Good article, but just want to mention two things.

1. If we see on a higher level, neural networks essentially find patterns in data. There is logic and there's intuition. Neural networks are good at developing intuition from data. And, search algorithms make use of heuristic functions that we can think of as intuition. So, in the future, we may see more applications using neural networks as a component inside a search algorithm, which essentially would be a step towards combining logic and intuition in a more sophisticated manner. We may even end up developing good approximation algorithms for NP hard problems.

2. Regarding the Silicon valley scene of not having enough labelled data available (scraping is easy, annotation is boring), the field is seeing good results in that direction. Self supervised learning is one of the frontiers of artificial intelligence today.