Interesting points. The problem with discussing the trolley problem as it applies to autonomous cars is that it conflates possibilities of the future with decisions that need to made today. Todays autonomous cars are not networked and not particularly smart. They are unlikely to know much about their own occupants, let alone information about the occupants of others. Their behavior in emergency situations is likely to be "avoid hitting/being hit by a solid object" and "if I have to hit a solid object, lets try not to make it a person or vehicle", although that second one may be future safety improvement. We can largely avoid the moral question of what is the moral behavior here (at least at in the short term) by pointing out that the machines aren't even capable of understanding the world well enough to adapt the behaivor, but are still (as the author points) far more staff in aggregate than non-autonomous systems.