The Epistemic Impact of Large Language Models on Policymaking

ABSTRACT Knowledge is central to policy sciences and for good policymaking. The rise of artificial intelligence (AI)—particularly large language models ( LLMs )—raises critical questions about their impact on how policymakers analyze data, design alternatives, predict outcomes, and implement solutions to address societal challenges. While much of the public policy literature highlights LLMs' potential to automate and transform policy cycles, less attention has been given to how LLMs reshape the epistemic foundations of public policy—namely, how knowledge is sourced, produced, and applied within policymaking and institutional systems—along with emergent epistemic risks. Building on Lasswell's functional and pragmatic view of the policy sciences and drawing upon concepts of LLMs as epistemic technologies, this article proposes a heuristic framework to articulate the epistemic impacts of LLMs on policymaking. Specifically, the framework explores how LLMs transform policy epistemics through first‐order effects—changes in knowledge production itself—and second‐order effects—shifts in power dynamics within policy networks. The article concludes by outlining a future research agenda to advance public policy scholarship in the age of AI.