Of hard drives and crashes

These days, in fact, airplanes do essentially fly themselves, with pilots and crew on board primarily to serve drinks and intervene in emergency situations. We let computers control much of our air travel, but ground transportation is still largely reliant on human drivers who navigate cars, buses, trucks and motorcycles, and with much more disastrous results than airplane autopilots.

There’s a lot that can go wrong with humans behind the wheel — they can be drunk, tired, distracted, rushed or reckless — similar to the drawbacks to human involvement in medicine, as discussed in last week’s artificial intelligence (AI) column. Just as medical AI machinery saves lives, so too can transportation AI technology. One especially promising and rapidly emerging option is the development of vehicles that drive themselves.

Though today’s airplanes are commonly piloted autonomously, we still don’t trust computers to navigate our cars. But cars are by far the more dangerous and damaging mode of transportation. In the United States alone, there are more than six million car accidents each year and more than 100 deaths every day due to accidents. There are far fewer airplane fatalities.

Simply put, we aren’t very good drivers. We could use some help.

Many technology experts now predict that help will come in the form of cars that drive themselves more carefully, reliably and safely than we can. In pursuit of this goal, the Defense Advanced Research Projects Agency (DARPA) established a Grand Challenge event in 2004 for autonomous vehicles to navigate a 150-mile route through the Mojave Desert. The most successful vehicle, developed by a team at Carnegie Mellon University, made it only 7.36 miles.

In the months following these disappointing results, a group of Princeton students approached operations research and financial engineering professor Alain Kornhauser about starting an autonomous vehicle project on campus, and they designed a car of their own for the 2005 Grand Challenge. In the 2007 Grand Challenge, Kornhauser led another student team. Six cars — not including the Princeton team’s vehicle — finished the entire race, a mere three years after the unimpressive 7.36-mile showing at the first competition.

With technology improving this rapidly, some researchers think we may be close to eliminating human drivers altogether.

“This is the future,” Kornhauser said in an interview last year. “It’s silly for us to sit there and steer these vehicles.”

Engineers face numerous technical challenges, however, in designing an autonomous car, including teaching it to steer, accelerate, brake and identify things like road obstacles, highway lanes, pedestrians and traffic lights.

Even when these mechanics are perfected, people will still harbor a deep-seated distrust of cars they don’t control.

The Aer Fungus joke says it all: When it comes to computers, we expect that everything will go wrong, we suspect they’re always about to crash, we doubt their reliability. Ultimately, we find it hard to trust them with our lives. It’s one thing to have the computer crash and lose your half-written term paper, it’s quite another to have the car crash.

“We have certain kinds of cognitive biases about technology, and it’s comforting to people to be physically in control of their car,” said Timothy Lee GS, a student at the Center for Information Technology Policy who recently wrote a series of essays on autonomous vehicles for arstechnica.com, a technology media site.

Get the best of the ‘Prince’ delivered straight to your inbox. Subscribe now »

“The idea that they might die because of a computer error will freak people out a lot more than the idea that they might die because of their own mistakes, but this bias will undoubtedly cost lives in the long run,” he said in an interview.

These biases stem in part from the frequent failings of the laptop and desktop computers with which we are so familiar. Because these machines serve such a different purpose from the ones that would navigate cars, they have been designed to optimize different characteristics. For instance, engineers who build home computers try to make the machines as fast and efficient as possible, since users value speed very highly. Similarly, laptop engineers emphasize compact design and lightweight hardware for portable models.

Optimizing these characteristics comes at the expense of others, however. So our personal computers may crash or freeze relatively often since these malfunctions don’t usually have very serious consequences; you may lose an unsaved document, but you’re unlikely to die. Autonomous vehicle technology, of course, will need to focus less on lightweight hardware and more on reliability and failsafe backup systems.

Still, anyone who has ever lost a hard drive can’t help reacting to the idea of computer-driven cars by wondering how we can possibly entrust our laptops with our lives.

To help us adapt to the idea of autonomous vehicles, the automobile industry may incorporate smaller pieces of AI technology into their cars, Lee predicted. As more of these sophisticated safety technologies are added, our cars may gradually come to resemble semi-autonomous and, eventually, fully autonomous vehicles. Thus, we would be spared the abrupt shock of switching directly to completely computerized cars.

“Pretty soon, you’ll start to see cars that do slightly more sophisticated things with artificial intelligence,” Lee said. “In a few years, we’ll have cars that can maybe calculate if swerving is a good idea in certain situations or whether or not to change lanes at a given moment.”

These changes are already underway, with common features like cruise control, backup sensors and GPS.

Beyond saving thousands of lives, widespread use of autonomous cars might have a tremendous positive impact on the environment. Self-driving technologies will make it much easier to rent vehicles, Lee said, so fewer people would actually own cars, and cities could employ a fleet of autonomous cars as very cheap taxis for residents.

In spite of these benefits, it will likely be some time before we trust self-driving cars and accept that humans are largely superfluous to transportation.

“The only reasons there are pilots is because people feel safer with pilots,” Lee said. Also, they make the loudspeaker announcements.

This is the fourth in a series of articles examining current and emerging artificial intelligence technologies and their impact on today’s world.