Title:Super-Nash performance in games

Abstract:Since the 1990s, artificial intelligence (AI) systems have achieved 'superhuman performance' in major zero-sum games, where winning has an unambiguous definition. However, most economic and social interactions are non-zero-sum, where measuring 'performance' is a non-trivial task. In this paper, I introduce a novel benchmark, super-Nash performance, and a solution concept, optimin, whereby every player maximizes their minimal payoff under unilateral profitable deviations of the others. Optimin achieves super-Nash performance in that, for every Nash equilibrium, there exists an optimin where each player not only receives but also guarantees super-Nash payoffs, even if other players deviate unilaterally and profitably from the optimin. Further, optimin generalizes and unifies several key results across domains: it coincides with (i) the maximin strategies in zero-sum games, and (ii) the core in cooperative games when the core is nonempty, though it exists even if the core is empty; additionally, optimin generalizes (iii) Nash equilibrium in $n$-person constant-sum games. Finally, optimin is consistent with the direction of non-Nash deviations in games in which cooperation has been extensively studied, including the finitely repeated prisoner's dilemma, the centipede game, the traveler's dilemma, and the finitely repeated public goods game.