TimS comments on How likely the AI that knows it’s evil? Or: is a human-level understanding of human wants enough?