Computing consensus: Language ideological work in
LLM
‐assisted deliberative democracy
Janet E. Connor Abstract
This article examines how the digital platform Polis and its experiments with large language models (LLMs) reconfigure democratic participation through a particular vision of consensus. Through analysis of media coverage of Polis and developers' public discussions of the platform, I argue that developers reframe consensus from a discursive achievement to a computationally surfaced pattern, turning deliberative democracy into a computing problem which LLMs are perfectly fit to solve. Under the guise of “saving” or “improving” democracy, Polis produces “illusions of consensus” that depoliticize purported public opinion and further abstract it away from the individuals assumed to have produced it.