Verbalized Bayesian Persuasion

• DOI: 10.48550/arxiv.2502.01587
• Publication Date: 2025-02-03
• Authors: Wenhao Li; Lin Yue; Xiangfeng Wang; Bo Jin; Hongyuan Zha; Bing Wang
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Artificial Intelligence (cs.AI); Computer Science - Computer Science and Game Theory; Computer Science - Artificial Intelligence; Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG)

Information design (ID) explores how a sender influence the optimal behavior of receivers to achieve specific objectives. While ID originates from everyday human communication, existing game-theoretic and machine learning methods often model information structures as numbers, which limits many applications toy games. This work leverages LLMs proposes verbalized framework in Bayesian persuasion (BP), extends classic BP real-world games involving dialogues for first time. Specifically, we map mediator-augmented extensive-form game, where instantiate receiver. To efficiently solve propose generalized equilibrium-finding algorithm combining LLM game solver. The is reinforced with techniques including commitment assumptions, obedience constraints, obfuscation. Numerical experiments dialogue scenarios, such recommendation letters, courtroom interactions, law enforcement, validate that our can both reproduce theoretical results discover effective strategies more complex natural language multi-stage scenarios.