Collective deliberation is central to both effective group decision-making and pluralistic AI alignment. I’m presenting our recent study in Science in which we trained the “Habermas Machine”, an AI mediator to facilitate human deliberation. Drawing on participants’ (N = 5734) opinions and critiques, the AI mediator iteratively generated statements aiming to capture shared ground on social and political issues. Participants rated the AI statements as more informative, clear, and unbiased than those written by human mediators, and many revised their own perspectives in response. Text-embedding analyses revealed that successful statements balanced dissenting voices with prevailing consensus. These findings, replicated in a large-scale virtual citizens’ assembly representing the UK population, highlight how AI-driven mediation can enhance deliberation, promote mutual understanding, and guide AI systems toward inclusive, pluralistic alignment by respecting the complexity of human viewpoints.