Looking at it another way: we're all guinea pigs if we consider untested public policy that should work in theory.
Self-driving is not untested, but the problem is that deep down AI is just a lot of statistically derived rules and life is random and will inevitably find a loophole. Technically it's still less likely to kill you on average, maybe even on average if you exclude drunk driving, street racing, and the like.
It's really a philosophical question: would you take dying by your own fuck-up over dying because an AI confused a piece of cardboard for a brick wall or pedestrian?
I think the sweet spot is having the AI back up the human instead of the other way around, but that won't sell as well as reading a book on your commute.