Bayesian Persuasion for Algorithmic Recourse

Watch a 17-minute conference talk from the Symposium on Foundations of Responsible Computing (FORC) 2022 exploring how Bayesian persuasion can be applied to algorithmic recourse in automated decision-making systems. Learn how decision subjects strategically modify their features to improve their chances of favorable outcomes when assessment rules are kept secret, and discover how this creates a game-theoretic scenario. Understand the benefits of providing action recommendations to decision subjects, including how both parties can benefit while decision makers can achieve significantly better outcomes. Follow the development of a polynomial-time approximation scheme for finding near-optimal signaling policies, overcoming the challenges of infinite variable optimization. Examine numerical simulations on semi-synthetic data that demonstrate the practical advantages of implementing persuasion in algorithmic recourse scenarios.