User blog comment:Nnn6nnn/Where Will Googology Go?/@comment-25337554-20180213032340/@comment-30754445-20180214074404

The concept of diagonalization can easily be taught to an AI.

The real challenge is knowing how to diagonalize. To do this you need to find a repeatable pattern in the original system, and this can be very difficult for humans as well.

Moreover, there's really no reason why an AI won't be able to do this. Indeed, the notion of a program that finds rules and patterns on its own, is practically the definition of AI. If your program can't do that, than it's just a really fancy calculator.