Humans aren't agents - what then for value learning? — AI Alignment Forum