That is why there is always a "fail-safe" maneuver which - in the case of humans - is to do what most likely will protect itself. It is hardwired, a reflex. If you use AI, you will have to program one of these maneuvers: In the event I can't decide what to do, what do I do?No driver can, and neither could a self driving car, do the necessary calculations in the short time allocated.
For my part, when I face a moral dilemma, I can always count on The Simpsons to show me the way. And here is how The Simpsons believe AI should react in extreme cases: