> the machine learning stuff presently in use is well tailored to very narrow, specific problems It's more accurate to say the machine learning in use now work

> the machine learning stuff presently in use is well tailored to very narrow, specific problems It's more accurate to say the machine learning in use now works on narrow classes of problems. But even that is a giant leap forward from teaching the machine to solve each problem individually. Teaching a car to drive very specific routes is one thing (and has been effectively done for decades now). Teaching a car to drive entire classes of routes is something else, and where we are now. In a decade or so cars will drive all classes of routes better than humans, and that will be something else again.