>"When people think about how AI might ‘go wrong’, most probably picture malevolent computers trying to cause harm. But what if we should be more worried about them seeking pleasure?"