COGNITION AND ECONOMICS PART 3 potx

In their concentration on individual incentives played out incontexts of imperfect information, they incorporate observations of the actualincentives encountered by scientists, and bring

ADAPTIVE CLASSIFYING SYSTEMS Thomas J McQuade

1 INTRODUCTION

As the cognitive sciences – particularly neuroscience, cognitive psychology,and a rejuvenated artificial intelligence movement that has largely abandonedthe model of the mind as a formal machine – have seen major developmentover the past quarter-century, it is inevitable that the findings thrown up bythis ‘cognitive revolution’ should be examined for their relevance to the un-derstanding of economic behavior This ongoing examination has tended toemphasize those characteristics of human cognitive capabilities that call intoquestion the descriptive adequacy of the rational-choice model, focusing ondepartures from individual rationality that may have economic consequences

at the market level.1Such a move may be the obvious one for an economistconfronted with this interdisciplinary challenge, but it is not the only one Thenew insights into the functioning of the brain can also be deployed in theunderstanding of complex systems in general – and of specific social ar-rangements in particular – and that is the direction taken here By criticallyexamining the systemic similarities and differences between the social ar-rangements of science and market, the aim is to show how a complex systemsapproach, inspired by developments in cognitive psychology but applyingthese at the level of the system rather than of the individual, can provide anew and useful way of understanding social systems

Cognition and Economics

Advances in Austrian Economics, Volume 9, 51–86

Copyright r 2007 by Elsevier Ltd.

All rights of reproduction in any form reserved

ISSN: 1529-2134/doi:10.1016/S1529-2134(06)09003-X

51

First, an important point of nomenclature The term ‘science’ is used here

to refer to the complex of people and institutions that make up the edge-generating activities of a scientific community rather than, as might bemore common, the knowledge itself that is generated within that socialsystem Similarly, ‘market’ refers to the complex of people and institutionsthat make up a community of buyers and sellers in a money economy Notincluded under ‘market’, however, is activity that takes place within firms,and not included under ‘science’ is the activity of academic teaching, both ofwhich would be described by the theory developed here as social arrange-ments with their own distinct institutional frameworks.2

knowl-In the context of those definitions, science is not a market,3but science andmarket are instances in the social domain of a general class of systems whichare characterized here as ‘adaptive classifying systems’ Now, even settingaside for a moment the puzzle as to what exactly is an adaptive classifyingsystem, this is certainly not the standard view in the economics of science Thebasic working hypothesis that the activities of scientists in the production ofscientific knowledge can be understood in market terms, deploying some var-iation of the method of optimization under constraint, has been conventionalwisdom ever since the pioneering forays ofNelson (1959)andArrow (1962).See, for illustration, the recent collection Science Bought and Sold – a rathertelling title – edited byMirowski and Sent (2002) It is true that the crudecharacterization of science as ‘the marketplace of ideas’ is not a feature ofmore modern work, and certainly writers in the ‘new economics of science’such asDasgupta and David (1994)hardly mention markets at all except inthe context of technological development (which they sharply differentiatefrom science) In their concentration on individual incentives played out incontexts of imperfect information, they incorporate observations of the actualincentives encountered by scientists, and bring to bear some sophisticatedeconomic tools in analyzing this new domain.4 But, in characterizing theinteractions as exchanges and invoking a benchmark of efficiency that if it is to

be meaningful at all must assume a complex of goods production and change in an idealized competitive environment, they are effectively charac-terizing the new domain as a type of market without actually using the word.But while it is perhaps inevitable that economists will tend, like the manwith a hammer who sees every problem as a nail,5 to take every socialarrangement they contemplate to be a market of some sort, this is notnecessarily the most helpful approach The obvious downside is that, byforcing our model of science into conformity with that of market, we willhave to downplay the differences, and if these differences include phenom-ena that are important to the operation of science then we impoverish our

ex-ability to understand science as a social system in its own right So the firstorder of business in this paper will be to examine the similarities and differ-ences between the social arrangements of market and science, and to illus-trate that, despite real similarities, the differences are indeed significantenough for one to be very dubious of the wisdom of treating science as aspecies of market – even as an ‘imperfect’ market.

A major thrust of this paper is, however, to develop a theory, not of whatscience is not but of what it is.6To that end, the defining characteristics of

an ‘adaptive classifying system’ are described (a theoretical construct that,interestingly enough, may first have made its appearance in a fairly obscurebook of Hayek’s – The Sensory Order), and it is shown how market andscience can be represented as different implementations of this more generalconcept The question as to what could possibly be the benefit of adoptingsuch an approach – an approach which seems, at first sight, to embody adeparture from methodological individualism uncharacteristic of an econ-omist – is raised and discussed, and applications are described that illustrateits ability to shed light on science and market phenomena that seem not to

be well handled by more standard approaches

2 SIMILARITIES AND DIFFERENCES BETWEEN

SCIENCE AND MARKETThere is no question that one can easily point to similarities between thesocial domains of market and science Both are populated by self-interestedpeople – self-interested, that is, in the sense of forming (and acting based on)subjective appraisals of the costs and benefits of their actions and plans, sothat behavior at the margin is sensitive to incentives In both, the peopleinvolved are constrained by scarcities of resources, by the inability to domany things at once, and by the cognitive limitations of their brains in thecontext of a complex environment.7 There are repeated, institutionalizedinteractions between the participants; interactions which can be quiteindirect and often involve complete strangers but are essential to the indi-vidual pursuit of happiness And, in both, specialization, competition, andentrepreneurial or calculated risk-taking behavior are major forces in theoperation and growth of the social network When one looks at the overallstructure of these two social arrangements, one sees no central locus ofcontrol in either case, and yet there is voluntary participation, driven by thepositive feedback from the subjectively perceived benefits of participation,and general adherence to the rules of interaction In cases in which the rules

are violated, both incorporate negative feedback processes that keep tion at a tolerably low level.8

defec-Yet, despite all these similarities, the differences are stark – and mental.9 Most obviously, in the domain of science, there are no marketprices And, since market prices are an emergent phenomenon of the marketsystem, their absence in science points to deep dissimilarities in the processes

funda-of interaction through which, in the case funda-of markets, market prices areformed The relevant institutions of interaction are in fact different, andthey are different for a good reason: the content of the interactions (thegoods, services, and money in the case of markets and the published articlesand citations in the case of science) have very little in common Publishedscientific articles, and whatever it is about their content that may be citable,are not regarded as property; the publication process necessitates interactionnot with those who may cite the article but with an editor assisted byreferees and therefore to call the publication–citation sequence an ‘ex-change’ is to take great liberties with the meaning of the word; and theacceptance of articles for publication may be based on appraised signifi-cance or interest or the author’s reputation or connections, but not onexpected profitability In short, the major form of interaction in science doesnot involve property, does not involve exchange, and does not involve eco-nomic calculation

Nonetheless, one might be tempted to say that, after all, whether or notscience is a market is really a matter of definition One can define key termssuch as ‘market’, ‘exchange’, ‘price’, ‘payment’, ‘investment’, ‘capital’,

‘property’, ‘product’, ‘efficiency’ – and even ‘economics’ – in a mannersufficiently general to encompass the phenomena observed in both domains.This has been the general device used by a long line of authors, including:

Nelson (1959) and Arrow (1962), who treat science as a duction process rendered suboptimal by the character of knowledge as apublic good;Radnitzky (1986),who applies cost-benefit analysis to the beh-avior of scientific researchers;Diamond (1988), who formalizes a ‘rationalscientist’ as a constrained utility-maximizer whose utility function includesaesthetic attributes of theories; Dasgupta and David (1994), who offer amore sophisticated analysis that recognizes the institutional peculiarities ofscience but who posit information asymmetries and principal–agent issues asinefficiencies in the subsidized production of knowledge (which is taken to

knowledge-pro-be an exchangeable commodity); andWalstad (2001, 2002), who seeks totransfer Austrian insights directly from market to science

It is a common assertion (not only in the economic literature citedabove, but also more generally) that any situation in which there is an

observable acknowledgement of value (including the scientific institution

of publication and citation) can also be considered an exchange.10 Thus,for example, we have the idea of there being two types of market – a

‘traditional market’ in which goods are exchanged and monetary marketprices are formed, and a ‘scientific market’ in which articles are publishedand sometimes cited and no market prices, monetary or barter, areformed.11 Ignoring the details of price formation in markets and the dis-connection between the acts of publishing and citing in science, the gen-eralization functions by identifying articles as products which are offeredfor sale and citations as payments in exchange for the use of those articles,

or, more precisely, for the use of the knowledge or information therein.The similarities already noted between science and market (in particular,the fact that both involve interactions between self-interested participants)make this a plausible move, and it comes with the analytical convenience ofthe ability to transfer wholesale concepts whose meanings and usefulnesswere established in the context of markets But, for all the attractiveness ofthe move, the returns to date have been slight12 and, with the notableexception of providing some impetus to the growing realization that sci-entists are as self-interested as anyone else, the main result seems to havebeen the contention that there are ‘inefficiencies’ in the arrangements ofscience – a conclusion that prompts some economists, including Arrow(1962)andDasgupta and David (1994), to pronounce ‘market failure’ andcall for (further) government subsidy or intervention.13

In any case, it is not good enough to say that the matter is only one ofdefinition, for definitions, and the analogies they cement, are not withoutconsequence Firm definitions are obviously necessary for clear exposition,but, often subtly and unobtrusively, they create a path dependence, illumi-nating some directions of inquiry while foreclosing others The definition ofscience as a type of market, for example, compels one to look for the analog

of marketable goods, and most authors on the topic argue that, despite theobvious problem of quantification, it is found in the knowledge or infor-mation content of the scientific publication But this cannot be so If there is

‘knowledge’ in a published article, it would have to be the author’s individualknowledge, but individual knowledge, being the current classificatory capa-bility of an individual brain, cannot possibly be a thing separate from theindividual involved If there is ‘knowledge codified as information’ asDa-sgupta and David (1994)describe it, with information regarded as a signalsubjectively appraised, one is out of that frying pan but has fallen into anearby fire – the need to account for, as a separate process, the emergence ofthe corpus of current scientific knowledge as a classification distinct from the

knowledge of the individual scientists who are parties to the informationtransfer This codified individual knowledge, the purported analog of amarket good, is not the final product, and it is a gross error to conflate it withscientific knowledge Scientific transactions are not simply a matter of ‘the-ory choice’ – as portrayed by Brock and Durlauf (1999) – in which thechosen bits of individual knowledge add up to the current state of scientificknowledge Scientists use aspects of each other’s work, modifying, adapting,criticizing, reinterpreting, and perhaps (from the point of view of the originalauthor) misinterpreting it as they develop their own work Repeated appli-cations of this process are observed to tend to result in a commonly acceptedconception (at least within particular schools, but sometimes in whole dis-ciplines, and especially in those disciplines where empirical reproducibility isconsidered to be significant),14 and this transformed conception, this tacitagreement as to the classification of phenomena in the subject domain (tem-porary and mutable though it may be) is what we call scientific knowledge Ifone wanted to look for an analog of scientific knowledge in the marketdomain, the spectrum of current market prices for goods and services would

be a reasonable candidate since both are emergent attributes of their spective systems, both have counterparts in the individual transactions of thesystem with which they are often confused,15and both provide information

re-to the participants of the system that is vitally useful for the pursuit of theirindividual ends But the definition of science as a type of market and theconcomitant need to identify the ‘science goods’ that are ‘exchanged’ in thismarket have led economists of science in another direction altogether.Science is not a market, but science and market, as social systems, havemuch in common They are related fraternally rather than filially To clarifythe nature of that relationship, we introduce the parental figure, the adaptiveclassifying system

3 ADAPTIVE CLASSIFYING SYSTEMS

The sort of ‘system’ we are focusing on is a mutable network of interactingcomponents that is sufficiently stable to be an identifiable entity, distin-guishable from its environment The system is open to its environment, sothat it can be affected by (and can itself affect) that environment, and so thedrawing of the system-environment boundary has an element of theoreticalconvenience to it Changes in the environment can threaten the integrity ofthe system and entities in the environment can, once incorporated into thesystem, strengthen its integrity and promote its growth The fact that the

system can and does change in a manner consistent with the maintenance ofits structural cohesion as a result of interaction with the environment is thereason for the adjective ‘adaptive’ The way in which this adaptation iseffected is particularly interesting – the system, in the course of its expe-rience of the environment, gradually modifies its internal structure (realized,

at its most basic level, as connections of varying degrees of permanencebetween its components) in such a way that it builds up, inherent in thestructure, a repertoire of useful classes of response that can be deployed asappropriate depending on the environmental situation In the most generalsense, then, the system is continually engaged in ‘classifying’ the phenomena

in its environment.16 Hence the term ‘adaptive classifying system’.17One might think that, given this list of unlikely sounding criteria, adaptiveclassifying systems, if they existed at all, would be hard to find They are notmechanical systems, since the connections between the components arevariable and impermanent and the components themselves are changed bytheir interactions They are not simple biological systems, like a bacterium,for example, whose adaptive capability depends directly on negative orpositive feedback evaluated against a built-in preference rather than on thecumulative modification of the interactions between its components Theyare not simply ensembles that adapt as a result of their environment’s se-lection pressure in promoting the existence and reproduction of some com-ponents and discouraging others, like a species or an immune system(although their components may, indeed, be subject to such selection).18Nevertheless, they are not rare Systems as diverse as ant or termite colonies– seeHofstadter (1979, pp 310–336),Tullock (1994), andSun (2002)– andthe brains of higher animals do fit the description And so, we claim, dovarious systems of human social interaction, including markets and science.Suggestions that economies and other social arrangements can be under-stood as examples of ‘complex adaptive systems’ have been made manytimes before in complexity theory – see, for example, Kauffman (1993,

pp 395–402)– and sociology – see, for example,Buckley (1998) It is also, ofcourse, implicit in Hayek – see, for example,Hayek (1967, pp 66–81), where

he asserts that ‘there is no reason why a polycentric order in which eachelement is guided only by rules and receives no orders from a center shouldnot be capable of bringing about as complex and apparently as ‘purposive’

an adaptation to circumstances as could be produced [in a more cally organized system]’ Kauffman (1993, pp 173–235; 1995, pp 71–92),discussing ‘the twin sources of order’, makes the important point that, inbiological systems, natural selection is not the only source of order, for thetendency in certain systems to self-organization is, in a sense, ‘order for

hierarchi-free’ By this he means that the formation of that sort of order is taneous’, a designation that will resonate with those familiar with the work

‘spon-of Hayek – see, for example,Hayek (1973)andBoehm (1994) But systemsinvolving some form of self-organization are members of a very large anddiverse set, and so the commonalities are likely to be of such a generalnature as to provide very little assistance in understanding particular socialsystems So, while endorsing Kauffman’s insight, we prefer to considerablynarrow the field so that more concrete things can be said about the structure

of the systems of interest Our purpose here is to push beyond suchprogrammatic statements by injecting more explicitness into the idea.The prototypical description of an adaptive classifying system, and thedirect inspiration for the generalization pursued here, is found inHayek’s(1952)explanation of how our brains are able to create the array of sensoryqualities by which we perceive events.19 In this remarkable and prescientwork,20the brain is characterized as a network of interconnected neuronsthat structurally changes as a result of the patterns of activity (in the form

of electrical impulses transmitted between connected neurons) that are duced in the network by incipient stimuli The central idea is that thismutable structure functions as a ‘map’ of the previously experiencedenvironment in the sense that it instantiates a classification of the stimulithat have impinged on the system from that environment The map is builtfrom experience, and is modified by strengthening of neuronal connectionswhen new experience confirms old and by the formation and detachment ofconnections when new experience produces activation patterns differentfrom those previously experienced The system is, in this very particularway, self-organizing

in-The mutability of the map is crucial to its ability to classify in-The networkpaths followed by impulses from stimuli that tend to occur together tend tobecome connected; conversely, connections rarely invoked tend to decay.This allows for an establishment of similarity and difference between stim-uli, and there is large scope for the building up of subtle gradations ofsimilarity and difference because the stimuli can induce activity in multiplebranching and converging neural paths so that there are very many pos-sibilities for the development (or decay) of connections at places whereconcurrent activations pass sufficiently closely to each other Further, sinceboth subject and object of classification are patterns of impulses, classifi-cations from one area of the network can be further classified in terms of theactivations those (classified) follow-on patterns induce in subsequent neu-ronal groups This resulting classification is, then, multiple in several senses– any particular stimulus can be a member of multiple classes, an assignment

of a particular stimulus to a class may change depending on the presence ofconcurrent stimuli, and classes (being represented in terms of impulses) canthemselves be further classified at subsequent levels Note that the ability toclassify is an emergent property of the system; it is not the property of anyneuron or small group of neurons, or of any particular interaction betweenspecific neurons.

The tendency to form connections between paths activated by rently experienced stimuli promotes the emergence, from a given stimulus,

concur-of an induced pattern concur-of impulses in the network characteristic concur-of thatstimulus and of other potential stimuli which have in fact accompanied it inthe past This pattern of impulses generated in the map by the currentstimuli can be described, therefore, as a ‘model’ of the current environmentbecause it is characteristic not only of the experienced stimuli but also of theusual implications of these stimuli The model is, in other words, anticipa-tory and embodies the system’s expectations of likely subsequent stimuli.And, since connections exist to motor neurons at many levels and theseconnections, like all connections in the map, have been developed as a result

of experience (phylogenetic or ontogenetic), the model can result in theselection of motor activity consistent with those expectations

The basic processes of classification described by Hayek as operating

in the brain, including particularly the formation of a mutable map of thebrain’s environment as experienced in the past and the ability of that map

to support an anticipatory model of current experience, have, we claim,their counterparts in adaptive social systems, implemented differently, ofcourse, but very similar in principle Social systems are brain-like in alimited but important respect – specifically, the interactions between theircomponents implement a classifying process on stimuli impinging onthe system, and this process can induce real changes in component behaviorand interaction that, in turn, engender adaptive reactions of the system as

a whole to changes in its environment In short, they are adaptive sifying systems.21

clas-4 MARKET AND SCIENCE AS ADAPTIVE

CLASSIFYING SYSTEMS

In order to convincingly identify a system of social interaction, be it market

or science, as an adaptive classifying system, it is necessary first to clearlydefine what constitutes the system in contradistinction to its environment,then to identify the structure of the system’s map, then to describe how a

model of the current environment can be generated in that map, then tocharacterize the type of classification that the system is performing of thefeatures of the environment to which it is sensitive, and finally to show howthe activation of that model can change the map in a way that refines thesystem’s classification.

Delineation of the boundaries of social systems is not a trivial matter, andthis is because there are several important differences between physical sys-tems such as brains and the social systems we seek to include with brains inthe category of adaptive classifying systems:

1 The most profound difference is that much of the structure in socialsystems is abstract The global institutions and conventions which con-dition all participants’ interactions are essential elements of the structure,

as are the more local personal habits and routines These institutions andhabits are not physical entities; nonetheless, they can be regarded ashaving causal efficacy.22Their abstract character may make them difficult

to identify and, at the very least, introduces a significant element of ory-dependency to any such identification The maintenance of the in-stitutions may be dependent on other social arrangements – for example,the contract and monetary institutions of the market system are depend-ent on the legal and monetary systems, and effects from these supportingsystems can thereby be transmitted to the market system At least for theexercise of boundary definition, however, these institutions can be taken

the-as given

2 The active components of any social system are people, and people can(and inevitably do) participate in multiple social systems – an academicscientist, for example, in addition to publishing and criticizing, may buygroceries, participate in the educational function of the university, vote in

an election, and defend against an harassment suit, all on the same day.Again, the identification of which actions belong within which system is atheory-dependent one

3 The people in social systems are not only the social analogs of the neurons

of the brain in that the transactions between them are the impulses of thesystem’s model, but they also function as the system’s sensory receptorsand motor effectors This means that there is no built-in localization ofsensory inputs or motor reactions; it also means that, to the extent thatpeople are changed by their experience in one system, the change canstimulate an input to another system For example, an economist whosecurrent scientific activity convinces him that, in the big picture, ‘exportingjobs’ is a healthy development may change his political behavior

4 People are vastly more complex than neurons – their brains, after all, arethemselves adaptive classifying systems of a highly specialized kind,capable of supporting purposeful and innovative behavior and thereforeexhibiting considerable flexibility in learning and adaptation Inputs to asocial system, then, can be generated (as noted above for the specificcontext of learning from experience in other systems) by such adaptation

at the personal level And, while these attributes can promote the mation of a wide range of cooperative interactions, they also introducethe phenomenon of competition which, as economists have long beenaware, is of the utmost importance in accounting for the activity in socialsystems and necessitates the identification, at least in general terms, ofpersonal motives that rationalize what the competition is about

for-5 The classifications produced by social systems are observable to thecomponent participants,23 and can serve as useful information to themwhich, in turn, may cause them to alter their behavior within the sys-tem.24This feedback effect is important in the maintenance and growth

of social systems, for it represents a benefit to participation

6 The terms ‘market’ and ‘science’ as used here do not refer to any ticular market economy or arena of scientific endeavor That level ofspecificity (‘the U.S economy’, for example, or ‘the physics community’)would be necessary in applications of the basic theory but would beunhelpful to the task of setting out the basic concepts The fact thatapplications might be dealing with more than one market or more thanone scientific domain does, however, raise the analytical possibility that,for example, part of the environment of the market of interest is anothermarket – a division into system and environment that could be effective ifthe amount of interaction between the two markets was relatively smallcompared to the activity in the market of interest

par-In the context of these subtleties of boundary delineation and structure,

we can proceed to characterize the particular social systems of interest,markets and science, in terms of map, model, and classification We definethe market to be the complex of people in their roles of buyers and sellersengaging in exchanges mediated by the institutions of property, contract,and money The market’s map is composed of the following elements:

1 The institutional framework of property, contract, and money – the damental and long-lasting institutions without which market activity on alarge scale would be infeasible.25

fun-2 The personal habits and routines that market participants have learned torely on to implement their plans People’s activities tend to follow

generally repetitive patterns of interaction, and although deviations mayoccur and action details may vary, the underlying routines are not, inordinary circumstances, subject to dramatic change.26

3 The market participants themselves, the active components of the system,whose buying and selling transactions involving goods and services con-stitute the impulses that animate the system, whose tastes and preferencescan be changed as a direct result of market experience, and whose ac-cumulations of wealth affect not only their capabilities for interaction buttheir tastes and preferences as well

The market’s model is the ongoing flow of transactions (characterized bytransfers of goods and observable exchange prices) between the marketparticipants These transactions are induced by stimuli from environmentalconditions conditioned by the preferences and creativity of the market par-ticipants themselves They follow transactional paths constrained by thecurrent structure of the map, and they result, indirectly, in a classification ofthe various stimuli currently impinging on the market system, a classifica-tion embodied in the array of market goods and their market prices.27Anindividual may intend to develop and sell a particular good, but no indi-vidual plans the overall configuration of marketable goods and services,related to each other (as an emergent result of market activity) as inputs andoutputs and as complements and substitutes of varying degrees An indi-vidual may deliberately set a particular price, but no individual plans theemergence of the spectrum of market prices that relate different goods andreflect overall appraisals of desirability and scarcity Yet, the market systemcould not survive as a coherent system unless these market goods and mar-ket prices were to some extent an operational reflection of actual resources,scarcities, needs, preferences, and the concomitant constraints imposed byother social systems.28

Market participants can observe not only their own transactions but thosearound them, and they can read reports of transactions others haveobserved (such as quotes of stock prices or pork belly futures) They are alsorecipients of advertising from vendors apprising them of potential transac-tions On the basis of their appraisals of this information, they can modifytheir own transaction repertoire, perhaps, for example, initiating transac-tions for a good that they see a lot of other people buying – transactionswhich, from the observer’s point of view, represent potentially appropriablegains In this way, novel local stimuli can have systemic effects, altering thesystem’s map Most such alterations will be at the level of changes in in-dividual preferences and minor amendments to personal routines, although

even relatively short exposure to stimuli that induce transactions in whichthere are large and obvious (at least to an entrepreneurial observer) unap-propriated gains can result in more substantive change And extended,unresolved exposure to such stimuli can result in changes to the more stableareas of the map – as, for example, during the 1100s, when contractualconventions supporting exchange were expanded to encompass negotiablecredit instruments,29a development with cascade effects, leading in turn tomany changes in commercial activity including, eventually, the emergence offormalized futures markets In any case, the array of marketable goods andservices, the quantities brought to market, and their market prices will ad-just as a result of even rather minor changes in the map and therefore willform a highly detailed and sensitive classification of the environmental in-fluences experienced by the system.

The potential of the market’s model to function in anticipatory mode can

be seen quite clearly in the operation of futures markets Transactions inthe corresponding spot markets, coupled with inputs relevant to trends inusage and circumstances of production (including, of course, the expecta-tions of individual market participants), condition the market prices infutures markets These futures prices represent the expectations of themarket system as to the future state of its environment,30 and these ex-pectations are continually adjusted as new information is processedthrough the system

A similar analysis can be carried through for science We define science to

be the complex of people in their roles of authors and readers of articlespertaining to the phenomena of the natural world (including human soci-eties) engaging in activities mediated by the institutions of scientificpublication and citation Science’s map is a stable but mutable structurebuilt from the following elements:

1 The institutional framework of publication and citation – the mental and long-lasting institutions without which science on a largescale would be infeasible.31

funda-2 The personal habits and routines that scientists have learned to rely on toimplement their plans These generally repetitive patterns of interactioninclude their organization into schools and groups and their patterns inselecting their usual outlets for communication

3 The scientists themselves, the active components of the system, whosetransactions involving publication of articles, use of information in pub-lished articles, and citation of information used, constitute the impulsesthat animate the system, whose tastes and preferences can be changed as

a direct result of their experience as scientific researchers, and whoseaccumulations of reputation affect not only their capabilities for inter-action but their tastes and preferences as well.

Science’s model is the ongoing flow of transactions (characterized

by publication in various forums and citation when invoking the work ofothers) between scientists These transactions are induced by stimuli fromenvironmental conditions (including experiments and observations) condi-tioned by the preferences and creativity of the scientists themselves Theyfollow transactional paths constrained by the current structure of the map,and they result, indirectly, in a classification of the various stimuli currentlyimpinging on the system, a classification embodied in the theoretical andtaxonomic corpus of established scientific knowledge32and in the generallyrecognized reputations of individual scientists.33An individual may intend

to develop and expound on a particular theory, but no individual determineshow, and in what form, the insights of this theory are incorporated into thecurrent body of scientific knowledge An individual may deliberately seekreputation, but no individual is in control of the emergence of the overallassessment of his reputation which reflects appraisals of the value and use-fulness to others of his contributions Yet, science could not survive as acoherent system unless the body of scientific knowledge was to some extent auseful classification of natural phenomena, and its operation wouldcertainly be seriously hampered if the general assessment of reputationpersistently ignored important contributions

Scientists are concerned not only with their own work and that of theircolleagues and competitors but with the work of those in related fields – theyprobably read (and absorb information from) many more articles than theycite (This is not to imply that scientists chronically avoid citation; it is simplythe case that, since citation only occurs in published articles employing thepublished results of others, there is a lot of scope for more subtle influences to

be absorbed from articles read.) They can also be quite attentive to who haswon prizes, which areas of research are ‘cutting edge’, and where the mostlucrative grants are to be had Based on their appraisals of this information,they can modify their own transaction repertoire, perhaps, for example, turn-ing attention to a phenomenon that they see as attractive to investigate due tolack of competition or availability of funding – transactions which, from theparticular scientist’s point of view, represent potentially appropriable gains Inthis way, novel local stimuli (unexpected observations, for example, or newsources of funding) can have systemic effects, altering the system’s map As is

the case for markets, most such alterations will be at the level of changes inindividual preferences and in personal routines, although even relatively shortexposure to stimuli that induce transactions in which there are large andobvious (at least to an entrepreneurial observer) unappropriated gains canresult in more substantive change And extended, unresolved exposure to suchstimuli can result in changes to the more stable areas of the map – as seen, forexample, in the emergence of working paper circulation networks and (morerecently) the trend toward internet posting of articles, both of which are re-sponses to the perceived costs of the lead times experienced in the conventionalpublishing process.34In any case, the current body of scientific knowledge willadjust in response to both major and minor changes in the map, and (perhapsmore slowly and less perceptibly) the complex of reputational assessments willchange also.

Science’s model can also function in anticipatory mode – as can be seen inthe operation of research groups or ‘schools’ and their training of graduateand postdoctoral students in the techniques, presumptions, and core ideas oftheir field The effect of school identification and training is to condition thescientists involved to be sensitive and receptive to certain inputs from theenvironment, to be selective with regard to the appreciation of contributions

of other scientists, and, generally, to view the stimuli they encounter throughthe filter of their school’s presumptions In this way, the organization ofscientific schools and training is an embodiment of systemic expectationsabout the character of the environment, based on previous experience Thisdoes not mean that surprises cannot occur and even lead to alterations in themap in the form of reorganizations of school affiliation or changes in coreideas But, as in the sensory domain where unexpected input can easily becompletely ignored, science can, for long periods, proceed in ignorance ofphenomena that have been detected but which have been filtered out ascontrary to expectations, and it is only with the experience of repeatedstimuli that adaptation finally occurs.35

In summary, both market and science are describable as adaptiveclassifying systems – self-organizing systems whose internal structure takesthe form of a mutable map which supports an anticipatory model of thesystem’s environment, a model that operates in terms of a classification ofevents in that environment and is the means by which the system adapts toits environment But, in the concrete implementation of map, model, andclassification, they are very different The following table,36summarizing theidentification of map, model, and classificatory function in science andmarket and juxtaposing them with the corresponding elements of neural

systems, shows clearly the different implementations of the same functionalelements:

interconnected

neurons

Current pattern oftransmissions withinthe existing neuronalnetwork

The order ofsensoryqualities aspersonalknowledge ofthe environmentMarket Network of people

exchange prices

The order ofmarket goodsand prices as

‘marketknowledge’ ofthe market’senvironment

Science Network of people

The order ofscientificknowledge

Science and market are both instances of adaptive classifying systems, butthe basic framework institutions are not the same, the patterns of relevantpersonal behavior are quite different, the major motivating influences on theparticipants diverge considerably once one gets more specific than simplycharacterizing them as ‘benefits’, and the emergent classifications havenothing in common beyond the fact that they are classifications

5 METHODOLOGICAL, EMPIRICAL, AND

DIAGNOSTIC PAYOFFSCharacterizing science and market as different implementations of adaptiveclassifying systems is one thing; showing how this might be a useful and

fruitful conceptual base from which to increase our understanding of socialsystems generally is quite another But, for the idea to be taken seriously, itobviously needs to be done Although treating potential applications inanything like reasonable detail is beyond the scope of this paper (to saynothing about getting ahead of actual work done), it is certainly possible topoint to specific areas of application that not only seem to be handledunsatisfactorily by extant social science paradigms in both economics andsociology and therefore are in need of analytical attention but also appear toinvolve just the sorts of phenomena that would be very suited to investi-gation and explanation in terms of adaptive classifying systems For com-pactness of discussion, these can be grouped into the categories ofmethodological, empirical, and diagnostic issues.

5.1 Methodological IssuesThe adoption of a systems viewpoint would seem, at first sight, to be animplicit rejection of methodological individualism and an espousal of agroup-oriented holism It may indeed lead one to reject a narrowly reduc-tionistic form of methodological individualism, but it is, most emphatically,not at all incompatible with a species of methodological individualism – onethat is conditioned by the ontological and epistemological challenges thrown

up by the complexity of the systems of interest The order that one finds inthis complexity suggests the usefulness of recognizing a series of ‘levels ofabstraction’ (not just one) at each of which the relevant phenomena aremost fruitfully described in terms of basic concepts appropriate to that level,while at the same time recognizing the causal and structural linkagesbetween these levels

Consider, for example, the brain There is little doubt that the activecomponents out of which the brain is composed are neurons But the rec-ognition that the brain is physically reducible to neurons is only one step inthe work of understanding how the brain works Certainly, the more that isknown about the characteristics of individual neurons the better, but it isalso relevant that these individual neurons operate in a context of anorganized structure built from neurons and other cells – a structure whichthe neurons themselves have had a significant role in creating and modifying

as a side-effect of their interactions And so the characteristics of this ture need to be understood if one is to understand the contextual constraints

struc-on neurstruc-onal activity, an inquiry for which the introductistruc-on of basic cstruc-onceptsappropriate for describing organization and structure is a most useful move.These structural concepts, such as spatial relationships between axon

bundles, are not directly reducible to the behavior of individual neuronsalthough they can, at a separate level of analysis, be understood in terms ofthe activities of neurons operating over time in particular contexts At yetanother level of abstraction, there seem to be large organized groups ofneurons that specialize in particular functions and interact with otherneuronal groups, and these present the need for theoretical constructspitched in terms of these interactions Finally, there is the level of the ex-perience of sensory phenomena in the context of the classification Hayek(1952)calls ‘the sensory order’, where the deployment of basic concepts such

as thoughts and emotions is appropriate and necessary.37 Now, all of thiscould have been phrased in terms of markets or science instead of brains(with the appropriate terminological substitutions), given the position thatmarkets and science are examples in the social domain of adaptive classi-fying systems The recognition of causal efficacy in the intermediate levels ofstructure in these systems with respect to both higher and lower levels sug-gests the more nuanced form of methodological individualism in whichcurrent individual behavior can only be understood in the context of theemergent results of past individual behavior that has affected both theinstitutional context and the physical environment of current behavior.38

A second methodological payoff, and perhaps an even more far-reachingone, is that the systems approach focuses attention on the puzzle of howadaptive systems can arise and flourish For example, rather than simplyassuming that rational self-interest constrained by market rules will tend toensure the production of societally beneficial outcomes, one can ask justhow such rules could come to coalesce in the face of the real possibility ofopportunistic defection and innovative avoidance of whatever constraintstheir less effective progenitors imposed An answer, given long ago by

Mandeville (1724)39but immediately deprived of the emphasis on ‘vice’ byhis Scottish successors, is that market and legal institutions emerged as aresult of (as opposed to ‘in spite of’) these behavioral characteristics – amajor impetus for change and development was precisely the personal need

to protect oneself from and to compete with such behavior.40 And themethodological lesson is that one might do well to look for the same phe-nomenon in other social systems Further, if one is enamored of designing(on a small scale, hopefully) institutional arrangements, one should take theMandeville Criterion into account and understand that effective institutionscannot be created and established on a once-and-for-all basis, but must bemutable in the face of defection in a positive, adaptive way, not simplystrengthening restrictive constraints but allowing for the incorporation ofinnovations inspired by competitive reactions to defection so that, over