Impact of LLM-mediated political discourse on democratic dynamics

Characterize how political discourse mediated by large language models influences public opinion and democratic processes, specifically by determining whether and to what extent such systems shift Overton windows, amplify extremism, or destabilize democratic deliberation.

Background

The authors note that LLMs increasingly shape political communication and discourse, yet the systemic effects of model-mediated content on democratic norms and opinion formation remain undetermined. They frame this as a critical gap that existing technical benchmarks do not address.

This problem sits within the paper’s broader call for accessible, psychometrically grounded evaluation of LLMs, extending measurement beyond toxicity and accuracy to values and societal impacts. Establishing how LLMs affect Overton windows and extremism is presented as essential for informed governance and policy.