AgentSociety¶
AgentSociety is a large-scale social simulator built on large model agents and first-principles. Through this platform, we can quickly create and manage agents in urban simulation environments, efficiently conducting modeling and simulation of complex urban scenarios. Thereby, AgentSociety, starting from the first principles of sociology, helps drive the transformation of social science research paradigms, promoting the development of sociology from behavior simulation to mental modeling, from static deduction to dynamic symbiosis, and from laboratory tools to social infrastructure.
The paper is available on arXiv:
@article{piao2025agentsociety,
title={AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society},
author={Piao, Jinghua and Yan, Yuwei and Zhang, Jun and Li, Nian and Yan, Junbo and Lan, Xiaochong and Lu, Zhihong and Zheng, Zhiheng and Wang, Jing Yi and Zhou, Di and others},
journal={arXiv preprint arXiv:2502.08691},
year={2025}
}
Features¶
🌟 Large Model-Driven Social Human Agents: Based on sociological theory, we build social agents with “human-like minds,” endowing them with emotions, needs, motivations, and cognitive abilities. These agents perform complex social behaviors such as movement, employment, consumption, and social interaction driven by these psychological attributes. We also support custom agents.
🌟 Realistic Urban Social Environment: It accurately simulates urban spaces critical to human social survival, replicating transportation, infrastructure, and public resources. This enables agents to interact under real-world constraints, forming a vivid social ecosystem.
🌟 Large-Scale Social Simulation Engine: Through adopting an asynchronous simulation architecture and the Ray distributed computing framework, it achieves efficient and scalable interaction and social behavior simulation among agents.
🌟 Social Science Research Toolkit: It comprehensively supports a range of sociological research methods, including various intervention techniques, data collection, and data analysis capabilities, facilitating in-depth social science research from qualitative studies to quantitative analysis.
Online Platform¶
We provide an online platform for AgentSociety, helping interested users quickly experience the simulation capabilities of AgentSociety.
Installation¶
pip install agentsociety
Refer to the Quick Start section for information on Prerequisites and Installation instructions.
In addition to the AgentSociety platform itself, we also provide some PyPI packages to extend the functionality of AgentSociety:
agentsociety-community: Community library for publishing custom agents and Blocks.
agentsociety-benchmark: Benchmark library for evaluating agent performance on various urban tasks based on the AgentSociety framework.
Use Cases¶
Visit GitHub Examples to see use cases.
Related Work¶
Based on the AgentSociety platform, a series of related works have been developed, including:
Jun Zhang, Yuwei Yan, Junbo Yan, Zhiheng Zheng, Jinghua Piao, Depeng Jin, and Yong Li. A Parallelized Framework for Simulating Large-Scale LLM Agents with Realistic Environments and Interactions, ACL 2025
Jinghua Piao, Yuwei Yan, Nian Li, Jun Zhang, and Yong Li. Exploring Large Language Model Agents for Piloting Social Experiments, COLM 2025
Nicholas Sukiennik, Yichuan Xu, Yuqing Kan, Jinghua Piao, Yuwei Yan, Chen Gao, and Yong Li. The Roots of International Perceptions: A Large-Scale LLM Simulation of US Attitude Changes Towards China, Submitted to AAAI 2026
Yuwei Yan, Jinghua Piao, Xiaochong Lan, Chenyang Shao, Pan Hui, and Yong Li. Simulating Generative Social Agents via Theory-Informed Workflow Design, Submitted to AAAI 2026
Jing Yi Wang, Jinghua Piao, and Yong Li. Does Reasoning Improve Rationality? Evaluating Reasoning-Enhanced LLMs Across Descriptive, Normative, and Instrumental Rationality, Submitted to EMNLP 2025
Contact Us¶
We sincerely invite scholars in the fields of social science, large language models, and agents to explore our platform. Researchers can contact us via email and submit their research proposals. Approved applicants will receive assistance and guidance from our team.
We welcome collaborative opportunities to advance social science research through our platform. Feel free to communicate with us via WeChat.
WeChat Group¶
