i think to prove it, you'd need to define what you mean by AGI.

There's something like the universal approximation theorem (https://en.wikipedia.org/wiki/Universal_approximation_theore...) for neural nets, if that's what you're looking for!