I've very mixed feelings with all the craze around deep neural network and learning. On one side, some results brings so much magic it's astounding and really looks promising. On the other side, anything that is not a use case of convergence from general to specific, like hand writing recognition, but general to general, like qualifying pictures with vocabulary, fails miserably (in a very entertaining way).<p>And the slip from neural networks to artificial neural networks to artificial intelligence we see on the broad news, really make it look like expert systems all over again. At first, it's said that it can solve any kind of problems, and then, we end up with a very narrow set of problems it solves reliably.