q6zwUeWkZL@OpenReview

Total: 1

#1 Self-Play $Q$-Learners Can Provably Collude in the Iterated Prisoner's Dilemma [PDF1] [Copy] [Kimi1] [REL]

Authors: Quentin Bertrand, Juan Duque, Emilio Calvano, Gauthier Gidel

A growing body of computational studies shows that simple machine learning agents converge to cooperative behaviors in social dilemmas, such as collusive price-setting in oligopoly markets, raising questions about what drives this outcome. In this work, we provide theoretical foundations for this phenomenon in the context of self-play multi-agent Q-learners in the iterated prisoner’s dilemma. We characterize broad conditions under which such agents provably learn the cooperative Pavlov (win-stay, lose-shift) policy rather than the Pareto-dominated “always defect” policy. We validate our theoretical results through additional experiments, demonstrating their robustness across a broader class of deep learning algorithms.

Subject: ICML.2025 - Poster