They may not be safer than human drivers. And they’ll make gridlock worse.
By David Zipper, The Washington Post
David Zipper is a Visiting Fellow at the Harvard Kennedy School's Taubman Center for State and Local Government, where he examines the interplay between transportation policy and technology.
The buzz is back for self-driving cars. Not long ago, blown forecasts left investors and tech enthusiasts feeling deflated: In 2016, Ford planned to offer self-driving taxis by 2021, and Lyft claimed it would start doing so even earlier.
But such unfulfilled promises now lie in the rearview mirror. “It’s beginning to feel like 2016 again,” Axios proclaimed last month. During the CES electronics show in January, General Motors, Chinese automaker Geely and autonomous-vehicle developer Mobileye all promised that individuals would be able to buy their own self-driving cars this decade, potentially as soon as 2024.
If that happened — emphasis on the “if” — it would be groundbreaking. Right now, there are no self-driving cars available for public purchase (even if Tesla misleadingly calls its driver assistance system “Full Self-Driving”). But if and when autonomous vehicles finally arrive en masse, should we cheer, or should we worry? It’s a good time to ask an even more fundamental question: What exactly is the point of self-driving cars? The answer, despite more than $100 billion in investment over the last decade, according to McKinsey, remains surprisingly nebulous. And that should trouble us — a lot.
One hundred and 20 years ago, early adopters of “horseless carriages” — what we now call automobiles — were often affluent men who flocked to the vehicles as a futuristic, high-tech form of entertainment. But it didn’t take long for their potential societal value to become clear, especially in urban areas.
At the time, American streets were filled with horses that required stables and regular feeding, and that left pavement covered in smelly waste. (Horses deposited about 1 million pounds of fresh manure on New York City streets every day, according to Christopher Wells’s book “Car Country.”) Cars replaced these messy animals and also allowed cities to spread out, mitigating dangerous overcrowding. While their drawbacks, like smog and traffic deaths, were substantial, automobiles offered real chances for advancement.
Like cars, autonomous vehicles were born not from public need but from technological opportunity. Phil Koopman, an engineering professor at Carnegie Mellon University who has worked on autonomous technology for more than 25 years, saw classmates developing prototypes that lumbered through Carnegie Mellon’s campus when he was a doctoral student in the late 1980s. “They weren’t solving a societal problem,” he says. “They were solving the ‘It would be cool if we could get cars to drive themselves’ problem.”
Early tinkerers didn’t focus too much on the eventual applications of self-driving technology, but the military did. Specifically, autonomous vehicles offered a chance to limit risk among men and women in uniform — a tantalizing possibility, especially in the wake of the invasions of Iraq and Afghanistan. The Defense Advanced Research Projects Agency (DARPA) organized the first Grand Challenge for autonomous vehicles in 2004 in the Mojave Desert, offering a prize of $1 million to the team whose machine first completed the 150-mile test track. No entrant came close: The farthest any vehicle got was just over seven miles. But the competition captured the imagination of the private sector, even though no one was sure what the civilian purpose of a self-driving car might be.
Enter Google. In 2007 the company hired Sebastian Thrun, leader of the Stanford team that won the second DARPA Grand Challenge, to launch a new self-driving-car initiative that would later form the basis of the company known as Waymo. Planning to deploy its technology on public roads, Google claimed that the rationale for self-driving cars was safety. In a 2010 corporate blog post, the company zeroed in on the 1.2 million annual traffic fatalities worldwide. “We believe our technology has the potential to cut that number, perhaps by as much as half,” Google claimed. It was quite a vision — and one whose appeal seemed unassailable from the political right or left. Other AV players followed Google’s lead, pointing to safety as the raison d’etre for their own self-driving cars.
In 2015 the National Highway Traffic Safety Administration offered those claims a tail wind when it published a memo stating that human error (along with other factors, such as flawed road engineering or dangerous car design) played a role in 94 percent of traffic collisions. Later, NHTSA’s nuanced finding was often boiled down to “94 percent of crashes are caused by human error” (a misinterpretation that the Utah Department of Transportation calls a “staggering fact”). Recognizing a windfall when they saw it, automakers and AV companies placed that 94 percent figure at the center of their marketing pitches.
Unfortunately, the potential safety benefits of self-driving cars are probably overblown. For one thing, many other problems cause crashes besides human error, such as confusing intersections and the blind spots of tall SUVs. But even with that significant caveat aside, it’s unclear whether an autonomous driving system will ultimately be safer than a person behind the wheel.
While self-driving cars won’t drive drunk or sleepy, the limitations of their machine learning will inevitably lead to mistakes that human drivers wouldn’t make. As an example, Koopman recalls a self-driving system a few years ago that struggled to identify the color yellow. “The system was missing bicyclists in yellow coats and construction workers in yellow jackets,” he says. “The system was 99 percent accurate — and it still missed them.”
Koopman believes that the safety hype around AVs is exaggerated. “There’s nothing I’ve seen showing whether AVs will be safer than humans in the short to medium term,” he says. “Machine learning is brittle, and it struggles with things it hasn’t seen before.” Especially in cities, seeing something unprecedented — like a woman chasing a duck with a broom, a scene a Google car apparently encountered — is quite common. (Notably, self-driving trucks on highways may be more viable than self-driving cars in cities because of the simpler road environment.) And system errors have already proved deadly: A prototype AV from Uber struck and killed Elaine Herzberg while she walked her bicycle across an avenue in Tempe, Ariz., in 2018. Afterward, Arizona’s governor suspended Uber’s ability to test its autonomous technology there, and two years later, the company sold its self-driving unit to Aurora, a start-up.
Police on March 21 released video footage taken moments before an autonomous Uber SUV hit and killed a 49-year-old pedestrian in Tempe, Ariz., on March 18. (Tempe Police Dept.)
Further complicating safety arguments is the question of whether car companies would design autonomous vehicles to break traffic laws. Until the federal government forced a recall on Tuesday, Tesla’s Full-Self Driving system allowed cars to automatically do illegal “rolling stops” instead of coming to a complete halt.
If road safety is the goal, there are already plenty of available technologies that automakers could invest in, rather than leapfrogging to fully autonomous vehicles. Take advanced driving assistance systems (ADAS), such as automatic emergency braking and pedestrian detection. Such features have already been shown to save lives, but their capabilities vary wildly from one model to another — assuming they’re available at all. According to the Insurance Institute for Highway Safety, automatic emergency braking wasn’t installed on 57 percent of the vehicles produced by Stellantis or 42 percent of those from GM during the 12 months ending last Aug. 31. Improving and standardizing ADAS is relatively low-hanging fruit compared with building self-driving cars, with more assured safety benefits.
Other proposed upsides of self-driving cars are also suspect. Will they boost productivity, allowing employees to take work calls and file memos from the road? That seems unlikely: One study concluded that only 1 in 4 drivers even wants to work during their commute. Could they free up space currently allocated to on-street parking? Maybe — though Amsterdam and Paris have shown that cities can do that without relying on autonomous vehicles. Will they expand transportation options for those who are vision- or mobility-impaired? Possibly in rural areas, but in cities and suburbs, people can already call a taxi or hail an Uber.
And then there’s the argument that autonomous vehicles can help fight climate change. Tara Andringa, the executive director of Partners for Automated Vehicle Education (PAVE), a national coalition educating the public about AVs, says the technology could provide sustainability benefits, “if we ensure that AV tech is limiting fossil fuels and that we are moving away from a system where each person has their own car.” Liberation from car dependence is part of the appeal of “robotaxis” and autonomous shuttles. Unfortunately, little evidence supports it. Rather than share self-driving rides, studies suggest that people want to travel in their own car, alone (which is consistent with companies’ struggles to make pooled ride hail profitable). Automakers have noticed; General Motors, for example, seems to be shifting away from a vision of shared autonomous rides toward one of personal car ownership.
A Ford Fusion with Argo AI autonomous driving software in New York in 2019. Some automakers are promising that self-driving cars will be available for the public to buy within a few years. (Mark Kauzlarich/Bloomberg)
That pivot should sound alarms for those concerned about climate change or the future of cities. While self-driving cars have advanced far beyond the playthings that Koopman encountered at Carnegie Mellon 30 years ago, their societal benefits remain speculative at best. Meanwhile their downsides are very, very real — especially if they become widely available to purchase.
To understand why, consider an experiment in Northern California a few years ago, in which 13 people were given a chauffeur to take them anywhere they wanted for a week, effectively replicating the experience of having their own autonomous vehicle. Freed from the hassles of driving, test subjects traveled a whopping 83 percent more miles than when they had to drive themselves.
A concept called the Jevons paradox explains what happened: When a thing becomes cheaper, people discover new ways to use it. Self-driving cars reduce the “cost” of driving — in terms of effort, if not dollars — and as a result, they will induce people to take trips that they would have otherwise foregone. Over time, people with self-driving cars could opt to move farther from the central city, worsening sprawl and leading to still more miles driven.
Even if self-driving cars are electric, those added miles would be disastrous for the environment. Whether powered by batteries or gasoline, all cars produce emissions from brake dust as well as from friction between tires and pavement. And manufacturing EVs and charging batteries both require power and materials, expanding vehicles’ total carbon footprint. If we’re serious about addressing climate change, promoting transit and cycling instead of driving offers greater potential than any kind of automobile technology, autonomous or electric or otherwise.
For cities, the new driving spurred by AVs would be even more ominous. Without some sort of restrictive policy like a vehicle-miles-traveled tax or decongestion pricing, overwhelmed streets could become mired in gridlock, particularly if owners of self-driving cars avoided parking fees by instructing their vehicles to circle nearby streets while they grabbed a sandwich or met with a client.
Concerned with safety, some Silicon Valley residents want self-driving cars off of their streets. (Jonathan Baran, Faiz Siddiqui/The Washington Post)
These are troubling arguments against self-driving cars, especially when placed against their dubious benefits. But when boosters need a trump card, they can always point to China. When Sens. Gary Peters (D-Mich.) and John Thune (R-S.D.) introduced legislation supporting autonomous vehicles last year, they warned that “the United States risks losing its technological leadership in the autonomous vehicle industry … unless it enacts policies to protect its leadership against the People’s Republic of China."
But does it really matter if China takes the lead in this realm? Global competition alone seems like insufficient justification to allow unfettered civilian use of a technology that could diminish American quality of life. After all, do the Chinese look with envy at American gunmakers like Smith & Wesson and Ruger, worrying that China’s strict gun control and lack of a domestic firearms market hinders them on the world stage?
With so much at stake in the future of self-driving cars, government officials should be keeping a watchful eye on new developments, ready to intervene to defend the public interest. Some do seem to assume that posture. Speaking at CES last month, Transportation Secretary Pete Buttigieg said that self-driving cars have “raised complicated, even philosophical, questions about safety, equity and our workforce.” But public officials often seem more eager to catalyze the arrival of AVs than to protect cities and citizens — and the planet — from their risks. In 2017, the House of Representatives passed the Self Drive Act, whose goal was “encouraging” the deployment of autonomous vehicles; that bill died in the Senate. Last year, the Senate considered Peters and Thune’s measure, which would have allowed the carmakers to request that up to 80,000 self-driving cars per year be exempt from established automotive safety rules.
Politicians’ support will come as a relief to the carmakers and tech companies whose market valuations rely on an expectation of widespread autonomous vehicle adoption. Companies can’t scale self-driving cars if state and federal regulators refuse to bend rules, such as requiring a driver to be behind the wheel.
It’s understandable that companies want to maximize shareholder return; that’s their role in a market economy. But automakers are still struggling to explain why, exactly, we should be excited about this technology, rather than alarmed by it. We shouldn’t let them off the hook unless we have a convincing answer.
By David Zipper
David Zipper is a Visiting Fellow at the Harvard Kennedy School's Taubman Center for State and Local Government, where he examines the interplay between transportation policy and technology. Twitter