I read Homo Deus recently and it's not too encouraging. I assume AI can keep building on what it learns and eventually learn to improve itself beyond what we can do. Then it's a matter of how much autonomy it has and how it wants to treat us. Science fiction has played out plenty of scenarios, but we cannot think like something that is way beyond us.
I just wonder how far off it is as people have been working on it since computers were invented. The current 'AI' chat bots are just mining text rather than creating something new, but when you effectively have access to new knowledge you can find things that have not been put together before. People will play with this stuff anyway.
!PIZZA