Oh, agreed, but you mentioned two use cases and kind of implied that humans have this innate ability to recognize when one might be a threat or not. Except that they don't, humans fail all the time in both of the cases you mentioned. The 'safer' approach, for both humans and autonomous vehicles, would be to travel at a rate that allows you to safely avoid something that 'could' happen. As I mentioned, humans are too impatient for this, which is one reason you see so many dead deer on the PA turnpike every year. The autonomous vehicle could avoid these, at the expense of some level of frustration for the human inside the vehicle.
Where's the tradeoff? The human overrides the system, and increases his chance of an accident.
bs