Learning to Cooperate and Compete in Diplomacy. Dr. Noam Brown, FACEBOOK. (CS Colloquium)

Dr. Noam Brown
Will lecture on

Learning to Cooperate and Compete in Diplomacy
AI has made incredible progress in purely adversarial games such as chess, go, and poker. However, the real world involves a complex mixture of cooperation and competition, sometimes with irrational or suboptimal participants, and in these settings past AI techniques fall apart. For this reason, Diplomacy, a popular game focused on negotiation and alliance-building, has served as a long-standing grand challenge for AI that requires radically different techniques compared to prior games and has major implications if AI algorithms eventually succeed. In this talk I will describe Diplomacy and cover recent research results that have allowed our group to achieve expert-level human performance in the no-communication version of this game via a combination of imitation learning on human data and self-play reinforcement learning. In particular, I will cover a new regret minimization technique that converges to a coarse correlated equilibrium regularized toward a human prior. I will conclude with a discussion of the challenges that await as research shifts toward versions of the game in which players can privately communicate and therefore correlate their actions.