The life-or-death case for self-driving cars

I have some bad news: You are almost certainly a worse driver than you think you are. Humans drive distracted. They drive drowsy. They drive angry. And, worst of all, they drive impaired far more often than they should. Even when we’re firing on all cylinders, our Stone Age-adapted brains are often no match for the speed and complexity of high-speed driving. There’s as much as a 2.5-second lag between what we perceive and how fast we can react in a vehicle traveling 60 mph, which means a car will travel the equivalent of two basketball court lengths before its driver can even hit the brake. The result of this very human fallibility is blood on the streets. Nearly 1.2 million people die in road crashes globally each year, enough to fill nine jumbo jets each day. Here in the US, the government estimates there were 39,345 traffic fatalities in 2024, which adds up to a bus’s worth of people perishing every 12 hours. The good news is there are much, much better drivers coming online, and they have everything human drivers don’t: They don’t need sleep. They don’t get angry. They don’t get drunk. And their brains can handle high-speed decision-making with ease. Because they’re AI. AI takes the wheel The average American adult will spend around three years of their life driving. If robots could take the wheel instead, well, think of all the Netflix shows we could stream instead. But the true benefit of a self-driving revolution will be in lives saved. And new data from the autonomous vehicle company Waymo suggests that those savings could be very great indeed. In a peer-reviewed study that is set to be published in the journal Traffic Injury Prevention, Waymo analyzed the safety performance of its autonomous vehicles over the course of 56.7 million miles driven in Austin, Los Angeles, Phoenix, and San Francisco — all without a human safety driver present to take the wheel in an emergency. They then compared that data to human driving safety over the same number of miles driven on the same kind of roads. The results of the study, almost certainly the biggest and most comprehensive research on self-driving car safety yet released, are striking. A master class in driving safety Compared to human drivers, the Waymo self-driving cars had: * 81 percent fewer airbag-deploying crashes * 85 percent fewer crashes with suspected serious or worse injuries * 96 percent fewer injury crashes at intersections (primarily because Waymo detects other cars running red lights faster than humans) * 92 percent fewer crashes that involve injuries to pedestrians. Had the typical human-driven fleet of cars covered those same 56.7 million miles, the Waymo researchers project it would have resulted in an estimated 181 additional injury crashes, 78 additional airbag crashes, and 11 extra serious-injury crashes. But the numbers really get eye-popping when you extend this data across all 3.3 trillion vehicle miles driven by humans in the US in a typical year. Back-of-the-envelope calculations suggest that if the same 85 percent reduction seen in serious crashes held true for fatal ones — a big if, to be clear, since the study had too few fatal events to measure — we’d save approximately 34,000 lives a year. That’s five times the number of Americans who died in the Iraq and Afghanistan wars combined. Don’t get in the way of progress Of course, there are plenty of caveats to the Waymo study and even more obstacles before we could ever achieve anything like what’s outlined above. In part because serious injury crashes are (thankfully) very rare, even 56.7 million miles isn’t long enough for researchers to be really sure that such crashes would occur significantly less often with robot drivers, so more data will be needed there. Waymo’s cars were also being driven largely in warm, sunny locations, operating in geofenced areas that had been heavily mapped by the company. It’s far less certain how they might do in, let’s say, the snowy streets of Boston in the winter. This is also a company-run study, though it has been peer-reviewed by outside experts. And even if we decided to go all in on AI drivers, actually producing enough autonomous vehicles to begin to replace human-driven cars and trucks would be an enormous undertaking, to say the least. Still, the data looks so good, and the death toll on our roads is so high that I’d argue slowing down autonomous vehicles is actually costing lives. And there’s a risk that’s precisely what will happen. Too often the public focuses on unusual, outlier events with self-driving cars, while the carnage that occurs thanks to human drivers on a daily basis is simply treated as background noise. (That’s an example of two common psychological biases: availability bias, which causes us to judge risk by outlier events that jump easily to mind, and base-rate neglect, where we ignore the underlying frequency of events.) This misapprehension is something I often see in news coverage and consumption, and it’s one of the reasons I started Good News. The result is that public opinion has been turning against self-driving cars in recent years, to the point where vandals have attacked autonomous vehicles on the street. And of course, given that nearly 5 million Americans make their living primarily through driving, any wide-scale movement to self-driving vehicles would bring significant economic disruption. But still, 34,000 lives saved on an annual basis would represent tremendous progress. Maybe, after about 100 years of trying, it’s time to give something else a chance behind the wheel. A version of this story originally appeared in the Good News newsletter. Sign up here! Correction, May 5, 10:40 am ET: A previous version of this post misstated the primary reason Waymo self-driving cars had 96 percent fewer injury crashes at intersections.

This moment demands more than headlines

The life-or-death case for self-driving cars