A Safety Tolerance for Autonomous Cars?

Personally, I've yet to even see a driverless car. At the moment, the New York state legislature is preparing to review bill A31, which would establish protocols for testing autonomous vehicles throughout the state. If passed, driverless cars could share roadways within a few months. New York will be an interesting testing ground for driverless technology, as traffic varies from congested metros to endless highways and rough terrain, sometimes with all four seasons experienced in a single day.

Not that I'm the least bit worried, though there could be some growing concern. On Valentine's Day last month, a Google autonomous car pulled out in front of city bus--it was the first case where a driverless car was clearly at fault for a collision. This came after several years and 1.3 million miles of road testing to date. Google engineers feel that the accident was a necessary lesson, and tweaked the car's driving algorithm to represent that large vehicles often behave differently in traffic than cars. Last November a Google car was pulled over for driving too slowly and impeding traffic.

In fact the majority of accidents with autonomous cars have been the fault of humans. In July 2015, all 14 incidents to date were due to human decision making. Tesla reported early issues with its autopilot system were because people expected too much from it. (Though perhaps they should have called it assisted driving, not autopilot.)

U.S. Transportation Secretary Anthony Foxx believes that autonomous vehicles could reduce traffic accidents by 80%. He said that autonomous vehicles can't be held to impossible standards, when they are clearly an improvement on human drivers. In January, his office announced a $4 billion investment in developing and regulating the technology.

It's a very real possibility that due to all these events, the timeline for integrating larger masses of autonomous vehicles has accelerated. More states are allowing driverless cars to be tested and operated within their borders. Google (amongst others) are making swift changes to their driving technologies.

But do driverless technologies need to be 100% accurate? They're likely already better than human drivers, and there is always going be some risk or tolerance. Google engineers spoke about how imperative it is to get real world feedback. How does this conversation shift when the first fatal accident involving a driverless car occurs?

That will ultimately determine if 2020 sees the first waves of driverless cars as Google plans, or if more demanding standards need to be implemented, which could push the technology out another 5-10 years.