# War and Peace (WarAgent): Large Language Model-based Multi-Agent Simulation of World Wars

**Authors:** Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, Yongfeng Zhang  
**arXiv:** [2311.17227](https://arxiv.org/abs/2311.17227)  
**Date:** 2023-11-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2311.17227v1/

## Key points

- Introduction of LLM-powered Multi-Agent System (MAS) for simulating historical events, particularly international conflicts
- Aim to unravel complexities shaping global order throughout history by modeling conflict and cooperation based on historical actors' characteristics and decision-making processes
- Importance of understanding mechanisms of war to unlock strategies for lasting peace
- Evolution of simulations in social science, leveraging LLMs to model complex behaviors and interactions
- Focus on World War I (WWI), World War II (WWII), and the Warring States Period (WSP) in Ancient China as case studies
- Key research questions include evaluating effectiveness of LLM-based MAS simulations, identifying triggers for war, analyzing network dynamics, assessing impact of variables on war and peace, conducting counterfactual analysis
- Use of cutting-edge AI technologies like WarAgent to offer data-driven insights for revolutionizing conflict resolution and peacekeeping strategies
- Potential application of AI in better comprehending human history and preventing future international conflicts

## Summary

In the era of groundbreaking advancements in Artificial Intelligence (AI) and Large Language Models (LLMs), a new frontier is being explored in computational social science research. This study introduces a novel framework of an LLM-powered Multi-Agent System (MAS) designed to simulate historical events, particularly international conflicts. By creating a dynamic environment where agents representing countries engage in conflict or cooperation based on historical actors' characteristics and decision-making processes, this research aims to unravel the complexities that have shaped global order throughout history. War and peace have long been intertwined in human history, influencing the course of civilizations. The intricate interplay between conflict and cooperation is often unpredictable, driven by various motives, strategies, and decisions made by nations. Understanding the mechanisms of war holds the potential to unlock strategies for lasting peace. While traditional methods of studying conflict through historical analysis offer valuable insights, they are constrained by their static nature and hindsight bias. The application of simulations in social science has evolved significantly over time, with early attempts limited by computational power and simplistic models. In contrast, contemporary simulations leverage LLMs to model complex behaviors and interactions across various domains. However, there remains a gap in applying these advanced technologies to simulate the nuanced nature of international diplomacy and war. This research addresses critical questions that challenge conventional understandings of historical conflicts by focusing on World War I (WWI), World War II (WWII), and the Warring States Period (WSP) in Ancient China as case studies. Key research questions include evaluating the effectiveness of LLM-based multi-agent system simulations in replicating historical strategic planning and decision-making processes, identifying critical triggers for war through simulations, analyzing network dynamics to understand alliances and conflicts, assessing the impact of variables on war, peace, and diplomatic interactions, as well as conducting counterfactual analysis. By delving into these questions using cutting-edge AI technologies such as WarAgent – an LLM-powered multi-agent AI system – this research offers data-driven insights that could revolutionize how we approach conflict resolution and peacekeeping strategies. Beyond historical analysis, these findings pave the way for utilizing AI to comprehend human history better and potentially prevent future international conflicts. The availability of code and data at \url{https://github.com/agiresearch/WarAgent} facilitates further exploration into this innovative approach towards understanding complex collective human behaviors within historical contexts.