Beyond Preferences in AI Alignment