their early stages of development they will have very different profiles of cognitive strengths and weaknesses (though, as we shall later argue, they could eventually overcome any initial weakness). Furthermore, the goal systems of AIs could diverge radically from those of human beings.
