Feynman was such a majadero. That or his pseudo-mathematical derivations of negative probabilities - like a Monte Carlo discovery generator, he occasionally said something which turned out to be right, at least approximately (e.g. our quantum computers have nothing to do with his idea, but he's still cited in every related text).
Regarding ML, many psychologists make career out of shouting that machines can't learn and never will. I think the results in RL are impressive and sooner or later they will realise learning as psychologists understand it. Using Feynman's catchy phrase, they are now at the level of a cargo-cult learning. I'd say they are now at the level of Leonardo da Vinci's flying machine - we need a plane. Termodynamics. New science. You get it? I'm sorry, I'm falling asleep.