Using AI to reduce distortive factors in Inquiry Dialogues

Inquiry Dialogues aim at warranted judgment rather than merely the expression of personal opinion. This aspiration makes them vulnerable to distortive processes that undermine the quality of the judgments they produce. In this discussion, I identify four such processes: informational cascades among participants, a lottery over facilitators, the participant lottery, and social pressure that suppresses the expression of unpopular viewpoints. I offer a theoretical defense of an intervention designed to address all four: having participants investigate their initial judgments in conversation with a carefully prompted AI chatbot before discussing those judgments with their peers. I argue that this intervention, which refines a noise-reduction technique developed by Yarmel and Gregory (2025), addresses each distortive process by approximating the kind of interlocutor that silent reflection, worksheets, and human facilitation cannot reliably provide at classroom scale. I defend criteria that prompts must satisfy, present a prompt developed through iterative testing, and report a proof-of-concept classroom exercise whose results were sufficiently promising to warrant systematic empirical investigation.