As the last 20 years of AI research has ignored semantics I can't see it helping in a significant way.

I suspect it will be a while after driving requires zero rigor (or even participation) before computers could program themselves.

Frankly, given how hard it is to explain to another human how to do something why should we expect smart computers to be easier to manage? Unless...we use a highly structured manner of issuing instructions...i.e. a programming language!