Clearer computer vision.
Vision is one of biology’s most complex processes. But that doesn’t stop Nikos Paragios from trying to bring this marvel of flesh and blood to the world of bits and bytes. He develops software that allows computers to interpret images more accurately, which could improve everything from medical diagnosis to driving.
As a professor at the École Centrale Paris, Paragios is a long way from the world of his childhood on the tiny Aegean island of Kárpathos, where he worked summers in a family-owned coffee shop, and there wasn’t a computer in sight. “But everyone said computer science is the future,” he recalls, so he headed to the University of Crete to study it.
Today Paragios is a leader in computer vision. Among his many projects is the mathematical modeling of hand gestures. The idea is to develop software to translate sign language into text, easing communication between the hearing and the deaf. The models could also allow drivers to simply point at icons printed on a dashboard–gestures that would be interpreted by onboard cameras and computers–rather than twisting knobs or pressing buttons.
Paragios is best known for his contributions to medical imaging. As a research scientist at Siemens in Princeton, NJ, he created software to automatically detect and define the boundaries of anatomical structures. Applied to magnetic resonance images of the heart, for example, the software highlights complex structures such as the coronary arteries, allowing doctors to pinpoint changes that can lead to heart attacks. Siemens is now integrating Paragios’s system into its MRI scanners.
Paragios continues to work on medical imaging technologies. With physicians at Henri Mondor University Hospital outside Paris, he’s developing software to diagnose muscle diseases called myopathies without painful biopsies. Paragios and his team are turning to a variation of MRI called diffusion tensor imaging, which measures the random motion of water in biological tissues (see “10 Emerging Technologies: Diffusion Tensor Imaging”). The goal is to create algorithms that can use this motion to determine the structure and orientation of muscle fibers, thus revealing developing myopathies.
But no matter its application, Paragios’s research is driven by his desire to “do something that brings great innovation and serves society.”
When your Internet communications absolutely, positively need to be anonymous.
A dissident in China uses Web-based e-mail to contact a journalist in Canada. An intelligence agency wants to surveil a foreign website. Like every operation on the Internet, these activities leave tracks. Online anonymity measures provide a way around this problem; one of the most advanced is Tor, or the Onion Router.
Computer scientist Roger Dingledine developed Tor under a contract with the U.S. Naval Research Lab; today, the software is distributed by the Tor Project, under the fiscal sponsorship of the Electronic Frontier Foundation.
To disguise Internet traffic’s origins, Tor plots a route through any three of more than 700 volunteer-run Onion routers around the world. It sets up a two-way link between the sender’s computer and the final router in the chain; data passed between them is encrypted in three layers, and each router in the chain peels off one layer along the way. Each data packet “remembers” only the address of the last router it visited. That way, even if the data is intercepted before the final router hands it off to the recipient, it’s difficult to trace back to the sender.
Editors’ note: This text is corrected version of the story that ran in the September/October 2006 print issue of Technology Review.
Keeping online collaboration simple.
37signals builds aggressively simple Web-based collaboration tools that help people manage everything from family to-do lists to big corporate projects. Hundreds of companies are using Ruby on Rails, the open-source software toolkit that the firm created, to quickly develop their own online applications. Jason Fried, the founder and president of the seven-person firm, is fond of saying, “It’s better to tell a short story well than a long one poorly.” In that spirit:
“Jason is immune to dogma and has much to teach. In 37signals, he has built an elegant company with elegant products based on the idea that less is more.” –Jeff Bezos, CEO, Amazon.com, and the first outside investor in 37signals
Beaming textbooks across Africa.
Growing up in Africa, Matthew Herren saw that many children in cash-strapped rural schools have to make do with textbooks that are decades out of date. Inspired to find a low-cost solution to the problem, he hit on the idea of using satellites to transmit up-to-date educational materials. He aims to establish the technology not through programs run by traditional aid organizations but through a series of self-sustaining businesses.
“We figured there had to be a way for technology to lower the cost” of providing books and other materials to children, Herren says (see “Development Powered by Education”). Herren, who is Swiss, turned to one-way satellite radio transmission because Internet access is unlikely to reach much of the African interior anytime soon. In a test last year, a Swiss foundation called BioVision installed a satellite receiver in a grade school in Mbita Point, Kenya, on the shores of Lake Victoria. The receiver downloaded textbooks onto its hard drive. The information was then transferred to handheld computers rigged with simple Linux-based software for book viewing. Sixty students received current classroom materials.
Herren is now trying to implement his scheme on a grander scale. Working with Bridgeworks, a venture capital firm in Zürich, he has already raised most of the $650,000 needed for seed money. With that capital, Herren hopes to launch a network of businesses across Africa that will sell and service the satellite receivers and handheld PCs. A country’s education ministry would hire one of these companies to provide and maintain its educational-download system–and slash the per-pupil cost of providing classroom materials by more than 20 percent.
If delivering educational materials by satellite works as hoped, the basic system could be used to provide remote villages with health and agricultural information. Herren notes that in many villages that lack even roads, all paths literally lead to schools.
A better operating system.
Operating systems are the software foundation for all computers, from laptops to servers–but most were designed decades ago, without Internet security in mind. Small flaws in the operating system or software on, say, a bank’s server could compromise millions of dollars’ worth of sensitive data.
To make such information more secure, computer scientist Eddie Kohler and his team designed Asbestos, an operating system that keeps private data from falling into the wrong hands even when other software on a computer has failed. Asbestos keeps personal data secure by “tagging” it with information about which programs or users can access it.
Usually, this sort of tagging requires a large amount of memory, but Kohler has structured the tag data to use minimal system resources. Initial tests have been promising, and Kohler hopes that within a few years, Asbestos will be an alternative to server operating systems such as Linux and Windows.
Super-high-resolution 3-D displays could change the way people look at everything from tumors to drug targets and natural gas deposits.
Before treating a tumor with radiation, an oncologist must decide how to direct the radiation beams so as to minimize damage to surrounding tissues without compromising efficacy. That kind of planning may soon become more accurate, thanks to true 3-D displays whose software Joshua Napoli helped pioneer.
The image on the left shows different perspectives on the same plan for treating a brain tumor with radiation: red marks the tumor site, blue marks critical structures such as the brain stem (which must receive as little radiation as possible), and green marks the proposed path of radiation beams. The images, 25 centimeters in diameter, are projected inside a somewhat larger dome; this “volumetric display” is made by Actuality Systems of Bedford, MA, where Napoli is head of software development.
Napoli’s software breaks up 3-D models generated by a computer into hundreds of frames that are projected onto spinning panels inside the dome, making a smooth, interactive image. The software has produced the highest-resolution volumetric 3-D displays in the world, Napoli says. Along with doctors at three hospitals, Actuality is studying whether radiation plans made using its displays are superior to those made using traditional monitors. Eventually, doctors could use the 3-D displays to view most types of medical scans.
Already, petroleum companies are using them to help visualize oil and gas deposits, and pharmaceutical researchers are using them to help picture how potential drugs interact with their targets. One of the biggest challenges, Napoli says, has been engineering the underlying graphics-processing software to work with real 3-D displays; displaying 3-D images this way requires a computer to process about 50 times as many picture elements as it would if the images were displayed on flat screens.
The man who opened up the map.
Rademacher’s new picture of the world–or at least of selected cities–took the Web by storm. Even Google employees wrote on a company Web page that his site, housingmaps.com, “blew our minds right off our shoulders.” Thousands of people realized that Google’s maps were a giant canvas on which they could doodle, taking the locations of crime scenes, favorite restaurants, or cheap gas stations and creating online tableaux for all to see. But more than that, Rademacher had shown a way to combine data and tools from completely different websites to create something new. One blogger called it a “mashup,” a word DJs use to describe the mixture of vocal and instrumental tracks from different songs, and the term stuck.
For Rademacher, there’s a moral to the story. Innovation is possible only when companies let you tinker with their creations. Too many good ideas are squandered, he says, because the tools needed to realize them are locked away: “To this day, there are very few technologies that are open.” Creating open technologies is Rademacher’s new passion. In September 2005, he left his job developing animation tools at PDI/Dreamworks Animation to pursue that passion at Google. His projects there, he says, are still “under wraps.”
How tags exploit the self-interest of individuals to organize the Web for everyone.
In 2001, a wonky Wall Street quantitative analyst named Joshua Schachter had a problem. In the late 1990s, he’d started a website called Memepool, which was a simple collection of Web links that he had found interesting, useful, or both. Over time, as Memepool’s users began sending in links they thought the site should feature, Schachter’s personal list of bookmarked Web pages grew to more than 20,000 entries, far more than any folder system could handle. To bring some order to the chaos, Schachter wrote an application called Muxway, which allowed him to manage his links by giving each a short label, or tag–enabling him to call up all the pages that were tagged, say, “Wi-Fi” or “math.”
People continued to view Schachter’s list of interesting links; but now, because of Muxway, those links were organized around tags. Pretty soon, about ten thousand people every day were stopping by. Schachter realized that even with (or perhaps because of) the deluge of information available on the Web, people were still hungry for good links, and they were interested in finding out what others thought was interesting. He also figured that if tagging was helpful to him, it could make storing and finding bookmarks easier for everyone else. So with that in mind, he rewrote Muxway, and in 2003 he launched it as a website called del.icio.us. Within a couple of years, hundreds of thousands of people were using del.icio.us, and it had metamorphosed into a system for organizing not just individuals information but the whole Web. Today it exemplifies the promise of what’s often called Web 2.0–websites and online applications that rely on user participation to achieve their greatest value.
At its core, del.icio.us is a bookmarking system: a place to store all those links that don’t fit in a “Favorites” folder. But it took off because it offers everyone what Muxway had offered Schachter: a way not just to collect links in one place but also to organize them. As people trawled the Web, they could tag interesting pages using whatever words they wanted, and del.icio.us would keep track of them all.
“You bookmark for one of two reasons: either you think you’re going to need that page again somewhere down the road, or you don’t have time to read it now, but you want to read it later,” Schachter says. “The challenge is, once you’ve got all these bookmarks, how do you manage them? The problem we’re really dealing with is memory and recall, and using technology to make your memory more scalable.”
Schachter deliberately avoided imposing any rules about how people could use tags. He knew it wouldn’t work: “If I went in there and said, Hey, you’re using that tag wrong, people would just tell me to fuck off,” he says. He also knew that letting people use their own tags–instead of choosing them from a menu he provided–would make del.icio.us more likely to be genuinely useful. Each person who uses del.icio.us is effectively coming up with an idiosyncratic system for classifying the Web: an article about, say, Dallas Mavericks owner Mark Cuban might be tagged “Mavericks” by one person, “crazy” by another, and “Mavericks” and “crazy” by a third. (Del.icio.us allows users to pin as many tags on a page as they want.) “If you’re trying to tag a page in a way that’ll get you back there someday, you want to use your vocabulary, not someone else’s,” he says.
Though del.icio.us has become a way for users to collectively organize information across the Web, it did not begin as anything so grand. Rather, it emerged as a way to help individuals manage their own information. “For a system to be successful, the users of the system have to perceive that it’s directly valuable to them,” Schachter says. “If you need scale in order to create value, it’s hard to get scale, because there’s little incentive for the first people to use the product. Ideally, the system should be useful for user number one.” This makes del.icio.us different from systems that rely on what economists call “network externalities”–meaning they’re valuable only if lots of people use them. It was hard to get the first person to buy a fax machine, because a fax machine is useless if you’re the only one who has one. But even for the first person to use del.icio.us–Schachter–it worked.
As it happens, lots of people found del.icio.us valuable right from the start, making it a proverbial grassroots hit. Schachter did no advertising, no marketing. But the site was so successful that in 2005 he quit his day job at Morgan Stanley, raised some money from outside investors, and launched del.icio.us as a regular business. Less than a year later, Schachter sold del.icio.us to Yahoo, where he now works in the Groups business, running the site full time.
Schachter’s original focus on the individual user has never wavered, and it remains essential to the way del.icio.us works. But as more and more people started to use the site, something interesting happened: when aggregated, all those individual tags created a useful system for categorizing Web pages. On the surface, del.icio.us doesn’t seem designed to do this, since each person makes his or her own tags, and there’s no overarching authority to maintain order. But even with no one in charge, the product of all the individual decisions of del.icio.us’s users is surprisingly well organized–and surprisingly intelligent. That is, if you do a search on del.icio.us for all the pages that are tagged with a particular word, you’re likely to come up with a remarkably good–and well-rounded–selection of related Web sources. In other words, although del.icio.us didn’t need lots of users to be useful, once it had lots of users, it became valuable in an entirely new way. Almost accidentally, it became an excellent tool for making sense of the Web.
What del.icio.us’s users were creating–without necessarily knowing they were doing so–was what technology blogger Thomas Vander Wal has dubbed a “folksonomy,” a flexible system of organization that emerges organically from the choices users make. We’re all familiar with the alternative, the kind of rule-bound, top-down classification scheme that Internet theorist Clay Shirky calls “ontological” in nature. The Dewey decimal system is an example: every object is assigned its place in a hierarchical system of organization, and every object is defined as, ultimately, one thing: a book goes in one place in the library and nowhere else. In a folksonomy, by contrast, definitions are fuzzier. With del.icio.us, the same Web page has many different tags, which often aren’t even related to one another, and no explicit rules are being followed. Web pages are therefore listed not in one place but in many places, and sometimes pages aren’t quite where you might expect them to be. So folksonomies are messier than “ontologies” are.
What del.icio.us has shown, though, is that folksonomies’ imperfections are outweighed by their benefits. In the first place, folksonomies are dynamic rather than static. A Web folksonomy thus allows us to reclassify content according to our changing interests. An academic paper that’s interesting today might be equally interesting a decade from now–but why it’s interesting, why people care about it, might be very different. A traditional categorization system has a hard time dealing with this: once the essence of an object is defined, it’s supposed to be defined for good. In a folksonomy, the reclassification happens almost automatically–as people start tagging the paper with new, more relevant tags, for example. Web folksonomies are also better at capturing the multiple meanings and uses that a given site has, rather than constraining the possible range of meanings. It’s useful, after all, to learn that many people have tagged stories about Mark Cuban “crazy,” in addition to indicating everything else that’s important about him. Finally, folksonomies are cheap. Imagine the labor and the time it would take to construct a traditional organizing system for all the pages on the Web, and then to maintain and update it. Then recognize that del.icio.us is producing a ceaselessly revised organizing system–at almost no cost.
The real magic of folksonomies–and the reason sites like del.icio.us can create so much value with so little hired labor–is that they require no effort from users beyond their local work of tagging pages for themselves. It just happens that the by-product of that work is a very useful system for organizing information. This distinguishes del.icio.us from other high-profile Web 2.0 sites like Wikipedia and Digg, which people contribute to without reaping any obvious personal benefit.
Schachter thinks the fact that del.icio.us does not rely on the selflessness of its users makes it more robust than it might otherwise be. “Im not a big believer in expecting a large number of people to act in an altruistic fashion,” he says. “You want to rely on people to do what they do.” The echoes of Adam Smith are unmistakable: del.icio.us is a system that, like a healthy market, turns individual self-interest into collective good.
Del.icio.us now has more than 300,000 registered users, and it generates as much traffic in a single day as it did in its entire first year. But even as tagging has become an industry buzzword that businesses are straining to associate themselves with, Schachter is confronting the fact that the vast majority of people on the Web don’t tag at all–and probably have never even heard of tagging. So how does he expand his sites audience? “You have to solve a problem that people actually have,” Schachter says. “But it’s not always a problem that they know they have, so that’s tricky.” He remains more focused on the site’s value to the individual than on its folksonomic aspects, because to him, helping individuals store and recall information is far more important than classifying the Web. And it may well be individual value that’s most likely to keep del.icio.us growing.
Regardless of what happens, Schachter has already shown that out of the seeming chaos of hundreds of thousands of independent and eccentric judgments, order and wisdom can emerge. And if you think about del.icio.us in terms of his idea of making memory scalable, he’s also helped create a rather remarkable social memory system, in which all of us are able to find more and better information than we would on our own. As Schachter puts it, “The one who stashes a page doesn’t have to be the one who ends up recalling it. Del.icio.us is a storer of one’s own attention. But it also means you can share it with others.” And that ability will only become more valuable over time. “The better you understand the world, the better you’ll do,” Schachter says. “I really think that in the end, more understanding wins.”
Perfecting peer-to-peer networks.
Ben Zhao wants to improve the Internet by letting new networks piggyback on top of it.
Zhao creates structured overlay networks. These peer-to-peer networks are like the ones file-sharing services use, routing data directly between individual users over the Internet. But instead of allowing any computer in the network to talk to any other, structured overlays have strict rules about which machines may talk to which. This allows them to route data more efficiently and to detect and sidestep failures more quickly. “Everything is about resiliency and recovery and robustness,” Zhao says.
As a graduate student at the University of California, Berkeley, Zhao wrote Tapestry, a series of networking protocols that was one of the first structured overlay networks. Some of his Berkeley colleagues built an application called OceanStore that uses Tapestry to provide cheap, reliable, global-scale data storage. Since then, Zhao and others have developed a host of programs that take advantage of Tapestry.
Now an assistant professor at UC Santa Barbara, Zhao is working to make structured overlay networks more secure. With companies such as Microsoft interested in the technology, his success could mean practical networks that manage and heal themselves.