人工智能体对实证主义社会科学研究范式的变革:本体论、认识论与方法论视角OACHSSCD

With the rapid development of generative artificial intelligence and large language models,AI agents have increasingly become an important research object and methodological tool in computational social science,exerting profound influence on the paradigm of empirical social science research.Traditionally,empirical social science has relied primarily on observation and statistical approaches to explain social phenomena.However,AI agents based on large language models possess capabilities such as semantic understanding,reasoning and decision-making,and contextual interaction.These capabilities enable researchers to simulate individual behavior and group interaction within virtual social environments,thereby providing new theoretical tools and analytical perspectives for social science research.In this context,drawing on the theoretical framework of generative social science,this study examines the structural impact of AI agents on the research paradigm of empirical social science. From an ontological perspective,AI agents transform the fundamental logic of social behavior modeling.Traditionally,agent-based modeling typically depends on pre-defined behavioral rules and mechanisms designed by researchers,with social systems simulated through rule-driven processes.The emergence of AI agents enables actors to generate behavioral decisions through semantic understanding and contextual reasoning,allowing social behavior modeling to shift from a rule-driven paradigm toward one characterized by cognitive generation.Within this framework,social science research is able to reconstruct processes of social interaction in virtual environments and to focus more directly on the generative mechanisms underlying the formation of social structures. From an epistemological perspective,the introduction of AI agents reshapes the structure of knowledge production in social science.The presence of AI agents introduces a new type of actor into the knowledge production process—entities capable of simulating human cognition and social interaction.Leveraging the semantic comprehension and reasoning capacities of large language models,AI agents can generate human-like behaviors and interactions within virtual social environments,thereby contributing to the reproduction of theoretical mechanisms and the interpretation of social phenomena.Consequently,the production of social scientific knowledge increasingly exhibits a form of human-machine collaboration,in which researchers design scenarios and modeling frameworks,AI agents generate behaviors and interactions within simulated contexts,and theoretical interpretation and mechanism analysis subsequently produce social scientific knowledge. From a methodological perspective,AI agents provide new pathways for social science research,extending research practices beyond reliance on observational data and statistical analysis toward scenario simulation and artificial society construction.As research tools,AI agents integrate,to a certain extent,the empirical research paradigm based on large-scale data analysis and the generative paradigm represented by agent-based modeling.Researchers can therefore simulate individual behaviors and decision-making processes at the micro level,construct specific social scenarios to explore mechanisms at the meso level,and build artificial societies at the macro level to observe the dynamic evolution of social structures.Through this integration,social science research gradually develops a multi-layered framework that spans individual behavior simulation and large-scale social system modeling. This paper argues that AI agents driven by large language models are promoting the emergence of a new generative research paradigm in social science.Rather than replacing traditional positivist approaches,this paradigm builds upon both data-driven empirical research and generative modeling traditions,integrating semantic cognition,behavioral generation,and structural feedback mechanisms to construct a research pathway capable of simulating social mechanisms in virtual environments.Such a framework provides new theoretical tools for the study of complex social systems,digital governance,and public policy research,while also expanding the interdisciplinary dialogue between social science and artificial intelligence.