Road to self- driving cars may be perilous

WASHINGTON – Some years ago, a journalist named Ryan Avent bet that his newborn daughter would never need to get a driver’s license because self-driving cars would be available before she reached her 16th birthday. While that bet still has a few years to run, it’s safe to say he lost. Come 2026, when his daughter turns 16, if she wants to move freely outside a few test zones and a handful of dense, walkable urban areas, she’ll need to learn to pilot a car.

Of course, automakers have made remarkable progress. Last October, I rode one of Waymo’s self-driving cars around Chandler, Ariz., and found the experience almost disappointingly uneventful. The service is also available in San Francisco, and last week the company announced it would be teaming up with Uber in the Phoenix area to provide rides and deliver food. None of this was possible in 2010, when Avent made his bet.

Yet the disappointments have been considerable, including some fatalities. Multiple companies either shut down or delayed autonomous driving projects after running into hurdles. All innovations involve a lot of failure, of course, but one still has to wonder why this particular one is taking so long and causing so many casualties.

Among the many answers, one of the most important is what can be called the either/both problem: The self-driving future will probably be superior to the human-operated present, which every year kills tens of thousands of people and wastes hundreds of person-hours per driver. But even an all-human system might be preferable to a mixed environment in which sometimes the cars are driving and sometimes the humans are.

This problem exists at both the micro and the macro levels. Inside the car, partial automation is worse than either full automation or none, because it requires the driver to stare at the road, hands on the wheel, ready to take over in an emergency. Unsurprisingly, watchdogs routinely chide drivers of semiautonomous vehicles for failing this responsibility – getting distracted or playing with their phones, allegedly sometimes even when they’re getting paid to keep their eyes on the road.

It will be a major challenge to keep drivers paying attention while the technology crosses the godawful gap between manual and fully autonomous operation. And that problem stands to get worse before it gets fixed, because as the cars improve, the drivers’ temptation to let their attention wander will keep growing. At the same time, their driving skills will begin to atrophy since they’re so rarely actually driving.

We could conceivably train drivers for emergencies on simulators, like pilots. But it’s hard to imagine getting ordinary drivers to log more time driving virtually if they also have to spend so many mind-numbing hours behind the wheel pretending to drive.

Meanwhile, a larger variant of the either/both problem will arise: Roads carrying both computer and human drivers might well be worse than those with all one or all the other. Self-driving cars have many advantages over human drivers: They never get tired or drunk or distracted. But they are also really bad at a lot of things we do easily, including predicting what pedestrians or other human drivers are likely to do. This is pretty fundamental to many basic driving tasks, such as turning left across traffic – something that self-driving cars still struggle with.

It’s easy enough to imagine a fully automated world in which everything – from signs to road design to the cars’ autonomous algorithms – are optimized for computer drivers. This world will undoubtedly be safer than the current one, and ideally, future generations will view traffic fatalities the way we view cholera: as a quaint abomination. But first, humans and computers will have to share the roads without always understanding what the other is doing.

Even if self-driving cars are safer than their human counterparts – as they might well be – their oddities stand out. Notice that the few known fatalities involving autonomous driving systems often become national news, while legions of horrific drunken-driving incidents remain local stories. As more self-driving vehicles take to the roads, capable of greater autonomy, expect this disparity to widen. At some point, people might start asking whether it’s worth the cost to push through the messy middle.

The answer is a resounding yes. Too many hours, and lives, are currently lost getting ourselves from one place to another. If we want to reach a time when we’re able to reclaim them, however, we’ll need to prepare for some expected turbulence en route.

––

Follow Megan McArdle on Twitter, @asymmetricinfo.