I think you missed the poster's overall point: how people perceive accidents matters. This is independent of what the actual causes are, and if there are less of them. When automation is involved, people may perceive that someone did something wrong, even though that automatic thing outperformed what a human could achieve.
I think manufacturers are already facing damages for those remaining 3000 fatalities. I assert "deep pockets" is the overriding factor.
I'd really like to get a sampling of those 3000 non-driver-error fatalities.
I think the term of art is "vehicle factors." Using that I found http://www.ncbi.nlm.nih.gov/pubmed/17916886 which suggests that manufacturers could address "the majority" of them, and I don't think I'm being crazy by suggesting self-driving vehicles already include a lot of them, since they aren't exactly Geo Metros.
Self driving cars will get into accidents that are not currently "vehicle factors" since they will be superior drivers, not perfect drivers. Just imagine the headline: "A computer killed my son."
I've long said that people would rather a human kill 1000 people than a computer kill one.
