So a claim how autonomous driving system beats the average would only tell us that it beats 5% of the human drivers.

Now, the way stats are massaged here is not even about "drivers", but miles driven, and this language is even worse. We'd need to make sure we are looking at human-driven miles in the same area, same roads, with similar cars.

>The median one might be better, but does it even matter? The average driver is still wreaking havoc.

Yes it matters. To be acceptable this technology needs to be at least in the same ballpark as a median-ish person on a median-ish day. Not some nonexistent average that is pulled down by the 1/X people who are drunk and the 1/Y who are from Socal and driving in Maine in a blizzard.

The fact that you basically never hear of "average non criminal driver" or "median law abiding driver" and that there is no real attempt at even standardizing a concept of normal drivers not engaged in bad behavior just reeks.

It's like the door is intentionally being left open for the same slight of hand as when people peddle some policy goal having to do with school shootings and back it up with statistics that are mostly normal crime. Or they are peddling some devious tax that will screw a whole lot of people, and they justify it with an average that's dragged way up by a few oddballs, or dragged way down by a bunch of zeros. Seems like the safety crowd and and self-interested industry are setting up to play off each other in a "recyclable plastic" sort of way.

Second off, what are you talking about that the "average driver is wreaking havoc"? The average driver is filing a collision claim every 15-20yr depending on who's numbers you believe. While I don't know the distance between average and median, either is a fairly high bar that Waymo and friends have to meet.