[com]

I don't think your point 1 is as true as you think it is. Anti-drink driving campaigns seem to work: http://www.thecommunityguide.org/mvoi/massmedia_ajpm.pdf describes studies that identified significant (>15%) reductions in alcohol related accidents, with economic payoffs during the campaigns of more than 20x.

[skyhook_mockups]

That's certainly a LOT better than I would have thought, although still depressingly short of the 100% reduction that computer controlled cars would give us.

[com]

I doubt we'd see a 100% reduction of accident injuries and fatalities with automated control of cars - diminishing returns due to other uncontrolled factors like pedestrians, dropped loads, pathological algorithm responses in unusual situations etc. Of course, spending 6x as much on road-safety campaigns as in the referenced studies won't deliver anything like a 6x improvement either.