A proposal for importing society’s values

Proposes using imitation learning on large language models trained on recorded deliberative democracy sessions to answer value-laden questions at scale. The approach involves recording human ‘mini-publics’ deliberating complex value questions with AI assistance, then training models to simulate these deliberations conditioned on diverse background perspectives. This enables low-latency, scalable approximations of democratic decision-making processes that would otherwise be prohibitively expensive to run with actual human participants.