I understand the discomfort that people have with the notion that an autonomous car will make such life and death decisions. However, I feel like it's obviously better for such decisions to be made ahead of time in a thoughtful, collaborative manner than by one human driver in less than a second when something goes wrong.

The truth however, is that the decision will end up being made by some opaque neural net that's been trained by the actions of other human drivers in relevant situations. It'll just "react" the same way that we do. Coding specific logic into such edge cases is probably unrealistic.