I wasn't really thinking about driving skills, but about equipment failure. There will always be a risk of equipment failure for automated machinery. Also, there's a risk of human error in the code governing driving.

One possible workaround is to have several people at some centralized location monitor driverless car performance. A monitorer could override the car and drive the car remotely in the event of equipment failure or bugs in the code.

But, maybe that's a pipe dream.

You say that like it's a sure thing. As excited as I am about the potential, I think it's still to be determined how well robots drive in real world environments, alongside humans, at scale.