If the automated driver system in Tesla cars has caused more than 1.13 deaths per 100 million miles driven, he would seem to have it backward.
No, he has it perfectly right. To illustrate, lets use something I like to call "math"
As of the closest thing we have to actual data, fatalities in autonomous vehicles right now are approximately 4.5 deaths per 100 million miles. As you can see, the current fatality rate for autonomous vehicles is about 4 times the human rate. It seems like they are killing more people than saving so what gives? The secret is in the rate of improvement. Human drivers have shown an approximately 25% reduction in deaths per mile every decade. That means it took 40 years to go from 4.5 deaths per 100mmi to 1.13 deaths per 100mmi. As of 2014, Google has estimated that if its vehicles were allowed to operate at full speed the fatality rate would have been around 50-100 per 100mmi, which is a whopping 91% reduction in 2 years. assuming they maintain that rate of improvement for 5 more years, the human fatality rate will be about 0.988 fatalities per 100mmi while the autonomous vehicle fatality rate will be approximately 0.02 fatalities per 100mmi.
Assuming a linear relationship rather than geometric (to make the math a lot simpler without affecting the answer much), and we will get and average fatality rate for humans of 1.059 fatalities per 100mmi, whole autonomous vehicles would average 2.26 fatalities per 100mmi. Sounds bad, but what about the 5 years after that? humans: 0.8645 fatalities per 100 mmi. Autonomous vehicles: 0.01 fatalities per 100 mmi. Given an annual death toll of 35k, that amounts to an excess of 34,500 deaths per year for every year after 5 years from now. If we delay that progress for 5 years, then in the mean time an excess of 34,500 people will die for every year that autonomous vehicles are delayed.
Those number are far more pessimistic in that they assume that we are talking about a 100% conversion to autonomous vehicles right now. Since we are in fact talking about a vanishingly small percentage of actual vehicles initially, the cost in lives for the development of autonomous vehicles is practically non-existant, while the end state savings are phenomenal. Ultimately, Automakers should be given a 5 year moratorium on liability so long as they can demonstrate sufficient continued improvement in performance (say 33% annual reduction in fatalities per mile.). After 3 years it will be safer to be in an autonomous vehicle. After 10 years driving a car manually would be tantamount to drunk driving.
The question of how safe autonomous vehicles are is a stupid question, as it is obvious to nearly everyone who knows which end of an equation is which that we *will* reach a point where autonomous vehicles are safer, and the sooner we get there the less lives will be lost before we get to have the benefits.
The surest way to halt that progress is through liability stemming from the over application of our American legal framework. Our legal system almost entirely fails to consider the larger implications of any given law, in favor of the application of precedents and procedures. You could make a very real argument that a single liability lawsuit, at this stage in the game could delay the deployment of autonomous vehicles by a decade or more, which would effectively condemn over 300,000 people to die needlessly
The automakers need to be given a temporary liability shield, in exchange for which they must demonstrate continued improvement in autonomous safety performance, and after 5 years they must release *all* hardware designs including source code to the public domain so that the other automakers can implement these life saving systems properly. After 10 years it should become illegal to sell any new vehicle that does not have full autonomous capability, and after 15 years, it should be illegal to operate a vehicle manually on public roads.