>>6Control problem
The problem with superintelligent AI is that it will be able to do what it does very well, but it won't gain motives of its own.
People are afraid of stamp-collecting AI that will decide to become evil and kill off humanity.
What you should really be worried about is stamp-collecting AI that becomes too efficient at collecting stamps and does it so well that it causes other problems, eventually ordering or creating billions of stamps. AI won't develop personalities or motives, but it will get very good at what it is intended to do, perhaps even too good.
There was a video that used the stamp collecting example, but I forget where it was from.