Jordan Pollack applies evolutionary theory to elementary education

Professor of computer science creates online games for children

First he taught computers to learn from people. Then he taught computers to learn from one another. Next, he taught computers to teach children. Finally—after learning some new ABCs himself—he taught computers to teach children how to teach one another. Now there are thousands of children in grades K to 8 learning to spell better and locate state capitals by playing his online computer games.

The free games, called BEEweb, run off servers in the basement of the Volen National Center for Complex Systems. Their creator, Jordan B. Pollack, professor of computer science and of the Volen Center, believes it’s possible to build machines that learn and evolve into entities of ever-increasing complexity and intelligence. It’s a quest, really, to find a mathematical and computable definition of evolution. The philosophical principle that drives his science—the conviction that there is a form of intelligence in nature that impels biological self-assembly that he can model in computer programming—led him to develop the BEEweb games in which children develop their own skills in spelling and geography, mathematics and spatial reasoning.

The games also encourage children to help other kids improve their skills, because in SpellBEE, GeograBEE, PatternBEE, and MoneyBEE, winning is based on cooperation rather than strict competition. Children learn something useful while interacting with one another safely online. A child logs on anonymously, chooses which game she wants to play, pings another player on the site, and the two trade spelling words or geography questions or set up money-counting questions. All play occurs within the strict confines of the game, so there’s no possibility of extracurricular conversation.

Cooperation trumps competition
“The first games we wrote for children were very simple; we tried to get them to type as fast as they could or spell better than other kids,” says Pollack. “Kids really liked it, but the teachers didn’t. As we did more research, we found out that the educational community hates competition in the classroom because it can lock in negative habits and lower children’s self-esteem, which decreases their motivation.

“For me—for my science—that was an interesting moment. Because the idea behind evolution is that there is a limited resource that species compete for,” he says. “So the idea that a system based on competition would lead to a negative outcome really surprised me. Society is organized by the principles abstracted from nature long ago. But in a classroom, it’s clearly a failure. To me this was a wake-up call.”

Pollack saw that he was missing a key concept in his application of evolutionary theory to artificial intelligence: competition can’t be the only objective. If strict competition in the classroom does not serve as a motivation for the less “fit” child to try harder, then we must not fully understand all the principles of evolutionary theory as applied to highly intelligent beings.

Teaching backgammon to robots
Along the path that led Pollack to develop the BEEweb games for children, he created some very smart machines that learn as they go and handily outwit human beings in the narrowly defined playing fields of backgammon and Tron (a game based on the 1982 sci-fi movie). Pollack’s goal was to create computers that possessed the ability to continue learning, but he found that after an initial burst of inventive behavior, these systems began slowing down and eventually reached equilibriums where all learning stopped. Once a computer got the best of its opponents, it rested on its laurels; it had no reason to get smarter and no reason to share its knowledge with other players.

In the late 1990s, Pollack put a backgammon-playing computer program that evolved through self-play on the Web, where it drew tens of thousands of human opponents. A student working with Pollack analyzed the scores and IP addresses of those human opponents and determined that their skills were improving over time. Next, Pollack did the same thing with Tron, a simple video game Web site where people played against a group of evolving computer programs. This time he saw some interesting developments in the computer’s “intelligence.”

“The computer programs actually improved and adapted to humanity, coming up with novel strategies that were hard for humans to cope with,” he said.

Pollack now had proof that computers learn through human interaction, and vice versa, and he wondered if he could develop an activity that would benefit people, rather than just entertaining them. That’s when he turned his attention toward elementary education, experimented with children’s educational games, and had his eureka moment about the role of competition in evolution. This also led him to a clearer understanding of what makes his game-playing computers tick.

What’s necessary, Pollack realized, is a system where all players don’t simply compete to be best. But players can’t be altruistic either. The answer: a model that rewards multiple objectives, not just competition. Each player takes a turn being teacher. Students get rewarded for answering correctly, and teachers get rewarded for asking questions the student can answer but not by making the questions so easy that the student fails to learn, or so hard that the student is repeatedly stumped.

“In the BEE games, the child playing ‘teacher’ will win by asking questions in the area where the student is almost ready to learn, not in the easy or hard zone, but in the zone of proximal development,” explains Pollack. “This motivates the teacher to do continual assessment, provides the student with appropriately diffi cult problems to learn from, and keeps the students engaged in educational activities.”

So, when an anonymous child in Michigan logs in to BEEweb.org and plays GeograBEE with an anonymous child in Texas, she learns to use the computer, picks up some new state capitals, becomes a bit of a teacher herself, and, most important to her, has some fun playing a game. Now that’s effective child’s play.

All you can BEE
Pollack says he dreams of getting BEEweb licensed into a large, “world-changing” program like Skype, the computer application that lets people worldwide make long-distance telephone calls over the Internet for free. According to his “back-of-the-envelope calculation,” in a single year, a couple of $2,000 servers in the basement of the Volen Center could provide the equivalent teaching hours of an entire school district.

“If one million kids each play an hour of SpellBEE or PatternBEE, we’ve essentially delivered one million hours of human teaching—the equivalent of a medium-sized school district, which might have a $1 million budget,” Pollack says.

Already BEEweb.org has 60,000 anonymous registered users, and each week the games get hundreds of additional visitors who drop by for a game or two. A Google map on the site shows that more than thirty states have registered users, with Michigan, Minnesota, and Texas outnumbering the rest. Some teachers register entire classes and use the games as regular teaching tools.

SpellBEE, GeograBEE, and PatternBee won Bessies (Best Educational Software Awards) for 2005–06 from the ComputED Center in Carlsbad, California.

—Denise Brehm