Sections 2.5 and 2.6 form the core argument of this chapter. These sections can be studied independently of each other, but I think it worthwhile to first take a look at how they work together to support Gaus’ main conclusion of Chapter 2, which is that the instrumentalist program fails. Gaus constructs a dilemma for the instrumentalist: In order to complete the instrumentalist project, one must either (1) adopt some reformed account of rational decisions in interactive contexts (formally, game theoretic contexts), or (2) show that societies will settle into following rules for guiding conduct in repeated interactions that coincide with moral rules, and neither (1) nor (2) is a viable option. Gaus argues that (1) and (2) are both nonviable options in sections 2.5 and 2.6, respectively.
I like to relate the instrumentalist program to the so-called reconciliation project, the attempt to show that acting morally is compatible with rational prudence. There are at least three nonexclusive ways one can try to complete the reconciliation project, which can be illustrated with alternative responses to a moral skeptic like Hobbes’ Foole (Leviathan 15) who claims that at least some of the time rational prudence requires one to violate moral requirements. In response, one can argue either that:
(a) The skeptic overlooks certain goods intrinsic to the practice of morality (a response in the tradition of Plato, Aristotle and Aquinas), or:
(b) The skeptic overlooks some of the options available both to her and to those with whom she interacts (a response in the tradition of Hobbes, Hume, and in our time Sugden, Skyrms and Binmore), or:
(c) The skeptic adopts the wrong account of rational choice in interactive contexts (a response in the new tradition initiated by McClennen and Gauthier)
Responses (b) and (c) obviously dovetail with Gaus’ sections 2.5 and 2.6. Of course, Gaus is discussing the instrumentalist program and not the reconciliation project, but it might be worth noting here that Gaus says little in Chapter 2 regarding supposed intrinsic goods of moral practice, and of course many including skeptics like the Foole deny there are any such goods. In any event, while I think it’s conceivable that an instrumentalist could believe in some goods intrinsic to moral practice (say, if one believes that following moral requirements is a necessary means towards achieving the desired end of reaching Heaven in the afterlife), Gaus keeps his focus in these sections on external goods like others’ labor that one could in principle obtain by means other than following moral requirements, and are more commonly associated with economic analysis.
In Section 2.5 Gaus gives his specific arguments against revisionist theories of game theoretic rationality, and he makes it clear that these arguments resemble some that have appeared before. (Perhaps I should add that I agree with pretty much all of Gaus’ discussion in this section.) Interestingly, Gaus considers only two specific proposals for a revisionist theory, namely, the evidential decision theory of Richard Jeffrey and the past-formed-disposition rule of Gauthier that I regard as a special case of Gauthier’s constrained maximization (Morals By Agreement 1986, Chapter V). I think Gaus believes that the two specific proposals considered here suffice to illustrate a moral general argument: Proposals for a reformed account of game theoretic rationality require us to give up certain fundamental dogmas of instrumental rationality (discussed in section 4.3) we really should not give up, so the revisionist program should be rejected.
Evidential decision theory recommends following C (cooperate) rather than D (defect) in Prisoners’ Dilemma when one believes one’s own choice is sufficiently correlated with her partner’s choice, same as evidential decision theory recommends choosing only the opaque box in Newcomb’s problem. This is tantamount to treating one’s own choice as if this choice is not causally independent of the choices of one’s partners. But Gaus, following Skyrms, reminds us that the players’ choices in a Prisoners’ Dilemma are causally independent (p. 76). Here is another way to another way to make the point (though Gaus does not explicitly make the point this way): To claim that one should choose a strategy on the grounds that this strategy is probabilistically correlated with the strategy one’s partner chooses is to in effect deny that the Prisoners’ Dilemma is a game in strategic form. (By definition, in a strategic form game the players must select their strategies without being able to observe the selections of their partners.) In this context, evidential decision theory would recommend that we don’t follow strictly dominant strategies, but we really have no good reason to give up on dominance reasoning. (Perhaps I should add that Jeffery himself did not think evidential decision theory recommends cooperation in the one-shot Prisoners’ Dilemma, though some evidential decision theorists do think this.)
Gaus next considers Gauthier’s proposed revisionist theory, illustrated by Gauthier’s analysis of the one-shot Farmers’ Dilemma game (illustrated on p. 78). Farmers’ Dilemma is sometimes referred to as sequential Prisoners’ Dilemma since here the game is an extensive form game of perfect information where if both follow C (help), both are better off than if both follow D (don’t), but as in Prisoners’ Dilemma there is a unique equilibrium where both follow D. (However, the underlying reasoning that leads the players to follow this equilibrium is strikingly different from that in the “ordinary” strategic form Prisoners’ Dilemma.) Gauthier argues that the two rational players can follow (C,C) if prior to play Player 2 adopts a disposition to follow C in case Player 1 follows C and Player 1 recognizes this and consequently chooses and consequently chooses C. Again, I view this is a special case of Gauthier’s constrained maximization, but here the surprise twist is that Gauthier argues that rationality requires Player 2 to act consistently with the disposition formed before the start of the game, so that if Player 1 has followed C then Player 2 should follow C, which does not maximize Player 2’s payoff, even though Player 1’s cooperation is now a “done deal”. This case of constrained maximization requires one to violate modular rationality.
Gaus does not press the most common line of criticism of constrained maximization. Gaus does not complain (as a great many others do complain) that even if one assumes that agents can adopt constrained maximization, there seems to be no way we could recognize when someone has formed the necessary dispositions, that is, we have no reliable way of distinguishing the constrained maximizers from the “straightforward maximizers” (Gauthier’s term for orthodox expected utility maximizers. Gauthier anticipates this objection in Morals By Agreement but many are not satisfied with Gauthier’s analysis.). Instead, Gaus argues against Gauthier’s proposal on the grounds that this proposal commits one to ignoring certain relevant information one has at the time of decision. A good example of this phenomenon is Gauthier’s own probabilistic variation of the Farmers’ Dilemma that Gaus summarizes on p. 83. According to Gauthier, Player 2 should follow C and thus be consistent with the pre-play intention to follow C even if the chance event places them in the branch where following C results in Player 2’s worst outcome and following D results in Player 2’s best outcome. Gaus finds this conclusion dubious. If Player 2 in fact arrives at the branch where following C results in her worst outcome, she now has information that should lead her to analyze the decision problem differently than at the time before play when she knew only the ex ante likelihood of arriving at this branch. As Gaus puts it, Gauthier’s revisionist project fails because Gauthier “could not find a convincing way of tying our goal-promoting choices to the past without making our lives go less well than they might go (p. 86).”
(1) Must an instrumentalist take one of the horns of Gaus’ dilemma? An instrumentalist could try to apply a revised account of game theoretic rationality to repeated games in a way that avoids Gaus’ most serious criticisms of the instrumentalist program. Kavka attempts something like this with his principle of disaster avoidance. (Kavka presents this idea in several works, including Moral Paradoxes of Nuclear Deterrence 1987). Disaster avoidance is designed for agents who cannot compute their expected utilities perfectly because they lack the ability to form numerically exact probabilities and utilities over the relevant alternative outcomes. Kavka thinks most of us lack this ability most of the time, so that disaster avoidance is a decision principle Kavka also attributes disaster avoidance reasoning to Hobbes, though obviously Hobbes lacked Kavka’s technical vocabulary. A disaster avoider chooses the option that minimizes her probability of suffering a disaster. (Obviously a disaster avoider needs to know enough regarding the relevant probabilities and payoffs to be able to choose according to this rule.) Disaster avoidance can be applied in both one shot games and in repeated games. For example, in covenant situations having a Prisoners’ or Farmers’ Dilemma structure, one should keep one’s promise in order to avoid the disaster of being excluded from the benefits of others’ aid in case one’s failure to keep the promise is detected and consequently punished. Disaster avoidance certainly has its critics, but this principle does less violence to our decision theoretic dogmas than do the revised theories Gaus does consider.
(2) Has Gaus considered the most important revisionist theories? For instance, Gaus does not discuss McClennen’s principle of resolute choice for acting in extensive form games (Rationality and Dynamic Choice 1990). Resolute choice recommends following certain strategies that can fail to maximize one’s own payoff in certain subgames if before play these strategies can be part of a payoff maximizing outcome for the entire game. Resolute choice also recommends occasional departures from modular rationality. Should Gaus also have discussed resolute choice and other related theories in this section?
(3) Should Gaus have considered in more depth the possibility that rationality recommends transforming the structure of the interaction problem? Gaus does consider this when he discusses some of Bicchieri’s analysis of how entering into a game like Prisoners’ Dilemma might possibly lead to the game being transformed into a coordination game. But Gaus also does not discuss a recent and quite different proposal for a reformed account of game theoretic rationality, namely, the team reasoning approach proposed by Michael Bacharach (Beyond Rational Choice 2006). The leading idea here is that team reasoners will transform a problem like Prisoners’ Dilemma into a single agent decision problem where C is the unique optimal choice among the alternatives C and D. Consequently, team reasoners will follow C whereas conventional game theoretical reasoners will follow D. Even if one accepts Gaus’ conclusions in section 2.5, could team reasoning succeed where evidential decision theory and constrained maximization fail? (I happen to think the answer is “No.”. I also think Gaus would reject team reasoning.)
(4) Do revisionist theories have some relevant explanatory power? Again, I agree with what Gaus has presented in this section (being an unabashed fan of orthodox game theoretic rationality myself.) But the revisionists (who I think are in the minority) defend their quite original program with great spirit. (There are large literatures that emerged in the late 20th century both defending and attacking some of the revisionist proposals, though my impression is that the revisionist program has fallen out of favor.) Perhaps more interestingly, some fans of revisionism argue that something like constrained maximization and/or evidential decision theory is at work when so many of us do cooperate in Prisoners’ Dilemmas or similar situations that are not repeatable or at best very unlikely to be repeated. There is a large and well-known body of experimental data on Prisoners’ Dilemmas and games like the Ultimatum game where a large fraction of the subjects follow “nice” strategies rather than what’s “rational”. A related practical phenomenon is tipping. In a great many cases we tip the people who serve us in restaurants, taxis and the like when we already have received their services and do not expect to meet them again. Why? Again, some defenders of revisionism suggest that these experimental and real world phenomena confirm their views. Do the revisionist theories merit a closer look, after all?
(5) Are the revisionist theories sufficiently general for purposes of grounding social morality? This is a line of criticism Gaus might have pursued. Throughout these sections Gaus focuses exclusively on the Prisoners’ and the Farmers’ Dilemma games to illustrate how the revisionist theories supposedly generate moral rules. Again I think Gaus is being generous since revisionists like Gauthier tend to focus exclusively on these particular games to illustrate the rationality of obeying moral rules. But what if the situation is structurally different from a Prisoners’ Dilemma. A number of authors have proposed using the Hawk-Dove game as a model of problems of property acquisition. Strategic form Hawk-Dove is similar to Prisoners’ Dilemma in that both players are better off at (C,C) than they are at (D,D), but in Hawk-Dove (C,D) and (D,C) are both strict equilibria. Extensive form Hawk-Dove has a unique subgame perfect equilibrium where Player 1 follows Extensive form Hawk-Dove has a unique subgame perfect equilibrium where Player 1 follows D and Player 2 follows C. What would the revisionist theories recommend in such games? One would think that the revisionist theories should recommend the players follow (C,C), same as in the Prisoners’ Dilemma. But this is not at all clear!