Stanford Professor Strives to Fix Driverless Cars’ Major Flaw: Ethical Decision-Making

It’s nothing more than a dune buggy on a cordoned-off street but it’s headed for trouble. A jumble of sawhorses and traffic’ cones simulates a road crew working over a manhole and the driverless car has to make a decision — obey the law against crossing a double-yellow line or break the law and spare the crew. It splits the difference, veering at the last moment and nearly colliding with the cones.

“I imagine that wasn’t the most comfortable experience for you,” Chris Gerdes, a boyish and bespectacled Stanford engineering professor, calls out to the slightly shaken passenger in the car.

Gerdes is suddenly causing a great deal of discomfort to automakers and tech giants. Raised in North Carolina in the shadow of the Charlotte Motor Speedway, he has been at the enthusiastic forefront of the driverless car movement, monitoring the brains of top racecar drivers in action and programming cars to imitate them. As he notes in his TED talk and other public appearances, he and his students have programmed their Audi race car, Shelly, to flawlessly make the 153 turns on the 12.4 miles of the Pikes Peak trail in Colorado — with no one at the wheel.

But as the autonomous car movement barrels ahead, Gerdes has gone from enthusiast to conscience, if not quite scold. He is raising questions about ethical choices that must inevitably be programmed into the robotic minds expected one day soon to be driving along the nation’s highways. And since Gerdes, who favors bluejeans and straight talk, is no tweedy Luddite railing against the evils of technology, the industry is paying attention; top executives are pouring into his lab in Palo Alto.

“Within the autonomous driving industry, Chris is regarded as Switzerland, he’s neutral,” said Patrick Lin, a philosophy professor at Cal Poly who spent a year working with Gerdes in his 7-bay garage filled with robot cars. “He’s asking the hard questions about ethics and how it’s going to work. He’s pointing out that we have to do more than just obey the law.”

On a recent day, Gerdes met separately at his lab with the chief executive officers of General Motors Co. and Ford Motor Co. That came about a week after he hosted a workshop on driverless ethics for 90 engineers and researchers, including from electric carmaker Tesla Motors Inc. and tech giant Google Inc., which has pledged to put out a robot car as soon as 2017. This year, Tesla will introduce an auto-pilot feature. GM will debut a 2017 Cadillac that drives hands free. Ford CEO Mark Fields says driverless cars will arrive by 2020.

Gerdes’ message: not so fast.

“People often say the technology is solved, but I don’t quite believe that,” he said in the conference room as his students nearby buried their heads under the hood of an autonomous Ford Fusion nicknamed “Trudy.” “There’s a lot of context, a lot of subtle, but important things yet to be solved.”

Take that double-yellow line problem. It is clear that the car should cross it to avoid the road crew. Less clear is how to go about programming a machine to break the law or to make still more complex ethical calls.

“We need to take a step back and say, ‘Wait a minute, is that what we should be programming the car to think about? Is that even the right question to ask?’” Gerdes said. “We need to think about traffic codes reflecting actual behavior to avoid putting the programmer in a situation of deciding what is safe versus what is legal.”

Gerdes, 46, who is training to be a racecar driver, was initially dismissive of the need to grapple with philosophy. Autonomous cars programmed with robot reflexes and precision were on track to drastically reduce the 33,000 U.S. highway deaths a year. Wasn’t that moral enough? But then three years ago George Bekey, co-author with Lin of a book entitled “Robot Ethics,” e-mailed Gerdes.

“My first thought was, ‘Ethics? Automated Cars? This seems like a bit of a fringe topic’,” Gerdes said.

He soon came to see both its significance and its painful complexity. For example, when an accident is unavoidable, should a driverless car be programmed to aim for the smallest object to protect its occupant? What if that object turns out to be a baby stroller? If a car must choose between hitting a group of pedestrians and risking the life of its occupant, what is the moral choice? Does it owe its occupant more than it owes others?

When human drivers face impossible dilemmas, choices are made in the heat of the moment and can be forgiven. But if a machine can be programmed to make the choice, what should it be?

“It’s important to think about not just how these cars will drive themselves, but what’s the experience of being in them and how do they interact,” Gerdes said. “The technology and the human really should be inseparable.”

Gerdes’ lab is a working garage, with shelves full of transmissions donated by Ford and an electric motor in a crate given by Google. Self-driving Audis, Nissans and Fords fill the bays, as his band of graduate students and industry veterans works on research projects for major automakers such as Volkswagen, one of 30 companies that underwrite the lab.

Gerdes’s racing roots run deep. An uncle who worked for Chrysler once designed a race car for Mario Andretti. At 10, Gerdes spent an afternoon in a shopping mall chatting with Dale Earnhardt, Sr., then a rookie driver who couldn’t attract a crowd.

When he wasn’t racing his slot car, Gerdes had his head in Isaac Asimov’s 1950s science fiction novels, which became the rulebook for robots that ethicists still use. Asimov’s first law: An autonomous machine may not injure a human being or, through inaction, allow a human to be harmed.

Once a month, Gerdes takes his crew and cars up to Thunderhill Raceway, four hours north of Stanford, to test the outer limits of autonomy. On pace to earn his racing license this month, Gerdes requires all his students to learn to take hot laps at Thunderhill. He believes that to automate driving, it needs to be understood at its most extreme.

As a result, the best and brightest in Silicon Valley flock to his garage.

“Chris’s mantra on research is: If it’s not cool, we’re not doing it,” said Sarah Thornton, a Ph.D. candidate in mechanical engineering who spun out twice while learning to race at Thunderhill.

In a Back-to-the-Future nod, Gerdes will unveil later this month a self-driving DeLorean nicknamed Marty.

It is clear, then, that even as he raises hard questions Gerdes believes in this quest.

“With any new technology, there’s a peak in hype and then there’s a trough of disillusionment,” Gerdes said. “We’re somewhere on that hype peak at the moment. The benefits are real, but we may have a valley ahead of us before we see all of the society-transforming benefits of this sort of technology.”