AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society

A research team from multiple institutions has developed AgentSociety, a groundbreaking large-scale social simulator that leverages large language models (LLMs) to create realistic simulations of human society. The system integrates LLM-driven generative agents, a detailed societal environment, and a powerful simulation engine to model complex social dynamics. In their experiments, researchers generated social lives for over 10,000 agents, simulating 5 million interactions both among agents and between agents and their environment. This represents a significant advancement in generative social science, replacing costly traditional experiments with scalable, replicable computational approaches.

AgentSociety serves as a comprehensive testbed for computational social experiments, focusing on five critical social issues: polarization, the spread of inflammatory messages, universal basic income policies, external shocks like hurricanes, and urban sustainability. These case studies demonstrate the platform's support for typical research methods including surveys, interviews, and interventions. The alignment between AgentSociety's outcomes and real-world experimental results validates its ability to capture human behaviors and underlying social mechanisms. This positions the platform as a valuable tool for social scientists and policymakers seeking to understand complex societal dynamics without the logistical challenges of traditional research methods.