The biggest myth about the development of self-driving cars

AI is sophisticated enough to take over the basic function of driving a vehicle, but driving entails so much more than just navigating a car. Human beings perform a vast variety of tasks in a car, and much of that job is dealing with other humans. Look at the role of the driver: only a part of his or her attention is focused on actually operating the vehicle. There are also ongoing interactions with others in the car—”Uh-oh, Michael is carsick and about to throw up! Can you turn down the music?” “Hey, you two kids had better stop fighting or I’m going to pull over.” Or, for example, if another car is about to cut you off, your passengers will automatically check to see if you are aware of the situation and are prepared to react.

The fact is, human beings won’t completely be out of the driver’s seat anytime soon. We still have the edge over AI because we think like other human beings. We anticipate that other drivers or pedestrians may not always follow the rules of the road, or they may behave in irrational ways. If we see children playing ball near the road, we will mentally take note and be prepared to stop if a child runs into the street to retrieve the ball. We know this not from a driving manual, but from life experience. We instinctively respond to situations in a very human way. Eventually, learning algorithms will be developed to imitate some of our gut reactions, but this will take time.

So far, there’s been little effort on the part of either the automobile industry or public safety officials to bring the public up to speed. When people hear the term autonomous car, they are often under the mistaken belief that the role of the human driver has been eliminated, that he or she is now relegated to the role of passenger.

“The biggest myth about automation is the more automation, the less you need human expertise. Actually, the more you automate, the more you need to educate, where, when, how etc.,” observes Bryan Reimer, PhD, a research scientist at MIT’s Center for Transportation and Logistics, a researcher in the AgeLab, and associate director of the New England University Transportation Center.

Dr. Reimer, who has studied driver behavior as it relates to automation, raises an interesting point: As our automobiles assume more and more of the tasks of their operation, there is a risk that human beings will lose some of the real-world driving experiences that make us seasoned drivers. This is fine as long as the car remains in control. But in a world of semiautonomous vehicles, the drivers may become less equipped to handle the complicated maneuvers that would stump the software. As Dr. Reimer notes, “We unfortunately will get worse at driving, as humans learn from doing. The less we do, the less we learn. That’s why a lot of the risks of this mixed system go up over time. If we’re no longer doing, we’re no longer learning. So, that means the future is one of novice drivers, and we all know that novice drivers are even more risky than established, trained drivers.”

So, there is a catch-22 here. If we lose our driving skills due to semiautomation, we won’t be able to be as effective partners with our semiautonomous cars. Furthermore, the very nature of semiautonomous vehicles may lull some people into a false sense of security.

In addition, when we get behind the wheel of a car equipped with lots of sophisticated automated features, human beings revert to behavior that can actually make us less safe. It has to do with what psychologists call cognitive load, a field of research that began with a study of mice in a maze. In 1908, psychologists Robert Yerkes and John Dillingham Dodson found that mice given a low electric shock were motivated to finish a maze, but if the shock was too high, the mice gave up. This led to the famous Yerkes-Dodson law we still refer to today; its bell-shaped curve shows the relationship between arousal and performance.

When you move to the left of the curve, you are less aroused, and when you move to the right of the curve, you are over aroused. Finding the right amount of stress is tricky: If stress levels rise too high, it can result in cognitive overload, and people, like mice, shut down. If, however, stress levels dip below a certain point, there is a risk of cognitive underload, which can also impair human performance. For optimal performance you need to land on the just right spot on the curve. But maintaining the right level of arousal in a car can be tricky.

For decades, it was assumed that automation induced underload and would make people drowsy or less alert. In some instances, that was true. In the late 20th century, researchers noticed that this was exactly what was happening to pilots in cockpits as the once-labor-intensive tasks of flying a plane were replaced with computerized navigation and safety equipment. But cognitive underload in an automated car looks different from, say, that in a cockpit. When we are understimulated and veer to the left side of the bell curve, we often become bored and look for some other, nondriving activity to occupy our time, such as texting, making a call, eating, or watching a video.

“And that’s where the fundamental premise of state management comes in, helping people to make better moment-to-moment decisions,” Reimer explains.

Car companies are beginning to add features to keep drivers more fully engaged during periods when the car is in control. General Motors’s 2018 Cadillac CT6 offers a Super Cruise system that enables “the first true hands free driving system for the freeway,” on some U.S. highways, but it has a camera embedded in the steering column that tracks the driver’s head position and eye moments to make sure that his eyes are on the road.

The ProPILOT Assist system offered on the Infiniti QX50 and Nissan Leaf takes a different approach to keep drivers engaged: in order to operate it, the driver must keep both hands on the wheel. These features help, but enabling cars to recognize and understand our emotions could also gauge human attention and mood, and keep the driver alert and ready to take the wheel.

The way we approach driver training will likely need an upgrade to accommodate the new world of semiautonomous cars. Perhaps we will need to periodically take refresher classes to maintain our driving skills, or practice on tracks like the Veoneer Smart City, which is filled with the kinds of unexpected challenges that occur in real life. But bottom line, it doesn’t look like we’re going to have to shut down our driving schools anytime soon.

Rana el Kaliouby is the cofounder and CEO of Affectiva. This essay was excerpted from her book Girl Decoded: A Scientist’s Quest to Reclaim Our Humanity by Bringing Emotional Intelligence to Technology.