关键词: 社会实验, 计算社会实验, 大模型, 社会模拟, 大模型智能体

Abstract:

Social experiments, as a typical research method in social sciences, aim to study specific social phenomena or the impacts of policies by observing the behaviors of individuals, organizations, or social groups in real or simulated environments. However, traditional social experiment methods often face challenges such as random bias, high costs, and ethical risks, making them inadequate to address increasingly complex research demands. Against this backdrop, computational social experiments have emerged, enabling researchers to conduct social experiments within computational simulation environments that are free from random bias, cost-efficient, and ethically manageable. Meanwhile, China is currently undergoing a critical period of transformative development, characterized by economic transition, intensifying social conflicts, diversified demands, and the challenges of globalization. These complex issues have surpassed the capabilities of traditional social experiments, placing higher demands on the scale, complexity, and authenticity of computational social experiments and their simulation environments. To address the dual needs of academic research and national strategies, there is an urgent need to construct large-scale social simulators capable of high-precision and large-scale simulations of complex social systems, supporting diverse computational social experiments and advancing the next generation of social experiments. This paper first introduced the concepts and methods of social experiments and discussed the necessity and application value of constructing large-scale social simulators from the dual perspectives of academic development and national strategic needs. It further reviewed the research progress of large language model technologies in simulating human behavior and analyzed their technical advantages in enhancing the authenticity of social simulations. Based on these insights, this paper proposed an overall framework for large-scale social simulators and validated its authenticity and broad applicability in social experiments in different fields through three typical cases: economic system simulation, social network simulation, and cognitive polarization simulation. Finally, this paper explored the future research directions and development trends of large-scale social simulators in areas such as the advancement of core simulation technologies, the construction of social science experimental platforms, applications in social governance, and the development of standards and policy regulation.

Key words: social experiment, computational social experiment, large language model, social simulation, large language model-empowered agent