Ownership: It's Impossible to Tell How Safe Self-Driving Cars Are

What Crash Test Ratings Really Say

  What Crash Test Ratings Really Say No matter how your car performs on the road or what its other capabilities are, its ultimate job comes down to safety: How does it protect you and your family in the event of an accident? No matter how your car performs on the road or what its other capabilities are, its ultimate job comes down to safety: How does it protect you and your family in the event of an accident?

Why It's Impossible to Tell How Safe Self-Driving Cars Really Are © Getty Why It's Impossible to Tell How Safe Self-Driving Cars Really Are

The case for self-driving cars rests largely on their propensity not to crash, and the generally accepted idea that they'd do better than harried, angry humans behind the wheel. For example, a 2012 study by KPMG and the Center for Automotive Research predicts that a self-driving fleet could eliminate 93 percent of crashes attributed to human error. Google isn't shy about noting that its self-driving cars have racked up more than 1.3 million miles of real-world testing since 2009 but have been involved in just 17 crashes, all but one caused by human error.

7 Of The Coolest Cars At Autocon San Francisco 2016

  7 Of The Coolest Cars At Autocon San Francisco 2016 ICYMI here are the coolest cars of Autocon San Francisco '16We'll make any excuse to go to northern California. It's very likely that someone up there is building a car that's uniquely NorCal. Just like Japan, a different region will often yield a different style. Autocon's second stop of '16 took place at the San Mateo Convention Center in Silicon Valley of San Francisco, Calif., and as usual brought out the finest cars in the area. But of course, there were some that stood out more than others. Here are seven of the coolest cars you missed out on during Autocon San Francisco.

The practical question, however, is: Just how safe do these vehicles need to be before we'll let them intermingle at large with human-driven cars on the highway? And here there's big problem, at least when it come to proving the case for autonomous cars scientifically. According to a RAND study released today, autonomous cars would have to be driven hundreds of millions of miles and, under some scenarios, hundreds of billions of miles to create enough data to clearly demonstrate their safety. "Our findings demonstrate that developers of this technology and third-party testers cannot simply drive their way to safety," the report says.

The study springs from a question its coauthor, RAND's Nidhi Kalra, asked herself while making a presentation on the question of how safe autonomous vehicles should be before we put them on the road. "My initial question was; If we assume they're perfect, how many miles do autonomous vehicles have to be driven to prove they're [safer] than human drivers?"

The Road to Ruin: Potholes Are Costing Drivers $66 Billion a Year

  The Road to Ruin: Potholes Are Costing Drivers $66 Billion a Year Anyone who has spent any time in a car recently knows that America’s roads are a mess. Research Research New Used New & Used Make (e.g. Mazda) Model (e.g. CX-5) And anyone who follow politics knows why: Repairing those creaking bridges and crumbling roads requires a lot of money: The Federal Highway Administration estimates that it would require $170 billion a year to significantly improve the country’s roads. (Click through the slideshow above to see the 10 states where broken roads cost drivers the most.

Nidhi recognized that no one knew. Nor did they know how many test-miles would be necessary to prove the safety of self-driving cars that, like Google's car, are very good but not quite perfect. She began making literal back-of-the-envelope calculations and then took the exercise to RAND's researchers and statisticians, who turned it into a full-on study. "We asked whether it was practical to answer these questions with test driving," Kalra says. "The answer is no."

The chief reason why this is so is that Americans drive like crazy. We rack up nearly 3 trillion miles every year, according to the Bureau of Transportation Statistics. In 2013, there were 2.3 million injuries reported, a rate of 77 injuries per 100 million miles driven. The 32,719 deaths from car crashes that year correspond to a rate of about one fatality per 100 million miles driven. "Even if you're not a statistician, you can say: if I want to prove through test driving that autonomous vehicles are safer than humans, I've got to drive them at least 100 million miles without a fatality," Kalra says.

When Scientology Sponsored Race Cars

  When Scientology Sponsored Race Cars Welcome to 1988's bizarre intersection between motorsports, religion, and sponsorship.​​

RAND concludes that under even the most aggressive test driving schedule you could put out there, existing fleets of autonomous cars would need decades and even hundreds of years to log enough miles to adequately show their safey vs. human-driven vehicles.

"We know that Google has driven [theirs] 1.3 million miles without a fatality," Kalra says. "That's commendable, but it's two orders of magnitude shy of what one would need to do to make some statistical claim about their performance with respect to fatalities." The study's co-author, Susan M. Paddock, adds: "Even if autonomous vehicle fleets are driven 10 million miles, one still would not be able to draw statistical conclusions about safety and reliability."

So what to do? Kalra thinks that we may have to put the technology on the road before we're certain how safe it is. If there's no plausible way to establish their safety in a laboratory setting; she speculates that the alternative may actually be putting AVs on the road in a controlled setting. "So we can learn in a very intentional way rather than learning through tragedy or mistakes that have to be rolled backward."

Follow MSN Autos on Facebook

The Leftovers: Camaro 1LE vs. M2 vs. Focus RS vs. 124 Spider Abarth vs. 718 Boxster S vs. 86 .
The Leftovers: Fighting the Sin of Omission with a Second Performance-Car ShootoutSuch was the sacrifice we made when choosing the field for our 2016 Best Driver's Car test. It was an embarrassment of riches in terms of jealousy-inducing vehicular choice. We could not bring 'em all for reasons ranging from available time and manpower to the width of the runway where we film the World's Greatest Drag Race. Being limited to 12 cars meant that to bring the Fiat 124 Spyder Abarth, we would have to cut the Aston Martin V12 Vantage S. Sorry, Fiat. And so the culling began.

See also