It’s a familiar frustration for ridehail users: you open the Uber or Lyft app, enter your destination, and discover that your intended trip costs several times more than expected. The culprit is surge pricing, one of ridehail’s most important and controversial innovations. Customers grumble about higher fares, but Uber and Lyft executives have insisted that surge pricing benefits them by attracting additional drivers, which allows the companies to fulfill more trips and reduce wait times. 

That justification makes intuitive sense, but it raises an awkward question about robotaxis, which are expanding across the US, from San Jose, California, to Washington, DC. If surge pricing is intended to expand the driver pool, why is it now being used by companies with driverless vehicles?

Waymo, which offers robotaxi trips in the Bay Area, Los Angeles, and Phoenix, charges surge pricing during peak times, as did Cruise, its now-defunct competitor. Assuming a robotaxi fleet is already fully deployed, higher fares cannot expand vehicle supply in the way they could for Uber or Lyft. Instead, riders simply need to pay extra, assuming they can afford to, or search for another way to travel.

Surge pricing, one of ridehail’s defining features, may need a rethink for an autonomous era. 

Uber began experimenting with surge pricing in 2012, and customers have been grumbling about it ever since. In 2014, one exasperated Aussie described the practice to Mashable as “price gouging at its worst.” (Price gouging is banned in many US states, but such laws typically kick in only during emergencies or natural disasters.) Screenshots of astronomical fares, like an $800 ride on New Year’s Eve in 2015, frequently went viral. Aware of the pushback, Uber and Lyft adjusted their app designs in recent years to conceal temporary price increases, but surge pricing (sometimes called “dynamic pricing”) has endured.

Harry Campbell began driving for Uber a decade ago. He now runs The Rideshare Guy, a publication devoted to ridehail, and The Driverless Digest, focused on the robotaxi industry. “At Uber, their number one [key performance indicator] from basically day one has been reliability,” he told me. “When you open the app, they want you to see cars available within three to five minutes.” Given the vagaries of trip requests and driver availability, keeping wait times within that window is no easy task.

Defenders of surge pricing argue that it convinces more drivers to work during times of high demand, which avoids lengthy wait times. “Surge pricing doesn’t just make rides more expensive,” James Surowiecki wrote in an article entitled “In Praise of Efficient Price Gouging” for MIT Tech Review in 2014. “It also expands the number of people who are actually able to get a ride.” The additional drivers allow fares to drift back toward normal levels.

But this supply-side narrative has always omitted part of the story. “Surge pricing also tempers demand,” Campbell said. “When people see that their ride is more expensive, they may not take it.” By deterring some potential customers, surge pricing makes it easier to serve those who remain. Would-be customers who can’t stomach the higher price are left to figure out a Plan B.

Voicing concerns about consumer protection, legislators in states like Massachusetts, New York, and Washington have proposed caps on temporary price hikes (and New Delhi, India, has imposed one). Surge pricing has become a generally accepted aspect of ridehailing.

And now it’s been adopted by Waymo, a company whose service is, apart from the empty driver’s seat, largely indistinguishable from Uber or Lyft. But while higher fares may convince part-time ridehail drivers to work during periods of high demand, surge pricing can do nothing to expand the tightly limited size of Waymo’s fleets. As of January, for example, the company operated only around 100 vehicles in Los Angeles.

“I think Uber and Lyft have a very strong justification for using surge pricing that gets more drivers on the road and gets you home,” Campbell said. “Waymo doesn’t have a good justification. They just say, ‘Hey, we’re charging you more because a lot of people want rides, even though we literally cannot add more vehicles to the fleet.’”

Surge pricing can’t attract additional robotaxi vehicles, but it does suppress rider demand, thereby narrowing the gap between requested and available trips during peak times. In an email, Waymo spokesperson Chris Bonelli wrote, “During busier times, temporarily increasing prices may help reduce demand and keep wait times reasonable for a good rider experience.” “Reasonable” is doing a lot of work there; Campbell shared a screenshot of Waymo wait times hitting 24 minutes in Los Angeles, where he lives.

Still, surge pricing’s ability to at least temper demand is enough for Brad Templeton, a consultant and veteran of the self-driving industry, to deem it useful. “The societal benefit is that you have scarcity instead of shortages,” he said. “If you really need a trip, you can get it — it’s just really going to cost you.” He drew a comparison with airline tickets that cost more during popular travel times like Thanksgiving weekend.

But Templeton acknowledged that surge pricing creates winners and losers, particularly if it cannot expand vehicle supply to soften price hikes. Those who can afford surge pricing will pay it; everyone else will have to find another way to travel — or forgo the trip entirely. 

“It does allocate more to the wealthy than the poor,” he said. “That may or may not match public goals” around fairness. This, after all, was the underlying critique of ridehail’s pioneering use of surge pricing, which the companies parried by noting how higher prices expand vehicle availability — something that Waymo and its ilk cannot claim.

Such tensions could dissipate if the supply of robotaxi vehicles becomes more flexible in the future. There are several ways that might happen.

In a March blog post and a recent episode of the Autonocast podcast, mobility investor Reilly Brennan divided the on-demand trip market into “base load,” consisting of trips taken during periods of typical demand, and “peak load,” representing those requested when demand temporarily spikes. 

One future scenario involves a fixed fleet of full-time robotaxis providing requested trips when demand is normal, while surge pricing during peak times encourages human drivers to grab their keys, thereby expanding the supply of vehicles (and reducing customer wait times). Such an arrangement could appeal to ridehail companies, which benefit from the lower cost of operations during non-peak times, as well as robotaxi companies, which can tap human drivers to add vehicle capacity when they most need it. The recently announced collaboration between Uber and Waymo in Austin suggests such a partnership may be plausible.

Brennan outlined another possibility that seems specific to Tesla: If the company’s promised Cybercabs become a reality (a big if) and its autonomous technology works reliably (ditto), the company could deploy its Cybercab fleet to fulfill base load demands while augmenting it during peak periods with personally-owned and self-driven Teslas, dispatched willingly by their owners when surge pricing hits a threshold of, say, $4 per mile. It’s a nice vision, but caution seems warranted given CEO Elon Musk’s failures to fulfill previous promises around self-driving tech.

Templeton believes robotaxi companies could accommodate more trips with limited fleets during peak times by offering customers discounts if they split their trip with strangers. Although ridehail’s experiments with shared rides have fizzled in part due to a lack of privacy, robotaxis might have more success if they use partitions to physically separate passengers from one another.

For now, at least, robotaxi companies like Waymo are free to charge whatever they like during peak periods, even though they can’t deploy additional vehicles to meet the higher demand. Templeton thinks that’s appropriate given the nascent stage of the robotaxi industry. “I think we should wait, watch, and learn,” he said.

After a brutal 2023, the vibes around self-driving cars are improving. Cruise, the industry leader whose vehicle was involved in a horrific San Francisco crash last fall, has rebooted under new management, while rival Waymo is expanding to serve broader swaths of the Bay Area and Los Angeles and Tesla is promising a new robotaxi service.

Although Americans say they remain wary of autonomous driving, boosters insist there is nothing to fear. In fact, they foresee roads full of self-driving cars that are both safer and cleaner than the status quo, a tantalizing prospect in a country where transportation is the largest source of greenhouse gas emissions and residents are several times more likely to die in a crash than those living in other rich nations.

Enticing though they are, such arguments conceal a logical flaw. As a classic 19th-century theory known as a Jevons paradox explains, even if autonomous vehicles eventually work perfectly — an enormous “if” — they are likely to increase total emissions and crash deaths, simply because people will use them so much. 

In the 1800s, coal was the sine qua non of economic development, essential for everything from heating to transport to manufacturing. In Britain, the country where the stuff first powered an industrial revolution, national leaders debated how concerned they should be about potentially depleting coal deposits. Some argued that supply would never be exhausted because improvements in steam engine designs would steadily reduce the amount of coal necessary to power a train, make a dress, or do anything else. Productivity gains would allow Britain’s coal resources to stretch further and further.

In his 1865 book The Coal Question, the economist William Stanley Jevons explained why he disagreed. Jevons drew from then-recent history to show that steam engines’ efficiency had led people to deploy more of them. “Burning coal became an economically viable thing to do, so demand exploded,” said Kenneth Gillingham, a professor of environmental and energy economics at Yale. “You have steam engines everywhere, and people are using them instead of water power. You actually use a lot more coal than you did initially.” Despite the improvements in steam engine design, Jevons argued, total coal use would continue to rise.

Today, the Jevons paradox describes a situation where greater efficiency in deploying a resource (such as water, gasoline, or electricity) causes demand for that resource to skyrocket — negating an expected decline in total usage. Electric lights are often cited as an example: people have responded to improved light bulb efficiency by installing so many more of them that there has been no decline in the total energy consumed by lighting. The Jevons paradox has become a bedrock principle of environmental economics, used to explain why efficiency improvements can backfire and cause the opposite outcome from what was intended. 

Its lessons can also illuminate transportation. Consider the projects undertaken by highway agencies to alleviate roadway congestion. Public officials often justify them by noting (accurately) that gas-powered engines are less efficient and release more pollutants if they are stuck in gridlock instead of moving at a steady clip. For that reason, they argue, highway expansions or traffic technologies that mitigate traffic jams will also reduce emissions. 

The Jevons paradox reveals a blind spot in such claims. If an added lane or new traffic technology does relieve congestion, more people will decide to drive due to a drop in the “cost” of using a car — in this case, the time sitting in traffic. Even if each car now produces fewer emissions due to faster travel speeds, these benefits could be overshadowed by the sheer number of new trips that would not have otherwise occurred. In other words: backfire. (The benefits of expanded highways are even more questionable when one considers the likelihood that rising car volumes ultimately force traffic to move as slowly as before — only now with more cars belching fumes as they inch forward. This phenomenon is known as induced demand.)

Now consider the case of autonomous vehicles. Seeking to win over skeptical regulators and members of the public, AV supporters frequently cite the supposed safety benefits from replacing the fallible humans sitting behind the wheel with technology that will never drive drunk, high, or distracted. Some also suggest that self-driving cars will reduce energy use and emissions since they will avoid the quirks of human driving that compromise engine efficiency. “The higher the proportion of AVs on the road, the smoother the overall flow of traffic ought to be, resulting in less energy-consuming stop-and-go traffic,” predicted a 2021 blog post from Mobileye, a technology company that claims it is “driving the autonomous vehicle evolution.”

Both of these supposed benefits are dubious; AVs’ computers may make driving errors that humans would not, and even if they run entirely on electricity, their software, hardware, and sensors require an enormous amount of power that generates its own emissions as it is produced. Still, it is reasonable to expect AVs’ reliability and efficiency to improve over time. For the sake of argument, let’s take a leap of faith and assume that an average self-driving car will eventually be both safer and cleaner than one driven by a human. Will total crash deaths and emissions then fall?

The Jevons paradox suggests we shouldn’t count on it.

As AV companies’ ads show, the raison d’être of autonomous vehicles is making driving easier and more pleasant, with passengers free to hold a work meeting, sing a song, or grab some shuteye. How do people respond when an activity becomes less onerous and more fun? They do more of it.

Similar to highway expansion, the availability of autonomous vehicles will likely lead people to take longer motor vehicle trips or opt for a car when they would have otherwise used transit, biked, or stayed home. The result will be a lot more (now autonomous) cars on the road. As the University of Virginia historian Peter Norton wrote in a prescient 2014 article, self-driving technology could lead people to “spend more total time in vehicles [and] use them for even more tasks.”

Norton, who teaches the Jevons paradox in his classes, told me that he wrote that article because he “was seeing smart engineers argue, to my utter astonishment, that [AVs’] efficiency against would only bring savings — with no counteracting costs. How they can continually deny this elementary fact is beyond me.”

Supporting his point, a recent paper from the Transportation Research Board concluded that “the likelihood of making additional trips increases” when autonomous vehicles are available, even if they are shared instead of owned. Since each self-driven mile creates some pollution and carries some risk of a crash death, the rise in total driving will counteract the theoretical climate or safety improvements over a single, otherwise identical human-driven journey.

The societal impact of self-driving cars looks even worse when considering second-order effects related to land use. Just as the ascent of car ownership fueled suburbanization in the 20th century, AVs could lead people to relocate to larger, less energy-efficient homes on the urban fringe, where car trips — now more tolerable — are longer.

At the moment, there are more questions than answers about the collective effects of AVs, which are currently available in only a handful of US cities. As self-driving companies pour billions of dollars into advancing their technology, it is impossible to know how safe and energy-efficient their products could eventually become. But the Jevons paradox suggests those are not the only questions to consider. Another, equally crucial one: how much more driving will AVs induce — and will those added miles swamp any possible upside?

Car companies love to explain how their research and development efforts will lead us toward “a world with zero crashes,” as General Motors puts it. Automakers like Stellantis and Nissan, among others, tout their efforts to develop “next generation technologies to make roads safer for drivers and pedestrians alike.” 

With American roadway deaths now exceeding 40,000 per year — including a surge of 10.5 percent in 2021, the fastest on record — these promises sound like salvation. 

The companies are referring to technologies, typically known as advanced driver-assistance systems (ADAS), that can manage aspects of the driving experience and intervene if the human behind the wheel makes a mistake. Such features include automatic emergency braking, lane keep assist, adaptive cruise control, and pedestrian detection. With billions of dollars invested, automakers, federal regulators, and safety advocates alike are bullish about ADAS’s potential to achieve “collision-free mobility,” as Honda puts it. 

But upon examination, these new features are hardly the panacea that their boosters imply. Some elements presented as safety enhancements (like lane keep assist) may be little more than driver conveniences. For now, at least, those technologies that could save the most lives (like pedestrian detection) remain deeply unreliable. And even if ADAS eventually works flawlessly, it is likely to have only a modest impact on annual traffic deaths. 

As the United States confronts a national crisis of traffic fatalities, carmakers and policymakers alike are focused on unproven and overhyped innovations. In reality, even the best technologies can’t compensate for the ways in which ill-conceived cars and poor street designs have made crashes more numerous and severe. We risk making our road safety crisis even worse by expecting car tech to bail us out. 

The core concept behind ADAS involves leveraging computer power to handle aspects of driving traditionally managed by a human — something that will be familiar to anyone who has used cruise control to make long highway trips a little less tedious. 

Over the last two decades, automakers have developed numerous features that deeply integrate technology into the act of driving, capitalizing on powerful sensors and cameras embedded in new vehicles. Some of these features, like pedestrian detection and automatic emergency braking, are intended for use only to prevent an imminent collision. Others, like lane keep assist and adaptive cruise control, can “ease the burden” of driving, as JD Power put it, by ensuring that the vehicle keeps pace with surrounding traffic and stays within its allotted road space. 

Products like Tesla Autopilot or Ford BlueCruise integrate these individual features into a multilayered system that can fully operate the vehicle on a highway — provided that the driver stands ready to take control if needed. That assumption is critical because these systems are not smart enough to manage a vehicle on their own. 

To ensure the driver stays focused, carmakers rely on driver monitoring systems designed to keep an eye on the tilt of a person’s face or their grip on the steering wheel. Should the driver’s attention wander, driver monitoring systems will issue a warning before ultimately forcing the driver to retake control of the vehicle.  

Industry groups sometimes call ADAS “partial automation,” a term intended to signify that, advanced as these systems are, they do not enable a car to become autonomous.

Although Elon Musk has claimed that Tesla’s Autopilot ADAS system could save half a million lives if universally deployed, there is scant evidence supporting that claim (or others about ADAS’s supposed safety benefits). The US Department of Transportation recently began collecting data about ADAS-related crashes, but it’s too soon to draw many insights.

David Harkey, the head of the Insurance Institute for Highway Safety (IIHS), is not impressed by what he has seen so far. “Partial automation systems may make long drives seem like less of a burden, but there is no evidence that they make driving safer,” he said in an IIHS blog post. “In fact, the opposite may be the case if systems lack adequate safeguards,” a reference to those all-important driver monitoring systems. 

Nevertheless, many car regulators and safety advocates are enthralled by ADAS’s potential to reduce crashes. In 2016, Mark Rosekind, then the head of the federal National Highway Traffic Safety Administration (NHTSA), extolled “technologies that promise a revolution in safety unlike any in the history of the automobile” (a quote that the Alliance for Automotive Innovation, the largest US carmaker association, places on its webpage). Cathy Chase, the president of Advocates for Highway and Auto Safety, said that “the universal adoption of these technologies will literally save tens of thousands of lives” in a 2021 letter to US Transportation Secretary Pete Buttigieg. 

But for the moment, ADAS remains very much a work in progress. A 2020 study by AAA found that the average system turned itself off every eight minutes, noting “instances of trouble with the systems keeping the vehicles … in their lane and coming too close to other vehicles or guardrails.” In a separate study of pedestrian detection, AAA found the feature to be “completely useless” at night, when 75 percent of pedestrians are struck. The European Transport Safety Council concurred, noting ADAS’s shortcomings in dark, wet, or foggy conditions. Even more troubling, automakers’ crucial driver monitoring systems can be cheated and do not work reliably. 

All of that being said, ADAS performance will likely improve due to the billions that carmakers are investing in its development, along with insights gleaned from reams of data collected from cars using these systems on the road. But even if the technology ultimately functions as intended, it may have only a marginal impact on the US roadway death toll.

Consider the need for a human to remain ready and able to take over the vehicle. Even if driver monitoring systems prevent a driver’s attention from wandering, what happens if road skills atrophy from lack of use? This is more than a theoretical issue; a study of truck drivers recently found that automated driving technology led to slower reaction times, and a report from the US Department of Transportation Inspector General raised concerns that automation could be degrading the skills of commercial pilots. 

Worse, those focused on ADAS’s technical capabilities risk overlooking its potential to encourage riskier driving. In 1975, Sam Peltzman wrote a seminal economics article examining the safety effects of state seat belt laws. Peltzman concluded that the mandated use of seat belts led drivers — secure in the straps across their waists — to take more risks behind the wheel, leading to injuries and deaths that negated those saved by the belts themselves. In effect, he argued that the new feature’s benefit was negated by behavioral adaptation. 

Today, the “Peltzman effect” explains why safety technologies in fields including healthcare and sports cause people to adopt riskier behavior. Applied to ADAS, the Peltzman effect predicts that drivers will become less careful, trusting the vehicle’s technology to protect them. But ADAS is not a fail-safe; physics limits its ability to abruptly halt a vehicle. For example, Mercedes claims that its system can prevent pedestrian collisions at up to 30mph and mitigate the severity of vehicular crashes at up to 45mph, but the company makes no promises above those thresholds. This isn’t a challenge limited to Mercedes: a recent AAA study found that carmakers’ automatic emergency braking systems prevented 85 percent of test crashes at 30mph but only 30 percent at 40mph. 

Already, a study by the IIHS found that the use of adaptive cruise control increased the share of drivers who broke the speed limit by 18 percent, and San Jose State researchers concluded that ADAS-equipped cars were more likely to crash into pedestrians or cyclists. These findings align with the Peltzman effect’s predicted shift toward unsafe driving, with those outside the vehicle bearing disproportionate risk. Such dangers could be exacerbated by drivers who overestimate ADAS’s capabilities, as more than half of Cadillac Super Cruise users seem to do, according to a recent IIHS study.

There is another reason these innovations could invite more roadway deaths, something both fundamental and easily overlooked. To understand it, note that Hyundai acknowledges on its website that “ADAS is not just about safety; it’s about providing convenience as well.” Indeed, features like adaptive cruise control and lane change assistance are largely intended to make driving more pleasant rather than to minimize crashes. 

How do people respond when an activity becomes easier or more enjoyable? They do more of it.

Just as comforts like air conditioning and radios induced car owners to drive more, the comforts of software-assisted driving will compel them to take additional trips and travel further (which may contribute to automakers’ excitement about it). One study has already found that Tesla owners using Autopilot drove about 5,000 more miles per year than those without it. All else being equal, additional miles driven bring additional chances to crash. 

ADAS’s overall effect on road deaths becomes murky when these countervailing forces are considered alongside its technological potential. The systems may prevent certain collisions that would have otherwise occurred while at the same time leading to a degradation of driver skills, riskier behavior behind the wheel, and a surge in total miles driven.

We can debate whether the overall effect on roadway deaths will be positive or negative, but those expecting a virtual elimination of crashes are likely to be sorely disappointed.

The good news is that if public officials are serious about reducing road deaths, there are plenty of compelling approaches that have nothing to do with technology, such as designing streets for slower speeds, building dedicated bike lanes and sidewalks, and ramping up transit service to entice people to switch from cars. (Riding in a bus or a train is orders of magnitude safer than being inside a motor vehicle.) Keeping habitually reckless drivers off the road would also be a significant step forward. 

Rather than placing so many eggs in the ADAS basket, regulators and automakers’ R&D teams could focus on other ways that safer car designs could reduce road deaths. Speeding, tied to some 11,000 deaths per year in the United States, could be curtailed with intelligent speed assist, which automatically warns or slows drivers who are exceeding the speed limit. Europe has already moved to require intelligent speed assist in passenger vehicles, but so far, Congress and the National Highway Traffic Safety Administration have shown no signs of following suit. (Among automakers, only Volvo has voluntarily installed the feature.) 

Automakers could also confront ways in which their previous design decisions have worsened America’s road safety crisis. For instance, car companies have sought to replicate the smartphone experience by replacing dashboard knobs with infotainment touchscreens. Regardless of whether you like car touchscreens (and many people don’t), the absence of tactile feedback forces drivers to look away from the road — something that is inherently dangerous when a multiton hunk of metal is hurtling down a road at 50mph. Research from Drexel University shows a growing number of crashes tied to infotainment, but so far, NHTSA has done nothing beyond issuing voluntary guidance (which carmakers have routinely violated) about the maximum amount of time it should take to complete a task on an infotainment system. 

Meanwhile, automakers have added girth to their SUVs and trucks, which now dominate the American car market (together representing over 80 percent of sales). A taller, heavier vehicle is more likely to injure or kill in a collision, especially when striking a pedestrian or cyclist. Acknowledging this risk, auto regulators in Europe, Japan, and Australia have incorporated pedestrian crashworthiness into their car safety ratings, known as the New Car Assessment Program, or NCAP. But the United States has not. 

Earlier this year, NHTSA announced a long-awaited update to the American version of NCAP, including a proposal to finally address risks borne by vulnerable road users. But rather than examining pedestrian crashworthiness, the agency chose only to evaluate ADAS’s pedestrian detection technology. In a press release, the agency credited itself with a “novel approach to tie technological change to reducing driver behaviors that contribute to many crashes.” The dangers of oversize SUVs and trucks were not mentioned. 

NHTSA’s focus on ADAS instead of SUV and truck bloat likely came as a relief to automakers who make fat profits from the sales of huge, pricey vehicles. Rather than discuss how their past choices have contributed to America’s burgeoning roadway death toll, they would prefer to paint a utopian picture of universally safe driving at some point in the future, once ADAS is perfected and widespread. 

Case in point: the NBC affiliate in the Washington, DC, region recently produced a segment exploring SUV “blind zones” that can make a child in front of the vehicle invisible to the driver. Asked for comment, the Alliance for Automotive Innovation, a carmaker industry group, tried to shift the focus away from dangerous SUV design and toward the promise of innovation. A spokesperson said that “vehicles continue to get safer as automakers across the board test, develop, and integrate new technologies that can save lives.”

That answer, of course, is a deflection. 

Lucas Peilert contributed research assistance to this article.