Abstract
The recent wave of artificial intelligence, epitomized by large languagemodels (LLMs),has presented opportunities and challenges for methodologicalinnovation in political science,sparking discussions on a potential paradigmshift in the social sciences. However, how can weunderstand the impact of LLMson knowledge production and paradigm transformation in thesocial sciences froma comprehensive perspective that integrates technology and methodology? Whatare LLMs' specific applications and representative innovative methods inpolitical scienceresearch? These questions, particularly from a practicalmethodological standpoint, remainunderexplored. This paper proposes the"Intelligent Computing Social Modeling" (ICSM) methodto address these issues byclarifying the critical mechanisms of LLMs. ICSM leverages thestrengths of LLMsin idea synthesis and action simulation, advancing intellectual explorationinpolitical science through "simulated social construction" and "simulationvalidation." Bysimulating the U.S. presidential election, this studyempirically demonstrates the operationalpathways and methodological advantagesof ICSM. By integrating traditional social scienceparadigms, ICSM not onlyenhances the quantitative paradigm's capability to apply big data toassess theimpact of factors but also provides qualitative paradigms with evidence forsocialmechanism discovery at the individual level, offering a powerful toolthat balances interpretabilityand predictability in social science research.The findings suggest that LLMs will drivemethodological innovation in politicalscience through integration and improvement rather thandirect substitution.