2006 Science as a Persuasion Game-1

8/22/2019 2006 Science as a Persuasion Game-1

1/14

E P I S T E ME 2 0 0 6 189

J E S S ZA M O R A B O N I L L A

SCIENCEASA PERSUASION GAME.AN INFERENTIALIST APPROACH*

ABSTRACT

Scientific research is reconstructed as a language game along the lines of Robert Brandomsinferentialism. Researchers are assumed to aim at persuading their colleagues of the validityof some claims. The assertions each scientist is allowed or committed to make depend on herprevious claims and on the inferential norms of her research community. A classification of the most

relevant types of inferential rules governing such a game is of fered, and some ways in which thisinferentialist approach can be used for assessing scientific knowledge and practices are explored.Some similarities and differences with a game-theoretic analysis are discussed.

1. Introduction

Unlike Trappist monks, who live under a vow ofsilence, scientists experience a furious frenzy ofverbal communication. The volume of scientificliterature, including books, journals, conferenceproceedings, pre-prints, abstracts, and laboratoryreports, has grown exponentially since the timesof Copernicus. This accumulation may seem anti-economical, especially because much of what ispublished will never be read by anybody (besidesthe referees and journal editors) or will be quicklydispatched with just a superficial glance.1 But I willnot try to analyse here whether scientific writingis economically efficient or not. (An argument willbe offered, nevertheless, that addresses the roleof unread papers in scientific communication.)My main goal is to outline an approach towardsunderstanding scientific research that takesinto account that communication is one of itsessential elements. On this approach, scientistsare seen as engaged in a type of gamewhosemain purpose consists in the persuasion of theircolleagues. Some attempts have been madeto analyse the behaviour of researchers from agame-theoretic point of view (e.g., Blais (1987),Bicchieri (1988), Goldman and Shaked (1991),Kitcher (1993), Luetge (2004), Zamora Bonilla(forthcoming a)).2 But here I will explore the ideathat scientific research is a language game asoriginally defined in the philosophy of languageand pragmatics since Wittgenstein (1953). My

preferred approach to the theory of languagegames is Robert Brandoms inferentialism(Brandom 1994, 2000), which integrates manydevelopments of previous theories. The structureof this paper is as follows. In section 2, I describethe basic elements of an inferentialist theory oflanguage, the central one being the notion of aninferential rule, and offer a brief comparison with

game theory. Section 3 sketches an inferentialistreconstruction of research as a persuasion game.Section 4 presents a classification of the mostgeneral types of inferential rules that constitutethe game of science. Section 5 discusses theorigin and evolution of scientific norms. Section6 explains how an inferentialist approach can beput to the service of a normative assessment ofscientific knowledge.

2. Language games and human action: aninferentialist approach

The notion of a language game has attractedthe attention of scholars from a wide variety ofdisciplines, but there is hardly a systematic andunified account of this notion. This is in contrastto mathematical game theory, where a wideagreement exists about the basic elements of agame.3 My choice of Brandoms inferentialismas a point of departure therefore seems to requireat least some justification, though a full defenceof it would demand more than a single paper.Brandom describes his theory as an inferential

Epist2_3_05_Zamora.indd 189Epist2_3_05_Zamora.indd 189 14/7/06 08:42:1514/7/06 08:42:15


2/14

190 EP IS TE ME 2006

Jess Zamora Boni lla

semantics grounded on a normative pragmatics.

The inferential semantics part means this: Insteadof understanding the meanings of linguisticexpressions as based on representations ofaspects of the world by pieces of speech or text,those meanings are primarily explained throughthe inferential links that each expression has withothers. The normative pragmatics part refers tothe idea that human actions (not only verbal ones)are understood as structured and motivated by thedeontic statuses of the agents. The semantic partbeing grounded on the pragmatic one meansthat it is not the semantic properties of languagethat determine what people can do with it, but

rather the other way around. The most importantaspects of Brandoms theory employed here arethe following:

a) To each participant in a conversation we canattribute a deontic score, i.e., an indication of thethings that, at every moment, she is committed orentitled to do (or not to do). Many, though by nomeans all, of these entitlements and commitmentsrefer to what one can, or has to, assert.

b) The agents normative scores evolveaccording to the inferential rules that are acceptedwithin the community of speakers. These rulesspecify how each score is modified by the actions

and assertions of each agent, as well as by otherevents.

c) Interaction between the agents basicallytakes the form of scorekeeping,4 i.e., not only byevery speaker undertaking specific commitmentsor entitlements, but also by her attributing themto others.

We can attempt to summarise these threeelements in a single formula. Consider a givenspeech community of n members. Let D be theset of all possible deontic scores this communityslanguage can express, and assign a number toeach element of D. Let d be an n-dimensionalvector, assigning to each individual one of thepossible scores. Let Ebe a collection of possibleevents, including the possible actions, A, eachindividual can perform (including assertions);E can also contain the null event, i.e., oneasserting that nothing new happens. Then wecan simply represent the inferential rules of aspeech community as a function of the form DnE D

n. Inferential rules are thus the kinematiclaws of deontic scores.5 Lastly, we would need

something like a dynamic law, expressing the

probability that a particular action is performedby an individual, given her and probably othersdeontic scores. (Formally, this law would be aprobability distribution over A conditioned by thevectors in Dn.)

This general framework suggests modellingsocial phenomena as a type of Markovianprocess. The main difference from a conventionalgame-theoretic analysis is that, in game-theoreticmodels, rational strategic choices are the forcebehind the decisions of individuals, whereasinferentialist models assume that people tend todo what they have to do. I will comment in the

next paragraph on why much of this differenceis more apparent than real, but I want toconcentrate now on an aspect of social actionthat inferentialism can illuminate better than game-theoretic or rational-choice models. It is the factthat rational action is basically action derivedfrom a process of reasoning. Rational-choicemodels simply assume away the real process ofdeliberation that agents carry out within their ownminds or within public conversation, and replacea description of this process by the assumption thatits outputcorresponds to the one that maximisesa utility function given beforehand. This has

proved to be a very powerful heuristic, but it hasalso encountered serious limitations, especiallywhen applied to situations where normativeconcerns are paramount.6 Inferentialist models,in contrast, assume that actions derive basicallyfrom commitments, and can thereby analysethe way these commitments are constructedthrough a deliberation process. This also allows anormative analysis of the process itself, i.e., of thearguments by which the decision is reached. Thisis particularly important in the case of science,where argumentation is essential.

The contrast between inferentialist and game-theoretic approaches is not absolute, however.In the first place, individual behaviour is oftennot fully determined by commitments. There aremany reasons for this indeterminacy: agents canbe entitled to choose between several options;commitment usually comes in degrees; someonecan be committed to perform two incompatibleactions; and agents can decide to break theircommitments sometimes. In all these cases,there is some room for strategic choice. In



3/14

E P I S T E ME 2 0 0 6

SCIENCEASA PERSUASION GAME

191

principle, it is possible to analyse these choices

game-theoretically by introducing the weightof normative commitment as an element of theplayers pay-offs.7

In the second place, we can simply interpretan agents patterns of reasoning and behaviour(i.e., the inferential norms followed by her) as astrategy in her game against the other membersof her community. In this case, it is reasonable toassume that patterns that often lead to non-Nash-equilibrium situations will tend to be substituted byother patterns. This also justifies the applicabilityof game-theoretic tools, particularly those ofevolutionary game theory, where agents dont

choose their strategy, but are rather assumedto be programmed by it. A different way ofexpressing the condition that the norms used in ascientific community must regularly lead to Nash-equilibrium social states is to say that these normsmust be incentive-compatible.8

Moreover, we should have an overtlyinstrumentalposition towards the game-theoreticand the inferentialist paradigms. The mostimportant point should not be whether they areright, true to the facts, or not (a difficult question,to say the least), but to what extent they generateilluminating models of social facts. Some recent

developments in the philosophy of science pointtowards the virtue of such a tolerant attitude (e.g.,Cartwright 1999, Nickles 2002).

3. Playing the game of science.

Like most language games, scientific communicationis systematically oriented towards persuasion,but, unlike other common language games,scientific communication is one in which theplayers goal is not primarily to convince othersto do something, but to make them accept somepropositions. This does not mean that practicalactions are irrelevant in the game of science. Onthe contrary, they are an essential part of it, butinferentialism suggests that we should consider therole of scientists material practices in their essentialconnection to the inferential rules that govern theircommunication practices. On this view, eachscientists main goal (but in no way her only one)is recognition, i.e., she wants her colleaguesto explicitly accept the claims advanced by her claims that may have the character of theories,

laws, models, facts, theorems, and so on. Of

course, most scientists dont pursue bare fame,but aspire to get recognition for having madean important discovery. I suggest interpretingthe phrase in inverted commas as the thesis thatit matters to scientists that the inferential rulesaccording to which they gain recognition areappropriate ones. (I shall return to this issuein section 6.) Perhaps some researchers do notcare about recognition at all, but I will assumethat, as a matter of fact, most do, and, moreimportantly, that scientific practice is basicallyorganised around the pursuit of this goal. The mostimportant problems for any recognition-seeking

researcher, and the reason why game-theoreticconsiderations are important, are, first, that thescientists colleagues are all simultaneouslytryingto do the same, and second, that the choice ofaccepting or rejecting ones presumed discoveriesis always made by ones colleagues.

To the rescue of our unfortunate researchercomes the fact that communication takes placeaccording to a system of inferential rules that,amongst other things, determine what claimssomebody is committedto make given what otherpropositions have been accepted by him before,or whether he has the obligation of accepting

them by default (e.g., any chemist is obligedto accept the claims contained in the standardperiodic table). So, what a researcher may tryto do is to find some claims that, according tothese inferential rules, and taking into accountthe commitments she and her colleagues haveundertaken before, a) she is entitled to make,and b)her colleagues are committed to accept.Scientific communication can be understood,then, as the set of moves each researcher makesin this game of looking for successful claims.9In the next sections, I will describe how theinferential rules of a scientific community can makeepistemically interesting a game like this. In theremainder of this section, I will concentrate on adifferent problem, directly related to the questionwith which I began, i.e., why do researcherswrite so much?.

In general, scientific norms do not allowanyone to accept two contradictory claims aboutthe same issue (or more than one solution to thesame problem). This entails that, for every claimaccepted by some researcher, there will be many



4/14



other claims that she does not accept, and the

competitive process of science entails that most ofthe claims that have been proposed will be nearlyuniversally rejected or ignored. If competition isvery strong, and if the norms are so demandingthat acceptable solutions to interesting problemsare very difficult to discover, it may easily happenthat many researchers are statistically condemnedto never making any important discovery (see note1). Hence, if scientific research has some appealfor recognition-seeking individuals, at least someother recognition-related goal attainable by themmust exist, different from the public acceptance ofthe validityof their discoveries. My suggestion is

that this secondary form of recognition consists inthe public acceptance of some pieces of work,not as right, but as good, i.e., not as a claimit is compulsory to accept, but as an argumentthat is coherent with the research communitysinferential practices.

In real science, this secondary form of scientificmerit usually takes the form of having your paperspublishedin good journals, whereas the primaryform of recognition consists in the content of thosepapers being cited and usedby your colleaguesin further papers. Usually, the better your papersare, the more probable it is that the claims you

have made in them become accepted, but thisis certainly not a necessary implication. First, adifferent solution to the problem you have tackled,still better than the one proposed by you, maybe invented. (This will not necessarily reduceyour works scientific quality, but will make fewerpeople accept it.) Second, you may happen todefend, based on reasons that are not particularlygood, a very good idea, which later becomeswidely accepted. (Perhaps you are so lucky thatbetter arguments in defence of your hypothesisare provided later by other people.) I haveintroduced the term external score to refer to thefirst, fundamental type of recognition, and internalscoreto refer to the merit derived from the coherenceof a scientists inferential practices with the rules ofher community (Zamora Bonilla, forthcoming a).A scientists global scoreis some aggregation ofboth. The particular form of aggregation dependson her own subjective valuation of both types ofrecognition, though it is reasonable to assume thatthe weight attached to the external score is muchhigher than that of the internal score.10 The game

of science proceeds, then, by each researcher

permanently trying to maximise her global score(which will often lead her to favouring strategiesenhancing the internal score), and this will have asan unintended consequence the determination ofher colleagues external scores, as the researcherhappens to be committed to accept or to reject theclaims proposed by them.

As we have just seen, the fact that a researchersglobal score is constituted by two differentelements entails that, in some cases, there willbe more than one possible strategy to increaseit. For example, one can either try to elaboratebold hypotheses, or decide to work within a more

conservative project. This possibility suggests aninteresting case for the application of game theory,as the reward a scientist can expect to gain fromher decision will depend on the choices of hercolleagues: if many of them choose an internalscore optimisation strategy, i.e., if they are tightlybound to accept any conclusion that you provefrom their previous assumptions, attaining a highexternal score will be easier for you, and henceit can be more profitable to choose an externalscore optimisation strategy. Figure 1 representshow a social equilibrium is determined when thiskind of interdependence occurs. The scale of the

graph represents the average ratio between theeffort devoted by each researcher to externalversus internal score enhancing activities. Thefunction fmaps each possible value of that ratiointo the average ratio that would emerge afterevery researcher decides what strategy is best forher given the average behaviour. The equilibriumr*is the fixed point, i.e., the point at which thereaction function crosses the identity line, i.e., whenthe communitys aggregate response to a givensocial state is just this state.11 Only at this pointare all the chosen strategies mutually compatible(which does not mean that all researchers willbe choosing the same strategy). Of course,the actual equilibrium cannot be determineda priori, as it will depend on the particularitiesof each scientific field and context. It would beinteresting to investigate empirically whether the ffunction differs between different research areas,countries, or times, and, if it does, what factorsthese differences depend on.



5/14

E P I S T E ME 2 0 0 6


193

4. A general classification of the inferentialrules of science

Seeing the inferential norms of any languagegame as kinematic laws of the deontic scoressuggests a straightforward way of classifyingthese norms:

(1) Some norms (argumentation rules) simplychange an existing score into a different one(e.g., they may add or remove a commitment),independently of whether something else happens.

This can be represented by placing the null event inthe left part of the formal expression of these rules,12

and the obligation or permission to make somespecific claim as the output of the rules. Logicalnorms are a prime example of argumentationrules, but context-dependent patterns of materialinference also belong into this category.

(2) Other norms (entry rules) determine howthe normative scores are affected by some eventsdifferent from the mere existence of previouscommitments and entitlements. An example isgiven by the rules which determine what claimsone is entitled or committed to make depending

on ones own perceptual experiences and thoseof others. These norms express how the worldaffects the things we are entitled or committedto assert or to deny. It is important to take intoaccount that, in general, non-linguistic eventswill only generate some new deontic status onthe basis of some previous state of the deonticscores.

(3) Lastly, some norms (exit rules) determinehow the claims included in the normative score of

one speaker (i.e., the propositions she accepts)

transform into the obligation to perform, or toabstain from performing, some non-verbal actions.These are the rules of practical reasoning, i.e.those determining how deliberation leads todecision making.

In the case of science, some further distinctionswithin each type of norm are useful. First, amongargumentation rules, we can distinguish betweennorms for the evaluation of claims and normsfor the choiceof claims. The former express thepossible virtues and fl aws of scientific claims(theories, hypotheses, empirical reports, andso on); these norms allow us to determine the

scale against which the quality of a scientificclaim can be assessed. The latter are those rulesthat command a researcher to accept or to rejectspecific statements according to their quality; thesenorms can also be interpreted as determininghow well justified the acceptance or rejectionof a particular claim is. In a nutshell, normsfor the choice of claims determine how goodthe accepted intellectual products of scientificresearch are, whereas norms for the evaluation ofclaims determine in what sense are they good(see Zamora Bonilla 2002).

Second, regarding entry norms we can

distinguish rules for evidence gathering fromrules about authority. The former determine whatclaims a researcher is entitled or committed to doafter the occurrence of some publicly observableevent (it is important to notice that the inferentialrules also determine what can legitimately countas such an event). The latter express what claims ascientist is entitled or committed to adopt becausesome otherscientists have adopted them before.Hence, rules for evidence gathering have the roleof determining the appropriate empirical methods,and rules about authority serve to identify who arethe experts about what. Both types of rules areessential for an efficient economy of trustwithinscientific communities.13

Lastly, exit norms can be interpreted as rulesfor resource allocation, and so they can beclassified according to the different types ofresources employed in scientific research: spacefor publication in journals, distribution of researcheffort within teams, distribution of money amongresearch projects, grants, prizes, jobs, and so on.It is important to notice that many actions (e.g.,

FIGURE 1



6/14



experiments) produce new events, which, in turn,

change the deontic scores.Figure 2 summarises how the rules of a

scientific community work. Starting from agiven (exogenous) state of the researchersdeontic scores (Dx), the occurrence of someevents (E) triggers its transformation into a new(endogenous) state (Dn). The latter commandsthe performance of some actions (A), whichgenerate new events. These, in turn, changeagain the endogenous state, which can also betransformed by the application of transformationrules, and so on. The presence of Dx reminds usthat no research starts in a normative vacuum, and

thus, new events can only attain some meaning(i.e., normative inferential strength) through theresearch communitys previous commitments. (Butthis does not entail that the old commitments willnecessarily be preserved by the operation of theresearch process.) In the end, all this will go onuntil a corpus of knowledge is produced, whichI shall identify here just with the commitments thatbecome generalised in the deontic scores of themembers of the research community, particularlywhen an equilibrium state is reached, i.e., whenthe inferential rules do not lead us to expect newactions or events to change the deontic scores.

Note that, according to inferentialism, it is notnecessary that an agent explicitly asserts a claimfor being really committed to it; it is not even

necessary that she believes in its truth. For her to

have a particular commitment, it is enough that itfollows according to the inferential rules acceptedin the relevant community and from her othercommitments. So, it is not necessary that everymember of the scientific community overtly acceptsa claim for this claim to constitute knowledge inthe sense expressed above.14

5. The origin and evolution of scientific norms

The next important question is why scientific rulesare the ones they are. I may admit that someinferential norms can be universal, but I am

pretty sure that most of them are contingent andcontext dependent. Different communities mayhave different practices, and these practices canchange as well, because most rules depend ona communitys knowledge, beliefs, instruments,or even institutions, and all these factors changepermanently. In the next section I will show whythe recognition of this variability is not an invitationto relativism nor to methodological anarchism.Now I will discuss the mechanisms determiningthat certain specific rules are actually in force ina scientific community. Curiously, although RobertBrandom has attributed a central role to the

communityas the holder of a languages inferentialrules, he has not discussed why diverse linguisticcommunities have such different inferential rules,i.e., why they have different concepts. Brandomsstrategy seems to be, instead, to scrutinise whichinferential rules constitute a possibility conditionfor any conceivable language game. Somethingsimilar could also be done in the case ofscience, and later in this section I will make somesuggestions in this direction. But it is interesting torefl ect first about what generalfactors determinethe adoption of some specific rules rather thanothers.

From an individual scientists point of view, itis true that inferential norms are given (as in mostlanguage games). Actually, by constituting themeanings of her claims and actions, the existingnorms will determine even what a single researcheris able to think and understand, and so it is unlikelythat she even considers the possibility of changingthose norms. But it is also a fact that rules dochange. How is this possible? The recognitionof this apparent paradox is what led ThomasFIGURE 2



7/14

E P I S T E ME 2 0 0 6


195

Kuhn (1962) to suggest that paradigm shifts were

more like sudden religious conversions thanlike the outcomes of rational arguments. A morerationalistic account can be offered, however,within an inferentialist perspective. The followingfacts are relevant for that account:

(a) Not all inferential rules are equally strictin determining understanding and action. Somerules are taken by the agents themselves asconventional, and hence more easily subject tovariation. Even in mathematics, we can drawa continuum of inferential strength from ruleslike those contained in a multiplication table tothose relating to appropriate heuristic methods,

or those of statistical inference. From Brandomsinferentialism, the starting point of the analysis isnot the inherent logical force of the rules, butthe way in which the community negotiates thenormative significanceof behaviour more or lessincoherent with the accepted norms. We can thenrestate this first point by saying that, within eachscientific community, not every violation of thenorms will be considered equally reprehensible.

(b) The existing norms do not necessarilyconstitute an exhaustive, nor even a consistentsystem. In many situations the norms can fail todetermine the next state of the deontic scores, or

can lead to contradictory commitments. Whenthis happens, it is reasonable that agents reactby considering the possibility of revising somerules. Again in Kuhnian terms, a way to solve aparadigm crisis is to eliminate those conceptswhose use led to anomalies.

(c) Actual inferential norms are usually not likealgorithmic programmes (although they seem likeones when represented in a mathematical model).Each individual agent has to construct, from herown experience, a subjective, usually tacit,interpretation of what the right rules are, andit may happen that the subjective interpretationsof different individuals are not entirely coherent.An agent can also realise that she was notunderstanding properly some rule. In caseslike these, negotiations to determine the rightinterpretation of the norms will arise. Again,this is illustrated by cases of Kuhnian scientificrevolutions, though this does not entail that theresolution of the crisis can not be the result ofreasonedarguments. (In some cases, the workingof conversation in figure 2 can lead to the adoption

of a new rule or the rejection of an old one.)

(d) As in many other social contexts, in sciencethe main goal of the training process is to endowyoung scientists with an appropriate mastery ofthe inferential norms. However, this process ofreplication of the rules from teachers to students,or from researchers to colleagues, is not necessarilyperfect and can give rise to mutations.

(e) Lastly, some norms can be more beneficialthan others for the members of the community, orfor some of them. This is obviously true for exitrules, i.e., for those that determine the allocationof resources and the distribution of benefits, butit may also happen with other types of norms.

For example, it makes an enormous differencefor recognition-seeking researchers what rulesfor theory choice are accepted within theircommunity, since each researchers probability ofdiscovering an acceptable theory will dependon what specific rule is actually established (cf.Zamora Bonilla, 2002).

These properties of the inferential normssuggest that there can be two generalmechanisms for their origination or transformation:conventional agreementand evolution. Scientistscan simply meet (physically or not) and decidethat some things must be interpreted or done in

such and such a way, or, more usually, they candelegate such a decision to a small number ofexperts. Consensus congresses, constitutions ofscientific societies, and international standards ormeasurement agencies, are examples of this typeof conventional setting of the inferential norms. Auseful approach to analyse the collective choice ofscientific norms is Constitutional Political Economy(see, e.g., Brennan and Buchanan 1985), whichcan be described as the application of gametheory to decisions about what game to play,i.e., what constitutional norms to adopt.15 Butevolutionary processes, in which every scientisttries to adapt her inferential practices to those sheobserves in her most successful colleagues, areprobably more frequent than explicit agreements.In this case, several theoretical approaches areavailable for modelling the evolution of norms(see Vromen 1995 and Nelson 2002 for a recentsurvey; Sperber 1996 and Blackmore 2000 offertwo interesting evolutionary approaches to thestudy of cultural items). Nevertheless, as individualdecisions under both mechanisms are driven by



8/14



how beneficial the adoption of a particular rule is

for each agent, given the rules adopted by others,it seems that a reasonable minimal requirementis that the chosen norms constitute a Nashequilibrium. Again, this does not entail that allagents will choose the same norms in equilibrium,only that nobody can make a better choice giventhe choices of the rest.16 Taking this into account,the most important step in an analysis of scientificrules is to identify how each possible norm affectsthe payoffs of the scientists whose behaviourwill be regulated by it, but this connects directlywith the question about the evaluation of scientificnorms, which is the topic of the next section.

6. Assessing the game of persuasion

Our reconstruction of science as a persuasiongame opens several avenues for the normativeevaluation of scientific knowledge and scientificpractice, either in specific case studies, orregarding general features of science. First, thereare two broad categories of items to be evaluated:the global processes of interaction betweenscientists (or between them and other agents), andthe particular pieces of argumentation they areusing. In other words, we can make a normative

assessment of the gamea group of scientists areplaying, or we can assess the particular movesthey are making within the game. Second, wecan adopt the stanceof an internal participantin the game, or of an external observer. Figure 3summarises the main general questions that canbe raised under each one of the four resultingpossibilities. Regarding the Arguments column,by adopting a participants perspective we mightask whether the behaviour of the scientists hasobeyed in an appropriate way the inferential rulesthey themselves have adopted. We would adoptin this case a position similar to that of the refereesin a sport game (hypothetically, at least). On theother hand, by adopting an external observerspoint of view, we could judge whether those rulesseem rational to us or not; for example, would wehave preferred that the game had been playedaccording to different rules? Regarding theGames column, the main difference betweenthe two different points of view consists inwhether we take into account only the interestsof the participants in the game (i.e., those people

having some choice to make in it), or the interestsof other agents as well. In the first case, oncewe reconstruct a scientific episode as a game,an obvious question is whether the choices ofthe participants constitute a Nash equilibrium ornot; in the latter case, we would be committedto assert that at least one of the participants hasbehaved in an irrational way. Another importantquestion is whether the situation is efficientor not,i.e., whether the agents would have got a betterpay-off by (more than one of them) having madesome different choices.17 Lastly, regarding thoseagents that are alienated from the game, in the

sense of having no significant power of choice,we might then investigate under what rules theoutcome of the game would have been moreefficient for them.

Of course, formulating relevant normativequestions of this kind and putting forwardappropriate answers is by no means an easy task,and nothing in the former classification saves thework of gathering an exhaustive knowledge of theparticularities of those cases we might decide toassess. The main point in having a generic schemalike the one I have just depicted, consists in showingthat such an evaluation is a legitimateenterprise,for it follows in an absolutely natural way from theunderstanding of the scientific process as a gameplayed according to some rules. A metaphorcan be useful here. According to the inferentialistapproach, the institutions most similar to scientificresearch are probably sports: both in sports andin science, individuals compete, either alone or inteams, to attain public recognition by performingin the best possible way under the constraints ofsome rules that, by providing the kinematic laws

FIGURE 3

What isevaluated?

Point of view

Arguments Games

Internalparticipant

Are the rulesapplied in anappropriateway?

Is the situationan equilibrium?Is it efficient?

Externalobserver

Do you findthe rules andthe argumentspersuasive?

How does thesituation affectother people?



9/14

E P I S T E ME 2 0 0 6


197

of scorekeeping, define what success consists in.

This competitive and regulated nature of sportsentails that the most important thing we can dowith them (besides playing them, of course) isanything but judging them.

In this sense, the approach offered in thispaper fits nicely with other approaches like thoseof Goldman (1999) and Koppl (2006), in whichthe primary goal of epistemological thinking isto assess the efficiency of cognitive practices,considered as social mechanisms for the productionof knowledge. The main difference to thoseapproaches would be that inferentialism allowsfor a more liberal and fl exible range of evaluative

stances. First, it does not necessarily assume thatthere is some unique epistemic goal to be attainedby scientific practices, nor that epistemic goalsare the only essential ones. Second, confl ict andcomplementarity is recognised not only within thegoals of a single agent, but also between the goalsof different agents, as it is evident in those casesdemanding a game-theoretic reconstruction. And,third, a distinction is clearly established between,on the one hand, the assessment of a scientificpractice with the standards of its participants and,on the other hand, an evaluation based on ourown preferred goals.

I shall end by discussing why, in spite ofits liberalism, the inferentialist approach is notcommitted to a relativistic interpretation of scientificknowledge. The following are some of the reasonswhy inferentialism can be seen as an objectivistapproach (they are presented according to thefour normative stances in figure 3):

(a) (Internal criticism of specific arguments).In contrast to Feyerabends famous slogan(Feyerabend 1975), in the game of science notanything goes. Inferential rules have real normativecogency for players, and so, when we evaluatewhether a scientist has made an appropriatemove according to her own commitments, ouranswers will be objectively right or wrong. Thisis not essentially different from when we judgewhether a particular action in a football matchhas been a foul or not, though perhaps the sameaction would have been correct in other sports(say, boxing), or in the same sport once the ruleschange (say, offside). It is true that successfulresearch sometimes violates some methodologicalnorms, but this can only be so if other collectively

accepted norms specify that this behaviour

constitutes a success.18(b) (Internal criticism of the game). Regarding

inefficient situations, their being so is also anobjective question: given the goals and preferencesof the agents engaged in a particular socialsituation, it may simply happen that a differentsolution would have been preferred by all theplayers. Scientific assessment in this case amountsto engaging in the same kind of arguments andnegotiations that the participants themselveswould address if they wanted to successfullymovetowards a more efficient situation.

(c) (External criticism of the game). Again,

this kind of evaluation consists in objectivelydescribing what specific benefits or disadvantagespeople different from scientific players themselvesexperience as a result of the research process,according to the actual values of the former.

(d) (External criticism of arguments). I haveleft this case for the end, because all the othertypes of criticisms depended on the (conditional)acceptance of the validity of some inferentialnorms, according to which the persuasion gamewas being played. It may seem that a relativiststance is compatible with the three previouspossibilities, for, after all, does relativism not

amount to the claim that everybody is free tohave her own evaluation standards? The kindsof evaluation I have considered until now canapparently be reduced to assertions like this isgood/right from the point of view of X, and bad/wrong from the point of view of Y, or so it seems.But, actually, most of my examples, particularlyin a and b, were of a subtly different kind, likeX is acting in a way contrary to the norms sheiscommitted to, or in a way contrary to the fulfilmentof her own interests. Furthermore, we can alsoevaluate the specific norms according to whichthe game of science is played each time. Ofcourse, this evaluation has to be based on theacceptance of some norms of a higher level, soto say, and these can be different from the onesthe players actually accept. This might seem aclear example of Kuhnian incommensurability, butI will try to show that it is not.

In the first place, a claim like according to thegoals and beliefs of group A, the set of inferentialnorms N is efficient, but according to the goalsand beliefs of a different group B, other norms,



10/14



M, are more efficient can be absolutely objective,

even if N and M are mutually contradictory. Thisis simply like comparing the rules of footballand those of gymnastics: both can be rationallydefended, and, what is more important, it isperfectly possible to conceive a rational debateabout which of the two sports to play in a specificmeeting (assuming there is only time, people, orspace for one).

In the second place, since argumentationabout rules necessarily makes use of higherlevel norms, it is likely that the number and extentof the possible criteria according to which thisargumentation takes place is severely limited,

i.e., it is likely that there are only a few possiblegeneral scientific values that serve as a finalguide in the construction of scientific arguments,as Kuhn (1977) himself recognised. The confl ictbetween different schools would again reduce,hence, either to their attaching a different weightto each of these values, or to the fact that thespecific circumstances of each research processmake it easier to satisfy some values than others.

In the third place, it is very likely thatlongstanding competition between researchersduring the last centuries has led science, as aglobal institution, to develop a set of general

criteria for the assessment and choice ofscientific claims that are extremely efficient in theproduction of theories, models and data of the

highest quality according to allthe values referred

to above, so that it can be practically impossibleto conceive some general criteria that are stillbetter. Those norms that are more specific andcontext dependent will surely be more sensitiveto the appearance of anomalies within concreteresearch fields.19

But I think that, over and above the recognitionthat different norms may have different degreesof efficiency in different circumstances (and thatthis is an objective question), inferentialism helpsto defeat relativistic approaches by insistingthat scientific claims are basically commitmentssupported by reasons. After all, evaluating a

scientific argument from an external perspectiveamounts to putting the following question: whatwould have been a more correct way of doing it,according to you? So, it is impossible to adopt aninferentialist stance without being committed to thethesis that there are some ways of doing researchthat are more appropriate than others. Except byhelping to identify some common argumentationcriteria, inferentialism does not attempt to find outwhat these right ways of doing science are, but ithelps to justify the objectivity of science by makingus recognise that the epistemic claims to whichwe are finally committed are not necessarily the

ones we would have wanted to defend in the firstplace, but the ones our reasoned dialogues haveled us to accept in the end.

References

Bicchieri, Christina. 1988. Methodological Rules as Conventions, Philosophy of the SocialSciences, 18: 477-95.

Blackmore, Susan. 2000. The Meme Machine. Oxford: Oxford University Press.Blais, M. J. 1987. Epistemic Tit-for-Tat,Journal of Philosophy, 84: 363-75.Brandom, Robert. 1994. Making It Explicit. Reasoning, Representing, and Discursive Commitment.

Cambridge, MA: Harvard University Press.Brandom, Robert. 2000. Articulating Reasons. An Introduction to Inferentialism. Cambridge, MA:

Harvard University Press.Brennan, Geoffrey, and James Buchanan. 1985. The Reason of Rules. Constitutional PoliticalEconomy, Cambridge: Cambridge University Press.

Cartwright, Nancy. 1999. The Dappled World: A Study of the Boundaries of Science. Cambridge:Cambridge University Press.

Cole, Stephen. 1992. Making Science: Between Nature and Society. Cambridge, MA: HarvardUniversity Press.

Feyerabend, Paul. 1975. Against Method. London: NLB.Goldman, Alvin I. 1999. Knowledge in a Social World. Oxford: Oxford University Press.Goldman, Alvin I., and Moshe Shaked. 1991. An Economic Model of Scientific Activity and Truth

Acquisition, Philosophical Studies, 63: 31-55.



11/14

E P I S T E ME 2 0 0 6


199

Jarvie, Ian. 2001. The Constitution of Science, Amsterdam and Atlanta: Rodopi.

Kitcher, Philip. 1993. The Advancement of Science. Oxford: Oxford University Press.Koppl, Roger. 2006. Epistemic Systems. Episteme2 (2): 91-106.Koppl, Roger, and Richard N. Langlois. 2001. Organizations and Language Games,Journal of

Management and Governance, 5: 287-305.Kuhn, Thomas S. 1962. The Structure of Scientific Revolutions. Chicago: The University of Chicago

Press.Kuhn, Thomas S. 1977. Objectivity, Value Judgments, and Theory Choice, in The Essential

Tension. Chicago: The University of Chicago Press.Lewis, David. 1969. Convention. A Philosophical Study. Cambridge, MA: Harvard University

Press.Lewis, David. 1979. Scorekeeping in a Language Game,Journal of Philosophical Logic, 8: 339-

359.Luetge, Christoph. 2004. Economics in Philosophy of Science: A Dismal Contribution?, Synthese,

140: 279-305.Mki, Uskali. 2004. Economic Epistemology: Hopes and Horrors, Episteme1 (3): 211-222.Nelson, Richard. 2002. Evolutionary Theorizing in Economics,Journal of Economic Perspectives,

2: 23-46.Nickles, Thomas. 2002. From Copernicus to Ptolemy: Inconsistency and Method, in Joke Meheus

(ed.), Inconsistency in Science. Dordrecht: Kluwer, 1-33.Parikh, Prashant. 2001. The Use of Language. Stanford: CSLI Publications.Rabin, Matthew. 1993. Incorporating Fairness into Game Theory and Economics, The American

Economic Review, 83: 1281-1302.Rubinstein, Ariel. 2000. Economics and Language. Cambridge: Cambridge University Press.Sen, Amartya. 1997. Maximization and the Act of Choice, Econometrica, 65: 745-79.Sperber, Daniel. 1996. Explaining Culture. A Naturalistic Approach. Oxford: Basil Blackwell.Sugden, Robert. 2000. The motivating Power of Expectations, in Julian Nida-Rmelin and

Wolfgang Spohn (eds.), Rationality, Rules, and Structure. Dordrecth: Kluwer, 103-130.Vromen, Jack. 1995. Economic Evolution. An Inquiry into the Foundations of the New Institutional

Economics. London: Routledge.Wittgenstein, Ludwig. 1953. Philosophical Investigations, Oxford: Basil Blackwell.Zamora Bonilla, Jess P. 2002. Scientific Inference and the Pursuit of Fame: A Contractarian

Approach, Philosophy of Science, 69: 300-323.Zamora Bonilla, Jess P. (forthcoming a). Science Studies and the Theory of Games, Perspectives

on Science.Zamora Bonilla, Jess P. (forthcoming b). Methodology and the Constitution of Science: A Game

Theoretic Approach, in Max Albert, Dieter Schmidtchen and Stefan Voigt (eds.), ScientificCompetition. Conferences on New Political Economy 24, Tbingen, Mohr Siebeck.

Notes* Some of the ideas in this paper were presented at a seminar on Methodology and the

Constitution of Science, in the Alpbach European Forum, August 2004, and in a seminar onScientific Competition at Saarland University, Saarbrcken, October 2005 (Zamora Bonilla,forthcoming b). I thank the organisers and participants in those events, as well three anonymousreferees of this paper, for their detailed comments. Financial support from Spanish governmentsresearch projects BFF2002-03353 and HUM2005-01686/FISO, as well as from UrrutiaElejalde Foundation, is acknowledged.

1 In a classical study, sociologist Stephen Cole documented that a 70 percent of the newresearchers in physics receive nocitations within the five years after their Ph.D., and only a 10



12/14



percent get five or more citations. Cf. Cole (1992), pp. 218 ff.2 See Mki (2004) for a discussion of other economic approaches to social epistemology.3 Parikh (2001) is perhaps the most systematic attempt of grounding the analysis of language

games on game theory itself. Rubinstein (2000) also uses game theory to illuminate someaspects of linguistic communication. Koppl and Langlois (2001) is one of the few applications oflanguage games to economic theory.

4 This idea was advanced by Lewis (1979).5 In a slightly more sophisticated way, we might represent deontic scores not as vectors, but as n

n matrices, in which the element dijwould represent the deontic score that individual iattributes toindividual j. Similarly, the elements of Ecould be substituted by ordered sets of n members, with eistanding for the event eas interpreted by agent i.

6 Cf., e.g., Sen (1997).7 Cf., e.g., Rabin (1993); but see Sugden (2000), pp. 124 ff., for an argument against the

reducibility of normative motivations to classical game theory8 I owe this appreciation to an anonymous referee.9 There can be defensive moves as well, i.e., one can give up a previous commitment to avoid

accepting a particular claim.10 Just for simplicity, a scientists internal score could be operationalised as a function of her

publishedpapers, whereas her external score would be a function of the favourable citations shereceives.

11 A sufficient condition for the existence of equilibrium is that fis continuous and that, for some valueof x, f(x)equals 0.

12 Recall that inferential rules were functions of the form D E D.13 This can be related to the distinction introduced in Kitcher (1993, ch. 5, 7) between

encounters with nature and conversation with colleagues. The most important differences arethat my classification is centred on the normative organisation of those encounters, and thatit factorises the commitments present in encounters with colleagues into two different types ofentities: argumentative and practical commitments (and so, we could called the former encounterswith language).

14 This can also related to Kitchers notion of a virtual consensus; cf. Kitcher (1993), ch. 3, 11.15 Jarvie (2001) offers a constitutional interpretation of the role of methodological norms in the

philosophy of Popper, particularly in connection with Sir Karls defence of liberalism.16 Bicchieri (1988) discusses the properties an equilibrium in the choice of methodological norms

must have to serve as a convention in the sense of Lewis (1969).17 Ltge (2004) offers some examples of Prisoner Dilemma games in the history of science. On the

other hand, co-ordination problems can be exemplified by cases when different schools employdifferent units of measurement or technical standards.

18 Perhaps we could interpret Feyerabends anything goes, not as telling that all research strategiesare equally functional under any circumstance, but as for every strategy, there are someconceivable circumstances where it is functional. But it is also doubtful that this less stringent claimis actually true, and, even if it were, it would not be very useful, for it says nothing about how toidentify the relevant circumstances.

19 There is another sense of incommensurability, according to which the members of differentscientific communities employ (or attach a very different meaning to) the same (or similar)concepts. I think inferentialism can provide an interesting explanation of this phenomenon, onethat hinders its purported relativistic consequences. The basic idea is that, by reducing meaningsto patterns of inference, each community is able to identify the patterns followed by the other, andcan make a comparison of the efficiency of each set of patterns according to its own goals.



13/14

E P I S T E ME 2 0 0 6


201

Jess P. Zamora Bonilla is professor in the Department of Logic, History, and Philosophy of Sci-ence of the National Open University of Spain (U.N.E.D.), Madrid, and academic co-ordinatorof the Urrutia Elejalde Foundation for Economics and Philosophy. His research has centered onthe economic analysis of scientific knowledge production, as well as the problem of scientificrealism and truth approximation.



14/14


Documents

2006 Science as a Persuasion Game-1