Human Drivers Will Kill 11 People While You Read This

Let’s not obstruct the technology that could have saved them

Senator Josh Hawley is calling for a ban on autonomous vehicles. So are labor organizations. They have valid concerns about job loss.

I lost a friend to a drunk driver. My wife and children were nearly propelled into a head-on collision after being rear-ended by a speeding, texting teenager. With safer robot drivers that exist today, none of this would have happened. I have concerns about not deploying autonomous vehicles.

Road safety is a personal issue for me. There’s a good chance it’s personal for you, too. That’s due to a fact that would be shocking if we hadn’t grown inured to it: each year, well over one million people are killed in vehicle crashes worldwide¹.

We’ve been hearing promises about self-driving cars for decades. The picture is muddied by the mixed safety record of vehicles developed by Tesla, Uber, GM’s Cruise division, and others. But Waymo (the robotaxi operator spun out from Google), playing tortoise to Tesla’s hare, has released data which paints a new picture: it is now within our grasp to virtually eliminate traffic injuries and fatalities.

Waymo’s Track Record is Astonishingly Good

In September 2023, Waymo published a study evaluating their vehicle’s safety. The study, carried out by the reinsurance company Swiss Re, found that Waymo vehicles experienced fewer than ¼ as many property damage claims as cars with human drivers. However, that was based on a limited track record. We all had to wait for more data.

Last December, Waymo and Swiss Re released another study, this time covering 25 million miles of autonomous driving. This study found “an 88% reduction in property damage claims and 92% reduction in bodily injury claims”.

As of today, Waymo’s safety dashboard reports 96 million miles driven autonomously (a 4x increase from that December report, highlighting just how quickly Waymo is expanding). How do things look with this larger data set? Pretty darn good:

I like these odds

While the earlier studies were performed by Swiss Re, this dashboard comes directly from Waymo; it would be nice to have some independent analysis. Fortunately, Understanding AI (a great source for analysis of progress in autonomous vehicles) is on the case. Writer Kai Williams dug through the NHTSA records for every Waymo crash reported between mid-February and mid-August – a total of 45 incidents. Almost none of these incidents are primarily the fault of the Waymo driver. In the majority of cases, another vehicle rear-ended the Waymo. Others were true oddities, such as the time when a vehicle being towed came loose and rolled backwards into the Waymo.

The Understanding AI analysis finds just four incidents in which the Waymo “driver” appears to have been at least partially at fault. In three, the car came to a stop on a narrow road, and a human driver tried to pass, hitting or scraping the side of the Waymo. In the fourth, a cat darted in front of the Waymo, which slammed on the brakes, triggering a rear-end collision.

It seems possible that the AI driver could have better handled those last four incidents. Perhaps it also contributed to a few of the other rear-end collisions. But I can’t help but think that if all of the cars involved had been driven by Waymos, none of these collisions would have occurred. They wouldn’t have tried to squeeze through a too-narrow gap left by a car stopped in an awkward position, and they would presumably have had sufficiently fast reflexes (and sufficiently cautious following distances) to avoid rear-end collisions.

Five of the 45 incidents didn’t involve failures on the part of any driver. In three cases, per Understanding AI, “an exiting Waymo passenger opened a door and hit a passing bicycle or scooter”. Two others involved mechanical failure – the truck which came loose while being towed, and an incident where one of the car’s wheels fell off!

In other words: Waymo vehicles are involved in 80-90% fewer collisions and injuries, and that amazing statistic actually undersells the case, because even on the rare occasions when a Waymo is involved in a collision, it’s almost always the fault of a human driver. In an all-Waymo world, the biggest remaining problem would be cyclists getting “doored”. And the technology is still improving rapidly; I expect further improvements in the few remaining injury scenarios².

In a future where most or all cars are driven by AIs at least as advanced as today’s Waymo driver, road deaths – including passengers, cyclists, and pedestrians – would be almost as rare, and seem almost as bizarre, as cases of scurvy. I am rooting for that future to arrive as soon as possible.

When An Ambulance Comes Through, You Clear the Way

We shouldn’t just throw the gates open to autonomous vehicles. So far, no other AV maker has demonstrated a safety record similar to Waymo’s. Given that we limit the rollout to proven vehicles, millions of workers could indeed lose their jobs, and the impact will land disproportionately on those least equipped to bear it. If the shock is not managed well with proactive government, industry, and educational support, the result will be bad for everyone. Other negative impacts may arise³.

Even if we wanted autonomous vehicles everywhere today, Waymo is far from ready. Their cars are still in testing for highway or snow driving, and Waymo carries out extensive mapping and testing in each new city before beginning commercial service. Maintaining their impressive safety record in a broader range of environments may require further work. Equipment costs are still high. The infrastructure to keep millions of robotaxis charged, cleaned, and maintained doesn’t yet exist.

But none of this can excuse delay. Once an AV manufacturer has demonstrated a safety record like Waymo’s, we should roll out the red carpet. Fast-track planning approval for maintenance depots and grid connections for charging stations. Figure out how to support displaced workers, instead of supergluing them to their current jobs. Treat issues as problems to be solved, not obstacles to be placed in the path of this lifesaving technology.

Think about the energy and urgency devoted to combatting drunk driving. Fallible human drivers overall cause about four times as many deaths as drunk drivers alone. The organization Mothers Against Drunk Driving strongly supports autonomous vehicles, for the very good reason that robots don’t get drunk. They also don’t speed, check their phones, or drive drowsy. And of course, the technology is only going to get more advanced from here.

We are told of many wonders which could be ushered in by AI. As I was drafting this, OpenAI CEO Sam Altman wrote:

Maybe with 10 gigawatts of compute, AI can figure out how to cure cancer. Or with 10 gigawatts of compute, AI can figure out how to provide customized tutoring to every student on earth.

It’s easy to become cynical about expansive visions of what AI might enable. We should recognize when the potential to change the world for the better moves from science fiction to demonstrated fact. It looks to me as if Waymo’s autonomous vehicles have crossed that threshold.

Eleven people died while you read this post. Eleven Waymos could have saved them. Let’s get those things on the road as soon as possible.

Thanks (as usual!) to Abi and Taren for feedback and images.

1. A recent WHO estimate puts the figure at 1.19M deaths per year. Tens of millions more suffer injuries, some disabling.

2. As Understanding AI notes, Waymo already has a “Safe Exit” feature that warns passengers of approaching bicycles, relying on the car’s 360° sensor package. That might already be reducing “dooring” incidents, but clearly it wasn’t sufficient in the three cases identified in this analysis. As Waymo gathers more data from real-world incidents, I suspect they will find ways to improve the warnings, for instance by sounding a more stringent alarm when a collision is imminent, or even momentarily preventing the door from opening.

3. In addition to issues of traffic, impact on transit, etc. (I’ve seen arguments in both directions), AVs will undoubtedly introduce both positive and negative impacts that are hard to foresee today. One foreseeable concern is the harm that a “hacked” vehicle could do; before AVs are deployed at mass scale, we should hold them to stringent cybersecurity standards. Historically, Google tends to take security quite seriously, and hopefully this will spill over to Waymo (a corporate sibling under Alphabet). However, also historically, the cybersecurity track record of manufacturing businesses whose products come to incorporate electronics – such as traditional car companies – is quite spotty.