I am more or less in the same boat as you. I do however think that a simple algorithm of "save people in the car" is better than "try to save objects identified

I am more or less in the same boat as you. I do however think that a simple algorithm of "save people in the car" is better than "try to save objects identified as people outside the car" because it is simple and in the long run likely very effective. Making computers make moral/ethical decisions is difficult, especially with poor inputs.