Collective Intelligence Quanitifed[sic] for Computer-Mediated Group Problem Solving

(http://arxiv.org/abs/cs/0412064)

I think this is a good premise for a paper, but a bad execution; too little data. I hypothesize that they had trouble writing the code for the experiment. If they had had many tasks, and many ways of combining people, then they could have had a much better chance of getting some positive results. As it is, the simplicity of the 8-puzzle combined with the lack of positive results (okay, the group solutions were slightly better than the individual solutions for the "hardest" problems; but mostly no positive results) greatly detracts from the interest of the paper.

I recommend that people do similar experiments, and write papers about them.