Reinventing Discovery_ The New Era of Networked Science - Michael Nielsen [36]
For example, in a 1989 follow-up to the original Stasser-Titus experiment, the group discussions were recorded so the experimenters could better understand how the groups came to their decisions. What they found was that instead of exploring all the available information, the groups spent most of their time discussing information they had in common. So, for example, if several people all knew that Best held an unpopular position on (say) dorm room visitation, there was likely to be a relatively lengthy discussion of that fact, and the information was likely to be mentioned again latin the discussion. But when someone in the group had a unique piece of information about a candidate, a piece of information that only they knew, the discussion of that information was usually perfunctory. That mattered, because in the original Stasser-Titus experiment, negative information about Best was often held in common by several members of the group, while positive information was often held by only a single member.
In 1996 another follow-up experiment was done, this time in a teaching hospital, asking groups to make medical diagnoses on the basis of video clips of patient interviews. Again, the information was partial: each person in the group saw only part of the video interview. The groups making the decisions included three people of different statuses a medical resident, an intern, and a student. Alarmingly, but perhaps not surprisingly, the groups paid much more attention to unique information held by the high-status medical resident. Unique information held by the interns and students was much more likely to be ignored.
These and many other studies paint a bleak picture for collective intelligence. They show that groups often don’t do a good job of taking advantage of their collective knowledge. Instead, they focus on knowledge they hold in common, they focus on knowledge held by high-status members of the group, and they often ignore the knowledge of low-status members of the group. Because of this, they don’t manage to convert individual insight into collective insight shared by the group. And that’s bad news if you’re trying to use collective intelligence.
The Limits to Collective Intelligence
Why are projects such as the Polymath Project, Kasparov versus the World, and the MathWorks competition so successful, while the groups in the Stasser-Titus and related experiments perform so poorly? To put it more precisely, why were the groups in the successful projects able to convert their best individual insights into collective insight, while the groups in the Stasser-Titus and related experiments failed to make this conversion? Was the difference due merely to differences in the processes used in the respective cases? Or is there some more fundamental difference, a difference that can’t be solved by an improved process, perhaps due to the nature of the problems under discussion?
To answer these questions, I want you to consider a little brain-teaser. I’ll give a verbal description of the puzzle, but the puzzle is rather visual, and you may find it illuminating to consult the pictorial explanation given in the picture and caption on the next page. You’re given an empty eight-by-eight chessboard, and asked to cover it with one-by-two dominoes, so that only two squares remain uncovered: the square in the bottom left, and the square in the top right. Can you do this? If so, how? If not, why not? You’re not allowed to stack dominoes, or break dominoes, or leave dominoes hanging off the edge of the board—everything in the puzzle statement is to be interpreted in the usual way. To further simplify things,