Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf
This addresses the challenge of using LLMs in incomplete information games like Werewolf for AI and social science applications, but it is incremental as it builds on existing LLM capabilities without novel model changes.
The authors tackled the problem of engaging large language models (LLMs) in communication games without tuning their parameters, proposing a tuning-free framework that uses retrieval and reflection on past communications, and demonstrated its effectiveness in playing Werewolf with emerging strategic behaviors.
Communication games, which we refer to as incomplete information games that heavily depend on natural language communication, hold significant research value in fields such as economics, social science, and artificial intelligence. In this work, we explore the problem of how to engage large language models (LLMs) in communication games, and in response, propose a tuning-free framework. Our approach keeps LLMs frozen, and relies on the retrieval and reflection on past communications and experiences for improvement. An empirical study on the representative and widely-studied communication game, ``Werewolf'', demonstrates that our framework can effectively play Werewolf game without tuning the parameters of the LLMs. More importantly, strategic behaviors begin to emerge in our experiments, suggesting that it will be a fruitful journey to engage LLMs in communication games and associated domains.