Information and Contract Design for Repeated Interactions between Agents with Misaligned Incentives

Nanda Kishore Sreenivas, Kate Larson

Keywords: multi-agent systems, information design, Bayesian persuasion, contracts, misaligned incentives

Abstract: We investigate repeated interactions between a decision-making receiver agent and an informed sender agent who cannot directly influence the environment. Our primary focus is to determine whether both agents can learn strategies to maximize joint reward, even when their incentives are not fully aligned. We illustrate that the sender learns an effective signalling strategy that the receiver learns to act upon. We further explore the use of contracts, where the sender sells its information to the receiver. Our findings show that the sender learns to extract surplus reward from the receiver in such scenarios.

Submission Number: 21