by Celeste Biever
November 25, 2004
For instance, the phone might learn that whenever you phone at least three friends after 5 pm, you will leave the office within half an hour and meet them in a particular bar. If you have a big presentation at work the following day, the phone could be programmed to warn you not to drink too much.
Or it could remind you to bring a particular CD that a friend wants. The phone, developed by Nathan Eagle and Sandy Pentland at the Massachusetts Institute of Technology, learns about its users' lifestyle by logging when they make voice and text calls or use phone applications like an alarm clock or phone camera. And using built-in Bluetooth short-range radio links it determines who you associate with by the proximity of their Bluetooth phones. The location capabilities afforded by cellular phone masts allow the phone to work out where you socialise, work and live.
The system is based on mobile message logging software called Context, designed in Finland by Mika Raento and a team of engineers at the University of Helsinki and the Helsinki Institute for Information Technology. Installed on the phone, Context logs the ID code of every Bluetooth chip that it passes, the location of every new phone mast it contacts, the number of every person phoned or texted, and every time an application is used. Each piece of data is time stamped and sent for storage on your network's servers.
The software learns by prompting users to enter where they are and what they are doing every time the phone moves into the range of a new cell mast- so it can associate activities like socialising, or working with certain locations. The system is being tested on 100 of Nokia's 6600 smartphones that have been programmed with Context and handed out to MIT students.
Data from them is continually being downloaded to a central server at MIT, where a pattern-recognition package works out the probability of what an individual is likely to be doing, which of their friends or colleagues they are likely to run into, and where they tend to be at any point in the day.
The software can also be used for "reality mining" Eagle says, reminding you of how long you spent working or partying in any one week, as well as answering questions like when you last saw a friend, what you did and who else you met. "No one has ever been able to capture this kind of data before, because cellphones and Bluetooth were never as ubiquitous," Eagle says. By analysing how often you meet your associates, the phone will be able to work out the strength of friendships. It might even pick up on a flirtation before you notice.
Epidemiologists and sociologists are excited about being able to glean this information. They will be able use it to track how people's interactions and activities affect how diseases spread and how social networks form. At present they rely on self-reported surveys, which are known to be inaccurate.
"This will revolutionise the field of social network analysis," says David Lazer of Harvard University, who analyses social interaction. "It gives you accurate measurements as opposed to just recalled behaviours."
This article appears in New
Scientist issue: 27 November 2004