How Artificial Intelligence Can Change Higher Education

Sebastian Thrun, winner of the Smithsonian American Ingenuity Award for education takes is redefining the modern classroom

Sebastian Thrun is turning his expertise in artificial intelligence to humans. (Ethan Hill / Composite Image: NASA; Google; Udacity)
Smithsonian Magazine | Subscribe

(Continued from page 1)

With Udacity, he says, he also wants to make education accessible to people with jobs, kids, mortgages. On the whiteboard table, he starts writing. “If you look at how life is arranged,” he says, “right now it’s play, then K-12 learn, all the way to higher ed, then it’s work, then it’s rest. These are our phases, they’re sequential. I want it to look like this,” he says, jotting a flurry of words so that “learn” is under “work” and “rest.” Why do we give up learning after college? And why, he asks, do universities give up teaching their students when they leave? “My HMO gives me a lifetime deal if I want, so why not my university?”

MOOCs offer the potential to make higher education more available, more affordable and more responsive to employers’ needs than traditional university degrees. But will they help inaugurate an “Athens-like renaissance” in education, as former Secretary of Education William Bennett has suggested? Coursera’s Ng says that online education may influence, rather than replace, traditional universities. “Content is increasingly free on the web, whether we like it or not,” he says. What MOOCs augur, he says, is the so-called “flipped classroom,” in which students watch classes online the week before and come to class “not to be lectured at,” but to actively engage.

Thrun believes online education is at the same sort of transitional moment self-driving cars were a decade ago—a moment that plays to his own problem-identifying strengths. Chris Urmson, the engineering head of Google’s self-driving car program, describes Thrun as someone “who has the insight to see when something needs to happen” but “isn’t purely visionary—he has the drive and execution to go and actually do it. Seeing the two mix in one person is rare.” (Thrun’s dual nature might be seen in the cars he drives: a Chevy Volt, the quintessence of quiet, left-brained efficiency, and a Porsche, that splashy emblem of ego, adventure and risk.) And Udacity speaks to another Thrun obsession: “For me, scale has always been a fascination—how to make something small large. I think that’s often where problems lie in society—take a good idea and make it scale to many people.”


Long before he was trying to tackle large, complex problems, Thrun tackled small, complex problems as a teenager in a small town near Hanover, Germany. On a Northstar Horizon computer, a gift from his parents, he tried writing a program to solve Rubik’s cube. Another program, to play the board game peg solitaire, involved what’s known in math as an “NP-hard problem”—at each step the time-to-solve grows exponentially. “I started the program, waited a week, it didn’t make any progress,” he says. “I realized, wow, there’s something profound, deep, that I don’t understand—that a program could run for millennia. As a high-school student, it’s not in your conception.”

At the University of Bonn, Thrun studied machine learning, but dabbled in psychology—“My passion at the time was people, understanding human intelligence.” In 1991, he spent a year at Carnegie Mellon under the tutelage of the AI pioneers Herbert Simon and Allen Newell, building small robots and testing his theories about machine learning. But even then, he was thinking beyond the lab. “I always wanted to make robots really smart, so smart that I wouldn’t just impress my immediate scientific peers, but where they could really help people in society,” he says.

He actually became an adjunct professor of nursing while developing robotic nurses at a Pittsburgh elderly-care home. Another early effort, a robot named Minerva, was a “tour guide” that welcomed visitors to the Smithsonian National Museum of American History. It was, says Thrun, a learning experience. “What happens if you actually put a robot among people? We found problems we never actually anticipated.” Visitors, for example, tried to test the robot’s ability. “At some point, people lined up like a wall, and hoped the robot would drive into an area where it didn’t know how to operate, like a nearby cafeteria,” he says. “And the robot did.”

In 2001, Thrun went to Stanford, where the Silicon Valley spirit hit him like a revelation. “In Germany there’s just many questions you’re not allowed to ask,” he says, “and for me, the core of innovation is for very smart people to ask questions.” In the United States, and particularly Silicon Valley, he found an “unbelievable desire” to ask questions, “where you don’t just go and proclaim something because it’s always been this way.” He wishes, he says, “that Silicon Valley wasn’t 2,500 miles away from Washington, D.C.,” that societal innovation could keep up with technical innovation. “We can’t regulate our way out of problems,” he argues, “we need to innovate our way out.”

It was in that spirit that he plunged into work on an early version of the car that would eventually make its way to Google. In 2007, he took a year’s leave from Stanford to help develop Streetview, Google’s 360-degree mapping feature. “It became an amazing operation, the biggest photographic database ever built at the time.” Then he assembled an AI dream team to make the self-driving car a reality (a version named Stanley, which won the 2005 DARPA Grand Challenge for driverless vehicles, is held by the American History Museum) and founded Google X as a skunkworks for developing products like the augmented-reality “Google glasses.”

Udacity may seem rather a departure for Thrun, but Urmson, his Google colleague, says that while it’s different on a “purely technical axis,” it shares with his other work the “opportunity to have this transformative impact.” There are other parallels. Thrun seems intent on hacking education the same way he hacked driving, drilling it down to its component parts, testing and retesting. “We do a lot of A/B testing,” he says, describing the technique, popular in Silicon Valley, for comparing two different versions of a web page to see which is more effective. “We have lots of data. We use it strictly for improving the product.” (He jokes that he even runs scientific tests on his 4-year-old son: “I gave him infinite access to candy the first day; the second, suddenly he didn’t like it anymore.”)


Comment on this Story

comments powered by Disqus