'defect' only applies to prisoners dilemma type problems. that is just one, very limited class of problem, and I would argue not very relevant to discussing AI inevitability.