TR10: Reality Mining

Sandy Pentland is using data gathered by cell phones to learn about human behavior.

By Kate Greene

Every time you use your cell phone, you leave behind a few bits of information. The phone pings the nearest cell-phone towers, revealing its location. Your service provider records the duration of your call and the number dialed.

Credit: Julien Pacaud

Multimedia •  Alex Pentland on the future of cell phones.

Who: Sandy Pentland, MIT Definition: Personal reality mining infers human relationships and behavior by applying data-mining algorithms to information collected by cell-phone sensors that can measure location, physical activity, and more. Impact: Models generated by analyzing data from both individuals and groups could enable automated security settings, smart personal assistants, and monitoring of personal and community health. Context: Cell phones are now sophisticated enough to collect and analyze data on personal behavior, and researchers are developing techniques that allow them to effectively sort through such information.

Some people are nervous about trailing digital bread crumbs behind them. Sandy ­Pentland, however, revels in it. In fact, the MIT professor of media arts and sciences would like to see phones collect even more information about their users, recording everything from their physical activity to their conversational cadences. With the aid of some algorithms, he posits, that information could help us identify things to do or new people to meet. It could also make devices easier to use--for instance, by automatically determining security settings. More significant, cell-phone data could shed light on workplace dynamics and on the well-being of communities. It could even help project the course of disease outbreaks and provide clues about individuals' health. Pentland, who has been sifting data gleaned from mobile devices for a decade, calls the practice "reality mining."

Reality mining, he says, "is all about paying attention to patterns in life and using that information to help [with] things like setting privacy patterns, sharing things with people, notifying people--basically, to help you live your life."

Researchers have been mining data from the physical world for years, says Alex Kass, a researcher who leads reality-mining projects at Accenture, a consulting and technology services firm. Sensors in manufacturing plants tell operators when equipment is faulty, and cameras on highways monitor traffic flow. But now, he says, "­reality mining is getting personal."

Within the next few years, Pentland predicts, reality mining will become more common, thanks in part to the proliferation and increasing sophistication of cell phones. Many handheld devices now have the processing power of low-end desktop computers, and they can also collect more varied data, thanks to devices such as GPS chips that track location. And researchers such as Pentland are getting better at making sense of all that information.

Story continues below

To create an accurate model of a person's social network, for example, Pentland's team combines a phone's call logs with information about its proximity to other people's devices, which is continuously collected by Bluetooth sensors. With the help of factor analysis, a statistical technique commonly used in the social sciences to explain correlations among multiple variables, the team identifies patterns in the data and translates them into maps of social relationships. Such maps could be used, for instance, to accurately categorize the people in your address book as friends, family members, acquaintances, or coworkers. In turn, this information could be used to automatically establish privacy settings--for instance, allowing only your family to view your schedule. With location data added in, the phone could predict when you would be near someone in your network. In a paper published last May, ­Pentland and his group showed that cell-phone data enabled them to accurately model the social networks of about 100 MIT students and professors. They could also precisely predict where subjects would meet with members of their networks on any given day of the week.