How would you prove that something was a real AI without centralization?
Proof of AI would likely rely on servers and be government or corporate controlled.
What would be the purpose of this in general?
To advance AI or to create a new warfare system?
So I guess then that it wouldn't need to be necessarily be AI. It would need to be a cryptographic task requiring neural network level processing, or abstract computation. Humans would be welcome to compete too, and might even stand a chance until supra-human neural nets became common. Consider that we already have superhuman neural nets in many specific tasks like Go.
So, it's basically still Proof of Work, but strictly performed by artificial and/or human
neural networks as opposed to ASICs and other mining devices? If yes, then it could be
done, and the AI network would improve over time, but it would only be specializing in
whatever the Work task is. This would also be centralized, I think, since you still need
to prove neural network is true, when participating.
Is there something specifically you want this neural network to be doing?
A type of specific Work? Brain mapping or something?
