Language models are now used as research tools and information gatekeepers. We study the ideological and moral priors they carry — and what those priors do to science and to citizens.
LLMs that annotate incivility or ideology may judge by ideological distance, distorting research on polarisation.
Assistants and summarisers frame political issues, and can narrow the perspectives people see.
LLMs embed latent ideological representations that systematically shape their outputs — affecting both science built on them and the information citizens receive.