Skip to main contentSkip to navigationSkip to navigation
A Tesla Motors Model S car.
A Tesla Motors Model S car. Joshua Brown was killed when his self-driving car drove itself at around 60 mph under the trailer of an articulated lorry that had turned across the road in front of him. Photograph: John G. Mabanglo/EPA
A Tesla Motors Model S car. Joshua Brown was killed when his self-driving car drove itself at around 60 mph under the trailer of an articulated lorry that had turned across the road in front of him. Photograph: John G. Mabanglo/EPA

The Guardian view on self-driving cars: attention needed

This article is more than 7 years old
Any system which requires humans to play backup to computers will be less safe than one which works the other way round

The death of Joshua Brown may well be remembered for as long as the death of William Huskisson, the member of parliament run over by George Stephenson’s Rocket in 1830. Mr Brown, a Florida businessman, was killed when his self-driving car drove itself at around 60mph under the trailer of an articulated lorry that had turned across the road in front of him. Tesla, the manufacturers, put out a statement explaining that neither the car’s cameras nor its driver had been able to detect the dazzling white side of the truck trailer against the dazzling Florida sky. It is easier, perhaps, to believe that the cameras failed while the driver wasn’t actually looking very hard.

The accident is more than a curiosity for several reasons. Self-driving cars might provoke as much of a social and economic revolution as the railways did. If that happens, Mr Brown will be, like Mr Huskisson, the first of many. But of course it is a claim of the technological enthusiasts that machines will save lives. Cars driven by artificially intelligent computer networks will be safer than those driven by fallible humans. And there is some statistical evidence that this might be true: semi-autonomous vehicles like the Tesla have now driven 130m miles with one fatality, whereas among all cars in the US there is a fatality every 94m miles. This is itself a remarkably low figure by world standards: in Mexico motor accidents kill far more people than the drug wars do. An optimist could argue convincingly that autonomous cars might save far more lives than the alternatives if they ever become truly mass-market devices.

The pessimist can put up good reasons why this should never happen. The first is technological: although it is clearly possible to make sensors and control systems which can detect almost everything that happens while driving along well-regulated roads, the difficulty of eliminating the last little portion of risks is considerable. A camera that can’t distinguish the side of a trailer from the sky behind it is one example. There will be other failures. Autonomous cars are notoriously lacking in the ability to interpret the human motivations behind the small aggressions when people jockey for position in traffic. Related to that is the problem of oversensitivity to disruptions that a human could instantly recognise. A car that can’t tell the difference between a paper bag blowing down the pavement in the wind, a dog running out into the road, or a small child doing the same, had better be programmed to slam on the brakes for paper bags. That will provide the passengers with exactly the kind of inconvenient experience they are paying to avoid but it is still clearly better than the alternative.

So who is responsible when a car that is driving itself kills someone? At the moment, the answer is the human driver. Tesla cars are not supposed to be driven without human attention – the driver is meant to be ready to take over at a moment’s notice. But this is transparently unrealistic. Plenty of drivers are inattentive at the wheel even in fully manual cars, and the experience of the car driving itself while you do something else must be one of the compelling pleasures of owning such a shiny gadget. The paradoxical lesson of this crash is that cars will be safer when they take more decisions away from humans.

This article was amended on 5 and 7 July 2016. An earlier version incorrectly described William Huskisson as the first railway accident victim in history. He died in 1830, not 1827.

Most viewed

Most viewed