Self-driving cars are the future.
Google’s efforts are some of the most visible: it has 58 self-driving cars on the road that have cumulatively driven 2.6 million+ kilometres since 2009.
After all, Google’s cars use a combination of cameras, radars, and lasers to “see” as far as two football fields away, with 360 degrees of visibility. Humans can only see about 120 degrees around them while they drive.
But that underlying idea — that today’s test cars are safer than human drivers — is “disingenuous” and an example of “technofetishism,” says Etsy CTO John Allspaw.
He believes that comparing self-driving car tests to human behaviour completely misses this point.
How I know the self-driving car report from Google is unhelpful: on non-self-driving accidents: “94% of them involving human error”
— John Allspaw (@allspaw) June 6, 2015
He says that using autonomous vehicle test driving as a comparison to human safety doesn’t work. He cites a recent Rand Corporation study that shows that autonomous vehicles would need to drive “hundreds of millions of miles and sometimes hundreds of billions of miles to demonstrate their reliability in terms of fatalities and injuries” compared to humans.
“The idea that self-driving cars are better than humans is bulls***,” he said.
Allspaw made his initial comments during a dinner attended by Business Insider, but followed up via email:
“Self-driving cars are getting better, but there’s a long way between now and the world that is promised, because safety is a complex phenomenon,” he writes.
“You can’t just extrapolate Google cars driving ~1.5 million miles under specific conditions (weather, topology, construction, traffic, accidents around it, etc.) to usurping the ~3 trillion miles/year under all conditions in the US.
“1.09 fatalities per 100 million miles is the current non-self-driving numbers.”
He advocates for new testing methods to demonstrate the safety and reliability of cars as well as the understanding that regulations need to evolve at the same time as the tech.
Although Allspaw doesn’t claim to know the details of how companies like Google or Tesla think about future testing, he does believe that anyone passionate about the future of self-driving cars needs to stop emphasising the idea of their current safety.
“In order to truly make progress on any sort of ‘autonomous’ tech (cars or planes or whatever), we have to take a holistic approach that helps understand what makes humans successful, not about looking for failures of human perception and solving for that,” he said.
“2014 had ~30k fatal crashes out of the 3 trillion miles travelled. We have to understand not how those crashes happened, but what makes the vast majority of them not happen. Luck is not a contributor, expertise is.
“Understanding human expertise is the key, not human frailty.”