Numerous studies have been done with companies operating "self-driving" fleets and the accident rate per million miles is reportedly 1000-times less than a human at the controls. (Alas I can't find citations). Of course it makes news when a self-driving vehicle crashes, yet media chooses to ignore the 50,000 people killed on the roads in the US each year.

Remember, the AI doesn't have to be perfect, it just has to be better than a human.

Obviously they are not there yet, but baby steps. The lane following and adaptive cruise control in my Tesla totally makes driving less arduous, but I'm a diligent driver who always monitors it. I believe most crashes were owners not monitoring their cars. I've seen videos of full self-drive on "dirty" streets (i.e. not intensely pre-mapped streets) and it's incredibly impressive. Like 95% as good as a human and much more cautious at intersections.

Vision-based AI is becoming excellent at interpreting the real world, and I'm reminded of that recently when I let a Roomba vacuum robot map my house and subsequently clean it with minimal impacts. OK, my house isn't dynamic like a street, but the vision tech is fascinating.

According to Musk, full-self-drive is always "next year". But I do believe it is within 20 years. Easily.