The question the column asks has 220.6 million miles of answer

Adrian Chiles asked in The Guardian on Thursday whether driverless cars will do any better than the humans they replace, and answered his own question with a doubt. That question has a published answer, and it runs to 220.6 million miles.

Waymo's safety hub covers rider-only driving through March 2026, split across Phoenix, the San Francisco Bay Area, Los Angeles, Austin and Atlanta. The city figures add up to 220.613 million miles, so the headline total checks out. Against a human benchmark for the same roads:

• 82% fewer crashes causing any injury, which is 707 crashes that did not happen.
• 94% fewer crashes causing serious injury or worse, 47 fewer.
• 93% fewer pedestrian injury crashes, 84% fewer for cyclists, 84% fewer for motorcyclists.

In rates rather than counts, the human benchmark for those cities is 3.91 injury crashes per million miles and the Waymo figure is 0.71, five and a half times lower. San Francisco produces the widest gap at 7.25 against 0.70.

Two things must be said about that data before anyone leans on it. Waymo publishes it, Waymo builds the human benchmark it is compared with, and the company updates the hub on the reporting schedule the US road safety regulator sets. And it covers one operator in five cities with good weather and mapped streets, which is not the same claim as driverless cars in general.

Even discounted for both, the empirical half of the column does not survive. The neural network doing the driving is not matching human drivers on injury crashes. It is well ahead of them, across six categories, in a dataset an order of magnitude larger than most published safety studies.

The half the numbers cannot touch

The other half of the column is a different argument, and the numbers do not touch it.

“But there's a difference between labour-saving and lobotomising.”

— Adrian Chiles, The Guardian, 10 September 2026

Adrian Chiles, The Guardian, 10 September 2026

He is writing about de-skilling, about what happens to people who no longer practise a difficult craft, and he is right that no crash statistic answers it. Automatic transmission and satnav both took something away, as he concedes about his own driving, and the case that driving is the line worth defending is a cultural one that data cannot settle either way.

Where the column goes wrong is in tying that argument to a factual claim it did not check. If driverless cars were roughly as dangerous as people, the de-skilling case would carry the whole weight. With 707 injury crashes on one side of the ledger, the trade is between a craft and an injury count, and the column does not present it that way.

Who asked for them

Chiles also asks who asked for these cars. The money did. Waymo and Zoox are expanding across US cities, and Uber put $100 million into Travis Kalanick's robotaxi startup this month. That is not a public referendum, and it is the only vote that decides whether the cars appear on a given street.

Read the column for what it does well. Nobody consulted drivers about any of this, the tech industry does spend brilliant minds on strange problems, and potholes really are still here. Just do not read the doubt in the middle of it as an open question.

Nothing here should be taken as financial advice; treat it as information to consider.