Polling people on ethical dilemmas and using their aggregated votes to train AI decision-making policies.