Not in itself, but as a tool, in the manner of better telescopes and microscopes. And people don't invest a lot of thought on approaches that are utterly infeas

Not in itself, but as a tool, in the manner of better telescopes and microscopes. And people don't invest a lot of thought on approaches that are utterly infeasible, until tech makes them feasible - plus, there's much learning in trial-and-error, experiments, tinkering, which you just can't do if it's too slow. Unfortunately, much of current statistical machine learning does not help its developers to see, unlike a *-scope. It's just a blackbox. See Chomsky vs Norvig http://norvig.com/chomsky.html http://norvig.com/chomsky.html