Otavio Good, leader of the San Francisco-based team that won the Defense Advanced Research Projects Agency’s Shredder Challenge earlier this month, doesn’t just do computer programming.
“I live it,” he told AOL Government in a telephone interview.
Good’s eight-person team, dubbing themselves “All Your Shreds Are Belong to U.S.,” correctly reconstructed more than 10,000 machine-shredded documents in five increasingly difficult stages to solve puzzles and capture DARPA’s $50,000 prize. (The team’s name is based on an Internet meme–“All your base are belong to us,” said Good.)
They outdistanced nearly 9,000 teams of problem solvers from around the world who had entered the competition. DARPA launched the competition in late October and announced the winner on Dec. 2.
Good said he heard about DARPA’s Shredder Challenge on a site called Hacker News. “I’d heard of other DARPA challenges before but had never taken part,” said Good (pictured above, far right. Also in the photo, left to right: Keith Walker, Winnie Tong, Luke Alonso, Zina Tebaykina, Sohana Ahmed.)
I figured the right way was right down the middle, where you would have a human placing puzzle pieces…but with the computer recommending pieces for them.”
Upon deciding to take up the Shredder Challenge, Good put together a team of crack computer programmers, who developed computer-vision algorithms that suggested the pairing of fragments, and “assemblers,” who then put the pieces together manually.
Good said he determined that this hybrid approach to assembling the pieces and solving the puzzles would work best.
“I was thinking there’s probably going to be some crazy people out there who would try do whole thing manually and there’s going to be some crazy people out there who would try to automate the whole the whole thing,” he said.
“I figured the right way was right down the middle, where you would have a human placing puzzle pieces as if they’re playing a jigsaw puzzle but with the computer recommending pieces for them. So that’s the way we went and that’s what worked.”
The team spent most of their waking hours outside of their day jobs putting pieces together, often collaborating on the project from different places on a Web site they had set up.
“We would go to work, finish at work, and then start working on this project until we went to sleep,” Good said. “We were in it to win it.”
“Lots of experts were skeptical that a solution could be produced at all, let alone within the short time frame,” said Dan Kaufman, director of DARPA’s Information Innovation Office. The winning team solved all five puzzles in about a month.
Altogether, the team spent about 600 hours on the solving the puzzles. “At the peak, we had all eight people placing pieces over the Internet,” said Good.
Good is the founder of Quest Visual Inc., a three-year-old San Francisco company that created Word Lens, an application that translates foreign text on an iPhone. A user simply points the iPhone at any printed text–English, Spanish or French–and the app instantly translates it on the screen without the need for an Internet connection.
Good doesn’t have a degree in computer science–or anything else, for that matter.
“I never went too far in college,” he said. “My mom taught me programming when I was probably in the second grade. I really enjoyed it and I’ve been programming ever since. Most of my career has been doing video games until about three years ago, when I started to work on Word Lens.”
For Good, who at 37 was the oldest member of “All Your Shreds Are Belong to U.S.,” the Shredder Challenge was perfect fit.
“It was right up my alley because it deals with computer-vision stuff and I know a lot of people who are really good programmers who do that kind of stuff,” he said. “I knew we could probably win it.”
DARPA’s challenges have spawned a broader wave of similar crowd-sourcing efforts across government, supported by a program known as Challenge.gov, a turnkey system that lets agencies administer their own innovation contests and challenges. Challenge.gov has helped 36 federal agencies launch 130 challenges over the past year.