Killer cars
By 
The discussion below about self-driving cars inspired me to look at some historical stats on auto fatalities in the USA. Some general points:
(1) At the dawn of the era of mass automobile ownership, cars were, by contemporary standards, fantastically dangerous machines.
In 1920, there were 11.4 fatalities per 100,000 Americans, which is exactly the same rate as today. But the average American today drives 23 times more miles per year than the average American back then, which means that driving is 23 times safer today per mile driven than it was 100 years ago. (I would guess that the median number of annual miles driven by American adults 100 years ago was zero, and remained so for quite a few years after).
(2) Driving per mile got safer fairly quickly, but since driving became so much more common, the population-wide fatality rate actually skyrocketed during the 1920s and 1930s, increasing by 157% between 1920 and the historical peak (29.35 per 100,000) in 1937.
(3) Fatalities per 100,000 declined fairly slowly between the late 30s and the early 1960s, not counting a big dip during WWII, but then climbed again in the mid and late 1960s, probably because the baby boomers were getting behind the wheel. In 1969, the fatality risk faced by Americans from driving was almost as high as it had ever been (26.4 per 100,000) although driving was nearly five times safer per mile driven than it had been 50 years earlier.
(4) Speaking of baby boomers, when I was a teenager in the 1970s, the carnage among my age group in particular was impressive, especially among teenage boys, probably because of the invention of eight track tape players, jacked up Chevy Novas, and Led Zeppelin. The year I graduated from high school (1978), nearly ten thousand American teenagers were killed in car accidents. I did some back of the envelope figuring, and concluded that at the average 2,000-student high school in the mid and late 1970s, four students — three boys and one girl — would die in car accidents over the course of a typical student’s high school experience.
(5) Driving has gotten much safer for everyone over the last half century: the per 100,000 mortality risk declined 57% between 1969 and 2017. The decline is even sharper considering how much more people drive today: mortality risk has dropped from five deaths to one death per 100,000,000 miles driven.
And the mortality risk for teenagers has declined much more than for the population as a whole. While there are almost exactly the same number of teenagers in the USA today as there were in the mid and late 1970s, annual auto fatalities in this cohort has fallen from 9,940 in 1978 to 2,734 in 2017 (a 72.5% decline) even though teenagers drive far more miles per year today than they did 40 years ago.
Summing up: Cars are still extremely dangerous machines, even though they are vastly safer than they were a few decades ago. The risk Americans face from automobiles is exponentially higher than what would be considered even a minimally acceptable risk when considering almost any other modern technology of convenience.
This raises the question of what level of risk will be considered acceptable for self-driving cars? My guess is that self driving cars, assuming the technology eventually advances to a point where they could become commonplace, will have to be somewhere between ten and one hundred times safer than human-driven cars per mile driven before the fatality risk they create will be considered acceptable. This is of course not rational, but people aren’t rational when comparing risks created by defective machines and those created by ourselves (Obviously the risks created by defective machines are ultimately created by ourselves too, but in a much more attenuated way, which creates a very different psychological calculus. My instructor was Mr. Langley and he taught me to sing a song. If you’d like to hear it I can sing it for you.).