Waymo's first fatal accident and what it means

It happened.

I’ve been writing about self-driving cars for seven years, and every time I wrote about safety, there was an asterisk. “No fatal accidents yet.” That asterisk is gone now.

A Waymo was involved in a fatal accident. I won’t detail the specifics here because the investigation is ongoing and the family deserves more than a blog post’s analysis. What I want to write about is the broader question that this moment forces us to confront.

The statistics

Waymo has driven billions of autonomous miles. The rate of incidents per mile is a fraction of the human driving rate. The NHTSA data shows that human drivers in the US are involved in roughly 2 fatal crashes per 100 million miles. Waymo’s rate, even with this accident, is significantly lower.

The autonomous vehicle is statistically safer than a human driver. That fact hasn’t changed.

Why the statistics aren’t enough

Here’s what I keep thinking about. If 40,000 people die in US car crashes every year with human drivers, and autonomous vehicles would reduce that to 10,000, that’s 30,000 lives saved. That’s a staggering improvement.

But those 10,000 deaths feel different. A human driver who makes a mistake is a tragedy we understand. A human who failed to brake. A human who looked at their phone. A human who fell asleep. We understand human error because we are human. We’ve all almost made the same mistakes.

A machine that kills someone is different. Not statistically. Emotionally. The machine was supposed to be better. The machine doesn’t get distracted. The machine doesn’t drink. When the machine fails, we don’t feel sympathy. We feel betrayed.

The hardest part of the self-driving transition isn’t the technology. It’s the psychology. Accepting a machine that kills fewer people means accepting a machine that still kills some people. And for the families of those some people, the statistics are meaningless.

What I think

I think autonomous vehicles should continue to expand. The data supports it. The safety improvements are real. The alternative (continuing to accept 40,000 annual deaths from human driving) is worse by every measure.

I also think we should grieve every life lost to an autonomous vehicle. Not minimize it. Not wave it away with statistics. Every death is a person, not a data point.

Both things can be true. The system is better. And the system isn’t perfect. Accepting that duality is the moral challenge of this technology.

I don’t have a clean conclusion. This is one of those posts where the question is the point. How do we accept imperfect safety for improved safety? How do we grieve the individual while celebrating the statistical?

I don’t know. I think it’s one of the hardest questions technology has ever asked us.

I’m sitting with it. It’s a heavy one.

Related thinking: