Fuzzy analogies will get you nowhere.

> And the issue of trying to prove N=NP is essentially trying to prove that there isn't a magical way to train a neural network with just one image of training data

No it's not. You can't do reasoning on bullshit inaccurate analogies.