Because leaving that to emergent behavior is silly. Why is the idea of training models on ethical reasoning so strange?