These sorts of edge cases are programming the computer to make decisions a human literally could not make because the time duration is so brief that a human couldn't think and react at all in time.
so while this feels like "mow down vs don't mow down" people, the real choice is:
1. human driver makes no decision within 300ms at best and whatever outcome fate decides is the result (worth noting that outside these philosophical edge cases, human drivers will be much more dangerous most of the time)
2. computer driver uses that 300ms to identify and take an action that would save driver. This is probably easier to do because the driver is stationary and their actions don't need to be predicted when factoring this
3. computer driver uses that 300ms budget to put the lives outside the vehicle ahead of the driver. People outside the vehicle are not stationary so their reaction has to be predicted which is harder and more expensive, so if the computer tried to prioritize them it would do a worse job of it
I think the marketing of this would be more successful if we literally said "we'll just randomize what happens inside the space where humans can't react so that we don't have to pick winners and losers", but that's of course a totally irrational position.
the most important thing is that regulations guide things to make car behavior PREDICTABLE so that cars can predict the behavior of other cars and people can also take their actions based on understanding of cars will act.