Simon (1990) InConsumer Behavior_Yesterday Today and Tomorrowvariants of Human Behavior

8/12/2019 Simon (1990) InConsumer Behavior_Yesterday Today and Tomorrowvariants of Human Behavior

1/20

www.annualreviews.org/aronlineAnnual Reviews
http://www.annualreviews.org/aronlinehttp://www.annualreviews.org/aronlinehttp://www.annualreviews.org/aronline


2/20

Annu. Rev. Psychol. 1990. 41:1-19

INVARIANTS OF HUMANBEHAVIORHerbert A. Simon*Department f Psychology,Carnegie-Mellon niversity, Pittsburgh, PEnnsylvania15213

CONTENTSPHYSICALYMBOLYSTEMS................................................................ 3ADAPTIVITY........................................................................................ 5Computationalimitson Adaptivity .......................................................... 5Reasoning nderhe OptimalityPrinciple ................................................. 6Computationaleasibility: Bounded ationality ........................................... 7RationalityWithoutOptimization............................................................. 8MECHANISMSORRATIONALITY.......................................................... 8Recognition rocesses.......................................................................... 8Heuristic earch................................................................................. 9Serial PatternRecognition..................................................................... 10Proceduralationality .......................................................................... 11THINKINGNDEASONING.................................................................. 11COGNITIVERCHITECTURE.................................................................. 13LINKAGESO OTHERARTSOF PSYCHOLOGY...................................... 14IndividualDifferences........................................................................... 14SocialPsychology................................................................................ 16CONCLUSION....................................................................................... 16

The undamentaloal of science is to find invariants, such as conservation fmassand energy and the speed of light in physics. In much f science theinvariantsare neither as generalnoras invariant s these classical laws. Forinstance, the isotopes of the elements have atomic weights that are nearly~TheUSGovernmentas the right to retain a nonexclusive, royalty-free license in and to anycopyright covering this paper.*This s the eleventh in a series of prefatory chapters written by eminent enior psychologists.



3/20

2 SIMONintegral multiples of the weight of hydrogen.Somenheritable traits of plantsand animals observe the classical 1-2-1 ratio of Mendel. The number offamiliar information chunks that can be held in short-term memorys approx-imately seven. It takes about 30 seconds to memorize an unpronouncablethree-consonant nonsense syllable, but only about nine seconds to memorizethree-letter word.Much iological knowledge s extremely specific, for biology rests on thediversity of millions of species of plants and animals, and most of itsinvariants apply only to single species. Becauseof inter-species moleculardifferences, even the important general laws (e.g. the laws of photosynthesis)vary in detail from one species to another (and sometimesamongdifferentindividuals in a single species). Onlyat the most abstract and qualitative levelcan one find manygeneral strict invariants in biology.Moreover,someof the most important invariants in science are not quan-titative at all, but are what Allen Newell and I (1976) have called lawsqualitative structure. For example, the germ heory of disease, surely one ofPasteurs major contributions to biology, says only something ike: If youobserve pathology, look for a microorganism t might be causing the symp-toms. Similarly, modemmolecular genetics stems from the approximatelycorrect generalization that inheritance of traits is governed y the arrangementof long helical sequences of the four DNA ucleotides.

Finally, in biological (including human) ealms, systems change adaptivelyover time. Simple change is not the problem, for Newton howedhowwe canwrite invariant laws as differential equations that describe the eternal move-ments of the heavens. But with adaptative change, which is as muchgovernedby a systems environment as by its internal constitution, it becomesmoredifficult to identify true invariants. As a result, evolutionary biology has arather different flavor from physics, chemistry, or even molecular biology.In establishing aspirations for psychology t is useful to keep all of thesemodels of science in mind. Psychology does not muchresemble classicalmechanics, nor should it aim to do so. Its laws are, and will be, limited inrange and generality and will be mainlyqualitative. Its invariants are and willbe of the kinds that are appropriate to adaptive systems. Its success must bemeasured ot by howclosely it resemblesphysics but by howwell it describesand explains humanbehavior.On another occasion (Simon 1979a) I have considered the form a sciencemust take in order to explain the behavior of an adaptive, hence of anartificial, system. By artificial I meana system that is what it is onlybecause it has responded o the shaping forces of an environment o which itmust adapt in order to survive. Adaptation may be quite unconscious andunintended, as in Darwinianevolution, or it maycontain large components fconscious intention, as in muchhuman earning and problem solving.



4/20

BEHAVIORALNVARIANTS 3Taking the artificiality of human ehavior as my central theme, I shouldlike to consider its implications for psychology. Moreover, since Homosapiens shares some mportant psychological invariants with certain nonbio-

logical systems--the computers I shall want to make frequent reference tothemalso. One could even say that my account will cover the topic of humanand computer psychology.PHYSICAL SYMBOL SYSTEMSAn mportant aw of qualitative structure underlies the information processingparadigm in psychology. The Physical SymbolSystem Hypothesis (NewellSimon 976)states that a system will be capableof intelligent behavior if andonly if it is a physical symbolsystem. A physical symbolsystem is a systemcapable of inputting, outputting, storing, and modifyingsymbolstructures,and of carrying out someof these actions in response to the symbols them-selves. Symbols re any kinds of patterns on which hese operations can beperformed, where some of the patterns denote actions (that is, serve ascommandsr instructions).We re all familiar with the physical symbol systems called computers.Computersstore symbols in the form of electro-magnetic patterns of somekind (quite different kinds in different computers); someof these patternsserve to instruct the computer what to do next (the stored program), whileothers contain numerical or nonnumerical nformation.Information processing psychology laims that intelligence is achievable byphysical symbol systems and only such systems. From that claim follow twoempirically testable hypotheses: 1. that computers can be programmedothink, and 2. that the human rain is (at least) a physical symbolsystem.These hypotheses are tested by programming omputers to perform the sametasks that we use to judge howwell people are thinking, and then by showingthat the processes used by the computerprogramsare the same as those usedby people performing these tasks. In making he comparisonwe use thinking-aloud protocols, records of eye movements, eaction times, and manyotherkinds of data as evidence.The physical symbolsystem hypothesis has been tested so extensively overthe past 30 years that it can nowbe regarded as fully established, althoughover less than the wholegamutof activities that are called thinking. Forstarters in reviewing the evidence, I would recommendNewell & Simon(1972), Simon(1979a, b, 1989a), and Anderson 1983). Readers can contin-ue the survey with numerous eferences they will find in those sources. Theexact boundaries of our present knowledge eed not concern us: The territoryin which the hypothesis has been confirmed is broad, encompassingmanyofthe kinds of activities that define human rofessional and scholarly work.



5/20


6/20

BEHAVIORALNVARIANTS 5Science suspended from skyhooks s not new, nor is it limited to particulardisciplines. Contemporary physics provides a prime example of skyhookscience in its continual movement ownwardo ever more fundamental and elementary particles, its greatest uncertainties lying always at the founda-tions.The separation of information processing from neural science presents animportant challenge to research, and a second great direction for explorationin psychology. What rrangementof neurons or neuronal circuits correspondsto a symbol? What s the physiological basis for the magical numberseven?By what mechanism does the presence of a symbol in short-term memoryinitiate or guide a mental action? The agenda, containing these and manyother items, provides work or both neural scientists and information process-ing psychologists, for the bridge will have to be built out from both banksbefore it can link in the middle.ADAPTIVITYLet menowput aside biological questions and return to human daptivity andits implications for the laws of psychology. A look at computer adaptivitymaycast some ight on the human ind. A computer, it is said, can only dowhat it is programmedo do (which maybe quite different from what theprogrammerntended it to do). Generally, it is not instructed to do specificthings at all (e.g. to solve a particular linear programmingroblem), butadapt its behavior to the requirements of a given task chosen from a wholepopulation of tasks (e.g. to solve any linear programming roblem lyingwithin given size limits). Then ts behavior n response o each task is adaptedto the requirements of the task, and it behaves differently, in appropriateways, with each task it is given. In short, it is an adaptive system.The adaptiveness of computers eads to a question that is the converse ofthe one raised above. Can a computer be programmed o do anything? Ofcourse not. Upper limits are set by the famous theorems of Grdel, whichprove that every symbolprocessing system must be, in a certain fundamentalsense, incomplete. It is a truth of mathematicsand logic that any program(including those stored in humanheads) must be unable to solve certainproblems.Computational Limits on AdaptivityFar more important than the Grdel limits are the limits imposedby the speedand organization of a systems computationsand sizes of its memories. t iseasy to pose problems hat are far too large, require far too muchcomputa-tion, to be solved by present or prospective computers.Playing a perfect gameof chess by using the game-theoretic minimaxing algorithm is one such



7/20

6 SIMONinfeasible computation, for it calls for the examination of morechess posi-tions than there are molecules n the universe. If the gameof chess, limited toits 64 squares and six kind of pieces, is beyondexact computation, then wemayexpect the same of almost any real-world problem, including almost anyproblem of everyday life.From this simple fact, we derive one of the most important laws ofqualitative structure applying to physical symbolsystems, computersand thehumanbrain included: Because of the limits on their computingspeeds andpower, intelligent systems must use approximate methods to handle mosttasks. Their rationality is bounded.Reasoning Under the Optimality PrincipleHistorically, human daptiveness (that is to say, rationality) has preoccupiedeconomists even more than psychologists. Modemmainstream economictheory bravely assumes hat people make heir decisions in such a way as tomaximize their utility (Simon 1979a). Accepting this assumption enableseconomicso predict a great deal of behavior(correctly or incorrectly) withoutever makingempirical studies of human ctors.If we wish to knowwhat form gelatin will take when t solidifies, we do notstudy the gelatin; we study the shape of the mold in which we are going topour it. In the same way, the economist who wishes to predict behaviorstudies the environment n which the behavior takes place, for the rationaleconomic ctor will behave in whateverway is appropriate to maximize tilityin that environment. Hence (assuming the utility function to be given inadvance), this maximizingbehavior is purely a function of the environment,and quite independent of the actor.

The same strategy can be used to construct a psychologyof thinking. If wewish to know owan intelligent person will behave n the face of a particularproblem, we can investigate the requirements of the problem. Intelligenceconsists precisely in responding to these requirements. This strategy has, infact, been pursued occasionally in psychology; he theories of perception of J.J. Gibson (1966) and John Marr(1982) exemplify it, as do someof the recentrational models of my colleague John R. Anderson (1989).Why ont we, then, close up the laboratory, frequently a place of vexinglabors and unwelcome urprises, and build a psychology of intelligence byrational analysis, as the economists have done? The answer, already sug-gested, lies in the law that I have called the Principle of BoundedRational-ity (Simon1989b). Since we can rarely solve our problemsexactly, the opti-mizing strategy suggested by rational analysis is seldom available. Wemustfind techniques for solving our problemsapproximately, and we arrive at dif-ferent solutions depending on what approximations we hit upon. Hence, todescribe, predict and explain the behavior of a system of bounded ational-



8/20

BEHAVIORALNVARIANTS 7ity, we must both construct a theory of the systems processes and describethe environments o which it is adapting.Computational Feasibility: Bounded RationalityHumanational behavior (and the rational behavior of all physical symbolsystems) is shaped by a scissors whose wo blades are the structure of taskenvironmentsand the computational capabilities of the actor.The study of cognitive psychology s the study of computational capabili-ties in the face of diverse asks. It is not a trivial detail but a fundamentalimitupon computation that human hort-term memory an hold only a half dozenchunks, hat an act of recognition takes nearly a second, and that the simplesthuman eactions are measured in tens and hundreds of milliseconds ratherthan microsecondsor picoseconds. These basic physiological constants de-termine what kinds of computations are feasible in a given kind of tasksituation and how rapidly they can be carded out (Newell & Simon 1972;Simon1979a). They are among he most important invariants that cognitivepsychology has discovered, accounting for manyphenomenaobserved inthinking and learning.Noting that computational imits must be a central preoccupation of cogni-tive psychologydoes not exhaust the complications of the subject. Wehavealso to take into account that thinking capacities are a function of skill andknowledge, tored neural structures in the brain. The expert can reach solu-tions that are unattainable by the novice, using computations nd kno~,vledgethat are simplynot available to the latter.A lightning calculator carries out elementary symbolic processes no morerapidly than a person with ordinary skills in arithmetic; empirical studiesreveal little or no difference in the speeds of their basic processes. Superiorityin computationderives almost entirely from superior knowledge f arithmeticfacts (e.g. knowledge f the multiplication table up to relatively large num-bers, or of the table of squares, or of primefactors), combined ith a superiorrepertory of computational trategies that save steps and conserve short-termand long-termmemoryapacity. In Chi et al (1988) the reader will find recentpapers on expert performance in a variety of tasks, including memory ndcomputationalfeats.A major way to relax the limits of bounded rationality is to store inlong-term memory nowledge and strategies that reduce the computationalrequirements of tasks. This would seem to add new plausibility to theargument or studying the requirementsof the task rather than the propertiesof the actor. But the argument till fails. In tasks of any complexity, knowl-edge and strategies do not allow the expert to find an optimal solution, butonly to find approximationshat are far better than those available to native(or naive) intelligence. A knowledge f the calculus allows its possessor



9/20

8 SIMONsolve manyproblems that could not be solved without it, but the domainofdifferential equations that cannot be exactly integrated in closed form vastlyexceeds the domainof those that can be.Rationality Without OptimizationThe wide-ranging attempts since the SecondWorld War o apply the optimiz-ing tools of operations research (linear programming, nteger and dynamicprogramming,queuing theory, and so on) to the decision problems of man-agement have underlined the computational complexity of real-world prob-lems, even relatively well-structured problems that are easily quantified.Using queuing theory, an optimumproduction schedule can be found for afactory that manufactures one or two products, using one or two differentpieces of equipment. Adding even one more product or piece of equipmentputs the problembeyondcomputational bounds for the fastest supercomputer.(An optimal class schedule for a university lies even further beyond he limitsof practical computation.)Yet factories (and universities) are scheduled every day. We re forcedconclude that methodsother than optimization are used--methods hat respectthe limits of human nd computer rationality. Perhaps the feasible methodsare specific to each specific situation, in whichcase it is hard to see whatcognitive psychology should say about them.

On the other hand, it is possible that somecommonroperties, derivingfrom human ounded ationality, are shared by the approximating procedurespeople use in manykinds of complex situations. If so, it is the task ofcognitive psychology o characterize these procedures, to showhow hey areacquired, and to account for their compatibility with the known omputationallimitations of the human rain.MECHANISMS FOR RATIONALITYLet me llustrate someof the mechanisms sed by human ounded ationalityto cope with real-life complexity. I will give just three examples rom a muchlarger number hat could be cited: processes used in problem solving byrecognition, processes of heuristic search, and processes for inducing sequen-tial patterns.Recognition ProcessesWenow know hat experts makeextensive use of recognition processes, basedon stored knowledge, to handle their everyday tasks. This recognitioncapability, based (by rough estimate) on 50,000 or more stored cues andassociated knowledge, allows them to solve manyproblems intuitively --that is, in a few seconds, and without conscious analysis. Recognizingkey



10/20

BEHAVIORALNVARIANTS 9cues allows experts to retrieve directly from memorynformation for dealingwith the situations that the cues identify. Recognition processes have beenshown o play a major role, perhaps the major role, in such diverse tasks asgrandmaster chessplaying, medical diagnosis, and reading. Introductions tothe evidence will be found in de Groot (1978), Simon 1979a) and Chi et(1988).Computer simulation models like EPAMFeigenbaum & Simon 1984)provide explanatory mechanismsor recognition-based expertise, including alearning mechanism or acquiring the stored chunks on which it is based.Alternative models are being developed n the form of parallel, connectionistsystems (EPAMs a basically serial system). The theoretical explanations andcomputer models assume processing speeds that are well within the knownhuman hysiological limits, and EPAM,s least, predicts a wide range of thephenomenahat have been reported in the verbal learning literature (includingthe times reported by Ebbinghaus or the learning of nonsensesyllables). Wecan regard intuition as a phenomenonhat has been rather thoroughly ex-plained: It is achieved through acts of recognition.Heuristic SearchWhat bout problemswhose olutions are not provided by immediate ecognition,but which equire analysis? Herealso, a number f the principal processes havebeen dentified and simulated.Collectively, they are usually called heuristic (orselective) search. When great space of possibilities is to be explored (andhumans ommonlyalk at searching spaces when he possibilities number ven inthe hundreds), earch becomesery selective. It is then guidedby various roles ofthumb,or heuristics, someof whichare specific to particular tasks, but some fwhich are more general (Newell & Simon 1972).If the task domains highly structured, the task-specific heuristics may every powerful, drawing upon the structural information to guide searchdirectly to the goal. For instance, most of us apply a systematic algorithmwhenwe must solve a linear equation in algebra. Wedont try out differentpossible solutions, but employ ystematic steps that take us directly to thecorrect value of the unknown.If the task domainhas little structure or the structure is unknowno us, weapply so-called weakmethods, which experience has shown o be useful inmanydomains, but which maystill require us to search a good deal. Oneweakmethod s satisficing--using experience to construct an expectation ofhowgooda solution we might reasonably achieve, and halting search as soonas a solution is reached that meets the expectation.

Picking the first satisfactory alternative solves the problemof makingachoice whenever (a) an enormous, or even potentially infinite, numberalternatives are to be comparedand (b) the problem has so little known



11/20

10 SIMONstructure that all alternatives wouldhave o be examinedn order to determinewhich is optimal. Satisficing also solves the common roblem of makingchoices when lternatives are incommensurable, ither because (a) they havenumerous dimensions of value that cannot be compared, (b) they haveuncertain outcomes hat maybe more or less favorable or unfavorable, or (c)they affect the values of more han one person. Thena satisficing choice canstill be made s soon as an alternative is found hat (a) is satisfactory alongalldimensionsof value, (b) has satisfactory outcomes or all resolutions of theuncertainty, or (c) is satisfactory for all parties concerned, espectively.Another weak mothod s means-ends analysis--noting differences betweenthe current situation and the desired goal situation, and retrieving frommemory perators that, experience has taught us, removedifferences of thesekinds.A small collection of heuristics, of which satisficing and means-endsanalysis are important examples, have been observed as central features ofbehavior in a wide range of problem-solving behaviors where recognitioncapabilities or systematic algorithms were not available for reaching solutionswithout search. The prevalanceof heuristic search is a basic law of qualitativestructure for humanproblem solving.Beginning with the General Problem Solver (GPS) in about 1958, a size-able number of computer programs have been built to simulate heuristicsearch in various task domains.With heir help, a rather detailed account hasbeen given of humanheuristic search, particularly in relatively well-structured domains hat call upon only limited amounts of domain-specificknowledge (Newell & Simon 1972). With these programs as foundation,other investigations have built processes that can create problemrepresenta-tions for simple situations, using natural language inputs to supply informa-tion about the problem and task domain.Serial Pattern RecognitionAbility to find patterns in sequencesof numbers, etters, or geometricfiguresis an important componentof human ntelligence (Simon & Kotovsky 1963;Kotovsky & Simon1973). The Thurstone Letter Series Completion Test andthe Ravens tests are examples of tasks aimed at measuring this component.Laboratorystudies of these tasks and computer imulations show hat success-ful humanperformance depends on a few basic pattern-recognizing andpattern-organizing capabilities. In extrapolating sequential patterns, for ex-ample, subjects notice when identical symbols recur, or when there aresubsequences of symbols that are successive items in a familiar list or alphabet.When ubsequences epeat in a sequence, subjects can notice this fact andtreat the repetitive subsequences as unitary components n a higher-level



12/20

BEHAVIORALNVARIANTS 11pattern. Thus, recursive or hierarchical patterns can be detected and ex-trapolated. These capabilities can be shown o be adequate for detectingpattern in complexpieces of music, and their sufficiency has been demon-strated by simulation programs capable of carrying out nontrivial musicalanalysis (Simon & Sumner 1968).Procedural RationalityProblem olving by recognition, by heuristic search, and by pattern recogni-tion and extrapolation are examples of rational adaptation to complex taskenvironments hat take appropriate account of computational limitations--ofboundedrationality. They are not optimizing techniques, but methods forarriving at satisfactory solutions with modest amountsof computation. Theydo not exhaust, but they typify, what we have been learning about humancognition, and they go a long way toward explaining how an organism withrather modest computational capabilities can adapt to a world that is verycomplex indeed.The study of humanbehavior in the face of difficult tasks showswhyweneed a theory of processes (procedural rationality) as well as a theory of therequirementsof the task (substantive rationality). A theory based only on taskrequirements could not tellus howbehavior depends on knowledge of rele-vant cues or strategies. It could not explain whywe satisfice instead ofoptimizing, or how we solve most everyday problems by recognizing cuesthat evoke their solutions. It could not give us a grasp of the range ofstrategies that maybe available for handling a particular task, or the differ-ences between expert and novice performance on the task. All these phe-nomenabecome nderstandable as we explore, by laboratory experiments andcomputer simulations, actual humanbehavior in a variety of task environ-ments.THINKING AND REASONINGIf we go back, say, to WoodworthsExperimental Psychology (1938),find that accounts of the human higher mental functions flow along twoquite different channels, representingdifferent intellectual ties to the adjacentdisciplines. Woodworthevotes two chapters to complexcognitive tasks: oneto problem solving, the other to reasoning (but titled Thinking ). Wood-worths own comment 1938, p. 746) is Twochapters will not be too manyfor the large topic of thinking, and we maymake he division according to thehistorical sources of two streams of experimentation, whichdo indeed mergein the more ecent work. Onestream arose in the study of animal behavior andwent on to human roblem solving; the other started with human hinking ofthe more verbal sort. In particular, research on problem solving had itsorigins in the disputes about trial-and-error versus insightful learning. Re-



13/20

12 SIMONsearch on reasoning derived from attention to theories of language and logicas models of thought processes.The problem-solving model, or metaphor, has generally been preferred byGestalt psychologists, with their commonelief that insightful thinking isnonverbal in nature, and by researchers in artificial intelligence, whohavefrom the beginning described thinking as heuristic search. The reasoningmodel, or metaphor,has generally been preferred by linguists with an interestin cognitive science and by philosophers whostray into this domain. Theydescribe thought processes in terms of propositions and logical manipulationsof propositions.A study of mutual citations would show hat communication etween thesetwo streams of inquiry has been poor. This reveals itself also in the differentprogramming anguages the two groups adopt when they simulate thinkingprocesses. The programminganguages associated with heuristic search arelist-processing languages like LISP and production-system languages likeOPS5. The programming anguages associated with reasoning are logic lan-guages (languages adapted to theorem proving in the predicate calculus) likePROLOG.The division is further reinforced by disagreement about the respectiveroles in thinking of sentences (or the propositions they denote) and imageryone or another kind. The reasoning metaphor views goals as described bysentences, derived from other sentences by processes similar to the processesof logic. The problem-solving metaphor views goals as achieved by se-quences of moves through a problem space. (The very phrase problemspace suggests the importance that is attached to a visual or spatialmetaphor.)Whenhe reasoning metaphor is used, information is expressed mainly indeclarative sentences. A small numberof rules of inference (like the rule ofsyllogism in formal logic) are used to derive new sentences from old. Theresearch tasks most commonly mployed o study human easoning are tasksof concept formation or tasks of judging the validity or invalidity of formalsyllogisms, the presumption being that human hinking consists in drawingvalid inferences from given premises or data.Whenhe problem-solving metaphor is used, information is expressed inschemas, which mayresemble interrelated sets of sentences or mayresemblediagrams or pictures of the problem situation. The problem situation ismodified by applying moveoperators, which are processes that change asituation into a newone. Nowadays,he moveoperators usually take the formof productions---condition-action pairs, C--> A. Wheneverhe information inshort-term memorymatches the conditions of a production, the actions of theproduction are executed. The execution of a sequence of productions accom-plishes a search through the problem space, moving from one situation toanother until a situation satisfying the goal requirements s reached.



14/20

BEHAVIORAL INVARIANTS 13All of these differences can be seen by comparing the corresponding two

chapters, mentioned above, of Woodworth(1938), and chapters 8 and 10Anderson (1985).

Determining to what extent human thinking fits the problem-solvingmetaphor and to what extent it fits the reasoning metaphor stands high on theagenda of cognitive psychology today. Of course, the answer may be both ofthe above, the processes of thought varying with the task domain and withlearned o~ innate differences among he thinkers. It is perhaps of interest thatJohnson-Laird, one of the leaders among those who emphasize the tiesbetween cognition and linguistics, has recently begun to describe thinking interms of mental models, an approach that lies muchcloser to the heuristicsearch paradigm than to the reasoning paradigm (Johnson-Laird 1983). Butwould hesitate to predict what this particular defection from the linguisticcamp portends for the future.COGNITIVE ARCHITECTUREThe whole congeries of mechanisms of human rationality must somehow beorganized in the human brain to work together in a coordinated fashion.Today, a good deal of effort of theorists in psychology is devoted to specify-ing the architectures that achieve this coordination. In this context, architec-ture refers to description of the cognitive system at an abstract, usuallysymbolic, level, and has little to say about the underlying biology of neurons.

Early information-processing architectures of cognition (e.g. Broadbent1958) emphasized memory boxes and their interconnections. Today,architectures specify organizations of processes as well as storage. Amongheproposals that enter prominently into current discussion are Andersons(1983) Act*, Newells (1989) SOARboth symbolic), and the connectionistsystem of McClelland & Rumelhart (1986).

Confining our discussion to symbolic architectures, while there are signifi-cant differences among them, none of them are incompatible with the recogni-tion, heuristic search, and pattern-induction mechanismsdescribed in the lastsection. For many purposes, it is sufficient to think of the whole cognitiveman as comprised of the following components:

Memories: short-term memoryf limited capacity (workingmemory),n associativelong-term memory,n EPAMiscrimination net to index the long-term memory, ndsmaller short-termmemoriesssociated with various sensoryand motormodalities. Sensory rocessorso extract features from timuli. Interpreters of motor ignals. EPAM: discriminationnet for sensory eatures that learns newdiscriminationsand chunks amiliar stimuli patterns. GPS problemolver hat employseuristic search,and hat can be usedby the learningsubsystem.



15/20

14 SIMON APatternnductionystemhat searchesor regular atterns n stimuli. Systemsor encodingatural anguagenput andproducingatural languageutput. Systemsor encodingo and rom mage-likeepresentations. Anadaptiveproductionystem,capableof creating new rocesses n the basis ofinformationained hroughnstruction, hrough xaminingorked-outxamples,ndbysolving roblems.This mayappear to be a lot of baggage, but all of the processing systemslisted are implementable as production systems that can be stored in theassociative long-term memory. Moreover, examples of all of these com-ponents have been simulated with computer programs, and their mutualcompatibility tested to some degree. For example, the UNDERSTANDys-tem (Hayes & Simon 1974) can encode natural language descriptionspuzzles into internal representations that are suitable problem paces for GPS.The ISAAC ystem (Novak 1976) can encode natural language statementsphysics problems into internal images, and use these images to producealgebraic equations, which t then solves.From his we mayconclude that, while many ssues about architecture arefluid at the present time, the knowledge we gain about architectures isunlikely to invalidate, or require major revision of, the knowledgewe havealready gained about component mechanisms like EPAM r GPS, or abouttheir roles in cognition.

LINKAGES TO OTHER PARTS OF PSYCHOLOGYContemporary information-processing psychology holds forth significantpossibilities for a greater unification among omainsof psychology hat arenowquite separate. It maybe possible to forge stronger links of cognitivepsychologywith the study of child development, with research on individualdifferences, with psycholinguistics, and with social psychology.There is already vigorous research in cognitive developmental psychologythat makes use of manyof the constructs discussed here. Computationallinguistics and psycholinguistics also have proceeded along parallel--andsometimeseven intersecting--lines. Some esearchers on individual differ-ences [the names of Hunt (1975) and Sternberg (1977) come immediatelymind] work within an information-processing framework. The linkages tosocial psychology are somewhatmore tenuous, but are beginning to form, aswe shall see. Without attempting a systematic review, I would ike to offercomments n several of these topics in particular, individual differences,social psychology, and psycholinguistics.Individual DifferencesTraditionally, the study of individual differences has employed sychometricmethodsof research. It has been motivated by interest in the nature/nurture



16/20

BEHAVIORALNVARIANTS 15controversy as well as by morepractical concernsof predicting and explainingschool and job performance. L. L. Thurstones Vectors of the Mind(1935),which characterized each individual by a vector of weights for individualtraits and predicted individual performance n specific tasks from he correla-tions between hese weights and the importanceof the corresponding raits forthe tasks, provides a template (or perhaps a caricature) for this viewindividual differences.Thinking-aloudprotocols, by providing rich information about the behaviorof individual subjects, have focusedattention oh the large differences in thesebehaviors (Ericsson & Simon 1984). While protocol analysis has provedmuchmore receptive to the study of individual differences than experimentaldesigns that take averages over sets of subjects, it has been ncorporated onlyincompletely into the current literature on individual differences. For ex-ample, Carrolls (1988) chapter on Individual Differences in CognitiveFunctioning, in the newStevens Handbookof Experimental Psychology,Volume , makes almost no reference to the new methods or the research onexpert-novice differences produced by them.Attending to the processes that subjects use in performing complex askshas enabled us to characterize the differences between expert and noviceperformance in many ask domains. In all of these domains, differences inknowledge which must include learned skills as well as factual knowledge)prove to be a dominantsource of differences in performance Chi et al 1988).Of course, this finding should not be taken to deny he existence of innatedifferences, but rather to account for their relative (quantitative) in-significance in explaining differences in skilled adult performance. Noonewould argue that any randomly selected person could be trained to playworld-class tennis; but one could argue, on the basis of the evidence nowavailable, that most normal humanbeings could become reasonably goodplayers with sufficient training and practice, and that none could becomeexcellent players without extensive training and practice.

Knowledgencludes knowledgeof strategies. A good deal of the researchon expert-novice differences has been aimed at understanding the strategiesthat experts acquire and apply, how hese strategies can be learned, and towhat extent they are transferable from one task domain o another. Hence, nthe contemporaryparadigms, the study of individual differences is closelytied to the study of learning and transfer of training. These ties, in turn,introduce a strong taxonomic spect into the study of individual differences,makingclear that a great many ask domainswill have to be analyzed beforewe can generalize safely about human kills.The new connections between skills and processes affect not only ourunderstanding of complexperformances, but also our interpretations of thesimple processes that underlie them. We re aware oday of the centrality ofshort-term memory imits to performance on many, if not most, cognitive



17/20

16 SIMONtasks, and we have long used George Millers seven chunks to characterizethose limits.But a chunk is not an innate measureof storage capacity. A chunk is anystimulus that has become amiliar, hence recognizable, through experience.Hence, the capacity of short-term memorys itself determined by learning,and can grow o vast size as individual acts of recognition access larger andricher stores of information in long-term memory. Two EPAM ystemspossessing the same basic structure can differ greatly in measured STMcapacity simply because one has a more elaborate differentiation net andassociated store of schemas han the other.Social PsychologyJust as individual differences find a natural place in information processingpsychology, so do social phenomena.To the extent that cognitive peformancerests on skill and knowledge, t is a social, rather than a purely individual,phenomenon. Language skills and skills in social interaction can beapproached within the same theoretical framework as knowledgeand skillsfor dealing with the physical environment.The recent workof Voss et al (1983) illustrates howcognitive and socialpsychology can mutually reinforce each other. He has studied howpeoplewho have different professional backgrounds and information approach thesame problem-solvingsituation. When sked to write an essay on agriculturalreform in the USSR,subjects who are experts in agronomyaddress them-selves to entirely different variables and strategies than subjects whoareexperts on Russian political affairs. Andboth of these groups of subjectsrespond quite differently from novices. Whenwe study expert behavior, wecannot help studying the structure of professional disciplines in our society.Cognitive psychologystill has an important task of studying the domain-independent components of cognitive performance. But since the perfor-mancedepends heavily on socially structured and socially acquired knowl-edge, it must pay constant attention to the social environmentof cognition.Many f the invariants we see in behavior are social invariants. Andsincethey are social invariants, many re invariant only over a particular society ora particular era, or even over a particular social or professional groupwithin asociety. Social variables must be introduced to set the boundaries of ourgeneralizations.CONCLUSIONLet mesummarize riefly this account of the invariants of human ehavior asthey are disclosed by contemporary cognitive psychology. The problem ofidentifying invariants is complicated by the fact that people are adaptive



18/20

BEHAVIORALNVARIANTS 17systems, whosebehavior is highly flexible. The invariants must be sought inthe mechanismshat allow them to solve problems and learn: the mechanismsof intelligence.

The Physical SymbolSystem Hypothesis, strongly supported by empiricalevidence, asserts that a system will be capable of intelligent behavior if andonly if it is a physical symbolsystem: if it can input, output, store, andmanipulate symbols. The hypothesis, and consequently information-processing psychology, describes intelligence at a symbolic, software,level, saying little about brain physiology. That fact need not impede ourprogress, nor has it, toward understanding at this symbolic level how aphysical symbolsystem like the brain achieves intelligent behavior.Because of the limits on their computing speeds and power, intelligentsystems must use approximatemethods. Optimality is beyond heir capabili-ties; their rationality is bounded.To explain the behavior of a system ofbounded ationality we must describe the systems processes and also theenvironments to which it is adapting. Human hort-term memory an holdonly a half dozen chunks, an act of recognition takes nearly a second, and thesimplest human eactions are measured n tens and hundreds of milliseconds,rather than microseconds, nanoseconds, or picoseconds. These limits areamonghe most important: invariants of intelligence.A majorstrategy for achieving ntelligent adaptation with bounded ational-ity is to store knowledge nd search heuristics in a richly indexed long-termmemory n order to reduce the computational requirements of problems.Experts use recognition processes, based on this stored, indexed knowledge,to handle their everyday tasks. When ecognition does not suffice, because agreat space of possibilities must be explored, they resort to highly selectivesearch, guided by rich stores of heuristics.When ntelligence explores unfamiliar domains, it falls back on weakmethods, which are independent of domain knowledge. People satisfice--look for good-enougholutions--instead of hopelessly searching for the best.They use means-ends nalysis to reduce progressively their distance from thedesired goal. Paying attention to symmetriesand orderly sequences, they seekpatterns in their environments hat they can exploit for prediction. Problemsolving by recognition, by heuristic search, and by pattern recognition areadaptive techniques that are compatible with bounded ationality.Several cognitive architectures have been proposed to account for theprocesses just described, but these architectures represent relatively modestvariations on a basic pattern that is widely accepted oday. This basic patterninvolves somesensory processors that provide input into short-term andlong-term memory,a recognition process that discriminates the featuresdetected by the senses, a problem solver that employs heuristic search, apattern induction system, systems for handling natural language, sys-



19/20

18 SIMONtems for handling image-like representations, and learning mechanisms thatpermit new processes and data structures to be constructed and stored inmemory.

The picture I have drawn of cognitive psychology and the invariants ofintelligence holds forth the promise of linking several parts of psychology thatnow mostly go their separate ways. Developmental psychology has alreadybeen strongly influenced by the cognitive revolution. The approach to in-telligence and individual differences is beginning to be modified in an in-formation-processing direction. But we are just beginning to see that, becauseof the strong dependence of intelligence on stored knowledge, cognitive andsocial psychology must be brought much closer together than they have beenin the recent past. When we have made these new connections solid, thechallenge will remain of bringing affect and emotion more centrally into thepicture.ACKNOWLEDGMENTSThis research was supported by the Personnel and Training Programs, Psy-chological Sciences Division, Office of Naval Research, under Contract No.N00014-86-K-0768; and by the Defense Advanced Research Projects Agen-cy, Department of Defense, ARPA order 3597, monitored by the Air ForceAvionics Laboratory under contract F33615-81-K-1539.Literature CitedAnderson, J. R. 1983. The Architecture ofComplexity. Cambridge MA:Harvard Univ.PressAnderson, J. R. 1985. Cognitive Psychologyand Its Implications. NewYork: FreemanAnderson, J. R. 1989. The place of cognitivearchitectures in a rational analysis. 22ndAnnu. Symp. Cognit., Dept. Psychol., Car-negie-Mellon Univ.Broadbent, D. E. 1958. Perception and Com-munication. New York: PergamonCarroll, J. B. 1988. Individual differences incognitive functioning. In Stevens Hand-book of Experimental Psychology, ed. R. C.Atkinson, et al, Vol. 2:813~52. NewYork:Wiley. 2nd ed.Chi, M. T. H., Glaser, R., Farr, M., eds.1988. The Nature of Expertise. Hillsdale,NJ: ErlbaumDe Groot, A. 1978. Chance and Choice inChess. The Hague: Mouton. 2nd ed.Ericsson, K. A., Simon, H. A. 1984. ProtocolAnalysis. Cambridge, MA:MIT PressFeigenbaum, E. A., Simon, H. A. 1984.EPAM-Iike models of recognition andlearning. Cogn. Sci. 8:305-36Gibson, J. J. 1966. The Senses Consideredas Perceptual Systems. Boston: HoughtonMifflin

Hayes, J. R., Simon, H. A. 1974. Un-derstanding written problem nstructions. InKnowledge and Cognition, ed. L. W.Gregg, pp. 167-200. Hillsdale, N J:ErlbaumHunt, E. B., Mansman, M. 1975. Cognitivetheory applied to individual differences. InHandbookof Learning and Cognitive Pro-cesses, ed. W. K. Estes, Vol. 1. Hillsdale,NJ: ErlbaumJohnson-Laird, P. N. 1983. Mental Models.Cambridge, MA:Harvard Univ.Kotovsky, K., Simon, H. A. 1973. Empiricaltests of a theory of humanacquisition ofconcepts for sequential patterns. Cogn.Psy-chol. 4:399-424Marl D. 1982. Vision. San Francisco: Free-manMcClelland, J. L., Rumelhart, D. E. 1986.Parallel Distributed Processing, Vol. 2.Cambridge, MA: MIT PressNewell, A. 1989. Unified Theories of Cogni-tion. The William James Lectures. Cam-brid.ge, MA:Harvard Univ. Press. Forth-commgNewell, A., Simon, H. A. 1972. HumanProblem Solving. Englewood Cliffs, NJ:Prentice-HallNewell, A., Simon, H. A. 1976. Computer



20/20

BEHAVIORAL INVARIANTS 19science as empirical inquiry: symbols andsearch. Commun. ACM19:111-26Novak, G. S. Jr. 1976. Computerunderstand-ing of physics problems stated in naturallanguage. Tech. Rep, No. NL-30. Austin,TX: Dept. Comput. Sci., Univ. TexasSimon, H. A. 1979a. The Sciences of theArtificial. Cambridge, MA:MITPress. 2rided.Simon, H. A. 1979b. Models of Thought, Vol.1. NewHaven: Yale Univ. PressSimon, H. A. 1986. The information process-ing explanation of Gestalt phenomena.Comput. Hum. Behav. 2:241-55Simon, H. A. 1989a. Models of Thought,Vol. 2, NewHaven: Yale Univ PressSimon, H. A. 1989b. Cognitive architecturesand rational analysis: comments.21stAnnu.Syrup Cognit., Dept. Psychol., Carnegie-Mellon Univ.

Simon, H. A., Kotovsky, K. 1963. Humanacquisition of concepts for serial patterns.Psychol. Rev. 70:534-46Simon, H. A., Sumner, R. K. 1968. Pattern inmusic. In Formal Representation of HumanJudgment, ed. B. Kleinmuntz. NewYork:WileySternberg, R. J. 1977. Intelligence, Informa-tion Processing, and Analogical Reasoning.Hillsdale, NJ: ErlbaumThurstone, L. L. 1935. Vectors of the Mind.Chicago: Univ. Chicago PressVoss, J. F., Greene, T. R., Post, T. A., Pen-ner, B. C. 1983. Problem solving skill inthe social sciences. In The Psychology ofLearning and Motivation, ed. G. H. Bower,Vol. 17. NewYork: AcademicWoodworth, R. S. 1938. Experimental Psy-chology. NewYork: Holt


Simon (1990) InConsumer Behavior_Yesterday Today and Tomorrowvariants of Human Behavior

Documents

Transcript of Simon (1990) InConsumer Behavior_Yesterday Today and Tomorrowvariants of Human Behavior