AgentsBench: A Multi-Agent LLM Simulation Framework for Legal Judgment Prediction

The justice system has increasingly applied AI techniques for legal judgment to enhance efficiency. However, most AI techniques focus on decision-making outcomes, failing to capture the deliberative nature of the real-world judicial process. To address these challenges, we propose a large language model-based multi-agent framework named AgentsBench. Our approach leverages multiple LLM-driven agents that simulate the discussion process of the Chinese judicial bench, which is often composed of professional and lay judge agents. We conducted experiments on a legal judgment prediction task, and the results show that our framework outperforms existing LLM-based methods in terms of performance and decision quality. By incorporating these elements, our framework reflects real-world judicial processes more closely, enhancing accuracy, fairness, and societal consideration. While the simulation is based on China’s lay judge system, our framework is generalizable and can be adapted to various legal scenarios and other legal systems involving collective decision-making processes.