And the answer is an obvious no, not even close, with the current state-of-the-art in deep nets. I could forgive an opaque model if it actually performed well.