Today is that grandest of holidays, OneWebDay. Billed as "Earth Day for the Web", people everywhere thankful for what the world wide web has given us are engaging in little projects to improve and honor it. Here's what I did...

My particular project was inspired by James Surowiecki's amazing work on the wisdom of crowds. It seems that, given the right circumstances, crowds can be remarkably intelligent—quite contrary to popular opinion that, while a person is smart "people" are stupid. I've seen other people replicate the "jellybean jar" phenomenon and thought I would give it a try at my office.

Out comes a really big jar and lots of malted milkballs. One hundred seventy-six (176) of them in fact, though I was tempted throughout the process to, uhm, abbreviate the total. The general idea is that, despite the wide range of guesses from the staff on how many milkballs are in the jar, the average of our guesses should be very, very close. Hopefully (since that’s the point of the project) closer than any one guess.

Next I go office to office, stopping people in hallways as need be, and give them the pitch: "Fill out the piece of paper with your name and your guess, using any method you wish for guessing—other than removing the top and counting them one-by-one, however it is of the utmost importance that you discuss neither your guess nor your strategy with any other staff member." Yes, it was a run-on sentence, but by the end of the morning, I had it down pretty well.

It turns out we had guesses as low as 86 and as high as 275. That's quite a range. Still, when I averaged all the guesses together, I got 178—only two away from the correct number! That beats both the closest guess (at 168) and the "panel of experts", meaning the average of the five closest guesses, (at 164). Yes, including the outlier "noise" actually made the guess more accurate.

When people are asked to make a decision, they do so with a certain amount of bias. If they discuss their decision with others, this bias spreads to others. However, in a diverse crowd operating as individuals, these biases cancel one another out, making the group more intelligent. I;'m sure that a memetic analysis would involve memes and anti-memes colliding and exploding.

The staff was overall very appreciative of the experiment/celebration and immediately saw its implications with how we both interact in group decision-making and how we receive feedback from our activist network. They were also quite happy that the lesson ended with the average being closer than any one guess, because that meant we got to split the malted milkballs.