The applications for these sleuthing technologies range from deciphering buying trends in retail outlets to identifying dangerous chemicals. Want to know how many razor blades are selling in Prague or Pittsburgh? Slap a radio tag a computer chip that allows a product to be tracked on its journey from manufacturer to consumer on every pack of Gillette blades, and you will get your answer in a hurry. That's the specialty of Alien Technology, an eight-year-old company based in Morgan Hill, Calif. The same tags can help track weapons too, and the U.S. Department of Defense just commanded its 43,000 sup-pliers to start using tags like those made by Alien. Worried about gas leaks from the furnace? Give Nanomix a call. The firm, based in Emeryville, Calif., is working on a miniature sensor, sometimes called an electronic nose, that detects hazardous chemicals in the environment. If these three firms were real detectives, Alien and Nanomix would be out on the streets doing surveillance, while Imagen would remain in the lab sorting through visual images for clues. Each of these companies plays its own role in the binary gumshoe game.
The thing all three have in common is digitization. Just about anything can be represented in digital form these days not just text, movies and music but poison gas, human faces, packs of razor blades you name it. Once information is digitized, it can be tracked, monitored and analyzed. The potential is evident in Lipson's vision of how people will use Imagen's algorithm-laden software. "Any data that can be represented as an image is amenable to analysis by Imagen's technology," says Lipson. "It is one of the very few technologies that can transcend different domains, from natural scenes to human faces to trademarks."
So far, Imagen has not been chasing terrorists or advising Hollywood starlets on their makeovers but concentrating on the far less glamorous but more commercially viable task of spotting defects on electronics boards before they end up in faulty cell phones or computers. Imagen's software is programmed to recognize patterns in much the same way the human brain learns to distinguish classes of objects (say, faces) as well as specific objects in that class like your best friend's face. That's done by teaching the software, through trial and error, the common patterns that all faces share plus the specific patterns that make your best friend's face unique and then by training it to recognize those patterns in different conditions like varying light.
What works for faces works for circuit boards too. Imagen licenses its software to Teradyne, a maker of automated test equipment based in North Reading, Mass. (The companies are close; indeed, Imagen has an office based at Teradyne's headquarters.) Teradyne delivers Imagen technology to manufacturing companies in Asia and the U.S. by integrating it into Teradyne test equipment that scans circuit boards for flaws. Simply stated, a Teradyne machine takes pictures of the circuit board and hands over the image to Imagen software, which scours the snapshot for imperfections. This business allows Imagen to make a profit, though for Lipson, there have been other rewards. She met her future husband and Imagen co-founder Pawan Singha during their graduate days at M.I.T. Today he serves on Imagen's scientific advisory board while working as an M.I.T. professor of neuroscience.
Although Lipson won't reveal financial details, she says Imagen is making a profit an impressive result for a company started with just $30,000 in prize money that Lipson won in 1997 as a doctoral student at M.I.T.'S Artificial Intelligence Laboratory. Imagen has not raised any venture capital either; indeed, Lipson claims to have spurned offers because, she says, working as a lone wolf, Imagen is free to develop its technology at its own pace and in several markets simultaneously. The company's only funding other than the M.I.T. prize and fees from Teradyne has been $100,000 from Alex d'Arbeloff, who until earlier this year was chairman of M.I.T. Corp., which oversees M.I.T.