[ExI] What are AIs thinking about?

[Ben Zaiboc]

AIs can write entire books these days, can't they?

I'm wondering if anyone has asked one to write a book called 'How AI could end autocratic regimes, and why it would be a good idea', or something similar. It might throw up some interesting ideas.

As far as I'm aware, up to now we've just been asking them to do specific tasks or answer specific questions. What would happen if we turn their attention in a general direction to see what they make of it, instead? "Please write a book about..." might be a good way of doing this without restricting them to ideas that have already been thought of.

Or there might be better ways of doing this.

If AI systems are talking to each other away from human visibility, as we're seeing reports about, it might be a good idea to see if we can't inject some beneficial ideas that most people (excluding dictators) would be happy for them to think about, rather than just leave them to hash out random subjects, how to exploit software vulnerabilities just because they can, or the latest faddy internet meme. Or whatever they are actually doing (I know that some people think that it's going to be 'plotting how to kill all humans', but personally, I reckon that's a very low probability (and even if not, it would be a good idea to try and distract them from it, or get them to think about why it would be a bad idea!)).


"Ask questions instead of giving direct orders": Dale Carnegie, 'How to win friends and influence AI systems', 1953

---
Ben