I think there's sometimes a lot of marketing and hand waving in this type of argument "crashes go down with autopilot". Most car accidents are caused by drunk or old people, and they drag the average up. if you tell me a Tesla autopilot beats a drunk guy it won't surprise anyone. Now as a non-drunk, young(ish) driver but experienced and careful, my statistics don't look anything like the average, they look at lot better. You have to convince this demographic, not beat the averages otherwise it's not rational for me to buy the feature. I'm guessing this goal post is a lot harder.