It's not supposed to make the problem disappear. The problem doesn't exist frequently enough to do anything with. How much code has to be put in for all of the

It's not supposed to make the problem disappear. The problem doesn't exist frequently enough to do anything with. How much code has to be put in for all of these edge cases? How much data needs to be learned to possibly and nowhere near reliably deal with these edge cases? How often will these edge cases occur? It's not worth the false positives, not by a long shot. You cannot make self driving cars perfect, that's not possible. You just need to make them better than human drivers.