Re: Minsky's "Weather analogy" pretty much nails the whoel "Deep learning" business.
The way I see it is this:
You have one entry point, and three exit points in a maze. There are millions of purmutations allowing one through. You observe a cockroach going into the single entry point over thousands and thousands of iterations and predict which exit they come out of.
You may be able to predict to a reasonable degree, using 'deep learning', which exit they come out of based on observation of the entry. But you wil never know why.