Title: Reverse Engineering Cooperation With Bayesian Theories Of Minds
Abstract: Human cooperation is distinctly powerful. We collaborate with others to accomplish together what none of us could do on our own, share the benefits fairly, and trust others to do the same. What cognitive processes underlie our social intelligence, and how do these distinct abilities contribute to the scale and scope of human cooperation? How can this knowledge be used to design cooperative artificial intelligence that is equally capable of interacting with people? I will present a computational framework based on integrating individually rational, hierarchical Bayesian models of learning together with socially rational, game-theoretic models of cooperation. In computational models and multiplayer experiments, I will show how this framework can explain how cooperative behavior is generalized: inferring the intentions and reputations of others, distinguishing friend and foe, and coordinating AI agents and people.
