A1274
Title: Learning and LLM methods for information design
Authors: Baoxiang Wang - The Chinese University of Hong Kong, Shenzhen (Hong Kong) [presenting]
Abstract: In real-world tasks, agents have their own goals and behave adaptively toward other agents. To thrive in those environments, an agent needs to influence others so their actions become more helpful and less harmful. Information design refers to such attempts to influence others by strategic communication of information. We discuss how the information strategy is learned through reinforcement learning. Powered by recent advances in LLMs, we further expand the problem to natural language spaces, which recovers the real-world complexity that was often oversimplified in previous works. We demonstrate LLMs' capacity for persuasion through game-theoretic learning, and discuss further implications on LLM reasoning and multi-agent LLMs.
