The companies selling driverless rides want you to focus on one number: the crash rate. And on paper, that number looks good. Waymo says its cars, across more than 170 million miles without a human at the wheel, are involved in far fewer serious-injury crashes than human drivers — over 90% fewer in its own analysis of federal crash reports. Take that seriously. It is probably the strongest card the industry holds.
But notice what that statistic does not say. It does not say the car wants you to live. It does not say anyone is accountable when it doesn't. It does not say you signed up to be a data point. "Safer on average" is a fleet-level abstraction. You do not ride as a fleet. You ride as one person, on one trip, exposed to one edge case the system has never seen before.
And the edge cases are not hypothetical. A Waymo struck a child near a school, causing injuries, and triggered a federal investigation. Regulators are separately probing how these vehicles behave around stopped school buses — the exact scenario every human driver is drilled to fear. Each of these is a moment where the math said "rare" and a real person met the remainder.
Here is the part the marketing never says out loud: you are in the experiment. Not the simulation. Not the closed test track. The live one, on a public street, with you in the back seat and no steering wheel to grab. The system is still learning. It learns from crashes. Some of those crashes will have passengers in them. The company calls this iteration. You might call it being the test.
Now ask the question that no crash rate can answer: when it goes wrong, who is responsible?
A human driver can be questioned, charged, sued, made to look you in the eye. A robotaxi offers you a support line and a software version number. The "driver" is a codebase owned by a corporation with a legal department whose entire job is to make sure the answer to who pays is not us. Indifference isn't a bug in this system. It's the architecture. There is no one inside who can be afraid for you, because there is no one inside at all.
This is what "doesn't care" actually means. Not that the car is malicious — it has no capacity for that either. It means there is no human stake in your survival.
A cab driver wants to get home tonight too; your safety and theirs are bound together. Sever that, and you are protected only by the parts of you the engineers happened to model, in the conditions they happened to anticipate, up to the point where the liability waiver kicks in.
None of this requires the cars to be more dangerous than people. It only requires you to understand what you are trading. You are trading a fallible human who can be held to account for a flawless-until-it-isn't machine that cannot be. You are trading a known risk for a novel one. You are trading a witness for a server log.
So before you tap that button, don't ask "are they safe?" The companies have an answer ready, and it's a good one. Ask the questions they can't answer cleanly.
// Query the system can't resolve
Q1
When this car kills someone, who stands trial?
RESPONSE — no value returned
Q2
Which failures has it simply never encountered yet — and will I be the first?
RESPONSE — unknown / out of distribution
Q3
Who decided I was ready to be part of this, and when did I agree?
RESPONSE — no record found
The robotaxi will get you there most of the time. It will do it without road rage, without a drink in its system, without a phone in its hand. Credit where it's due.
But it will also do it without a single reason to value the one thing you can't get back. It is very good at driving. It will never once care whether you arrive.
Decide accordingly.