Earlier quoted context omitted.
A lot of people here aren't going to like it, but the only reasonable way out I can see is to eventually socialise ownership and control of AI. I don't see an alternative that isn't really bad.
Can you explain some of these alternatives that are so bad?
Another possibility is that, once AI exceeds human performance in all economically useful activities, including high-level planning, governance, law enforcement, and military actions, it discovers that the benefits of keeping humans around aren't worth the costs and risks.