The I.Q. of the Crowd

by Jim Schnabel

October 1, 2010

One of the basic findings in modern psychology is that people who are good at one kind of mental task tend to do well at others. A person thus has a “general intelligence” level, which can be quantified, for example on I.Q. tests. Researchers now report that a group can have a general intelligence too. Surprisingly, this collective ability appears to depend more on how “tuned in” group members are to one another than on their average individual intelligence.

“We found that individual intelligence had no more than a weak relationship with collective intelligence,” says the study’s first author, Anita Woolley, an assistant professor of organizational behavior and theory at Carnegie Mellon University’s Tepper School of Business in Pittsburgh, PA. The research was first reported online in Science on Sept. 30.

“It’s a neat idea to look at the intelligence of a group in this way,” says Chris Frith, a cognitive neuroscientist at the University of Aarhus in Denmark who has worked in this area.

Stephen Kosslyn, a social cognition researcher at Harvard, notes that the finding in 1904 of a general intelligence for individuals has made a huge impact in the field of psychology. “It is possible that the discovery of a general collective intelligence factor will have a similar impact,” he says.


Greater than the sum of its parts

In the study, Woolley and her colleagues tested 120 people for general intelligence as well as personality-related characteristics, and then randomly assigned them to 40 three-person groups. Each group had to perform a diverse set of group tasks over several hours, such as solving visual puzzles, brainstorming, and making collective moral judgments.

The researchers found that each group’s performances on one of these tasks tended to correlate with its performances on the other tasks. Using a statistical technique known as factor analysis, they determined that a single underlying factor explained nearly half of the variation in groups' task performances. A similar technique has been used previously in individuals to show that a single factor—the basis for I.Q.—influences individual test scores to about the same degree.

Naming this collective intelligence factor c, Woolley and her colleagues then found that it strongly predicted each group’s performance on a new, more complex task—collectively playing checkers against a computer. By contrast, neither the average nor the maximum individual intelligence within each group was a significant predictor of this performance.

To confirm these results, the researchers did a follow-up study with nearly 600 additional people who were divided into teams that varied in size from 2 to 5 members. Each team now performed a wider range of group tasks, and each subject was evaluated with an alternative measure of individual intelligence. The results were essentially the same.

Analyses indicated not only that individual smarts didn’t matter much to group intelligence, but that measures of group cohesion, motivation, and satisfaction also didn’t matter.

The sturdiest predictor of group intelligence, the researchers found, was a measure of how evenly group-members contributed. “The groups in which everybody was speaking in more equal amounts tended to be collectively more intelligent than the groups in which one person was dominating the conversation,” says Woolley.

Another significant predictor of higher group intelligence was a higher individual average score on a social-sensitivity test known as “reading the mind in the eyes”—a measure of the ability to intuit another person’s feelings from an image of his or her upper face. Related to a key aspect of social cognition, also sometimes known as theory-of-mind ability, the test was originally designed to screen for people with autism spectrum disorders, who typically score poorly on it.

“It also differentiates among people in the more normal end of the continuum,” says Woolley. In particular, women tend to score more highly on this test than men—and thus Woolley’s team found that the groups with a higher proportion of women tended to have higher collective intelligence.

To Kosslyn (whom Woolley consulted in the early stages of her study) it makes sense that collective intelligence depends so heavily on group members’ ability to connect with one another. “In my view, each of us comes to rely on other people to help us fill in for our weaknesses,” he says. “We set up what I call social prosthetic systems, in which other people help us to extend our intelligence and, equally important, discover and regulate our emotions. I suspect that social sensitivity is one factor that affects how easily and effectively people can set up these social prosthetic systems.”

So far there are few studies in the scientific literature that bear directly on these questions. One that appeared in Science in August, however, came up with broadly similar results. A team of researchers, whose senior author was Chris Frith, found that pairs of subjects did better than either alone, in a decision task based on simple visual perception—but only under certain conditions.

“It critically depended on their being able to discuss with each other when they disagreed on what the visual signal was,” says Frith.

The subjects also had to be relatively evenly matched in ability to obtain the two-heads-are-better-than-one effect. When greatly mismatched, the pair performed worse together than the better pair-member performed alone.

According to Frith, the communication within these pairs transmits not only each person's estimates concerning the visual stimulus, but also his or her confidence level:  “So you have to have some estimate of not only how good people are, but how much insight they have.” His group is now working on a follow-up study of what team-members actually say during these interactions.