Note on Superrationality
November 30, 2020
In game theory, a superrational player is a perfectly rational (that is, utility maximising) player with the added characteristic that they treat all other players as superrational, where superrational players always play the same strategy in the same scenario. This idea (coined by Douglas Hofstadter in a chapter of Metamagical Themas) models an agent that reasons according to Kant’s categorical imperative: to act only according a maxim that can be willed as a universal law.
Unlike a perfectly rational player, in a prisoner’s dilemma a superrational player will cooperate. This is striking to me, as it gives some texture to a thesis I’ve been mulling over for some time: that morality can be understood as a technology for beating prisoner’s dilemmas.
Within normative ethics, the categorical imperative is often presented as a procedure for deciding what’s the right thing to do, a question which frames the standard disagreement between deontologists and consequentialists. But the CI can also be approached from a metaethical direction. Irrespective of one’s position on normative ethics, it seems reasonable to claim that the CI represents our best intuition about how the moral ought is used. If this is the case then superrationality can be understood as a structure of agency implicit in the rules of the moral language-game, which is to say cooked into the use-conditions of moral vocabulary. As players of the moral language-game, we encounter one another and ourselves as superrational, simply in virtue of playing that particular game.
If this is right then it permits a reversal of the dominant order of explanation. Rather than taking our constitution as individual moral agents as a metaphysical given, then asking how we should act or interact, it can be suggested that our constitution as moral agents is a product of our participation in certain social-linguistic practices. This is a large part of why I think Kant often gets undersold in surface-level introductions to normative ethics. It should be remembered that his formulation of the CI takes place not in the context of normative ethics, but in a discussion of the metaphysics of morals. The CI is not just a procedure for answering the question “what’s the right thing to do?”—it belongs to the background practices that make this question possible to ask in the first place.
What is found in the constellation of concepts that form Kant’s metaphysics of morals—the authority to bind oneself to a universal law, the responsibility to treat others as ends rather than means, and the CI—is a statement of the primacy of the interpersonal. Without the interpersonal recognitive practices codified as implicit rules of the moral language-game there is no moral agency, full stop. From the Kantian perspective to treat another as a means to an end, even to a moral end (as the utilitarian is accused of), is not just to act immorally—it is to abandon the moral terrain completely. To do so is to change the game from one whose implicit rules materialise the universal rationality of the superrational agent to one in which instrumental rationality reigns. This is, in effect, to break the social-linguistic technologies foundational to moral agency itself.