Sure, but in the above user's hypothetical, that would mean that in such an area a concerned human driver with a greater ability to predict general human behavior would have a statistical safety improvement on the autonomous vehicle, which might not understand, for instance, that since it's a Friday night and the big game just ended and I'm in the city center I should be more careful than usual because there will be more intoxicated people.

Which is a lot of high level reasoning and inference with information from a variety of sources which aren't on the face of it strictly related to the driving task.