About Creating AIs with Human-Compatible Values, with Mark Stefik