PUNT? Professor Thomas Malone (standing) coaches CCI student researcher Jason Carver through the first quarter of play.
Making money watching Tom Brady? Sign me up!
That’s what I thought to myself, after seeing a June Craigslist post entitled “MIT will PAY you to watch football.”
“Think you know something about football?” the ad said. “Then come to our lab, watch a game, and see how well you can predict what the teams will do.”
The invitation was issued by MIT’s Center for Collective Intelligence (CCI), whose broadest mission is to conduct research on how new communications technologies are changing the way people work together. CCI is the first university-based research center in the United States to focus specifically on the study of collective intelligence, a bafflingly oblique discipline that studies, among other things, how people think and function en masse toward a common result. Understanding those dynamics can help you to predict everything from elections to Wall Street trends.
The theoretical forecast models used by collective-intelligence experts are sometimes called “prediction markets” — a set of algorithms, for example, based on past performance, by which a computer might predict air-travel volume. (In nearly every election when they’ve been used, prediction markets have been more accurate than polls.)
But people sometimes know things algorithms don’t. A computer might have predicted accurate air-passenger stats for September 9, 2001, but would have failed miserably, using the same prediction market, a week later. Meanwhile, any fool with a heartbeat could have easily foreseen a major decrease.
But what’s that have to do with football?
The core research question in CCI’s football experiment seeks to answer how people and computers can join forces so that, collectively, they act more intelligently than any person, group, or computer has ever done before.
Those taking part in the MIT experiment watch a video of a college football game (no Tom, to my dismay), which researchers periodically pause so that subjects can participate in a prediction market to forecast the next play. In addition to people, computer “agents” (developed by wiz-kid Jason Carver as part of his master’s thesis) also participate, basing their predictions on previously transcribed data about identical game conditions (e.g., first down, 10 yards to go, etc.) and simple rules for determining what a football team is likely to do in different situations.
CCI Director Thomas Malone, the Patrick J. McGovern Professor of Management at MIT’s Sloan School of Management, chose football plays because he thought they were an interesting analogy for other kinds of predictions — a company trying to dope out what a competitor is going to do, or a general speculating on the enemy’s next move. Of course, football — in addition to being a somewhat more sanitized arena than the field of battle — is also a good way to get butts into the lab seats in sports-crazy Boston.
“We expect to compare the accuracy of predictions in three conditions: humans only, agents only, and both humans and agents,” says Malone. “We don’t know yet how the results will come out, but we think one particularly interesting result would be if humans and agents together do better than either alone.”
CCI will be running experiments throughout the summer. No word yet on whether Bill Belichick has signed up.