When I was a teenager, I saw a comedian who I considered ancient, which likely meant he was sixty or so (teenage me was a little shit). His schtick was that his age made him humorously incompetent. He best bit, in my little shit opinion, was “I don’t understand the concern about elderly drivers. I’ve not had an accident in decades. Lots of accidents seem to happen behind me, but I’m fine.” Every time someone hypes the alleged safety of self-driving cars, that line comes back to me.
Timothy Lee, whose work I generally like, had a piece at Ars Technica that tried to make the case that at least some self-driving cars were already arguably safer than human driven cars. The headline was “Are self-driving cars already safer than human drivers?” The piece, though, was a disappointing exercise in repeating the comedian’s set.
I know that authors generally don’t choose their headlines, but the arch “just asking questions” tone was definitely prevalent in the article. Lee made a point of highlighting that Cruise has much farther to go than Waymo, to his credit, but in other ways he was much too credible. First, the data he relied upon was entirely self-reported. Obviously, the incentives are not aligned with being entirely honest. And we know that Google and Facebook (the parent companies of these self-driving cars) do not have records of complete honesty. Nor do car companies in general, nor do their main self-driving competitors. Even if they wanted to be entirely honest, there is a lot of pressure for them to fudge things enough to keep their cars on the road to improve. Finally, even if you assume the companies are completely honest, they aren’t required to report every kind of incident.
Lee builds an impressive edifice on top of that rather rickety structure, but the problems run even deeper. To mangle this metaphor even further, the subbasement is flooded: he doesn’t take into account the very real systematic problems that self-driving cars have to overcome. Streets are built for human drivers and the human ability to handle ambiguous situations. The cars assume a perfectly functioning modern infrastructure. The rather common occurrence of a wireless network being overwhelmed by a music festive caused fleet of self-driving taxis to melt down and freeze in place. Fortunately, they didn’t prevent anyone from getting access to emergency help, but that is a real risk.
We already know that self-driving cars are a danger for emergency vehicles, because emergency vehicles introduce ambiguity into the driving experience. And in the last couple of days, the worst case appears to have happened — a self-driving car got confused and blocked an emergency vehicle long enough that the delay contributed to the death of the person being helped. Cruise claims that the ambulance had an alternate route, but I find that hard to believe and more to the point, if a human had been driving the taxi, an alternate route would have been a moot point.
Looking at just the number of accidents per mile that is entirely the fault of the self-driving cars is not a sound measure. You need to look at the entire eco-system — how their odd behavior causes problems for other drivers, how their inability to handle ambiguity drives problems for emergency vehicles, how they need almost the entire infrastructure rebuilt for their specific needs. Looking at just one small corner of the puzzle will lead to inaccurate understanding.
The argument is that these self-driving vehicles will eventually be safer for all of us and that we need to allow them to make mistake son the roads in order to learn enough to get to that point. No one quite comes out and says this, but the implicit in that argument is that the damage they do, the deaths they cause, are worth the cost so that we can all benefit later. I do not find that a convincing argument.
First, I have yet to see much evidence that these vehicles are advancing fast enough to ever get to the point where they are safe, in the whole picture, to justify their damage they do while getting to that point. Second, it is clear that in order to function, almost everything about our driving infrastructure is going to have to be changed to accommodate these vehicles. If we are going to spend that kind of money, there are much better ways to spend it. From the perspective of the environment and public safety, we would be much better off reshaping our transportation infrastructure around public transportation and walkable cities rather than self-driving cars. If they cannot slide seamlessly into existing streets — and there is almost no evidence that they can — then they are a moral and practical dead end.
Driving augmentation systems have unquestionably been a boon to driving safety. Focusing on improving those systems would be work worth doing. But pretending that the leap form human augmentation to human replacement is a difference of degree rather than difference of kind is a kind of dishonesty. And by refusing to look at the whole picture when making your arguments, you lose trust and send yourself down roads that lead to dead ends and dead people.
Leave a Reply