Because the software industry has such a sterling reputation for delivering complex systems that are nonetheless extremely reliable!
Or at least, they should. Even the very newsworthy recalls of recent years have only had consequences for very small numbers of people, across millions/tens of millions of vehicles (which still needs addressing, but the amount of problem being mitigated per recall dollar is growing smaller over time).
(The early evidence is that they treat it as a serious engineering problem, but I'm sure no hands on the wheel needs a lot more complexity to fail safe than anything that is shipping)
The interesting questions are: - what is the likelihood of failure? - what are the likely failure modes? - are they designed to fail safe? - are there backups to the fail safes? - are there backups to the backups?
And of course complex systems will always have unpredictable failure modes. Failures due to interplay with complex external systems that couldn't possibly be predicted.
And so on and so on.
If you see a car moving weird, you can make stuff to reduce the risk of an accident happening, when a self driving car suddenly turn and kill you, you weren't able to do anything.
Now I love the intellectual challenge to it, however I don't think we'll ever see the today's situation only with self driving cars, way too many cars. My guess is self driving stuff will improve a lot public transportation and people will only travel that way.
Here is a fact: the systems will fail from time to time. Maybe a lot.
The reaction to deaths that occur due to automated systems will be like Biblical wailing and gnashing of teeth. It will be utter outrage.
Americans' (perceived) outrage to tragedy seems to be to usually be totally out of sync with the reality of the threat. It won't matter if they are safer statistically. In addition to more efficiency, the driverless cars need to have the best-designed fail-safe systems ever constructed or we will not see widespread adoption within decades, at least in the US.