Commentary: Self-Driving Cars Likely to Bring Spike in Crashes Before Benefits Seen

July 11, 2016 by

Every car on the roads is an accident waiting to happen. It’s about time Silicon Valley woke up to the fact.

The boom in autonomous-driving technology offers the prospect of one day reducing the roughly 90 percent of road accidents caused by so-called human factors - drivers or pedestrians who are drunk, or distracted, or tired. But Tesla’s first fatality from a car crashing in its self-driving mode is a sad reminder that the industry has a long way to go, and perhaps needs to slow down.

A two-ton car traveling at 70 miles per hour (113 kilometers per hour) carries about the same energy as a kilogram of TNT exploding, so it shouldn’t be a shock to learn that maneuvering hunks of metal along crowded highways at speed carries risks. About 32,719 people were killed in road accidents in the U.S. during 2013, according to the World Health Organization, or one person every 16 minutes. Even if self-driving vehicles could reduce those fatality rates by 90 percent, they still would be causing more deaths than the Sept. 11 attacks, every year.

And getting to that point will involve climbing a mountain of road deaths. As Gadfly wrote in April, there’s a longstanding dynamic where new transport technologies tend to produce spikes in fatalities before engineers learn the lessons of real-world accidents and make systems safer than the status quo. Driverless cars are unlikely to be an exception.

Rand Corp.’s Nidhi Kalra ran the numbers on this and estimates that to prove even a 20 percent improvement in safety would require self-driving cars to put in 5 billion miles on the roads – equivalent to driving to the planet Neptune and back, or running a fleet of 100 autonomous test vehicles 24/7 for 225 years.

That’s obviously not practical, and waiting for statistical proof might even result in avoidable deaths if an excess of caution were to hold back the adoption of technology that ultimately is safer.

Still, Silicon Valley’s condescension toward traditional automakers for their caution over autonomous driving needs to stop. One might also expect a pause in the tendency for carmakers’ forecasts to advance from an adoption date around 2025 to 2020 or even sooner. (Tesla’s Elon Musk has said the technology will be ready by 2018.)

Google’s autonomous car has had difficulty distinguishing between rolled-up balls of newspaper and rocks in the road, while Tesla’s crash occurred after the car’s autopilot failed to spot a tractor trailer crossing the roadway because its white side was hard to distinguish from the sky behind it .

A more realistic outlook may also have knock-on effects on the capital flowing into the industry. Startups in the automotive and transportation sector have attracted $21 billion of funding over the past five years, according to data provider CB Insights.

It’s likely that only a small slice of that money will have gone to autonomous cars - the bulk of it was probably to Uber. But the likes of Zoox, Drive.ai, and New York-traded Delphi Automotive, which have obtained permits to test self-driving autos in California, could find that capital is more reluctant to bet on a quick return than it was previously. Internal autonomous-driving teams at companies such as BMW, Toyota, General Motors and Volkswagen may also find that managements have more pressing priorities.

It’s possible that Elon Musk, too, will pause. He’s already got a lot on his plate: The planned merger with SolarCity received a lukewarm reception from investors, and the U.S. highway safety regulator is looking into a potential issue with the Model S’s suspension . Beyond that, Musk plans to send a spaceship to Mars in 2018 and has a non-commercial interest in a 760 mile-per-hour levitating train.

Tesla is unlucky to suffer the first fatality from autonomous driving, but the tragedy contains a salutary warning: Even the best transportation technologies regularly kill people.

The Silicon Valley approach – roll out a beta version and let users figure out potential problems - was always going to be a risk when sent out on the roads. Highways are complex and dangerous places, and no quantity of sensors or processing power is going to change that.

This column does not necessarily reflect the opinion of Bloomberg LP and its owners.