Microsoft AI CEO Mustafa Suleyman says that smarter-than-human AIs capable of self-improvement, complete autonomy, or independent goal setting would be "very dangerous" and should never be built. He says others in the field "just hope" that such an AI would not harm us.