The trouble is when we swap "human error" for "machine error". Because humans screw up all the time but a machine is going to fail in a way that'll likely be very obvious to us. That is going to be take an emotional toll, I think, and will be a hurdle to get over for self driving cars.
No one wants to be told "sorry but your family died because of a bug in the car's software told it to make a left into a semi".
AI and perfect AI are different things.
I said AI wouldn't have the issues of human error, distracted and drunk driving. It opens up all new types of errors, sure, but it wouldn't have the biggest killers of people. So unless we screw up pretty badly here it almost has to be lower mortality rates.
This is a big one that most people seem to miss: A self-driving car doesn't have to be better than an alert, conscientious human driver to be better than 90% of the actual complacent, distracted, tired humans on the road.
I don't know why I have to point this out, but: betting is different than knowing.