:format(webp)/cdn.vox-cdn.com/uploads/chorus_image/image/63705282/photo.0.1462601078.0.jpg)
What to do when an academic study says your robot cars might be more dangerous than regular cars? Hire your own academic!
That’s what Google did.
On Friday, a study came out from Virginia Tech, commissioned by the Internet giant, claiming to offer the first apples-to-apples comparison between conventional vehicles and self-driving ones. The study looked at Google’s autonomous cars, which have logged over 1.3 million miles, and found that they have a lower crash rate (3.2 accidents per million miles) than national rates (4.2 crashes per million).
The research comes two months after another study that said quite the opposite. That one, from the University of Michigan, looked at Google’s cars, plus self-driving vehicles from Delphi and Volkswagen, and found they are more accident prone.
This is a major fixation of Google’s self-driving car unit, which is set to become its own Alphabet company that will — everyone (like, everyone) in the industry believes — be in the business of robot taxi fleets. Google has reported each of its incidents, noting that they are all caused by human error. (Often, it seems, curious drivers of other cars are distracted or lured closed by the weird robots on wheels.)
But regulators and consumers will only be watching autonomous vehicles more closely. And anything that suggests they are dangerous throws a wrench in Google’s plans.
The Virginia Tech report points to a big flaw in earlier studies: Most drivers don’t report minor accidents. Google does. When that’s factored in, self-driving cars are safer.
But the Michigan study also controlled for that and found similar results. More importantly, that study didn’t have a strong enough statistical significance — i.e. the findings could be bunk.
Still, it seems Google finds reports like these worth the time and money to shoot down. After all, you can’t bring a knife to a robot fight.
This article originally appeared on Recode.net.
