If the operator is in control of the car, liability should be straightforward. Similarly, if the operator of the car is an overseer but has the ability to step in and avoid programmed behaviour it should be straightforward. If, however, there is no manual override and all people in the vehicle are "just passengers" then I would not like to be the owner of such vehicles in the US. Might turn out to be rather expensive when the first child/family/whatever gets run over.
> Might turn out to be rather expensive
Wait, so your main concern is about the financial aspect?
yeah, as I was discussing liability not the ethical nature of auto-driving. There are technical issues as to whether the code could be created properly and ethical issues about how true AI might end up decide that killing the passenger(s) was better than killing a crowd. That's a different discussion.