Kantian Ethics and the Prisoners' Dilemma - SSRN papers

Eastern Economic Journal, 2009, 35, (137–143) r 2009 EEA 0094-5056/09 www.palgrave-journals.com/eej/

Kantian Ethics and the Prisoners’ Dilemma Mark D. White Department of Political Science, Economics, and Philosophy, College of Staten Island/CUNY, 2800 Victory Blvd., Staten Island, NY 10314, USA. E-mail: [email protected]

The prisoners’ dilemma game stands as a seminal case of the conflict between individual and collective rationality. Some scholars have suggested that Kantian duty-based ethics can prevent the suboptimal outcomes associated with the game. I argue that this claim is too strong, because Kant’s moral theory does not entail specific duties requiring cooperation in prisoners’ dilemma games. I support this argument in terms of Kant’s categorical imperative, and also with reference to his distinction between perfect and imperfect duties. Eastern Economic Journal (2009) 35, 137–143. doi:10.1057/eej.2008.20 Keywords: prisoners’ dilemma; game theory; ethics; Immanuel Kant JEL: A13; C71

INTRODUCTION The prisoners’ dilemma game stands as the prototypical example of conflict between individual and collective rationality: through mutual pursuit of self-interest, all players end up worse off than if they had behaved otherwise.1 It has been suggested that ethical behavior, as opposed to self-interested decision-making, would help solve prisoners’ dilemma problems: most prominently, Amartya Sen, describing the classic bank robber interpretation of the game, has written that ‘‘it is indeed easy to see that it will be difficult to find a moral argument in favor of confession by the prisoners’’ [1974, p. 77], implying that any ethical system (other than ethical egoism) would demand cooperative behavior in prisoners’ dilemma situations, including Cournot oligopolistic competition, private contributions to finance public goods, and arms races, to mention just a few. As an example of such an ethical system, many scholars have focused on the moral philosophy of Immanuel Kant, the 18th century German philosopher best known for his emphasis on duty and autonomy. They argue that application of Kant’s ethics, as represented by his famous categorical imperative, would require that the players in such games cooperate rather than deviate, and therefore reach the Pareto-superior outcome. In this note, I argue that Kant’s moral theory, while commonly thought to be very demanding, does not in fact guarantee a solution to the prisoners’ dilemma. This is due to the fact that Kant ruled out many actions (and omissions) as immoral, but did not specify precise positive moral requirements of the sort that would correct the behavior that creates the suboptimal outcome. I will demonstrate this in terms of two versions of Kant’s categorical imperative and with reference to his distinction between perfect and imperfect duties.

Mark D. White Kantian Ethics and the Prisoners’ Dilemma

138

THE FORMULA OF AUTONOMY OR OF UNIVERSAL LAW The most widely known version of the categorical imperative is the Formula of Autonomy or of Universal Law: ‘‘act only on that maxim through which you can at the same time will that it should become a universal law’’ [Kant 1785, p. 421]. This formula represents the familiar ethical standard of universalization, which holds that any rule or plan of action one adopts must be consistent with everyone else’s adopting it as well. To apply the categorical imperative to the prisoners’ dilemma, we must ask: would this rule forbid the behavior that leads to the suboptimal prisoners’ dilemma outcome? I will use the term ‘‘deviation’’ to refer to the dominant strategy in a prisoners’ dilemma game (which leads to the suboptimal outcome), and ‘‘cooperation’’ for the dominated strategy (which leads to the optimal outcome).2 Can a plan, or ‘‘maxim,’’ of deviation be universalized without inconsistency? There are two understandings of consistency in Kant’s ethics: consistency-in-conception and consistency-in-thewill [Kant 1785, p. 424]. Consistency-in-conception is a test of logical consistency, which, for instance, rules out lying because universal lying would destroy the trust on which successful lying depends. In the prisoners’ dilemma, is one player’s deviation logically inconsistent with the other player doing the same? Certainly both players can, and often do, deviate simultaneously, so there is nothing logically inconsistent there. (Of course, this behavior results in inferior payoffs, but that does not impact the logical consistency of mutual deviation.) Consistency-in-the-will asks if the agent can rationally will that her maxim be universalized. This test generates duties such as beneficence by ruling out a maxim of indifference to others: since everyone will need aid at some time, no one can rationally will that everyone is always indifferent towards others. Rather than internal logical consistency, Kant wrote that, under this test, maxims, when universalized, must not contradict ‘‘objective ends, which depend on motives valid for every rational being’’ [Kant 1785, p. 427], namely humanity (possessing an inviolable dignity) and our own survival, but not our everyday preferences and inclinations. Willing universal indifference endangers our own survival, so it fails the consistency-in-the-will test. But what about deviation in the prisoners’ dilemma game? Unless survival is at stake, it is difficult to see how universal deviation is inconsistent with either objective end. So if we rule out such extreme circumstances, we can conclude that neither understanding of consistency derived from this formula of the categorical imperative prevents deviation in the prisoners’ dilemma game. Sen comes close to the contradiction-in-the-will test when he writes that ‘‘certainly neither prisoner would like that confessing becomes a universal practice, and the only universal law that each prisoner would like is that everyone should refuse to confess’’ [1974, p. 76, emphasis added]. But here he conflates ‘‘willing’’ with ‘‘liking’’ (or ‘‘wanting’’), which is not how Kant intended the moral law to be applied. Sen’s statement of the rule is based on preferences or inclinations, what one ‘‘likes’’, while Kant’s categorical imperative is based purely on reason divorced from personal considerations. Indeed, Kant [1785, p. 430n] criticized the Golden Rule — ‘‘do unto others as you would have them do unto you’’ — for the same reason, that it is grounded in subjective inclination rather than objective duty. While this wording may reflect simply a lack of precision of Sen’s part — for he was writing chiefly for economists, not philosophers, and he certainly has a much more elaborate conception of wants, preferences, and choice than this — a similar misunderstanding of the universalization formula of the categorical imperative has Eastern Economic Journal 2009 35


139

been repeated often in the literature on altruism and public good financing. For instance, Bilodeau and Gravel understand Kantian ethics to mean ‘‘compelling an individual to undertake any action which we would want everyone else to undertake’’ [2004, p. 646, emphasis added], and cites many papers that share this conception, including Laffont [1975], one of the earliest attempts to integrate Kantian ethics explicitly into economics. But Bilodeau and Gravel recognize that this understanding of Kant’s moral theory may not be accurate; in fact, they point to Wolfelsperger [1999] and Ballet and Jolivet [2003], who emphasize the frequent misuse of Kantian ethics in economics.

PERFECT AND IMPERFECT DUTY Perhaps we should look at the nature of duties themselves for a different perspective on the problem. There are two types of duties in Kant’s ethics, perfect and imperfect. Perfect duties, also called narrow duties, are normally negative, and rule out certain actions that fail the consistency-in-conception test, such as lying. We saw above that deviation in prisoners’ dilemma games cannot be ruled out by the consistency-inconception test, so ‘‘do not deviate’’ cannot be a perfect duty. Imperfect duties, also called wide duties, are often positive, and result from failures of the consistency-inthe-will test. They demand only that certain attitudes or ends be adopted, such as helping others, but do not require any specific action in their pursuit. In other words, imperfect duties allow for latitude in their execution, and their performance can be adjusted to accommodate other duties, including duties to oneself. This belies the common belief that Kantian ethics demand tremendous altruistic sacrifices; as Kant wrote, ‘‘How far should one expend one’s resources in practicing beneficence? Surely not to the extent that he himself would finally come to need the beneficence of others’’ [1797, p. 454].3 Can a duty not to deviate in prisoners’ dilemma games be a specific instance of the duty of beneficence? Not deviating will increase the other player’s payoff regardless of her action, so it would certainly be considered helpful, and may be considered a virtuous act. But at the same time it comes at a sacrifice to yourself (since you would be choosing the dominated strategy), and you are not required to make that sacrifice unless you judge that duty to have a ‘‘stronger ground of obligation’’ [Kant 1797, p. 224] than duties to yourself (a standard on which Kant unfortunately did not bother to elaborate). So while cooperation in a prisoners’ dilemma game may be nice, kind, admirable, or noble, it is not required; even a duty-based concern for the other player does not mandate that one sacrifice his own well-being by avoiding the self-interested dominant strategy of deviation. The essential problem with relying on Kantian ethics to prevent prisoners’ dilemma outcomes is that while the categorical imperative rules out many actions as immoral, it does not specify any precise actions that must be taken to be moral. All duties, perfect and imperfect, are ultimately negative in nature — for instance, the ‘‘positive’’ duty of beneficence is truly a negative duty, specifically requiring that we not be indifferent to others, but without telling us exactly what we should do positively in the process. (Even the ‘‘perfect’’ duty not to lie does not specify what should be done instead, such as tell the truth, evade the question, or simply remain silent.) Therefore, since negative duties do not demand specific action, we cannot thereby derive a strict moral obligation to cooperate in prisoners’ dilemma games based on them. Eastern Economic Journal 2009 35


140

But if we expand the strategies available to the players, there may be a way to imply a perfect duty to cooperate, specifically by allowing both players to promise cooperative behavior before the game is played.4 Kant uses promise-keeping as one of the four primary examples of ethical behavior, based on the consistency-inconception test: as with lying, if promises are not kept, trust will be compromised and promises would lose their efficacy [1785, pp. 402–03].5 Therefore, there is a perfect duty not to break one’s promises (or not to make false promises in the first place). So if the players make promises to each other to cooperate, then they are bound by their duty to keep these promises, thereby avoiding the prisoners’ dilemma outcome. However, since there is no duty to make such a promise in the first place (because such a duty would be based on a more general duty to cooperate, which does not exist), there is still no perfect duty that requires the players to choose the cooperative action in prisoners’ dilemma games. But even ruling out a perfect duty to make promises to cooperate, such a practice can provide a way for players to avoid the prisoners’ dilemma outcome: if each player promises to cooperate, contingent on the other player making the same promise, then both players would be bound by their promises and would therefore cooperate. But there still remains one problem: a promise to cooperate is effective in eliciting similar behavior from other players only insofar as the player making the promise is known to be a Kantian — after all, anyone can promise to cooperate, which is ‘‘cheap talk’’ in a single-play game (and may be generated by purely self-interested behavior in an infinitely repeated game). For these reasons, the Kantian duty of promise-keeping cannot be relied on to ensure cooperative behavior in prisoners’ dilemma games, but can provide a method for Kantian agents to resolve prisoners’ dilemma games in their own self-interest.6

THE FORMULA OF RESPECT FOR THE DIGNITY OF PERSONS A more humanistic perspective on Kantian ethics is provided by another formulation of the categorical imperative, the Formula of Respect for the Dignity of Persons: ‘‘so act as to use humanity, both in your own person and in the person of every other, always at the same time as an end, never simply as a means’’ [Kant 1785, p. 429]. The negative aspect of this formula, to not use persons simply as a means to our own ends, results in perfect duties, such as ‘‘do not lie’’ and ‘‘do not steal.’’ But this formula also has strong positive implications, for we are not just prohibited from using others as means, but we are also required to treat them as ultimate ends, and therefore act to promote their interests when possible, resulting in imperfect duties such as beneficence. Is there any sense in which deviation in a prisoners’ dilemma game uses the other players simply as a means, and not at the same time as an end? It is true that one player’s deviation always raises his payoff and lowers the other player’s payoff (regardless of the other player’s action), but does that imply that the first player used the second simply as a means to his benefit? If it did, it would imply that one firm uses its competitor when it lowers its price to gain market share, increasing its own profit ‘‘at the expense’’ of the other firm’s profit; or that a firm in a imperfectly competitive industry uses its customers due to its ability to raise price above marginal cost. Would these actions be forbidden by Kant’s injunction to respect the dignity of persons (assuming we can extend this concept to firms as well)? Eastern Economic Journal 2009 35


141

While these actions may fall into some readers’ personal conception of using someone, Kant had a much more precise notion of what it means to use someone merely as a means and not at the same time as an end. To respect another person’s dignity, Kant wrote you must treat her as an autonomous, rational being; specifically, she must be able to — though not necessarily want to — rationally assent to your ends. For instance, in the context of the duty not to make false promises, Kant described what it means to treat someone otherwise: [T]he man whom I want to use for my own purposes by such a promise cannot possibly concur with my way of acting toward him and hence cannot himself hold the end of this actiony. [A] transgressor of the rights of men intends to make use of the persons of others merely as a means, without taking into consideration that, as rational beings, they should always be esteemed at the same time as ends, i.e., be esteemed only as beings who must themselves be able to hold the very same action as an end. [Kant 1785, pp. 429–30, emphasis mine] Note his emphasis on possibility and ability, not preference or desire: the other person does not have to actually share my end or agree with it, and can in fact abhor it. Kant only requires that she be able to rationally assent to it, even if she chooses not to, for it is the fact that she can choose to, not whether she would or not, that is essential. For this reason, coercion and deception are the primary examples of using someone simply as a means, because they both render her incapable of making a choice regarding the other person’s end.7 Coercion obviously rules out choice altogether, and deception keeps the other person unaware of my true end, denying her the ability to assent to it. But if she can make a choice without coercion or deceit on my part, then I have treated her as a rational, autonomous person, and have considered her as an end as well as a means.8 So the firm that, without coercive or fraudulent behavior, lowers prices to steal business from a competitor treats that competitor with respect (even though it lowers its profit). Asymmetry of power, as in the imperfect competition case, does not imply violation of respect either; although he may be resentful of the higher price charged, the customer has the choice whether or not to participate in a transaction with the company and recognize its end of profit maximization, as long as it has not acted fraudulently.9 The same goes for the prisoners’ dilemma game: each player’s end is assumed to be his or her own self-interest, to which the other player can certainly assent (even if the players have no regard for each other, or despise each other). Recall the conception-in-the-will test, which requires that an agent be able to rationally will that a maxim be universalized. In this context, a player’s end must be universalizable in the sense that other players can possibly assent to or share it (whether they actually do or not).10 Since the standard definition of the prisoners’ dilemma game rules out coercion (through free choice of actions) and deception (through common assumptions on information), it is implied that the players are treating each other with the respect due to rational, autonomous persons.

CONCLUSION While Kantian ethics may encourage cooperation in a prisoners’ dilemma game, it does not require it; in other words, the categorical imperative permits players in Eastern Economic Journal 2009 35


142

prisoners’ dilemma games to choose their dominant strategies of deviation despite the resulting suboptimal payoffs for both. Because the categorical imperative does not demand specific actions, but only forbids certain actions (perfect duty) or requires general attitudes (imperfect duty), it does not support a strict duty of beneficence towards the other player. Furthermore, deviation does not involve using the other player simply as a means and not at the same time as an end, because it does not impair their autonomy or choice through coercion or deceit. As I noted above, some readers may have a stronger definition of what it means to use another person, but Kant’s understanding of it was rather narrow, which supports my argument that Kantian ethics is not strict enough to guarantee cooperation in prisoners’ dilemma games. Economists seeking a moral system that would prevent prisoners’ dilemma outcomes will have to turn to more specifically demanding ethical theories, such as utilitarianism (which is already the ethical basis of most mainstream economic analysis). Indeed, unlike utilitarianism, Kant never intended to provide definite rules for ethical decision-making, but instead offered a general framework to guide persons’ judgment in moral dilemmas.11 Due to this ambiguity, Kantian ethics cannot guarantee a hard and fast solution to prisoners’ dilemma problems.

Acknowledgements I thank Alex Viskovatoff, Elias Khalil, Deirdre McCloskey, and the rest of the participants at the International Network of Economic Methodology session at the Allied Social Science Association meetings, and the Eastern Economic Association meetings, at which a very early version of this paper was presented. Special thanks must go to Gil Skillman and two anonymous references for the Journal, who provided exemplary commentary and criticism. The usual caveats apply.

Notes 1. This game is alternatively referred to as the prisoner’s dilemma and the prisoners’ dilemma. I choose the latter to emphasize that the problem only arises in strategic, multi-player situations. 2. Keeping the description of the game general avoids the complication of judging the intrinsic morality of deviation and cooperation in any specific game. For instance, deviation in the standard bank robber tale — telling the truth and cooperating with the authorities — is the more ethical choice from society’s point of view, but deviation in a public good financing problem — free-riding — is not. 3. See White [2004] for an economic model of choice incorporating Kantian ethics based on the distinction between perfect and imperfect duties. 4. This may run up against the standard assumption of no binding agreements in non-cooperative game theory. But I take that assumption to rule out third-party enforcement, not commitment on the part of the players themselves, which is impossible for self-interested agents in a one-shot or finitely repeated game. 5. See Hill [1991] for more detail on the Kantian ethics of promise-keeping, to oneself as well as to other people. 6. I thank Gil Skillman for pressing me to develop this point further. 7. For extended discussion, see O’Neill [1985] and Korsgaard [1986]. 8. This statement holds even if the choice situation reflects incomplete information, or constraints resulting from legal or institutional coercion, neither of which are an agent’s responsibility towards others, and therefore do not affect the respect with which one treats another person. 9. For an extension of this logic to antitrust law, see White [2007]. 10. Ideally, all of Kant’s versions of the categorical imperative are equivalent, so ancillary principles of each are related as well. 11. For more on this point, see White [forthcoming]. Eastern Economic Journal 2009 35


143

References Ballet, J., and P. Jolivet. 2003. A Propos de l’E´conomie Kantienne. Social Science Information, 42: 185–208. Bilodeau, M., and N. Gravel. 2004. Voluntary Provision of a Public Good and Individual Morality. Journal of Public Economics, 88: 645–666. Hill, Jr., T.E. 1991. Promises to Oneself, in Autonomy and Self-Respect. Cambridge: Cambridge University Press, 138–154. Kant, I. 1785. Grounding for the Metaphysics of Morals, translated by James W. Ellington. Indianapolis: Hackett Publishing Company (1993). _______ . 1797. The Metaphysics of Morals, translated and edited by Mary Gregor. Cambridge: Cambridge University Press (1996). Korsgaard, C.M. 1986. The Right to Lie: Kant on Dealing with Evil, reprinted in Creating the Kingdom of Ends. Cambridge: Cambridge University Press (1996), 133–158. Laffont, J.-J. 1975. Macroeconomic Constraints, Economic Efficiency and Ethics: An Introduction to Kantian Economics. Economica, (n.s.) 42: 430–437. O’Neill, O. 1985. Between Consenting Adults, reprinted in Constructions of Reason: Explorations of Kant’s Political Philosophy. Cambridge: Cambridge University Press (1989), 105–125. Sen, A.K. 1974. Choice, Orderings and Morality, reprinted in Choice, Welfare and Measurement. Cambridge, MA: Harvard University Press (1982), 74–83. White, M.D. 2004. Can Homo Economicus Follow Kant’s Categorical Imperative? Journal of SocioEconomics, 33: 89–106. _______ . 2007. A Kantian Critique of Antitrust: On Morality and Microsoft. Journal of Private Enterprise, 22: 161–190. _______ . (forthcoming). In Defense of Deontology and Kant: A Reply to Van Staveren. Review of Political Economy, 21. Wolfelsperger, A. 1999. Sur l’Existence d’une Solution ‘Kantienne’ du Proble`me des Biens Collectifs. Revue E´conomique, 50: 879–901.

Eastern Economic Journal 2009 35