> a good fit on one problem, made with a universal approximator, is not an explanation. It can be just a compressed description,

I couldn't have said it better.