If you think about it, isn't logic boiled down to making the best recommendation? When we make a decision its our logical minds making a recommendation based upon stimuli, previous similar decisions and that elusive 'intuition'. Neural networks already can do the first two, although the third might not ever happen and could be the distinguishing factor between human intelligence and artificial intelligence.
The problem is of course that neural networks currently can be 'the ultimate recommender' as you say, in only one highly specialized task i.e. image recognition, personal recommendations like music, or even self-driving cars. As far a I know there is no way to have a neural network learn two separate skills. Yet anyways.
The big change that I'm hoping will be figured out is how to connect all the specialized neural networks. Having a neural network that once presented with a problem 'recommends' another neural network to best solve it. The person/company that figures that out, no small task, will be an instant household name if not already.