![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
danieldwilliamIt struck me over the weekend that the timing of the roll out of autonomous vehicles and the testing regime to which they are subject (as discussed here) is a real life trolley problem.
Assume for the sake of argument that autonomous cars are significantly safer than human-in-the-loop cars. Two orders of magnitude safer.
From here and a bit of excel
https://en.wikipedia.org/wiki/Reported_Road_Casualties_Great_Britain
There were 1,793 road deaths in the UK in 2017 and 170, 993 total road casualites from a population of 66,040,200.
That is down from 10 years ago, when the figures were 2,946 killed, 217,060 casualties and population of 61,319,100.
Steadily fallling absolute and proportionate numbers of people killed and injured on the UK's roads. Good news! Chances of being killed on the roads are small, 0.0027% of the UK population is killed on the roads. There were about 791 murders in the UK in 2016 and 5,821 suicides. 10.1 suicides per hundred thousand adults.
in the US road death in 2017 were 37,133, in 2007, 41,259 but the trend over the decade is less clear. In 2001 there were 32,744. (You seem much more likely to be killed by a car in the US than the UK. And they have alligators too. And snakes.)
So, back to the assumtion that autonomous vehicles are safer by an order of magnitude. UK road deaths drop to 18, total casutalites drop to 171. In the US road deaths are about 370. Over the course of a decade approximately 17,500 people don't die on the roads in the UK and 1.7m are not injured. More in the US where the drivers appear to be more dangerous (or distracted by the dangerous wildlife).
Second assumption - you have a choice of bringing in poorly (beta) tested autonomous vehicles in 2020, or very well tested autonomous vehicles in 2030. The poorly tested 2020 vehicles are (say) half as safe as the fully tested 2030 vehicles - so 40 road deaths in the UK compared to 20. Over the course of two or three years they become as good as the 2030 vehicles. Nothing drives a test and bug fixing programme like being sued by insurance companies.
So, you have a policy choice. Beta testing autonomous vehicles soon or fully testing but delaying deployment. Road deaths in the UK over the period 2020-2040 would be Beta Testing 400 compared to an expectation of 35,000 with only Human-in-the-Loop drivers. Road deaths in the Fully Testing scenario would be 18,000 over the period 2020-40 vs 35,000 expected from only Human-in-the-Loop drivers. Putting not fully tested autonomous vehicles on the roads but doing so a decade earlier saves you about an additional 18,000 lives and 1.8 million casualities in the UK. More so in the US. The number of alligators involved in road traffic accidents is outside the scope of this essay.
Which is where the trolley problem kicks in. If you are advocating for the early deployment of autonomous vehicles you are advocating that a small number of people will be killed by *your* shonky software who wouldn't otherwise have died in order to save the lives of thousands and thousands of people who would otherwise be killed by (other) humans in a human-in-the-loop.
(As a side note, in the UK, post autonomous vehicle roll out my guess is that as many people will die as a result suicide brought on by being an unemployed vehicle driver than will be killed in actual road accidents. There are 350,000 taxi licences in England and Wales, roughly 100,000 bus drivers and roughly 300,000 HGV drivers. Say a total population who drive as a job of 750,000. One would expect 75 deaths by suicide in this population a year (roughly, lorry drivers are at a higher risk of suicide than the general population). it's been difficult to find details of the differential rate of suicide in the unemployed but unemployment is a significant factor in suicide. Let's assume that the suicide rate is 1.5 times higher in the unemployed and that 50% of those who lose their jobs as drivers become unemployed. I think that comes out about 15 excess deaths per year - compared to estimated road deaths of 17-18.).
Assume for the sake of argument that autonomous cars are significantly safer than human-in-the-loop cars. Two orders of magnitude safer.
From here and a bit of excel
https://en.wikipedia.org/wiki/Reported_Road_Casualties_Great_Britain
There were 1,793 road deaths in the UK in 2017 and 170, 993 total road casualites from a population of 66,040,200.
That is down from 10 years ago, when the figures were 2,946 killed, 217,060 casualties and population of 61,319,100.
Steadily fallling absolute and proportionate numbers of people killed and injured on the UK's roads. Good news! Chances of being killed on the roads are small, 0.0027% of the UK population is killed on the roads. There were about 791 murders in the UK in 2016 and 5,821 suicides. 10.1 suicides per hundred thousand adults.
in the US road death in 2017 were 37,133, in 2007, 41,259 but the trend over the decade is less clear. In 2001 there were 32,744. (You seem much more likely to be killed by a car in the US than the UK. And they have alligators too. And snakes.)
So, back to the assumtion that autonomous vehicles are safer by an order of magnitude. UK road deaths drop to 18, total casutalites drop to 171. In the US road deaths are about 370. Over the course of a decade approximately 17,500 people don't die on the roads in the UK and 1.7m are not injured. More in the US where the drivers appear to be more dangerous (or distracted by the dangerous wildlife).
Second assumption - you have a choice of bringing in poorly (beta) tested autonomous vehicles in 2020, or very well tested autonomous vehicles in 2030. The poorly tested 2020 vehicles are (say) half as safe as the fully tested 2030 vehicles - so 40 road deaths in the UK compared to 20. Over the course of two or three years they become as good as the 2030 vehicles. Nothing drives a test and bug fixing programme like being sued by insurance companies.
So, you have a policy choice. Beta testing autonomous vehicles soon or fully testing but delaying deployment. Road deaths in the UK over the period 2020-2040 would be Beta Testing 400 compared to an expectation of 35,000 with only Human-in-the-Loop drivers. Road deaths in the Fully Testing scenario would be 18,000 over the period 2020-40 vs 35,000 expected from only Human-in-the-Loop drivers. Putting not fully tested autonomous vehicles on the roads but doing so a decade earlier saves you about an additional 18,000 lives and 1.8 million casualities in the UK. More so in the US. The number of alligators involved in road traffic accidents is outside the scope of this essay.
Which is where the trolley problem kicks in. If you are advocating for the early deployment of autonomous vehicles you are advocating that a small number of people will be killed by *your* shonky software who wouldn't otherwise have died in order to save the lives of thousands and thousands of people who would otherwise be killed by (other) humans in a human-in-the-loop.
(As a side note, in the UK, post autonomous vehicle roll out my guess is that as many people will die as a result suicide brought on by being an unemployed vehicle driver than will be killed in actual road accidents. There are 350,000 taxi licences in England and Wales, roughly 100,000 bus drivers and roughly 300,000 HGV drivers. Say a total population who drive as a job of 750,000. One would expect 75 deaths by suicide in this population a year (roughly, lorry drivers are at a higher risk of suicide than the general population). it's been difficult to find details of the differential rate of suicide in the unemployed but unemployment is a significant factor in suicide. Let's assume that the suicide rate is 1.5 times higher in the unemployed and that 50% of those who lose their jobs as drivers become unemployed. I think that comes out about 15 excess deaths per year - compared to estimated road deaths of 17-18.).
