What if we were actually trying to use technology to make cars safe?
For the avoidance of doubt, I don’t think that the primary of even secondary goal of autonomous car companies is road safety. A lot of people making the case for autonomous cars talk about how safe they will be, contrasting that to human drivers today. And it is sort of tautologically true—as the blogger Atrios has noted—that if autonomous cars work well enough to be deployed they will be safe. The risk of incidents for the companies producing them is too high for them not to deliver safe vehicles. But autonomous cars are a solution to the problem of automobile safety in the same sense that automobiles themselves were a solution to the problem of runaway carriage horses. If what you want is purely to stop death and injury via automobile, an autonomous car is an over-engineered bank-shot solution which brings along plenty of problems of its own.
I'm a technologist, for better or for worse, so I think a lot about how you can solve problems with technology. I've thought a lot about what it would look like if you were trying to devise a technological solution specifically for traffic safety. That is, what are the products and technologies that could plausibly be sold that would reduce or eliminate traffic deaths. Sort of a Vision Zero upgrade package. It turns out, a lot of the technologies that would be necessary already exist. Some are already part of regulatory mandates (albeit not in the United States). Some of the necessary technologies don't yet exist, but could be easily created with no particular technological innovation.
When I do this thought experiment I give myself a couple of ground rules. First, I'm trying to come up with technologies that would be basically congenial to drivers. I think this is important. Not because I think that the experience of driving a car is sacrosanct, or that it’s inappropriate to force people to drive a certain way.
Building, selling and driving cars are activities necessarily mediated by thousands of restrictions, regulations and constraints. The idea that owning and driving cars on public roads represents anything like "freedom" is pernicious fallacy. But if you come up with a solution that drivers and auto companies hate, the political and practical headwinds will usually be insurmountable.
The framework I like to work within, as I've mentioned before, is harm reduction. While we make it easier not to drive, but we accept that some people will drive nonetheless. How do we put the tools in place so that if people do choose to drive, they can do that without harming themselves or others? So for this exercise, solutions that are—or feel—punitive won’t work.
I’m not really interested in solutions which involve law enforcement or surveillance, like red light cameras. As Sarah Seo points out in her excellent book Policing the Open Road, the intersection of law enforcement and automobiles has been catastrophic for minority populations in this country. Solutions that defer to the discretion and judgment of law enforcement, which is what we have now, lead in this country to bad outcomes for minority populations.
Third, and maybe most importantly, I am only considering, in this thought experiment, technologies in which the human driver is fundamentally the operator of the car. Autonomous systems that work reliably enough to take over driving for humans with in an emergency situation do not presently exist. More important than that issue, though, is the fact that with a few (extremely important) caveats, humans are incredibly good at driving. This is easy to miss when you consider the chaos caused by drivers in the real world. Almost 40,000 people died in automobile crashes in 2020. There are horrifying incidents every single day involving cars crashing into pedestrians and other cars. Intuitively, it seems like autonomous driving systems could hardly fail to be better than human drivers.
What’s missing in that intuitive calculus is that its considering all drivers at all times. The overwhelming majority of accidents are caused by a human driver who is not driving to their best ability. Alcohol is a factor in a huge number of crashes. Excessive speed. Road rage. Fatigue or inattentiveness. If you take those out of the equation—which is to say, if you can convince human drivers to follow the law and drive to the best of their ability—you eliminate almost every crash. Skilled, attentive, calm human drivers are vastly more facile than the best autonomous systems, and will remain so possibly forever. Insofar as autonomous systems are capable of doing a better job of driving than a human it's because they never get fatigued or upset or inattentive and because they can be deployed with failsafe systems.
So the question of how to use technology to improve road safety really doesn’t resolve to how to replace human drivers. It resolves to a question of how to convince human drivers to operate to the best of their own ability. You don’t need a state-of-the-art autonomous vehicle to do that. The technology to do it mostly already exists. Those failsafes in autonomous cars exist. It isn’t technologically difficult.
One obvious place to start is speed governors. These have existed on cars for well over a hundred years. Many ebikes and scooters have them. In fact, many production cars have them, just set to a speed (often 120 or 160 miles per hour) where most people never hit the limits. There are a couple problems with traditional speed governors. One obvious problem is where to set the maximum speed. If you set it to eighty miles per hour, you'll eliminate a lot of highway speeding, but you won't do anything to stop dangerous driving in cities. If you set it to twenty-five miles an hour, the car is useless for highway travel. In addition, people make (mostly specious, but anyhow) safety arguments: sometimes it's necessary to accelerate out of a dangerous situation.
This is a place where technology can help. One of the features of autonomous and other vehicles that is now very robust is localization. Geofencing allows for features to be enabled only in certain areas, and modern mapping systems know the speed limit at every moment. Also, the electronic throttle in modern cars allows for very fine-grained control of how the vehicle responds to the gas pedal. It is easily possible to have a throttle response curve that changes depending on speed, acceleration, and the speed limit. Taken together, these systems allow for adaptive geofenced speed control: a vehicle that responds as normal until you get past the speed limit, at which point keeping the pedal down has gradually diminishing returns, such that you have to press it further and further to keep accelerating and, eventually, to keep the same speed. If you let off the pedal and slow down below the speed limit, the responsiveness returns.
People have been trying to add something approximately like this kind of system to European automobile regulations for years. It's called Intelligent Speed Assist. They’ve finally succeeded. There are an enormous number of caveats, though, not least that the regulations as currently written only call for a brief audible warning in a system that can be turned off.
The level of regulation involved in mandating these features in Europe seems—to me, at present—unthinkable in the United States. But I don’t think that’s the biggest problem. Audible warnings make for terrible safety systems. The way they work is by being annoying; if there's something in a car that's annoying the driver and the driver can turn it off the driver will turn it off. And a system that is designed to annoy drivers that can't be turned off will be battled at every turn by the car companies who are trying to sell vehicles that people want to drive. I don’t think Intelligent Speed Assist needs to be unpleasant. It could operate invisibly, as a feature.
You'd need some other method to get people to install these systems. You'd need to figure out how to get people to want to. One thought is that there are some subset of people who would much prefer to obey speed limits but feel constrained in doing that by social pressure. Maybe those people would be happy to install a system that would make obeying the speed limit easier and more pleasant. It’s one less thing to think about. It sounds nice to me. Or you could offer a feature to parents worried about teen drivers.
This is one example. Add a few fairly straightforward additional features—traffic signal and stop sign assist, driver attention monitoring, ignition interlocks—and you have a car that’s easy and fun to drive in which you won’t kill or injure anybody.
Emerging from the reverie of this thought experiment, I know of course that I’m kidding myself. We could design safety assist systems that are simple and pleasant to use, and which would eliminate the vast majority of traffic deaths and injuries. That part is true. The problem is, people aren’t interested.
We as a society would rather create a trillion dollar industry of, essentially, vaporware autonomous cars. The idea of avoiding tens of thousands of preventable deaths a year is seen as punitive meddling, no matter how seamlessly the technological approach is. The idea of a vehicle that makes it easy and effortless to be safe and obey the law is unthinkable. You have to couple that to the unrealistic fantasy of a robot butler to drive you everywhere you need to go before you can get any traction.
