This project is funded by the Center on Long-term Risk, the research wing of the Effective Altruism Foundation (Award no. AWD019091).
This project aims to extend the methodology of empirical game-theoretic analysis (EGTA) in directions driven by and contributing to dynamic bargaining/negotiation problems. One motivation is to design and evaluate robust intelligent bargaining agents that can successfully participate in practical, multilateral, multi-attribute negotiation with varying degrees of knowledge on the environment and the strategies and preferences of other self-interested participants; the annual Automated Negotiating Agents Competition (ANAC) offers a suite of complex bargaining scenarios and can serve as a test-bed for such automated agents. This application domain will also advance fundamental research on the technical aspects of EGTA for extensive-form games (or, more generally, sequential games with incomplete/imperfect information), coupled with powerful deep reinforcement learning (dRL) techniques.