Center for Social Information Sciences (CSIS) Seminar

Risk-aversion and bounded rationality are two key characteristics of human decision-making in behavioral economics. Risk-averse quantal-response equilibrium (RQE) is a solution concept that incorporates these features, providing a more realistic depiction of human decision making in various strategic environments compared to a Nash equilibrium. We study RQE in both general-sum normal form games and discounted infinite-horizon Markov games. For normal form games we adopt a monotonicity-based approach to show the uniqueness of RQE and its Lipschitz continuity with respect to the payoff matrices as long as players' degrees of risk aversion and bounded rationality exceeds some threshold. For Markov games we first define the risk-averse quantal-response Bellman operator and prove its contraction under further conditions on the players' risk-aversion, bounded rationality, and temporal discounting. We finally present Q-learning and Actor-Critic based MARL algorithms with scalable adaptation using neural-networks that provably converge to the RQE for all Markov games.