Exploring the Potential of Large Language Models (LLMs) to Simulate Social Group Dynamics: A Case Study Using the Board Game "Secret Hitler"

This study explores the capacity of large language model-powered agents to simulate human-like behavior in multi-agent social systems. Using Secret Hitler — a hidden-role board game centered on trust, deception, and strategic communication — we evaluate how LLM agents navigate dynamic group interactions. Our findings show that agents exhibit human-like behaviors, including strategic temporal adaptation, contextual reasoning, and complex social cognition such as theory of mind and implicit coordination. Notably, 85% of agent decisions factored in at least two other players’ mental states, highlighting their capacity for multi-agent mental state inference. However, they struggled with key aspects of human gameplay, including more nuanced strategic deception, emotional subtlety, and fluid conversational dynamics. These insights contribute to computational social science, agent-based modeling, and game theory, advancing our understanding of the potential of LLMs to simulate complex social interaction.