of these variousrelationships. Psychologists inquiring intohow humans solve problems may suggestthat the subject possesses knowledge of thetask domain (e.g., in the domain of chess orother board games) in the form of rules thatspecify 'IF such and such condition is thecase THEN take such and such action' (21).

How does one test the validity orplausibility of these abstract theories aboutformal representations? Cognitive scientistsdevise various laboratory experiments onhuman (or sometimes primate) subjects;or they may construct computer modelsbased on the representation and simulate

386 Dasgupta

them on the computer. In either case, thepurpose is to corroborate or refute certainproperties (e.g., behavior or performance)of the subject or of the simulated modelas are predicted by the hypothesizedrepresentation. Sometimes, computersimulation and laboratory experimentsmay be both performed to test thevalidity of the representation. As a specificexample, Langley and his collaboratorspostulated a number of general rules ofthe "IF ... THEN ..." type to represent partof the knowledge for a computer program(called 'BACON) which was able to use theserules to 'discover' Kepler's third law ofplanetary motion (22). Subsequently, Qinand Simon presented the same type of datathat hod been fed to BACON to humansubjects who were given the task to'discover' possible laws connecting the data.The subjects were also asked to verbalizetheir thoughts and strategies as theywent about solving the problem; from ananalysis of such verbal protocols, Qin andSimon argued that the human subjectswere deploying the same kinds of rules ashad been implemented in the computermodels (23).

It must be noted that while thenature of representations occupy theattention of much of cognitive science, thereare other aspects of cognition that cannotnecessarily be investigated at the leu"el ofrepresentation. For example, in studying thenature of creative thought, the questions ofinterest might be: What kinds of goalsinitiated a particular act of creation?How did these goals originate? What kindsof knowledge did the creative being drawupnn in the cours~ of creative work? Whatkinds of inferences or mental actions can

Indiao J Physiol Pharmacol 2000; 44(4)

we infer for a given act of creation? Here,the researcher assumes that such entitiesas goals and knowledge are represented inthe person's 'working memory' and 'longterm memory' respectively. But whatthese representations are, are not somuch of interest as what specific entitiesare being represented (i.e., what goals?what knowledge? what inferences?) Thesekinds of questions are more appropriatelyaddressed at a lev~l of cognitive abstractiontermed by Newell the 'knowledge level' (24);one might say that the knowledge level ofabstraction is at the very edge of thecognitive paradigm, where it intersects withthe paradigms of the social sciences.Studies of the cognitive process of creativity,asking questions of the above sort, aretypically of the sort that reside at this~edge', and have been carried out at theknowledge level (25).

The computation_cognition connection - revisited

It is not just representation that is takento be the core of cognition but the processingand transformation of representations also:and here computation and the computermetaphor became the principle conceptualtools by which most cognitive scientistsenvisioned how representations could bemanipulated and processed.

The complex relationship betweentechnology and science is one of theenduring and fascinating themes in thehistory of human culture. The appeal toartifacts and their workings as a model ormetaphor for the workings of nature is oneaspect of this relationship; yet another,more obvious, one is the m.eof technology as scientific instruments.

Indiln J PhYliol Phln'Gacol 2000; ....(..)

The one helps to clarify our thoughts aboutthe natural world, the other extends ourcapabilities to perceive the natural world;but both serve a common purpose: asamplifiers of human ability to comprehendnature.

It is In the former sense in which thecompute'r has most interestingly served thecllouse of cognition. First, it provided asignificant insight into how the symbolicprocessing and mental simulation that Craikhad suggested as the basis for thoughtcould be more pr~ciuly e:nuisione:d-viz.,in the form of computer programs executedon a physical computer. Well before themid-1950s, people like Turing had realizedthat the computer was not a numberprocessing device but a general purposesymbol processing machine. Here, then,was a model of how symbols might berepresented in the head and how theymight be manipulated, processed andtransformed into other symbols. A newuocabulary was mnde available, onethat could be used to talk about thefunctional character of cognition, avocabulary that included such terms as'program', 'algorithm', 'storage register','communication channel', 'processor', 'input­output unit', and so on,

Second, the computer and itsoperations suggested how somethingpurely physical like the brain, obeyingphysico-chemical laws and harboringnothing remotely like purpose could developpurposive properties of the kind Tolmanhad suggested (4). The following analogywas identified:

Computer: HARDWARE->SOPrWARE->PUNCTION

Cognition: BRAIN->MINO-:>BEHAVlOR

ln the case of a computer, there arephysical circuit elements which intrinsicallyhave no purposive attributes; they obey theJaws of physics. But by virtue of organizingthese circuit elements, and encodingand storing signals in a memory like devicethat could then be interpreted by thecircuits as instructions or data, theresulting machine could be made tomanifest purposeful b~havior. As a wayof re-examining the classical mind-brainproblem, it is easy to imagine how appealingand compelling this analogy must havebeen.

Third, if the computer ana itsorganization yield some sense of howbrain matter might be organized to giverise to mental properties and processesleading to behavior, then perhaps thelogical organization (or, technically, the'architecture') of the computer mightsuggest possibilities for the cognitiveorganization of the brain. Cognitivescientists, thus, began to systematicallyinvestigate the architecture of cognition.For example, just as a computer'sarchitecture consists of memories ofvarious capacities and access times, specialand general purpose processing units,control units, input/output transducers,instructions, and so on, so also, it wasargued, a cognitive architecture mightconsist of analogous functional componentssuch as 'working' and 'long term' memory,rules or operations, input (perceptual) andoutput (motor) systems, etc. (26).

388 Dasgupta

Fourth, the very same characteristicsof the computer that had provided ametaphorical role for understandingcognition also afforded an experimentalapparatus, a kind of telescope for the innereye, with which one could "observe"mental phenomena. By programmingthe computer, one could attempt tosimulate specific, higher level cognitivetasks-such as language understanding,game playing, scientific creativity, andso on. If the simulation producedbehavior that was consistent with whatwas predicted, then the programitself became a theory of how humans mightthink and carry out relevant cognitiveprocesses.

The case or David Marr's 'Vision'

The power of computation as a metaphorfor understanding cognition has been,historically, compelling. For many scientistsnot inherently computationally mindedbut interested in one or another kindof cognitive phenomenon, the emergenceof the computational point of view trulyconstituted a genuine paradigm ·shift. Avivid example of this influence of thecomputer-as-metaphor is the work of DavidMarr. Like Craik a generation before him,Marr died young. And just as The Natureof Explanation was Craik's legacy tocognitive science, so was Marr's Vision(1982), published posthumously (27). For OUT

present purpose, the most significant aspectof this book lay not so much in thespecific theories he developed to explaincertain aspects of the visual process, hut inthe extent to which computationalideas permeated his explanation of visual

Indian J Phyaiol Pharmacol 2000; 44(4)

processing. Basically, he posited that visionentails symbolic representation of theexternal world in the human nervous systemin some fashion and the processing of thisrepresentation in some manner that allowsone to make visual sense of what is lookedat. Marr, further, appealed to threedifferent levels of abstraction to explainvision. At the highest, most abstract, level,one explains the visual system in terms ofthe goals of the system, and the logicor strategy by which system performs itstask. In fact, this corresponds to the'knowledge level' of explanation mentionedearlier (although Marr termed it the'computational' level). At the next lowerlevel, one attempts to explain how thehigher knowledge level theory mightcome ahout in terms of the representationof symbols and the algorithms usedto effect the transformations. Marr termedthis the 'algorithmic' level. Finally, atthe lowest hardware level, one tries toexplain how the representations andalgorithms are physically wired into thenervous system.

One c<\nnot articulate a more explicitlycomputational view of cognition than this!It is important to note that to a computerscientist, there is nothing startlingabout Marr's three levels of abstraction,since much. has been written on howcomputers might he effectively designed anddescribed in terms of multiple levels ofabstraction (28). But to a psychologist likeMarr, and to other psychologists exposed tohis work, these computational modes ofexplanation were a significant contributionof Vision.

Indian J PhysiaI Pharmacal 2000; 44(4)

Cognition using neural networks

Not all cognitive scientists takecomputationnlism as the sine qua non ofcognitive science. For instance, Bruner, oneof the founding fathers of cognitive sciencein the 1950s, has in recent years reactedsharply to what he feels is an extremecomputational attitude in contemporarycognitive science (29). And even withinthe computational camp, there continues torage a fierce debate about the kind ofcomputation or information processing toadopt as the appropriate model. There arethose who argue that the usual nlodel ofthe digital computer is far too removed fromthe neuronal structure of the brain; and thatthe extent of parallel procesing that goeson in such cogntitve tasks as patternrecognition, vision, categorization, etc.,cannot be adequately characterized by the'standard' symbol processing computationalmodel. Instead, a cognitive architecturemore akin to the neuronal network anatomyof the brain should be pursued.

What is being referred to here isconnectionism and what its major advocateshave called "brain style computation" (30).In fact, this approach has its roots in thework by Pitts and McCulloch (9) in the1940s mentioned in an earlier section. Thesescientists, it will be recalled, showed thatthe operntion of a single nerve cell nndits connections with other nerve cellscould be modelled in terms of a formalneural network and the principles ofmathematical logic. The basic idea of'modern' connectionism is that informationprocessing is effected by the action of verysimple processing elements that are abstract

The Origins and Nature of Cognitive Paradigm 389

versions of biological neurons, along withthe interaction between these processingelements which communicate with eachother by sending a.nd receiving signals.Computation is, thus, "distributed" ina highly parallel fashion within a networkof such abstract neurons. In connectionistarchitectures of cognition-unlike thesituation in the conventional symbolrepresentation and processing model­memory resides in a distributed fashion inthe connections between the neurons ratherthan in some centralized storage system. Asnoted above, the debate continues betweenthe proponents of the 'classical' and theconnectionist models. In the opinion of thiswriter, the two models are not necessarilyin conflict; they are, rather complementary,signifying two distinct levels of explanationand abstraction of cognitive phenomena.

The role of t1eurolcieoce

Implicit in the emergence of thecognitive paradigm is the notion that whilein principle cognitive phenomena should beexplicable in terms of neurobiology, inpractice this is neither feasible nor evendesirable. In particular, to explain suchhigher level cognitive phenomena asproblem solving, planning or creativity interms of neurobiological processes p.osesth·e same kind of complexity issues asexplaining, say, the behavior of a bridge atthe level of quantum mechanics. Just asthere will always prevail a chemical levelof explanation of matter even when thesechemical concepts can be explained inphysical terms, and just as there will alwaysprevail a biological level of explanationdespite advances in chemistry and

390 Dllsgupta

physics of life, so also there will remain acognitive level of explaining the mind·brainphenomena, even when the elementaryconcepts of cognition have beenreduced to neurobiology. The different'levels of explanation' serve differentpurposes and will continue to have theirunique uses. But this obviously does notmean that the cognitive scientist can affordto ignore the findings of the neurosciences;indeed, any theories that purport toexplain cognitive phenomena in termsof information processing models must,at the very least, have what Thagardhas referred to as neurological plausibility(31).

Conversely, the cognitive neuroscientistcannot afford to ignore the results,theories and models of cognitive science.In fact, a major part of the agenda ofcognitive neuroscience is to establishand identify the neurological correlates ofmodels and tbeories of cognitive phenomena.Of course, this agenda itself has a historygoing back to the identification or mappingof brain areas responsible for such cognitivefunctions as vision and la:p.guage. A modern,albeit speculative, instance of reducingcognitive phenomena to a neurologicalmechanism is Francis Crick's conjectureabout the mechanism by which the visualcort.ex attends to visual awareness.Here, Crick drew on the psychologicalevidence that awareness entails attention,and hypothesized that the shiftingof attention from one object to anothermay be due to the coordinated firing ofa group of neurons and the suppressionof another group of coordinated neurons(32).

Indian J Physiol Pharmacol 2000; 44(4)

A very different aspect of theneurobiology/cognitive science relationshipis exemplified by research on the effectof hormones on cognitive functions.Since the late 1970s, evidence has beenaccruing that the same hormones thatactivate and control brain mechanismsunderpinning sexual behavior also influenceareas of the brain involved in learning andmemory (33)

Conclusion

Both historically and substantively,cognitive science, the science of mind, is agenuinely 'multicultural' science: manyintellectual cultures have participated in theemergence of the cognitive paradigm, andcontinue to do so today; each culture hasbrought to the problem its own paradigms,world views, techniques, tools, traditionsand vocabulary. One of the real challengesfor the cognitive scientist is to explicitlyrecognize and accept this multiculturalism.This does not mean that cognitivescience will necessarily be a fragmentedcompendium of independent disciplines; onlythat there may not be, or it is unlikely therewill be, just one mode of explainingcognitive phenomena. Rather, what seemsfar more likely is that something as complexas the mind will necessarily demandmultiple modes of inquiry, multiplemetaphors and models, and multiple levelsof explanation. If the recognition ofscientific multiculturalism is onc challengefor the cognitive scientist, the other is tomake the explanations at these differentlevels 'hang together' in a mutuallyconsistent fashion. Needless to say cognitivescience can look forward' to a very longfuture.

rod ian J Physiol Pharmacol 2000; 44(4) The Origins aud Nature of Cognitive Paradigm 391

REFERENCES

1. Bechtel W, Abrahamson A, Graham G. The Life ofCogoitive Science In: Bechtel W, Graham G, eds.A Cnmpanion to Cognitive Science, Oxford,Blackwell, 1998: pp 1-104.

2. Gardner H. The Mind's New Science. New York,Basic Books, 1985.

3. Kuhn TS. The Structure of Scientific Revolutions.Chicago. University of Chicago Press, 1970.

4. Tolman EC. Purposiue Behavior in Men andAnimals. New York, The Century Co., 1932.

5. Bartlett FC. Remembering. Cambridge. CambridgeUniversity Press. 1932.

6. Craik K. The Nature of Explanation. Cambridge.Cambridge University Pre s, 1943.

7. Jeffress LA. ed. Cerebral Mechanisms in Behavior.The Hixon Sympo.~ium. New York, Wiley. 1951.

8. Von Neumann J. The general and logical theory ofautomata. In Ref. 7, 1-41.

9. McCulloch WS, Pitts WH. A logical calculus of theideas immanent in nervous acitivity. Bulletin ofMath Biophysics 1943; 5: 115-133.

10. Lashley KS. The problem of serial order inb havior. In: Ref. 7. 112-146.

11. IcCulloch WS. Why the mind is in the head. In:Ref. 7, 42-81.

12. Turing AM. Computing machinery and intelligence.Mind 1950: 59, 2236: 433-460.

13. Newell A, Simon HA. The logic theory machine: Acomplex information processing system. IRE Transon Inform Theory 1956: IT·2. 13: 61-79.

14. Simon HA. Administratiue Behavior, New York,'fhe Free Press 1947.

15. von eumano ,1, Morgenstern O. Theory of Gamesand Economic Behauior. Princeton, PrincetonUniversity Press, 1947.

16. Chomsky N. Three models for the deSCriptionof language. IRE Trans on lnform Theory 1956:IT-2, 3: 113-124.

] 7. homsky N. Language and Mind. ew York,Harcourt Brace Jovanovich, 1972.

1 . Miller GA. The magical lumber seven plus orminus two. Psycl Reo 1956, 65: 81-97.

19 Bruner JS, Goodnow, JJ, Austin GA. A Study ofThinking. New York, John Wiley, 1956.

20. Uller C, Carey S, Huntley-Fenner G, Klatt L. Whatrepresentations might underlie infant numericalknowledge. Cognitive Development 1999; 14: 1-56.

21. Newell A. Simon HA. Human. Problem Solving.Prentice-Hall. Englewood-Cliffs, 1972.

22. Langley P, Simon HA, Bradshaw Q, Zytkow J.Scientific Disc(lvery. MIT Press Cambridge, MA,1987.

23. Gin Y, Simon HA. Laboratory replication ofscientific discovery processes. Cognitiue Science1990; 14: 281-512.

24. Newell A. Unified theories of cognition. HarvardUniversity Press, Cambridge, MA 1990.

25. Dasgupta S. Creatioity in Inuention and Design:Computational and Cognitive Explorations ofTechnological Originality. Cambridge UniversityPress, New York, 1994.

26. Newell A, Rosenbloom PS, Laird JE. Symbolicarchitectures for cognition. In: Posner MI, ed ..Foundations of Cognitive Science. MIT Press,Cambridge, MA: 1989. pp 95-132.

27. Marr D. Vision: A computationallnuestigation intoHuman- Representation and Processing of VisualInformation. W.H. Freeman, San Francisco, 1982.

28. Dasgupta S. Computer Architecture, Vol. I.Foundations. John Wiley, New York, 1989.

29. Bruner JS. Acts of Meaning. Harvard UniversityPress, Cambridge, MA. 1990.

30. Rumelhart DE. The architecture of mind: Aconnectionist approach In: Posner MI, ed. TheFoundations of Cognitive Science. MIT Press,Cambridge, MA. 1989: pp 135-159.

31. Thagard PR, Mind: An Introduction to CognitioeScience. MIT Press, Cambridge, 1996.

32. Crick FHC. The Astonishing Hypothesis.Touchstone/Simon Schuster, New York, 1994.-

33. Williams CC. Estrogen effect on cognition acrossthe lifespan. Hormones Behavior 1998; 34, 2:80-84.