>But I've yet to see any acknowledgement that this is useful. I think that the reason why this isn't acknowledged (by people who should know) is likely to be a

>But I've yet to see any acknowledgement that this is useful. I think that the reason why this isn't acknowledged (by people who should know) is likely to be a political one. Examining the solution that I typically seen given for this problem should explain why. First, the assumption is that driving algorithms are updated to account for any edge case (for instance, an unpredictable animal jumped out this time) after it occurs, so that case is then guaranteed to not re-occur. This presents a problem: anyone who's ever worked with software before knows that that's an impossibility until a sufficient number of incidents happen in nearly identical ways, or until a fix is manually applied (which itself could take a very long time). But we fall back on the second part of our argument, statistics. The fact that you're going to crash into something you wouldn't before is outweighed by more than one other person not crashing in a place where the computer can predict and avoid a crash more effectively than a human. This too, presents a problem, if you also consider the implication that the availability of manual-driving cars will likely be degraded in some way after the introduction of self-driving ones (from having increased cost/regulation/insurance premiums or simply being banned entirely). If we discount these "weird" situations and fail to make allowances for them, people will get hurt where they wouldn't otherwise (even if the local or global sum of deaths is reduced). The benefits skew in favor of cities where traffic behavior is/will be much more consistent. So it's best if local knowledge can't be useful- why complicate the matter or feed skeptics talking points if you don't need to?