Self-driving cars have an emissions problem.

A recent study by researchers at the Massachusetts Institute of Technology measured the energy consumed by powerful onboard computers that serve as the brains of autonomous vehicles. They found that widespread global adoption of self-driving cars would generate an additional 0.14 gigatons of greenhouse gas emissions per year—as much as the nation of Argentina.

The researchers highlighted a serious problem, one that has received scant attention in discussions about our supposedly autonomous future. Still, the authors didn’t frame their conclusion as an indictment of self-driving cars, but as a reason to build AV computers that are less thirsty for power. “This has the potential to become an enormous problem,” Soumya Sudhakar, a coauthor of the study, told Dezeen. “But if we get ahead of it, we could design more efficient autonomous vehicles that have a smaller carbon footprint from the start.”

I hate to be the bearer of bad news, but Sudhakar is too optimistic. Even with a miraculous reduction in their computers’ energy consumption, self-driving cars are poised to send global emissions skyrocketing. Assuming AV technology ultimately works (admittedly a big “if”), a 160-year-old theory of human behavior suggests it will compel people to drive much more than they otherwise would have, generating emissions as they go. Importantly, that will be true even if AVs are also EVs, running on electric power rather than gasoline. Although EVs emit no tailpipe emissions, they still produce greenhouse gas through the generation of electricity needed to charge their batteries, as well as from their manufacturing processes.

When you consider the net effect of many more miles driven, without combined trips, the problem of power-hungry AV computers looks like a modest part of a much larger problem. Even if the computer efficiency challenge is “solved,” self-driving cars would be a climate catastrophe.

To understand why, we can turn to an influential book published in Great Britain in 1865. In The Coal Question, William Stanley Jevons argued that ongoing improvements in coal mining would only fuel demands for further production. The reason: Any productivity gains would cause the price of coal to decline, compelling people and businesses to find new ways to use it. That increase in coal demand would lift its price—and ultimately lead to more coal production.

That idea, now known as the Jevons Paradox, is illuminating when applied to transportation. Consider highway widening projects, which can consume billions of taxpayer dollars with the goal of “ending” congestion. Those hopes are seldom fulfilled. Instead, gridlock returns, as thick as ever, after the project is complete.

The Jevons Paradox shows what highway engineers keep getting wrong. Newly added lanes make traveling during peak times “cheaper” (in time, rather than money), which leads people to adjust their travel plans. Some decide to drive at rush hour when they otherwise would have left home at 5 a.m., or perhaps taken public transit. Others decide to embark on trips that they would not previously have taken at all. People will continue to shift their travel plans toward the newly expanded roadway until congestion is as bad as it was before—only now with many more vehicles.

The Jevons Paradox carries ominous implications for self-driving cars’ impact on climate change. By freeing drivers from the tedium of watching the road, autonomous vehicles will make car travel easier—not unlike the experience of whizzing along a recently expanded roadway. People will respond by traveling farther and taking more trips than they would have otherwise. Emissions from transportation, already the largest source of greenhouse gases in the U.S., will inevitably rise.

Already researchers have shown that advanced driver-assistance systems, which handle aspects of driving but still require a person to be engaged behind the wheel, are leading people to drive more. (ADAS is often seen as a stepping stone toward fully autonomous vehicles.) One study found that Tesla owners with Autopilot, the company’s ADAS product, drove around 5,000 more miles per year than those without it. Self-driving cars are poised to increase car use even more, since they free passengers from needing to pay attention to the roadway at all.

A cleverly designed University of California, Berkeley, study sounds a warning bell about our self-driving future. Researchers gave 13 people access to a chauffeur who would drive them wherever they liked for a week, effectively mimicking the experience of owning a self-driving car that is always and immediately at one’s beck and call. During that week, subjects drove a whopping 83% more miles than they had the week before.

Over time, self-driving technology will induce even more travel, as people gradually adjust their residential and work locations to incorporate the enhanced conveniences of car trips. Proximity to a central city will decline in importance, leading many to opt for bigger homes (which, of course, require more power to heat and cool) in peripheral exurbs. Another word for this trend: sprawl.

The average car trip from these outlying locations will be longer, and transit and bicycles will be less likely to serve as climate-friendly alternatives. Notably, emissions attributable to self-driving car trips will be higher than those from public transportation or a bike even if all AVs are fully electric.

While EVs don’t produce tailpipe emissions, the carbon footprint of their manufacturing process is substantial (often greater than a gas-powered car), and producing electricity to charge their batteries generates further emissions, as does the disposal of the vehicle at the end of its life cycle. University of Toronto researchers concluded that even a best-case scenario of EV adoption would not prevent a 2-degree Celsius increase in global temperatures; a shift away from cars in favor of transit, biking, and walking will also be necessary.

At this point, a defender of self-driving cars might interject and say, “But wait! Autonomous vehicles will be shared, not owned—which will improve their efficiency and reduce emissions.” Indeed, shared trips sound good (who is against sharing?), and AV companies like Cruise and Zoox have developed self-driving pods designed for multiple people to ride together in comfort. But don’t let these vehicles fool you: Should self-driving technology become widely available, all signs suggest that the vast majority of trips will be taken privately—not with strangers.

To understand why, consider shared-ride-hail services like UberPool and Lyft Shared, which have struggled to catch on despite whopping corporate investments. Those who have tried these products are likely to see why.

Shared-ride-hail trip times are notoriously unreliable, since a rider cannot know whether another passenger’s routing will take them far out of their way (or if they won’t be matched with anyone at all). Worse, if the stranger sitting next to the rider reeks or seems sketchy, there’s no easy escape. In fact, riding mass transit, which generally costs a fraction of a shared-ride-hail trip, is in many ways a superior experience: A bus or train’s routing is fixed, leading to more predictable arrival times, and a passenger can easily move to another seat if a person nearby makes them uneasy.

Like shared ride hail, shared robotaxi services would assumedly cost much more than transit while providing a service that is of dubious quality. In fact, some travelers, particularly women, could find robotaxi services worse than shared ride hail, since the presence of a driver provides a measure of protection against other passengers’ inappropriate or unsafe behavior.

Notably, car companies and AV developers have refrained from committing to providing only shared passenger services. That makes sense; after all, the available market for their products is far greater if self-driving cars are individually owned. And research suggests that private ownership of AVs is precisely what most people want—societal impacts notwithstanding.

So we are looking at a future in which self-driving cars are owned, not shared, while being used to take many more trips—and to travel much farther—than would have happened if a driver were sitting behind the wheel.

As the MIT researchers said, the hefty power demands of today’s AV computer systems might be a fixable problem. But either way, the devastating environmental toll of self-driving cars looks like a feature, not a bug.