Your second paragraph is very intriguing. I never really thought about this. I wonder if people will actually be able to restrict usage though. Its software, and historically it has been hard to restrict it. Of course cloud based systems have two advantages, software is hidden behind the API and they have really powerful systems. But the former requires a single lapse in security to leak and latter just requires time till consumer hardware can catch up. If I use past data to predict future (which might be a bad idea in this case), it might be almost impossible to restrict AI software.