“There's a set of fundamental underlying techniques—Markov modelling and things like that—that tie across these domains,” he says. “Most of my expertise in this area has to do with algorithmic techniques for analyzing data across space and time. Much as a video is a sequence of still shot images over time, we can also look at the evolution of texts and network structures.”

Some of Huttenlocher's research has focused on the problem of object recognition, which will result in new abilities that allow people to systematically search for images.

“Right now you can search text easily, but it's very difficult to search images unless you happen to have put some textual labels on them,” says Huttenlocher. “Those textual labels are whatever you happened to think to write down.”

The overall goal is to program computers to categorize images without relying on human textual cues. Visual recognition systems also have the potential to revolutionize other aspects of human-computer interaction, such as allowing users to employ physical gestures to control computers, instead of keyboards and mice. The fact that the human brain can categorize objects and recognize gestures proves that the tasks are feasible in principle, notes Huttenlocher; the challenge has been to implement such visual abilities on a machine in an efficient manner.

“You can imagine how this stuff, in another couple of decades, can really change the way we interact with machines.”

In the meantime, Huttenlocher is assisting social scientists who want to make sense of information networks. Along with other faculty such as Jon Kleinberg and Geri Gay, he is a participant in Cornell's Institute for Social Sciences (ISS) theme project called “Getting Connected”.

This new project examines how information networks can give us cues about human social relations, which are complex and shifting and difficult to study.

“Many online interactions actually produce permanent and semi-permanent information artifacts,” says Huttenlocher. “And those information artifacts themselves have a network structure. We can study [those artifacts] rather directly.”

To that end, he's been helping to refine techniques that will help social scientists to measure such phenomena as the diffusion of opinion in the blogosphere and the temperature of sentiment in online discussion groups, without needing to employ armies of human readers to pore over individual Web sites.

“What we're trying to do is to use our ability as computing and information scientists to find out what we can learn about social networks,” says Huttenlocher. “That's a big project.”