Optimin achieves super-Nash performance

TL;DR

This paper introduces super-Nash performance as a new benchmark for AI in mixed-motive games and demonstrates that the optimin solution concept guarantees super-Nash payoffs in all n-person games.

Contribution

It proposes a novel benchmark and solution concept, optimin, for evaluating AI performance in complex social interactions beyond zero-sum games.

Findings

Super-Nash performance can be achieved in all n-person games.

The optimin solution guarantees super-Nash payoffs even under unilateral deviations.

A new benchmark for AI in mixed-motive settings is established.

Abstract

Since the 1990s, AI systems have achieved superhuman performance in major zero-sum games where "winning" has an unambiguous definition. However, most social interactions are mixed-motive games, where measuring the performance of AI systems is a non-trivial task. In this paper, I propose a novel benchmark called super-Nash performance to assess the performance of AI systems in mixed-motive settings. I show that a solution concept called optimin achieves super-Nash performance in every n-person game, i.e., for every Nash equilibrium there exists an optimin where every player not only receives but also guarantees super-Nash payoffs even if the others deviate unilaterally and profitably from the optimin.