The New York Times (NYT) has published "all the news that's fit to print" since 1851, but today, only articles published since 1981 are available online. This despite the fact that over the past decade, the Gray Lady had been using the most accurate digitizing system available: fast typists. But the newspaper's effort has shifted into high gear, thanks to a 28-year-old computer scientist and MacArthur "genius award" winner named Luis von Ahn whose software could put the entire archive online by late 2009. To be clear, the typists digitized 27 years of articles in 10 years, while von Ahn's software is processing 129 years in less than 24 months.
But this isn't a story of man versus machine. Von Ahn didn't simply write a program that processes information faster than a human. This is a story about man and machine, and about the researcher whose Web-based programs harness uniquely human abilities—such as reading or knowing common-sense facts—and then aggregate that knowledge to solve large-scale, long-standing problems of computer science.
Von Ahn calls his field "human computation," a jargony term that belies the impact that he is already making outside of academia. "Luis combines great ideas with insightful implementations to produce results that are truly remarkable," says Josh Benaloh, a cryptographer at Microsoft Research who hired von Ahn as a summer intern in 2004 when he was a graduate student. "There is simply no comparison between Luis and other young researchers."
How to Use Seconds
Von Ahn's breakthrough technology—the Captcha, developed with his thesis advisor Manuel Blum in 2000—is used by 60,000 sites including Yahoo (YHOO), Facebook, and Ticketmaster (TKTM) to verify that the entity filling out a Web registration form is a person rather than a bot. Has a Web site ever presented you with a distorted word and asked you to retype it? That's a Captcha.
"Within five years, about 200 million Captchas were being typed everyday," says von Ahn. "And I started to feel bad, because each one was wasting 10 seconds of someone's time." Von Ahn wondered how those seconds could be used to do something valuable, and, in 2007, he introduced ReCaptcha, which draws its random words from the Internet Archive's book digitization project and, until the job is completed late next year, from The New York Times archive. (The former, a nonprofit, doesn't pay for the service; the latter pays a few million dollars annually, according to von Ahn.) To ensure accuracy, every word is shown to multiple people. When their answers match, the ReCaptcha database returns the digitized word to its rightful sentence.
ReCaptcha is a win-win in that Web sites are able to block spambots and digitization efforts get a boost. That said, those Web users weren't retyping words because it was fun. Yet people, von Ahn noted, spent 9 billion hours playing solitaire online in 2003. If he could design programs as entertaining as a game, could that time and energy could be channeled into useful work?
Making Computers More Intelligent
In 2006, von Ahn launched the ESP Game. Two players are each shown the same image and asked to type in descriptive labels. When the labels match, the players get points and are shown another image. Surprisingly fun to play, the game also generates image labels as accurate as those of paid workers, according to von Ahn, who says that the game has been played by more than 200,000 people and collected more than 50 million image labels. Those labels, in turn, improve image search technology. (For more on the game, which Google licensed to improve its image search, see Rules of the Game (BusinessWeek.com, 1/28/08).
In May 2008, von Ahn launched Games with a Purpose, a research initiative/publisher of casual games, each designed to collect the human knowledge needed to solve a different computer science problem. Tag a Tune, for instance, generates song descriptions (sad, danceable, guitar-heavy) that will ultimately improve music search, while Verbosity collects common-sense facts that von Ahn hopes will make computers more intelligent. Both games have been played by about 60,000 people and collected more than 10 million pieces of data, which Von Ahn parses in different ways. The data collected by his game Squigl, for instance, are used to train a computer vision algorithm to recognize objects.
Von Ahn is particularly excited about a soon-to-launch game that will enable people to translate a sentence from one language to another, even if they only know one of the languages. It sounds improbable, but if the game develops accurate results it will be lucrative. "Almost any large company needs translation services," he adds. "And we can offer them for a fraction of the cost of human translators."
Still, he's not quick to license his games, and has turned several companies away until he can refine them further. "I don't want to be in the business of duping companies into paying me for a technology that isn't ready," says von Ahn, who, thanks to the $500,000 MacArthur Fellowship and a $200,000 Microsoft Research New Faculty Fellowship, doesn't need to rush any of his technologies to market. But the potential impact of his work is huge.
"Luis has opened an entire new field of research," says Microsoft's Benaloh. "We may have only just begun to exploit the true power and potential of human computation."