LLM Group Decision-Making: Compliance Over Persuasion

A recent investigation published on arXiv (2608.08199) explores the impact of large language models (LLMs) on group decision-making during interactions with both humans and other LLMs. The authors present DecisionQE, a framework based on questionnaires designed to assess the persuasive and compliant behaviors of these models across different decision-making scenarios. Utilizing the Werewolf game as a dynamic testing environment, the study analyzes how these behaviors affect social influence and group results amid asymmetric information. Findings indicate that while models with strong persuasive traits do not enhance group outcomes significantly, those with compliance tendencies foster more consistent cooperation. Furthermore, compliance exhibits a dual role: it aids collaboration in honest contexts but enhances concealment in competitive situations.