Large-Scale Agent-Driven Social Simulation
August 22 (Friday) 3:30 PM-5:30 PM
Location: Nanjing Hall, 3rd Floor

Introduction:
       Li Yong is a tenured professor and doctoral supervisor in the Department of Electronics at Tsinghua University, and a member of the Ministry of Education's Changjiang Scholars Program. He has extensive research experience in data science and intelligence, having published over 10 papers in Nature journals such as Nat. Sustain., Nat. Mach. Intell., Nat. Hum. Behav., and Nat. Cities, as well as over 100 academic papers in CCF A-level international conferences and journals such as ICLR, NeurIPS, KDD, and WWW. His papers have been cited over 31,000 times, and he has received six Best Paper/Nominated Paper awards at international conferences. He has been selected as a global "Highly Cited Scientist" and a recipient of the National "Thousand Talents Plan" for Young Talents. He has also received the IEEE ComSoc Asia-Pacific Outstanding Young Scholar Award, the First Prize of the Ministry of Education's Science and Technology Progress Award, the First Prize of the Hubei Provincial Technological Invention Award, the First Prize of the Institute of Electronics' Science and Technology Progress Award, the Second Prize of the Institute of Electronics' Natural Science Award, and the Wu Wenjun Artificial Intelligence Outstanding Youth Award.

Report Title: Large-Scale Social Simulator

Report Introduction: 

        my country is currently in a critical period of transformation and development. Complex issues such as economic transformation, social contradictions, diverse demands, and the impact of globalization are becoming increasingly prominent, placing higher demands on the intelligent, scientific, and precise nature of social governance. Therefore, there is an urgent need for large-scale, accurate simulations of complex social systems to support the precise formulation and evaluation of social governance policies. Amidst breakthroughs in data, computing power, and algorithms, the "large-scale social simulator," once a grand vision only conceived by sociologists, has finally reached its AlphaFold moment. Against this backdrop, we propose the "Large-Scale Social Simulator AgentSociety." This simulator comprises three core modules: first, a social agent with human-like intelligence that integrates emotions, motivations, and cognitive abilities to enable individual decision-making to more closely align with real human behavior; second, a high-precision urban environment simulation system that accurately models the constraints and feedback of physical space and social resources; and third, a large-scale social simulation engine that enables efficient and scalable agent interaction and social behavior simulation. To validate the application value of AgentSociety, we conducted a series of computational social experiments, encompassing intervention experiments, interviews, and questionnaires. Focusing on four scenarios: opinion polarization, the spread and governance of inflammatory messages, the impact of universal basic income policies, and the social dynamics of hurricanes, we explored in depth the evolution of individual and group behavior under different social mechanisms. The experimental results were highly consistent with real-world results, validating not only the effectiveness of AgentSociety in social simulation but also its potential as a low-cost experimental platform for policy evaluation and social governance research. To further enhance the openness and usability of AgentSociety, we have simultaneously developed a visual online experiment platform that supports zero-code experimental design and simulation operations, lowering the technical barriers for social science researchers and policymakers, and promoting interdisciplinary research collaboration and the widespread application of intelligent governance tools.

Introduction:
        Qu Jingjing is an Associate Research Fellow at the Shanghai Artificial Intelligence Laboratory and the founder of the Epitome AI social experiment platform. Her current research focuses on AI for Social Science, trustworthy human-machine collaboration, human-feedback large-scale model thought chain intervention, multi-agent feedback intervention, and AI scenario innovation and risk control. She has been deeply involved in several major AI governance initiatives in China and Shanghai, and has led numerous projects under the Ministry of Science and Technology's 2030 Plan and provincial and ministerial-level projects.

Report Title: Epitome: A New Tool for Human-Robot Collaborative Experiments

Report Introduction:

        AI is a cultural-social technology (Science, 2025). How to better integrate AI with society, achieve specialized and interoperable integration in specific scenarios, and ensure trustworthy interactions requires a significant infusion of social science knowledge and extensive social experiments for testing and verification. AI for social science and the social science of AI research face challenges such as high technology development costs, difficulty organizing across disciplines, and complex experimental implementation. Currently, there are no tool platforms to support this type of research. Therefore, we are launching Epitome: the world's first open platform dedicated to the deep integration of AI and social science. The Epitome platform, through seven modules, provides a comprehensive, one-stack experimental solution spanning basic models, complex application development, and user feedback. The classic control-comparison causal logic used in social science experiments is embedded within a multi-layered human-computer interaction experimental environment, encompassing conversations, group chats, and multi-agent virtual scenarios. Its user-friendly canvas-style interface allows researchers to easily design and run complex experimental scenarios, enabling systematic research on the societal impact of AI and the exploration of comprehensive solutions.

Introduction:
        Lu Tun is a professor, doctoral supervisor, and vice dean of the School of Computing and Intelligence Innovation at Fudan University. He is also the deputy director of the Shanghai Key Laboratory of Data Science, the director of the Center for Social Computing Research at Fudan University, and a visiting scholar at Carnegie Mellon University (CMU). He is currently a distinguished member of the China Computer Federation, secretary-general of the Collaborative Computing Committee, and deputy director of the Collaborative Information Services Committee of the Shanghai Computer Federation. His research interests include social and collaborative computing, human-intelligence collaboration and interaction, crowd-intelligence collaboration and systems, large-scale intelligent agent simulation and deduction, and intelligent governance of digital societies. As the project leader, he has been responsible for numerous National Natural Science Foundation projects, Ministry of Science and Technology Key R&D Program projects, 863 projects, and Shanghai Municipal Projects. His research results have been published in leading conferences and journals such as CSCW, CHI, UbiComp, NeurIPS, ICLR, WWW, SIGIR, IEEE TKDE, and ACM TOIS. He has co-received the Best Paper (or nomination) award at several international academic conferences, including CSCW. He has served as an AC for CHI and CSCW, co-chaired the program committees of several well-known academic conferences at home and abroad, and served as an associate editor and editorial board member for several academic journals at home and abroad.

Report Title: RECOSIM: A Precise and Scalable Online Community Recommendation Simulation Framework

Report Introduction:

        Recommender systems, as the key engine of modern online communities, not only meet users' personalized interaction needs but also profoundly influence the distribution of topics and relationship structures within the community, guiding the self-renewal and continuous evolution of the community ecosystem. Recommendation interaction simulation, with its advantages of process visibility and experimental controllability, has become a crucial tool for modeling, understanding, and optimizing online community recommendation ecosystems and their dynamic evolution. This report will introduce RECOSIM, a framework for online community recommendation interaction simulation. This framework systematically decomposes the complex and diverse user-recommender system interaction process into five core modules: encoding model, decoding model, activity model, scoring model, and generation model. By designing mechanisms such as "post-encoding vectorized simulation + on-demand decoding during analysis" and "synchronously aligning interaction preferences with activity," RECOSIM overcomes the bottlenecks of existing methods in terms of cross-scenario generalizability, multi-round accuracy, and large-scale efficiency. We conducted extensive experiments on diverse online communities, including Sina Weibo and Zhihu, and systematically validated RECOSIM's effectiveness through multi-dimensional evaluations, including module-level and overall evaluations, and individual accuracy and group stability. Furthermore, this report will analyze the impact of different recommendation strategies (such as collaborative filtering, hot topic recommendation, and social recommendation) on user activity, topic attention distribution, and social network structure during community evolution. Finally, we will explore the application potential and future development directions of RECOSIM in academic research, production, and algorithm regulation.

Introduction:
       Lv Peng is a professor and director of the Center for Social Computing Research at Central South University. He is also vice president of the Wuhan Institute of Artificial Intelligence at Peking University and has been elected president of the Asian Society for Social Simulation (ASSA). He holds a Ph.D. in Sociology from Tsinghua University and a postdoctoral fellow in the Department of Automation at Tsinghua University. He holds a joint Ph.D. from the University of Chicago, USA, and has been a visiting scholar at Seoul National University, South Korea, and the Korea Advanced Studies Foundation (ISEF). His research focuses on intelligent social governance, large-scale social simulation, computational social science, and artificial intelligence culture and art. He has published numerous high-quality papers as first author or corresponding author in renowned journals such as Nature, Sociological Research, and Chinese Public Administration. He has been selected as a "Special Researcher" by the Cyberspace Administration of China, an Outstanding Young and Middle-aged Expert by the State Ethnic Affairs Commission, a National Key Social Science Expert, and a Young Changjiang Scholar by the Ministry of Education.

Report Title: Large-Scale Social Simulators and Community-Level Social Simulation