Iv российский философский конгресс

1   ...   17   18   19   20   21   22   23   24   25

Аркадий Блинов (Австралия). Комментарий к докладу проф. Марковой

Arcady Blinov (Australia). Comments on Prof. Markova’s Paper


What I most sympathise with in Prof. Markova’s paper are the following three things:

(A) Her raising the question about the sources of novelty of scientific knowledge:

If we want to have really new knowledge, we cannot simply deduce it from knowledge already existing. In this case we should to admit, that the received knowledge in some kind already contained in previous, then what is new in it? At the same time, if a new knowledge is not generated by an old one where does it appear from? Is it possible to admit that the new knowledge in a science arises from what is not a science, from what is behind its limits?

The question is, I think, both legitimate and interesting. One possible answer to this question, as Prof. Markova indicates in her paper, has been suggested by some of the sociologists of knowledge; the answer is along these lines:

At the end of the last century the powerful tradition of sociological interpretation of scientific knowledge and its social construction was formed […]. The principle of novelty is observed in this case strictly; really, the content of knowledge is determined exclusively by set of all circumstances (not necessarily bearing a relation to science), which were available in the period of obtaining a result in laboratory or in any moment of historical development of natural sciences (research such as case studies). Such situation is unique, it cannot be reproduced in other place and at other time, it cannot be a repetition of what already was.


(B) The fact that Prof. Markova indicates a problem that the above answer forces on us, namely, the problem of the lack of “stability” in science:

Difficulties in interpretation of science, however, do not become less. Main of them - on what basis all these separate events can be considered as belonging to science? Why are we fully confident that newly received results are capable to be included logically in the already existing scientific knowledge? A question arises: may be this infinite variability of knowledge destroys any stability of science in space and time, "dissolves" it in the boundless world outside the science? Such threat exists, and the picture of investigations of science developed to the end of the last century demands, at least, serious specifications, and even essential changes. All questions which interest researchers considerably change. The problem of a logic commensurability of a new theory and an old one does not cause disputes any more, a new theory does not destroy its predecessor, they coexist as independent from one another. Such concepts as the truth of scientific knowledge, its objectivity, evolutionary development and development through revolutions gradually disappear from the works of researchers of natural sciences. Together with these problems interest to the idea of development of scientific knowledge in time is lost also.

I fully agree that the problem is real and grave.


(C) The fact that in search of a satisfactory solution of the above problem, Prof. Markova appeals to a mathematical model, namely, to René Thom’s mathematical theory of catastrophes:

The biology always gave a rich material for the decision of philosophical problems of development, evolution, sharp, mutational changes. As R.Thom's book shows, the modern biology can help with the decision of problems which are put forward now by the philosophy, sociology, and history of science during interpretation of scientific knowledge by them. It is important, certainly, that for a substantiation of the author’s ideas he has chosen mathematical discipline about surfaces and forms, geometrical topology. This circumstance emphasizes an orientation of interest of the author to the consideration of biological problems in space, more likely, than in time. And such change of orientation fully corresponds, in my opinion, with those changes which occur last decades in philosophical interpretations of science.

This appeal to mathematics also finds my full sympathy because being, as I am, a long-standing believer in what is, rather inexactly, called ‘exact philosophy’, I think that mathematical models constitute the ultimate source of satisfactory explanations in any branch of human intellectual endeavour, philosophy included.

But when Prof. Markova begins to explain exactly why Thom’s theory is a good source of support for the stability of science, our ways part. Actually they parted even a little earlier. What I mean is that Prof. Markova seems to proceed on the assumption that (1) when such notions as truth and objectivity of scientific knowledge were active and prominent in philosophy of science, then the job of securing the stability of science was performed by them; (2) now that those notions have been obliterated, mainly as the result of criticisms from the practitioners of sociological analysis of science, the stability of science is endangered; (3) to re-affirm the endangered stability in the absence of the obliterated notions of truth and objectivity, one should seek some further notions that could do the same job – such, e.g., as can be found in the conceptual repertoire of Thom’s theory of catastrophes.

Now, I agree with (1), but I have difficulties to fully agree with (2) – or, to be more precise, with the first part of (2) that suggests that the notions of truth and objectivity of scientific knowledge have been obliterated, mainly as the result of criticisms from the practitioners of sociological analysis of science. It is quite understandable that one gets exactly this kind of impression when one reads through the flow of case-studies publications made under the banner of the Strong Programme (the SP). But I think that actually the overall situation is a little more complicated than that, one salient fact here being that there is a rift between programmatic points on the notions of truth and objectivity made by the originators of the SP, on the one hand, and the way those notions are actually treated by those working sociologists of scientific knowledge who see themselves as champions of the SP. I think the rift deserves some attention.

To begin with the programmatic points, a characteristic collection of such points regarding the notion of truth can be found in David Bloor’s Knowledge and Social Imagery, pp.32-39. Below is my reconstruction of some crucial ones:

(1) The notion of truth should be kept strictly separated from the factor of the actor’s sensory experience of the physical world.

(2) One should, further, distinguish between philosophical analysis of the concept of truth and an analysis of social functions performed by that concept.

(3) On the philosophical side, Bloor’s treatment of truth is uncompromisingly anti-correspondence. Actually, it comes to a peculiar blend of instrumentalism, coherentism and pragmatism.

(4) On the social-functions side, ‘truth’ and ‘falsity’ are the labels assigned to theories and beliefs. One of the central jobs that are done by such labels is to associate authoritative judgments to those theories and beliefs.

(5) Authority is a distinctively social category, and its prevalent social source is power. Hence, any discussion of truth in sociology of knowledge should be presumably held under the rubric of social causation.

(6) Since ‘truth’ and ‘falsity’ are just labels that people assign to theories to elicit suitable value attitudes toward those theories from other people, ‘the strong programme enjoins sociologists to disregard [those labels] in the sense of treating both true and false beliefs alike for the purposes of explanation.’⁷¹ This is Symmetry Principle.

(7) As to the factor of sensory experience of the physical world, which has now been practically isolated from the notion of truth, Bloor has these points to make:

(i) it is only commonsensical to take that factor into account by the practitioners of the SP;

(ii) taking that factor into account is quite compatible with the tenets of the SP;

(iii) in addition to that, because it is one of several causal factors in knowledge (for the sake of brevity, I will call it the objectual factor henceforth), its role is to be understood by seeing how objectual causes interact with other causes, first of all with social ones.⁷²

Such is the programme, as far as the notion of truth is concerned. I would say that there is a potential in it to be implemented in such a way as to do justice to objectual causes and, by the same token, to temporal stability of science. But the way this programme has been actually implemented so far, proved to be quite different from what the programme seemed to promise, and this is what I call the rift between the programme’s potentialities and its actual implementation, the latter being such as to justify Prof. Markova’s concerns as to whether justice is being done by science studies to temporal stability of science. Indeed, more often than not, individual case studies in the sociology of scientific knowledge are carried out in such a manner that objectual causes tend to be almost entirely ignored or at most to be done lip service to.

What is responsible for the rift, then? My answer is, ‘The blame should be laid mainly at the door of (i) a couple of potentially misguiding points in the SP itself, and (ii) the actual lack of a certain discovery the advent of which was, probably, expected by the SP initiators, but which has not materialised yet.’

The two potentially misguiding points are as follows:

(A) Points (4) and (5), from the above list of programmatic points on truth, when taken together, may leave one with an impression that Bloor takes it that truth is a purely social, - indeed, a socially manufactured, - phenomenon. The impression is false; in one place Bloor bluntly says that if something of the sort were true, ‘[i]t would be a disaster for the strong programme.’⁷³ But being false, the impression still lingers and can misguide. I thing it lingers because, to my knowledge, in no one place Bloor (or, indeed, any other initiator of the SP) says in so many words what I think is a most obvious and also a most important truth about truth, namely, that it is a child of (at least) two parents, one objectual and the other, social. Because the objectual parentage of truth was not properly emphasized in the writings of the SP initiators, this lack of emphasis may have influenced the rank-and-file practitioners of the SP, thus being partly responsible for the rift.

(B) Take now the Symmetry Principle, which is point (6) in the above list. With a proper application of Charity Principle, I think one ought to give it an interpretation along these lines: ‘Sociology of scientific knowledge would apply a uniform explanatory format in all cases. The same explanatory format would do the explaining of, say, true and false beliefs.’ And it would, then, be better dubbed ‘Uniform Explanation Principle’, then ‘Symmetry Principle’. But as it is, the talk of ‘the same type of cause’ instead of ‘the same explanatory format’ can be (and I am afraid actually was in many real-life cases) dramatically misleading, namely, leading people to slip into something like this inference: (i) social causes just must be there in (at least some) sociology-of-knowledge explanations, so (ii) social causes should explain at least some beliefs, now (iii) the same type of causes should explain all beliefs, (iv) therefore, it is social causes that should explain all beliefs. Objectual causes are conspicuously absent from the conclusion, as they are, more often than not, absent from case studies of the SP practitioners.

Let me stress that I do not mean that this type of fallacious argument was operative in the minds of the SP initiators, but as I said, the wording of their programmatic points is prone to mislead people to it, and unfortunately, the SP initiators’ later clarifications of their initial pronouncements do not quite dispel the misleading impression, either. It’s a pity actually, because it would have sufficed to do a very small thing to annihilate the misleading effect; and that small thing is to substitute the above mentioned Uniform Explanation Principle for the original Symmetry Principle. Then, the fallacious argument would be killed (or, rather, pre-empted) in this way: (i) social causes must be part of an explanatory format to be used in the sociology of scientific knowledge to explain at least some beliefs; on the other hand, (ii) the same explanatory format should explain all beliefs; (iii) thus, take any belief, and social causes would be part of the explanatory format that explains that belief. Nothing fallacious with the argument this time, and nothing wrong with the conclusion either, because as it is now, it only claims part of the overall space of the explanatory format to be filled by social causes, leaving other parts of it to be filled by some other kinds of causes, presumably, by objectual causes in the first place. The fallacious argument above is so closely tied up with the Symmetry Principle that I am tempted to dub it ‘the Symmetry Principle Fallacy’. (Let me repeat that I am not trying to suggest that the author of the Symmetry Principle was at any time a victim of that fallacy. Actually, I think that he was not.)

(B) To the discovery, then, that has not materialized as yet. Consider again the programmatic point (7, iii) above: Because objectual causes constitute just one of several causal factors in knowledge, their role is to be understood by seeing how they interact with other causes, first of all with social ones. The point is of immense importance. Actually, a generalisation of it would apply to any scientific discipline dealing with various kinds of causes. Moving from science studies to mechanics, I can imagine Galileo making a similar point on the phenomenon of free-fall: ‘Because gravity constitutes just one of several causal factors in free-fall, its role is to be understood by seeing how it interacts with other causes, first of all with air resistance’. And up he came with his discovery of the Parallelogram Law for combining forces.

Now, this is exactly a functional counterpart of Galileo’s Parallelogram Law that, for the time being, is absent from the field of the sociology of scientific knowledge and/or of science studies. It just has not been discovered yet. And this non-discovery is all-important. To clearly understand exactly how important it is, consider the state of infantile helplessness mechanics was in before the advent of the Parallelogram Law; actually, it barely existed as a discipline. I think we should not succumb to the illusion that the sociology of scientific knowledge (or science studies, for that matter) has already transcended its infantile pre-Galilean stage; it cannot do so until its practitioners know how to take account of the combined effect of objectual and social causes (plus any further kind of causes that may occur to be relevant) – until, that is, some sociological analogue of the Parallelogram Law is discovered. Until that happens, the above mentioned rift will persist, because even if the desirability of analysis of interaction of objectual and social causes is preached on the programmatic level, such analysis cannot be successfully practised on the level of case studies in the absence of a more or less rigorous technique for performing it. What is actually happening in the sociology of scientific knowledge on the level of case studies in the absence of an adequate analytical tool for investigating interaction of causes is, when translated back into the terms of mechanics, that people (i) describe in minutest possible details the trajectories of individual cases of free-fall, (ii) make enlightened guesses about possible impact of air resistance on this or that turn in the trajectory, and (iii) from time to time do lip service to the contribution of gravity. They are just not in a position to do any better, being ignorant about the Parallelogram Law for combining forces!

Coming back to Prof. Markova’s proposal concerning catastrophe theory (which is point (3) of (my reconstruction of) her assumption): I think we should not abandon hope to effectively redress objectual causes in their rights, which in turn would do justice to the notion of temporal stability of science. But before that is done, - before, that is, a science-studies analogue of the Parallelogram Law is discovered, - it is too early to decide whether such mathematical models as catastrophe theory can be of any substantial help in this respect. It will actually depend on the nature of that yet undiscovered analytical tool for analysing interaction of causes⁷⁴.

Аркадий Блинов (Австралия). Знание и социальная субоптимальность

Arcady Blinov (Australia). Knowledge and Social Suboptimality


One traditional subject in social epistemology is organization of cognitive labour. A meaningful discussion of this subject seems only to be possible in the presence of some assumptions that are intrinsically normative. It is so because most discussions of organization of cognitive labour focus on such questions as (1) ‘How should a society organize cognitive labour?’, or else (2) ‘What is an optimal organisation of cognitive labour in a given society?’

Both questions seem to assume that what is at issue is a problem of maximisation of a value. What kind of value is to be maximised? Well, it depends on what is taken to be the main epistemic value. Typically, it is knowledge, or true belief, or some closely related notion.

Interestingly, whenever the subject of optimal organization of cognitive labour is discussed, the discussion proceeds on an assumption that the society in question will benefit if the epistemic value is successfully maximised.

For the sake of a label, I will dub this the Monotonicity Assumption (the MA), meaning that it implies that the function that assesses a social utility of knowledge (or true belief, or whatever is deemed to be the main epistemic value) is monotonic: the greater the index of the epistemic value in question, the greater its social utility. (Or, at the very least: The latter will never decrease with an increase of the former.)

For instance, Alvin Goldman in his book Knowledge in a Social World (1999) seems to tacitly accept the Monotonicity Assumption. But tacitly or otherwise, there is still little doubt that he does accept it, because otherwise there seems to have been no point in spending ¾ of the book discussing how to promote truth in various social settings. Presumably, whatever is being promoted is deemed to possess a social utility.

Frederick Schmitt, on the other hand, comes up, in his introduction to Socializing Epistemology (1994), p.18, with an explicit expression of a link between maximising an epistemic value (in this case, knowledge) and promoting a social utility: ‘Human society is possible only if individuals coordinate and institutionalize their inquiry in such a way as to obtain knowledge.’ Whoever assents to this thesis comes very close, indeed, to an acceptance of the Monotonicity Assumption.

Steve Fuller’s interpretation of what counts as optimal in the domain of social organization of cognitive labour is less transparent. For example, the first pages of his book Social Epistemology (1988) make it clear that he relativizes the concept of optimality to certain kinds of knowledge product that are desired by a society. Such a relativization seems to avoid a commitment to the Monotonicity Assumption in its categorical form.

Be it as it may, there is little doubt that the MA is shared, tacitly or explicitly, by many authors in social epistemology.

One remarkable thing, in this connection, is that even with those authors who do share the MA, it is only very rarely, if ever, that one can find a discussion of reasons for or against the acceptance of the MA. In this respect, its status bears a resemblance to that of what R. G. Collingwood calls, in his Essay on Metaphysics, ‘absolute presuppositions’, some characteristic features of these latter beasts are that (i) they are as general as any proposition can be, (ii) they are typically taken for granted, (iii) it is even only very rarely that they find an explicit expression, but (iv) they are, nevertheless, a fundamental feature of the science of a specific epoch and culture.

I am not going to take any definite stance toward Collingwood’s theory of absolute presuppositions. My main contention in this paper is much more modest, namely, I will argue that even if it may be meaningful to assign the status of absolute presuppositions to some general statements, the MA cannot be among them. The reason is that the MA’s scope is not general enough. Actually, there can be constructed a formal proof to the effect that the MA is only valid within a limited area of applications. Beyond the borders of that area, there is no guarantee that, given a social situation, the growth of knowledge has a positive (or even a zero) social utility; it may well prove to be the other way around, that is, the growth of knowledge (or of some other related epistemic value) may rather become a social disutility.

If that contention is correct, it seems to have important implications for the whole tenor of current discussions of optimal organization of cognitive labour in social epistemology. For one thing, a social epistemologist who seeks an answer to the question ‘What is an optimal organisation of cognitive labour in a given society?’ will have to begin his inquiry with a clarification of his meaning of the term ‘optimality’. Does she mean, by ‘optimality,’ a maximization of an epistemic value or else a maximization of general social utility? Given the limitations on the scope of validity of the MA, the clarification becomes vital because, as it will be argued below, in some kinds of social situations that are both important and frequently encountered in real life, there can be clashes between epistemic values and the general social utility. Quite bluntly: Sometimes, it may be disadvantageous for a society to optimize its knowledge production processes.

In the remaining part of the paper, I will

(1) give a general account of how epistemic rationality may clash, in a suboptimal social situation, with rationality all things considered;

(2) construct and discuss a game-theoretical model of such a clash;

(3) discuss the significance and scope of possible applications of the game-theoretical model in question;

(4) trace some implications of the above points for the issue of organization of cognitive labour as discussed in social epistemology.


1. Epistemic vs aggregative rationality

I borrow from Richard Foley his characterisation of the distinction between epistemic and aggregative rationality⁷⁵. The characterisation is this:

All judgments of rationality are judgments about how effectively an individual is pursuing some goal. However, such judgments are commonly elliptical. For one thing, they commonly fail to make explicit what goals are in question.⁷⁶ To avoid confusion one should strive, when pronouncing a judgment of rationality, explicitly to relativise it to a goal. When the goal in question is epistemic, then the judgment is one of epistemic rationality.

What goals are epistemic? Foley takes it that there is only one purely epistemic goal, namely, that of now believing true beliefs and now not believing false beliefs. I think that I am not prepared to agree with the 'only one' part of Foley's claim, but it does not matter for my purposes here. What matters is that we all seem to have more or less clear intuitions about which goals are epistemic and which not. For example, two further goals are unmistakably epistemic, though not necessarily purely so: (ii) acquiring as much knowledge as possible; (iii) developing one’s reasoning ability (or more generally: cognitive abilities at large) as high as possible.

On the other hand, one can have more than one goal simultaneously. Then judgments about how effectively he is pursuing the whole constellation of his several (weighted) goals are judgments of rationality, all things considered, or, to have a regular adjective, aggregative rationality.

It should be clear that the two notions of rationality - epistemic and aggregative - are distinct. More than that, it is prima facie possible that on some occasions the two clash with one another: say, a belief which is rational, a.t.c., for an individual on a specific occasion to maintain may not be epistemically rational for him, on the same occasion, to maintain.


2. The game of the Good Jailer: An epistemic dilemma for the prisoners

It is relatively safe to ignore the distinction between the two rationalities when treating one-agent contexts. Actually, there is a decision-theoretic result, namely, Savage-Good theorem, that guarantees impossibility of a conflict between aggregative rationality and one variety of epistemic rationality under some well-specified conditions: In a situation where a single utility-maximiser is to take a decision, new information can never be harmful for her, given that the information is correct and costless.

Admittedly, even remaining within the domain of one-agent contexts, one can think of a situation like that of Pascal’s Wager where it is rational, all things considered, for the individual to come to maintain a belief which is epistemically irrational for her to maintain. But what makes this possible is the unusual assumption that there exists a being who (i) is endowed with the supernatural ability of having immediate access to the agent’s mind, and (ii) who can reward or punish the agent for having this or that belief.

The picture changes dramatically when we move from one-agent to many-agent [= interactive] situations, that is, to the domain of Game Theory. The fact that the value of knowledge can be negative in an interactive context is well-documented in the literature.⁷⁷

Let me come up with a situation that is quite paradigmatic in this respect, but which, to my knowledge, was never discussed in the literature. The situation is a variation on the famous Prisoners' Dilemma. The Prisoners' Dilemma is this: On suspicion of having jointly committed a crime, two persons, say Ann and Peter, have been detained and put into separate cells so that they are unable to communicate. Common knowledge for both is at least this: If one confesses while the other does not, he who has confessed will be immediately set free for helping the investigator. The other will be put away for ten years. If both confess, both will be put away for nine years. If both keep silence, both will be locked up for a year for a misdemeanour, since there is not enough evidence to support the more serious suspicion.

The names of the two strategies on Table 1 are abbreviations for 'cooperate' and 'defect', respectively. As is well known, the unique Nash equilibrium for this game is that both players should defect. This implies that it is rational for each player to defect, which is also supported by the fact that, for each player, D strictly dominates C. So if they are rational, they will both defect and spend in jail nine years each.

		Peter
		C	D
Ann	C	-1, -1	-10, 0
	D	0, -10	-9, -9

Table 1

Such is the standard Prisoners' Dilemma. My variation is this: Suppose that Ann and Peter are members of a gang which is governed in a democratic fashion. In particular, a couple of days after Ann and Peter's arrest there took place a general meeting of the gang. The only item of the agenda was a proposal to consider collaboration of a jailed member of the gang with the investigator as a capital offence which is to be punished by death. If the proposal has been adopted by the meeting, and this has become common knowledge between the two players, then of course this knowledge should result in a drastic change of their strategic situation. Suppose, for the sake of smooth calculation, that each of the two players assesses the negative utility of their own death as equal to 50 years in jail. Then the new situation is represented by Table 2:

		Peter
		C	D
Ann	C	-1, -1	-10, -50
	D	-50, -10	-59, -59

Table 2


Now both the logic of Nash equilibrium and that of strict dominance recommend that each should cooperate. So if they are rational they will both cooperate (that is, keep silence) and spend in jail one year each.

Unfortunately, the players do not know the poll's result, but being old-standing members of the gang as they are, they know the mentality of their fellow gangsters, so that they share the belief that is represented by subjective probability of .5 that the meeting has adopted the proposal and subjective probability of .5 that the proposal was not adopted. The fact that they share this belief is common knowledge between them. As can be easily calculated, this still leaves them, qua maximisers of expected utility, with recommendation that each should cooperate.

Let us dub the resulting game ‘The PD/CP under the Veil of Ignorance’, where ‘PD’ stand for ‘Prisoners’ Dilemma’ and ‘CD’ for ‘Capital Punishment’. Of course, the PD/CP under the Veil of Ignorance is a typical game with incomplete information in Harsanyi’s sense.

So far so good. But this is not the end of the story. Suppose now that the two prisoners are offered one more option. It happens that one of their jailers, out of sheer sympathy with the two hapless creatures, comes up with a suggestion. He can inquire and report to them about the meeting's result. To handle the issue with perfect equity, though, he will report either to both - if each opts to learn, or to neither - if at least one of the two opts to remain ignorant. The offer and the fact that both detainees completely trust the jailer's information are common knowledge between them.

Call the resulting game ‘The Kind Jailer’. To grasp its formal structure, consider first the game ‘PD/CP without the Veil of Ignorance’ which is exactly like ‘PD/CP with the Veil of Ignorance’ except that the veil of ignorance [= the two-member information set] is removed: whichever option (that is, the PD or the CP) the Nature chooses, the two prisoners will learn the choice. Now, the Kind Jailer is the game in which the two prisoners start with having a choice between playing the PD/CP with, or without, the Veil of Ignorance. Their initial (simultaneous) move is voicing their preferences between the two options. After that move, they proceed to play the PD/CP without the Veil of Ignorance iff both preferred to do so at the initial move. Otherwise, they play the PD/CP with the Veil of Ignorance.

Now, as to the Kind Jailer, the main question is 'What is it rational for each prisoner to do: accept the jailer's offer or refuse it?' The question is elliptical, which cannot be tolerated given the situation at issue. We should make the goal explicit, and this will result in at least two different complete (non-elliptical) questions:

Q1	What is it rational for each prisoner - say, for Ann, - to do relative to the epistemic goal of acquiring as much knowledge as possible?
Q2	What is it rational for each prisoner - say, for Ann, - to do relative to the whole constellation of her goals, that is, what is it rational for her to do, all things considered?

If each values knowledge positively, but sufficiently lower than freedom and/or life, and this is common knowledge between the two, then the answers to the two questions differ, the answer to Q1 being 'Accept the offer', and the answer to Q2, 'Refuse it'. This is so because under ignorance, aggregative rationality recommends each to cooperate, which results in one year in jail for each. On the other hand, given their subjective probabilities, the offer brings the 50-out-of-100 risk that they will come to common knowledge that the meeting failed to introduce capital punishment, and then it will be aggregatively rational for each to defect, which will keep each in jail for nine years. The risk being too high, aggregative rationality recommends each to remain ignorant.⁷⁸ Thus, shared ignorance, and even shared epistemic irrationality, can easily be a boon rather than a bane, all things considered, in an interactive situation.


3. The game of the Good Jailer: Possible generalisations and applications

The paradigmatic situation of the Good Jailer derives its significance from the fact that it seems to be generalisable along no fewer dimensions than the original Prisoners’ Dilemma. Let me cite some crucial dimensions:

(1) It generalises to other epistemic goals, that is, to other varieties of epistemic rationality. For example, there is a result⁷⁹ to the effect that if, in a finitely repeated Prisoners’ Dilemma, there are bounds (possibly very large) to the complexity of the strategies that the players may use, then there is a Nash equilibrium that yields a payoff close to the cooperative one. Now, if we try and put a real-life interpretation on this mathematical result, then one realistic reason why the players’ strategies should be of limited complexity may be that the players have the epistemic imperfection of being of low intelligence: they are just not intelligent enough to think of and implement very complex strategies. Under this interpretation of the result at issue, simple-mindedness is on a par with incompleteness of knowledge in the sense that it is an epistemic imperfection that, when shared by all the participants, can be beneficial in Pareto-inefficient interactive contexts. Consequently, it may occur, under suitable interactive circumstances, that it is aggregatively rational for all the participants to jointly indulge in a corresponding variety of epistemic irrationality, e.g., that of refraining from developing one’s intelligence as high as possible.

(2) Secondly, exactly in the same way in which its core component, the Prisoners’ Dilemma, does, the Good Jailer generalises to the situations with more than two players, which makes it relevant for the whole range of problems of collective action.

(3) Thirdly, the precise pattern of the situation can vary, the only invariant required being Pareto-inefficiency of the core situation. It is a straightforward observation that for every Pareto-inefficient interactive situation, there exists a way of augmenting it with a stage of epistemic preplay such that the augmented situation is Pareto-efficient, but the cost of restoration of Pareto-efficiency is that part of the Pareto-efficient Nash-equilibrium path of the augmented game is for all the participants to jointly commit an epistemic irrationality of some sort or other⁸⁰.

As to possible applications, my contention is that, given all possible generalisations of the Good Jailer, the formal models of its kind can provide a clue for explaining some important sorts of empirically observable irrationalities in human thought and reasoning – in the same way in which the formal model of the Prisoners’ Dilemma provides a clue for explaining some important kinds of real-life strategic situations.

Given the limitations of this paper, I will cite just one, but important, area of possible application. There is a problem in current economic theorising which is highly relevant both to cognitive science and to the theory of rationalities in conflict I am discussing here. The problem is that, more often than not, real-life markets are imperfect in that sense of perfection that has been ascribed to them by neoclassical theory. One implication is that the agents’ beliefs begin to matter, whereas they were irrelevant under the assumption of perfection. Now, the question is ‘Why is it that very often belief systems that determine the choices of real-life market agents happen to be less than rational epistemically, being myths, taboos, prejudices and other such theories that can be grouped under the umbrella term of ideologies?’⁸¹

I think that an interesting answer can be found along the lines of the theory of rationalities in conflict, the rough outline of the answer being this: Imperfect markets fail to guarantee Pareto-efficiency. But as we have seen, if an interactive situation is Pareto-inefficient, then a jointly committed epistemic irrationality of the right sort can be a remedy. Ideologies may happen to be exactly such sort of epistemically irrational belief systems that, when maintained by all or most of the participants, compensate for the Pareto-inefficiency of the initial market situation.

In other words, some sorts of empirically observed epistemic irrationality may happen to have an impeccable rationale: they render services to aggregative rationality. And there seems to be no reason why this format of explaining irrationality could not be transferred from imperfect markets to further areas of collective action and even to coordination problems with several equilibria.


4. Some implications for the issue of organization of cognitive labour

Because (i) clashes are possible between epistemic rationality and general social utility in suboptimal social situations, and (ii) those clashes may result in the phenomenon of compensatory collective epistemic irrationality (CCEI), the whole of the issue of optimal social organisation of cognitive labour, in general, and some aspects of this issue, in particular, should be seen in a novel (and perhaps somewhat unexpected) light.

Of course social suboptimality and, hence, CCEI influences different kinds of social knowledge production in different degrees. With natural science, the degree of influence seems to be at its lowest. But already social sciences hardly can be said to be entirely impervious to such influence. Interestingly, as early as 1916, in his Trattato di sociologia generale, the Italian economist and sociologist Vilfredo Pareto made a crucial distinction between the epistemic value of a social theory and that theory’s social utility. Moreover, he is reported to have quipped ‘that he hoped his Treatise would not be read too widely, since this would help undermine necessary moral values’. Even though Pareto, to the best of my knowledge, never specified the nature of social conditions under which the distinction would grow into a clash between the two values, - mind you that in the 1910s, there were no such analytic tools as game theory, or such models as the Prisoners’ Dilemma, - nevertheless the very fact that in his quip Pareto mentioned moral values suggests that his theoretic intuition was focused precisely on those manifestations of the conflict between epistemic values and social utility that have the phenomenon of social suboptimality underlying them (which, curiously, has later been named after him: ‘Pareto-suboptimality’).

Both Pareto’s distinction and Pareto’s quip are important and relevant to the theory of CCEI because the cognition area in which the conflict between the two rationalities and the accompanying phenomenon of CCEI have most chances to emerge is precisely the area of moral knowledge. The link between social suboptimality (as instantiated in the Prisoners’ Dilemma) and moral norms has become a subject of detailed analysis in Ullmann-Margalit’s book The Emergence of Norms (1977). On the other hand, the same year of 1977 saw the publication of John Mackie’s Ethics, where he developed his error theory to the effect that wide-spread belief in an objective property of ‘to-be-doneness’ in morality-related things is erroneous. Thus, as early as more than 25 years ago there were established two theoretically important links, namely, (1) a link between social suboptimality and moral norms, on the one hand, and (2) a link between morality and epistemic imperfections (= Mackie’s ‘error’) – or even blunt epistemic irrationality, on the other hand. But it is only now, in the light of the theory of CCEI, that the nature of a third link, - one between those two links, - begins to get transparent.

Thus, it is precisely moral knowledge and the whole constellation of cultural beliefs that supports moral beliefs that become a most natural area of application (far from being the only one) for the theory of CCEI.

Still, what are the implications of the theory of CCEI for the issue of optimal social organisation of cognitive labour? Well, the most obvious one can be easily read off the structure of the only Nash equilibrium in the game of the Kind Jailer. To maximize their epistemic value (= their stock of knowledge), both players should vote, on their first move, for an inquiry into the situation in their gang. That is, given the circumstances, to optimally organize their cognitive labour amounts for them to jointly voting for an inquiry. But this choice lies on an out-of-equilibrium path and, thus, is not optimal, all things considered! This rift between what is cognitively optimal and what is optimal all things considered (which, in such social settings, tends to coincide with what is optimal in the sense of maximizing general social utility) is at the very heart of Pareto-suboptimal social situations.

To formally distinguish between the two senses of optimality, let us introduce two different predicates: ‘optimalE’ for the optimality that corresponds to maximizing epistemic value, and ‘optimalSU’ for the optimality that corresponds to maximizing general social utility. Then, the implication under discussion can be expressed by saying that (i) in situations that display social suboptimality, the issue of optimal social organisation of cognitive labour splits into two different issues: (1) the issue of optimalE social organisation of cognitive labour, and (2) the issue of optimalSU social organisation of cognitive labour, and (ii) in such situations, an optimalSU social organisation of cognitive labour typically consists in a suitable way of undermining the prospects of optimalE social organisation of cognitive labour.

For example, in the situation of the Kind Jailer, for the players to arrange an optimalSU social organisation of cognitive labour, it takes them to deliberately miss an opportunity of arranging an optimalE social organisation of cognitive labour, that is, the opportunity of running an inquiry.

For the sake of another illustration, suppose that Pareto’s quip is literally true. Under that assumption, for the relevant society, part of an optimalSU arrangement for organization of cognitive labour may consist in banning all dissemination of the content of Pareto’s Treatise, which of course is a partial disruption of any optimalE social organisation of cognitive labour.

Methods of disrupting optimalE social organisation of cognitive labour may differ across social situations. One more interesting example can be found in Joshua M. Epstein’s discussion of his agent-based computational model that generates the stylized facts regarding the evolution of norms. (See his paper ‘Learning to be thoughtless: Social norms and individual computation’, Center on Social and Economic Dynamics Working Paper No. 6, revised January 2000. Forthcoming in Computational Economics.) Epstein’s conclusion is that his model captures a real-life phenomenon that has been essentially ignored, namely that the amount of individual thought--or individual computing--is often inversely related to the strength of a social norm. Suppose that Epstein’s conclusion is correct. Then his result can be interpreted as one more method of undermining optimalE social organisation of cognitive labour in order to secure optimalSU social organisation of cognitive labour, which may consist in a co-ordinated constellation of individual decisions to earmark as little time as possible for reflections on social norms.

Admittedly, all my examples and illustrations in this paper are rather simplistic. But so is the famous Prisoners’ Dilemma, which is not deemed to detract from its heuristic and theoretical value. It is rather a rule than an exception that a theory should begin with some simpler-than-life idealisations. What matters, rather, is whether the initial idealised model manages to grasp the core structure of a real-life regularity. In case of the Kind Jailer model, a promising fact is that, actually, the KJ is just the Prisoners’ Dilemma, developed one step further, and the direction of development seems rather natural, so that the nesting model, that is, the KJ, has all the chances to inherit some valuable features of the nested one, the PD – in particular, its immense generalisability and a large variety of real-life situations for which the model can serve as a prototype.