The Nightmare Scenario
This is something I've believed for decades. I grew up in one of the last true streetcar suburbs in the country, a bucolic (and now unaffordable) place which puts paid to the lie of the necessity of car dependence for pleasant (if slightly boring) suburban living. I was in no hurry to get my license as a teen. Then and now my favorite way to get around a city was by bicycle. Riding a bicycle in a city teaches you in a vivid and existential way the chaos and violence always latent in interactions between automobiles and the world around them. In my adult life I lived for several years in Los Angeles, a vibrant, omnicultural megalopolis visibly writhing against the catastrophe of automobile. As many things as I loved about living in California, when I returned back east to Boston—fairly glossed as an oft-resentful cultural backwater with lousy weather—my day-to-day life was far more satisfying because I didn't have to drive.
One way to explain the value proposition of autonomous cars—particularly robotaxis—is that they will make taking trips by automobile in urban centers more pleasant, more seamless, and less aggravating. This is a goal that I have never supported; some of the implications of aiming for it could lead to a wave of urban desecration as painfully misguided as that which accompanied the advent of the automobile. Attempting to avoid that outcome was my primary personal justification for the work I did in autonomous cars.
When I think about the rollout of autonomous cars it is almost entirely about how to avoid replicating and expanding the great original tragedy of the automobile. What do I mean by that? When automobiles became widely available it became clear that they would need dedicated infrastructure. Streets, once shared relatively freely between horse-drawn vehicles, pedestrians, streetcars and bicycles, became automobile-only thoroughfares. Sidewalks, once a gracious amenity for pedestrians wanting a raised, dry surface for strolling, became narrow, restrictive alleys, walled off and often obstructed by cars: a grudging concession to the foolish few who still walked. Crosswalks, in theory a space where control could be at least temporarily ceded back to people on foot, instead resemble the fording of dangerous rapids. The particular tyranny of cars' inherent dangerousness and massive physical presence boxed out people outside their cars from the great majority of their cities' available public space. At the same time, the drivers and passengers of those cars were trapped in their own noisy, dirty congestion. To mitigate the problems with cars—a goal many working autonomous vehicles are earnestly committed to—the aforementioned are the problems you should be solving. Instead, autonomous vehicles could easily make things much worse.
How could autonomous cars make car dependent-cities worse? As I described last time, my central insight about autonomous cars was that they lacked theory of mind: the ability to understand what people in the world were thinking. I had this insight as I was crossing a road with my young daughter, walking her to preschool. I was watching a car in the road, trying to determine if they understood that we both had the right of way and were claiming that right. Autonomous cars, I realized, would be unable to play their part in that kind of negotiation. Either they would have to stop and wait for any pedestrian that could possibly cross in front of them (the situation that mostly obtains now) or they would have to proceed through pedestrian interactions with inadequate information, the companies responsible for deploying them hoping for the best.
This is not a tenable situation. As I thought through what would happen if autonomous cars lacking this ability were deployed widely, I could see only chaos. I didn't—and still don't—think that the companies deploying these vehicles could tolerate the risk of proceeding with inadequate information. A crash of a human-driven car is a possibly-understandable tragedy. A crash of a Waymo autonomous car is a human being injured or killed by a robot owned by a multinational corporation. The brand risk of incidents is too high for these companies. This played out in practice with Uber ATG's tragic killing of Elaine Hertzberg in Arizona. While every effort was spent on many fronts to blame the safety driver of the vehicle—even subjecting her to criminal prosecution—the blowback on Uber ATG was sufficient to end their internal efforts to develop an autonomous vehicle outright.
So if the autonomous vehicle companies can't take the risk of their vehicles proceeding with inadequate information, and the necessary information -- theory of mind -- is an aspect of the world they are blind to, they have no choice but to stop for any pedestrian that might get in their way. This situation—which is how things are with current deployments—isn't tenable either. Nobody will be interested in riding in a robotaxi that drives halting, circuitous routes and regularly stops in situations where a human driver never would. The promise of a smooth, private, perfectly relaxing autonomous robotaxi trip will be dashed against the rocks of physical reality.
That situation of thwarted expectations bears a notable resemblance to the situation in the early part of the 20th century. Then, as now, a new transportation modality was being developed and promoted primarily to the well-off: those who could afford automobiles back then, those who can afford taxis or Ubers as a primary mode of transportation now. Then, as now, the eager early adopters of the transportation modality saw it more for what it was promised to be—a vehicle of surpassing technical sophistication whisking you to your destination in unprecedented style and comfort—than what it actually was, a dangerous and unreliable new technology with predictably negative knock-on effects. Then, as now, municipalities were eager to cater to the needs and desires of the socially connected elite who were most invested in this technology.
In one of our early pitch meetings, the venture capitalist to whom we were pitching asked why it was even necessary to have autonomous vehicles co-exist with humans. Soon enough, he said, all vehicles would be autonomous, and you would no longer have this problem. Well, we answered, at the very least, pedestrians will still exist. His suggestion was that they could be shunted onto newly-built underpasses and overpasses, so they would never have to share the roads now given completely over to autonomous cars.
I wouldn't say that this solution, such as it is, to the problems with deploying autonomous vehicles is generally regarded as optimal in the autonomous vehicle industry. But it's out there. And if your perspective is that we should be trying to reclaim our cities from cars, it's something like a vision of Hell. Having given our streets almost entirely over to cars driven by humans, we would now surrender them for our own nominal convenience to robots. Instead of being trapped as drivers of cars in cities, we would be trapped as passengers. The roads would stay choked with traffic, the environmental cost of cars would stay fixed, but alternate modalities of transportation—walking, biking—would be rendered practically impossible.
This isn't the only way it could go. The scenario I've laid out here is a vivid dystopia, but not a necessary one. You can tell a story where autonomous cars are a part of cities that have been reclaimed for human-scale uses. I’ve told that story myself, and even partially believe it. But the dystopian “solution” of giving our roads over to autonomous systems outright has been mooted seriously with some regularity, not just by that one VC, and the historical parallels with the advent of the automobile are clear to see.
It is an eminently possible nightmare scenario, and it’s what I think about every time somebody argues that autonomous cars are likely to be a solution to the problems human-driven cars create.
