Nearly 40,000 people will die on American roads this year, but virtually none of them will be killed by a driver who, in a split second, makes a rational (or even instinctive) decision to hit one particular person instead of another individual or group. 99.99% of all accidents are caused by a driver who is, by definition, out of control. That's why people who are doing real work on AVs in particular or road safety in general look on 'trolley problems' as the province of amateurs.
But America has made a trolley problem-style choice when it comes to road safety. One fork led towards European style driver training, licensing standards, driver testing, and vehicle inspections; meaningful deterrents to distracted driving; and fuel prices that encourage public transit and discourage large personal vehicles. The other fork led to U.S.-style (nearly nonexistent) training & testing, and inspections of lumbering SUVs & pickups; half-hearted attempts to reduce texting while driving or increase helmet use by motorcyclists. It’s America’s policy to ignore the economic externalities associated with low fuel prices, which include what amounts to a subsidy for ridiculously large and dangerous SUVs and trucks.
Comparing trends in mortality rates between other developed nations and the U.S., it’s clear that the fork we chose accounts for at least half those 40,000 deaths.
American unwillingness to insist on reasonable driver & vehicle standards is the most powerful argument for allowing/encouraging autonomous vehicles on U.S. roads. Developing AVs (or even ADAS) is not, to be sure, the most cost-effective approach to improved road safety; it's just the one approach that might work in the land of 'live free, or die'.
The development of AVs opens up a lot of interesting topics for pseudo-philosophical discussion. The trolley problem is not one of them. But articles like the New Yorker’s recent take on trolley problems might really hurt the deployment of AVs and American road safety, by exposing AV companies to unnecessary liability.
As far as I know, no human driver has ever been charged with a crime, or found liable, for an instantaneous, trolley problem-style response during an accident. The closest thing to a classic trolley problem – and an accident that occurs too regularly on U.S. roads – happens when a driver on a two-lane road sees an oncoming car and instinctively steers away from it, to the right, collecting a pedestrian or bicyclist on the right shoulder. In those cases, drivers almost always get off lightly or scot-free. Juries rarely say, “You should’ve let yourself be side-swiped.”
As frustrating as that is, for those of us who defend vulnerable road users, the decision to veer away from a perceived threat is a natural instinct. If we refuse to train drivers well enough to anticipate those problems and avoid them, it’s not surprising that courts don’t punish that instinct.
It will be a long time before AVs make U.S. roads significantly safer, but that will happen long before American courts or bureaucrats hold drivers to a higher standard. I believe that, Uber’s experience notwithstanding, we’ll eventually see a day when autonomous vehicles, or improved ADAS, will anticipate and avoid almost all accidents I described above. If AVs reduce fatalities by 90%, will it matter if, once a decade, some autonomous vehicle actually chose to kill one person on a bicycle to avoid a woman pushing a baby carriage?
I think not. But if courts are influenced by pop culture’s ridiculous obsession with trolley problems, we may well see a day when AV makers are held responsible for decisions we never attempted to parse when they were made by human drivers. That liability could slow the deployment of AVs and will impact the cost of AVs, as manufacturers wrestle with the reality of being held responsible for the actions of their AI systems.