Not all human values come through equally in training AIs. New research sheds light on this and how we should tackle it in ...