Ken Binmore claims that when he asked some philosopher colleagues about Kantian ethics, some replied that Kant thought we should cooperate in a prisoners’ dilemma (PD) (Natural Justice, p.viii). He does not cite any particular academics here, though there is a quote from Amartya Sen in Binmore’s Playing Fair (p.26) which could be taken to provide some support for the idea. The claim that Kant thought we should cooperate in a PD has since been repeated by Herbert Gintis, who cites only Binmore as a source (‘Behavioral ethics meets natural justice’, p.6), and I have also heard the claim repeated by academics at UQ and QUT. It is not entirely clear whether the claim is that we should cooperate in all PDs or just some PDs, but the intention is clearly to suggest that Kant’s views reduce to absurdity in one way or another. My purpose here is to dispute this claim (it would take a much longer article to conclusively debunk it), but also to examine how someone could come to see it in Kant’s writing. This is neither a full explanation of Kant’s views, nor is it meant as a defence of his views, so interested readers are encouraged to explore the topic further themselves.
What is not Disputed
The PD wasn’t conceptualised in Kant’s time, though there are discussions of game-theory-like mixed-motive interactions throughout the work of the early social contract theorists – probably most famously in Rousseau’s Stag Hunt. Game theory was formalised in the early 20thC, and the PD was described in the mid 20thC and named somewhat later. So attributing this belief to Kant requires ‘reading it into’ his writings, he obviously did not state it explicitly himself.
It will help to distinguish the PD as an abstract (formal) concept from the dilemma of the (worldly) prisoners held in separate rooms by police. Conflating the two seems to be the source of some of shock and outrage against Kant that Binmore’s claims have stirred up. A formal PD has a dominant strategy – this is analytic. But this understanding is of only limited use to you if you are a non-sociopath who is actually in police custody and deciding what to do next. If we are sticking purely with the formal model of a (one-shot) PD, then it is true by definition that those in it would be irrational to cooperate, and to say otherwise is to misunderstand or attempt to re-configure the model. But there is much more scope in the worldly case studies to enquire about what one should do and why. In the worldly cases there are at least three questions to answer: what are our values, what should our values be, and given our values what is it rational for us to do? Game theory tends to deal directly only with the last of these. So if the fictional prisoners see defection as the clearly best option, we might still be able to say that they are not reasoning with regard to their best reasons (in Kantian terms – their being interested only in their own escape from police coercion is a failure of autonomy, in some sense). But we cannot say (to return to the formal model) that given a certain all-things-considered utility function, someone actually has a different utility function – and if Kant said so, then that would certainly be a problem.
I am not accusing Binmore here of conflating the formal PD and a worldly one – just pointing out that, as Kant could not have been talking explicitly about a formal PD, we should be careful about diagnosing a call for ‘cooperation’ in his work as a call for ‘cooperation’ in a genuine PD. (I am not aware of any specific examples of worldly PDs given by Binmore that Kant would allegedly call for cooperation in, so I have to guess my way through this, to some extent.) I want to examine why someone might attribute the view that one should cooperate in a PD to Kant. In these cases, I think there is ample room for disagreement with Kant on whether a certain situation counts as a genuine PD, and this disagreement can entirely account for any sense that cooperation is irrational.
Kantian Personhood and Rationality
The first question is when a Kantian person ends up in a PD in the first place. The PD is a dilemma designed for homo economicus, but if this is not the model of rational choice that Kant is using, then it is unclear what the relationship is between his account and the PD.
One theme from Kant which is still going strong in contemporary discussions of morality is the link between freedom and moral accountability. If we do not have control over our actions, we can hardly be held accountable for them, or rationally hold ourselves accountable through feelings of guilt and shame (so the neo-Kantian argument goes). The keys are two inescapable features of being a (typical) person: that we are moral, and that we can reason. Having a sense of moral right and wrong requires a sense of control over one’s actions, and so a choice regarding whether to act on our utility function. (I think Kant is right to hold that this sense of freedom need not depend on any controversial neurological assumptions – but that is a tangent that I won’t pursue here.) The fact that we can reason also means that our action is a product of choice – we can be asked to account for our behaviour, and indeed to some extent to account for our personality. If we are asked to account for our status as homines economici, it will not do to simply reply that homo economicus is a useful model of behaviour for economists, or this is how we would expect people to be given a certain evolutionary model of our ancestry. Or at least, it will not do to give this reply if we care about the question and our interlocutors’ reasons for asking it. (Some doubtless will care more than others.) To Kant, a person cannot be considered free if their actions are wholly explicable in terms of causes and effects – the action is just a product of circumstances, and so neither moral nor immoral, whatever its good or bad effects. To participate in morality, we need something more – and for Kant, this is an ability to govern our behaviour in accordance with rules derived rationally – with rationality in some sense sitting outside of the world of cause and effect, in much the way that truths of mathematics do. Note that I am here talking of two different explanations of behaviour which are helpful, compatible, but not reducible to each other – this may be a hard pill to swallow for some social scientists with physics-envy, but unless a real incompatibility can be demonstrated then both explanations seem to offer legitimate lines of academic enquiry.
The rules which Kant argues for are not here merely ‘rules of thumb’, but rather strong constraints on action. There are some things we may do, some things that we physically cannot do, and some things that we do not do because we see them as strictly inappropriate: no matter how well they might benefit us materially (say), they are at odds with the moral persona that allows us to be morally accountable. Allegiance to the rules is here looks a bit like what psychologists today call a ‘sacred value’, though the hyper-rationalism of Kant seems at odds with the view of the modern psychologists. It is disputable whether this Kantian ideal of practical rationality (or, similarly, a holder of ‘sacred values’) can be modelled as a homo economicus. The dominant view, I take it, is that to accept anything like Kantian ethics, we need to accept values which are of a very different sort than the maximalist ones of most of economics. For example, it is not clear that a rule-governed person sees breaking a rule twice as being twice as bad as breaking a rule once. This is disputable – Gerald Gaus, using a model of Stanley Benn’s, has tried to demonstrate that a preference for rule-following can be modelled as a sort of economic utility (e.g. Gaus, The Order of Public Reason, §9.4). This is done as part of a quasi-Kantian account of personhood and social morality.
If the Kantian model is at odds with homo economicus, then its relationship to the formal PD is prima facie indeterminate. Game theorists say that homines economici should defect – they do not say that chairs or equations or colours or poems should (as none of these can participate in a PD). I can see no reason, though, why a Kantian person would not sometimes end up in a PD and therefore find it rational to defect – homines kantii can certainly engage in maximalist reasoning and can certainly recognise their own ends as distinct from those of others. But of course, what sort of worldly scenarios will count as PDs will vary between the Kantian and homo economicus models.
The other option is that a Kantian person can be modelled as a homo economicus. Curiously, this is a possibility which Binmore mentions, but does not deal with very well. In Playing Fair (p.149) he proposes a Kantian actor who rejects the maxim “Never use a strongly dominated strategy.” In order to make sense of this seeming incongruity, Binmore separates the ‘personal inclinations’ of the Kantian actor from their inclination to follow the rules. The Kantian uses ‘personal inclinations’ to identify that a PD (and so a strongly dominated strategy) exists, but acts on the rule (‘Cooperate!’) anyway, in line with their full set of preferences (making us question, surely, whether it really is a strongly dominated strategy at all). Binmore explains: “I shall assume that, when placed in situations that seem like the Prisoners’ Dilemma to a game theorist, a person who acts on Kantian principles will play dove himself when he knows that his opponent is to play dove.” Why? “Otherwise there would be no grounds for game theorists and Kantians to disagree.” (my emphasis) So some Kantian intention to cooperate in a ‘PD’ is seemingly both a premise and a conclusion here. And the situation is only a PD at all from the perspective of a narrow (moral preferences excluded) model of homo economicus, a move which the Kantian obviously rejects but which Binmore also (rightly) rejects elsewhere (e.g. Playing Fair, p.19).
In short, we can only model a Kantian who wants to cooperate in a PD by building in some cognitive dissonance – they are either a homo kantus who mistakes themself for a homo economicus, or a broad homo economicus who we have mistaken for a narrow homo economicus. But it is not clear why we would want either of these models.
The Social Contract
Both Kant and Binmore are, in some sense, social contract theorists. But this is a label applied to many different sorts of projects. Here are three questions addressed through this sort of framework: Why should we have morality (why have a sense of right and wrong)? Given a sense of right and wrong, what should the specific content of our moral code be? What does an ideal/acceptable political order look like? These are not at all the same question, though lessons gleaned from trying to answer one may come in handy answering the others. As far as I can tell, Binmore is primarily interested in the third question, but when citing Kant’s work primarily focuses on answers to the second question, and is similarly baffled and outraged by work on the first which he can’t easily integrate into his own project.
Game theory has given us a new vocabulary with which to discuss the problems of social life and the point of moral and political theory. It is common now to translate the claims of early social contract theorists into the language of game theory. (A popular but very crude one: Hobbes has been represented as suggesting that without government we are all in a PD with regard to each other, so we should get an external force (the sovereign) to force us to cooperate, such that we no longer have the option of defecting and are no longer in a PD.) But even more interestingly, it helps us to better appreciate how the different social contract theorist’s questions may relate to each other. A worldly PD is a problem, in that we are led to sub-optimal outcomes. There are numerous responses available if these worldly PDs are prevalent – including political (i.e. governance) solutions, changing ones’ own values, or changing one’s patterns of social interaction (c.f. tit-for-tat), or of course learning that it is rational to defect and just going ahead and doing that every time. And each of these solutions comes with its own problems, including new PDs and other sub-optimal outcomes, and so new responses to these new problems. On the social contract theorist’s view, one reason for a moral and political order is to lower the incidence of PDs; it is not to punish us by demanding unconditional cooperation when we are (genuinely) in one.
If there is a criticism to be made of Kant’s Groundwork here, perhaps it is that it treats as foundational a certain conception of moral personhood, rather than allowing this to be critiqued in light of the conclusions derived from it. That is, it focuses too much on the second question, and without enough regard the first and third. He does not sufficiently examine the costs of his account of morality, and revise his account of rational personhood accordingly. I am genuinely uncertain how much bite this criticism actually has – I would want to go back and read Kant much more thoroughly before affirming it as true. But I have the impression that the neo-Kantian writers are conscious of the potential for this problem and are determined to avoid it.
So Would Kant Say That We Should Cooperate in a Prisoner’s Dilemma?
(The short answer is ‘no’.) The formal PD is an abstract model with an ex hypothesi right answer, not a situation in which we are actually called on to make a choice. We need something more worldly in which to discuss which of Kant’s social-contract-derived rules of behaviour should be in play (though I note that White in ‘Kantian Ethics and the Prisoners’ Dilemma’ has a different take on this). If we want to treat Kant as a sort of moral sage and obey his suggestions unquestioningly in apparent PDs – and I do not recommend this – the ‘right’ way to behave will still vary from situation to situation. Even if we take the label literally, and focus on a story of two prisoners in police custody, we still may not know enough. Does cooperation (with one’s fellow prisoner) require lying? Kant famously didn’t like lying. Does defection involve a breach of trust? Maybe that trust should never have been established in the first place, but maybe it should still be honoured. It will doubtless matter whether the people involved are opportunist con-artists, mistakenly imprisoned innocents, or Sophie and Hans Scholl. This will affect the (Kantian) right thing to do, but should in the same manner affect whether a genuine PD exists at all. Undeniably though, there will be some cases in which reasonable people feel that they are in a PD, but a Kantian denies that they should ‘defect’. We may disagree with the Kantian in some of these cases, but we should at least recognise that it is not an outrageous or stupid claim.
Final Note
It is probably time to say something about why I have chosen to dispute at length a misunderstanding that I have no real stake in. One reason is that I am trying to get a better understanding of Kant’s work, and writing it out tends to be the best means of doing this. I also had the impression that this misconstrual of Kant is becoming widely believed among academics outside of philosophy schools, but I could find no resource available to give to people who believed it, so I thought I’d better create one. The topic also sits at an intersection of a few trends in academic work which bother me. These all relate to attempts to present a ‘theory of everything’ on the basis of one’s own area of specialisation. Binmore is a game theorist, and he seems to want all of us to be so, all of the time. Hence passages like the following: “But for me, consenting just means saying yes. After all, is it not always true that we agree to something because we perceive the alternatives as being worse?” (Playing Fair, p.16, emphasis in original). I need hardly say that this is a horribly unidimensional account of an important moral concept – by which I mean not just a moral-philosophical concept but one which everyone needs in order to be socially competent. Second, if one has all of the answers, then of course that means that all of the other answers must be wrong. The attempt to trash Kant’s reputation seems to have gotten some traction among social scientists, but really adds nothing of value to Binmore’s body of work – the books would be much better if Kant’s Groundwork was not mentioned at all. But when one is defending a theory of everything, one presumably has to debunk everyone else, no matter what they are actually talking about (or recruit them into service, as Binmore tries to do with Hume). Finally, and relatedly, there is a sense in this sort of dispute that some people are Right and others are Wrong, where being Wrong is not simply a matter of having reached wrong conclusions, but of one’s work being so comprehensively corrupt that we need to quarantine it. This is not a productive attitude to have. Kant’s worldview was more theistic and (in a sense) teleological than that which is fashionable in academia today, but that need not precluded his work on autonomy and the social contract (for example) from being of great contemporary importance.
[Images are of Boggo Road Jail, Dutton Park, Queensland, July 2018.]
The ArachneanMetaxion Point 