I'd picture that as a hybrid system that uses some old AI ideas together with some new AI ideas. AlphaGo which combines a neural network-based player with a MCMS player that simulates a large number of games all the way to the end, is a good example.

The intellectual north star of this is

https://en.wikipedia.org/wiki/SAT_solver

and might be a better model than the "semantic web" in that the semantic web has revolved around description logics like OWL that implement a limited subset of first-order logic that is always decidable. One trouble is that OWL-based systems can't handle arithmetic at all, given a height and a weight you can write a rule that says anybody over 6 feet is tall but you can't say a person with a BMI over 30 is obese because OWL can't add, multiply and divide. But something a little more expressive that searches efficiently in practice because it gets "hunches" from a neural net could be useful but it will not evade the limits on computation and logic that Godel, Tarski and Turing warned you about.