Enhancing Decision-Making with Large Language Models through Multi-Agent Fictitious Play

The paper introduces Multi-Agent Fictitious Play (MAFP), a novel multi-agent system paradigm designed to tackle decision-making tasks characterized by stance entanglement, where agents' decisions are interdependent. MAFP employs a game-theoretic approach, allowing agents to iteratively adjust their strategies based on the collective decisions of others, thereby enhancing decision quality and robustness. Evaluation results show that MAFP surpasses existing baselines in terms of tournament strength and robustness, highlighting its potential utility for practitioners dealing with complex decision-making scenarios in AI applications.