Vox Populi

Have you ever played a game where you were asked to guess how many cookies or pieces of candy were in a jar?  Did you come close to the answer?  As it turns out, you would have had a much better chance of guessing the correct answer if you had been part of a large group with everyone submitting a guess and the final answer being an average of all the guesses.   

   
All the way back in 1906, the English polymath, Sir Francis Galton, attended "The West of England Fat Stock and Poultry Exhibition" in Plymouth.  While there, he discovered a "guess the weight" of a cow game, similar to the ones described above.  The game attracted more than 800 people, all of whom paid half a shilling to write down their guesses.  They wrote their guess on a ticket (along with their name and address).  Galton apparently became interested in the guesses, because he paid the ticket master for all of the tickets at the end of the game.  Unfortunately, only 787 of the tickets were usable.  He averaged all of the guesses together, which resulted in an overall average guess of 1,207 pounds.  The actual weight of the cow was 1,198 pounds - the average guess was only 9 pounds off!  He published his results in the journal Nature (the original article can be found here).  The "crowd" showed a lot more "wisdom" compared to the individual participants.

Whether you call it the "wisdom of crowds", "collective intelligence",  "crowdsourcing", or "vox populi", it really doesn't matter.  What is well-established is that large groups of people are collectively smarter than individuals when it comes to problem-solving, decision-making, innovating, and predicting (see last year's post, "How groups make can make better decisions...").  The theory is that individuals are inherently biased - taking the average from a much larger group of individuals theoretically eliminates this bias and noise, leading to more accurate decisions.

It’s not quite that simple though.  As I mentioned in a post last year, there is something known as the "common information effect", which could place groups at a disadvantage relative to individuals.  Here, groups lose the purported benefits of diversity and the "wisdom of crowds" effect because the individuals in the group focus primarily - almost exclusively - on the information that they share with each other.  Similarly, groups are subject to social influence effects (individuals adopt their peers' opinions and copy their solutions to problems), particularly under conditions of something known as network clustering (like-minded individuals tend to work together more).  Together, these effects decrease the benefits of group diversity and can lead to premature convergence on a suboptimal solution to a problem.

So then, which interpretation is correct?  Do groups really make better decisions than individuals?  And if so, does context matter?  We are left to question whether it is possible to leverage the benefits of group decision-making ("the wisdom of the crowds" effect) without the accompanying loss of group diversity due to social influence effects and/or the common information effect.

It is with all of this in mind that I want to talk about a study ("How intermittent breaks in interaction improve collective intelligence") published by the Harvard Business School professor Ethan Bernstein with his colleagues Jesse Shore from Boston University and David Lazer from Northeastern University.  They conducted a set of experiments in which three-person groups tried to solve the famous traveling salesperson problem.  The problem here is to find the shortest travel distance between 25 different cities (it's a surprisingly difficult problem that requires higher level mathematics).  Each group completed the task seventeen times (each "round" lasted just under one minute), so theoretically they could improve their performance by learning from past rounds.  

One set of groups never interacted with the other groups, working on the traveling salesperson problem in complete isolation (i.e. no social influence effects).  Another set of groups constantly interacted with the other groups (i.e. social influence effects were present all the time), while the final set of groups interacted only intermittently with the other groups.  Based upon the previous research summarized above, Bernstein and colleagues predicted that the groups in which members never interacted would be the most creative, coming up with a lot of different solutions to the traveling salesperson problem with varying degrees of quality - some of the solutions would be really good and others would be really bad.  They also predicted that the groups that constantly interacted would find an overall higher quality of solutions (groups would learn from each other, so there would be potentially fewer "bad solutions" as observed with the groups that never interacted).

The results of the study were consistent with their predictions.  The groups working alone and never interacting came up with some really great solutions to the traveling salesperson problem, but they also came up with some really terrible solutions as well.  They also found that the groups that were constantly interacting with other groups came up with a better overall solution to the problem on average, but they found a solution in each round less frequently.  

The best results were found in the groups that intermittently came together - the best of both worlds, if you will.  They came up with an optimal solution more frequently, and the best solutions were generally much better (i.e. shorter distances between the 25 cities) compared to the other sets of groups.  In other words, when the effects of social influence are intermittent, it provides all the benefits of group decision-making (potentially with the additional benefit of collective learning between the different groups) without the associated costs.

Based on these results, Bernstein argues that organizations should be redesigned to intermittently isolate people from each other's work for the best collective performance.  When asked about the study in an interview, he said, “It provides real instruction for business practice because that’s not where we’re going.  We’re not naturally going to be in a world of intermittency anymore. We’re naturally in a world of always on.”  The conventional thinking is that we should be leveraging the "wisdom of crowds" by connecting groups as much as possible.  However, this study suggests that we should also be giving groups time by themselves to work on problems.

Bernstein's co-author, David Lazer adds, "People are used to thinking and believing that we want to maximize how much people learn from each other; we want to maximize transparency.  I don’t want there to be zero communication, but I do think that transparency and communication do have downsides, especially for certain kinds of problems; problems where you need certain kinds of creativity and you want to avoid rapid convergence because that convergence reduces how much the group explores and considers alternatives.”

There are examples of this concept of intermittency in the business world.  For example, "sprints" are used by agile teams, where people come together for a period of time and then separate again.  One of the growing trends is for companies to build open floor plans, yet even in these open spaces, they create nooks where workers can isolate themselves when needed. 

The switch to remote work during the COVID-19 pandemic has undoubtedly changed the way that we work together.  I spend more time in hybrid meetings now than I ever did before the pandemic, and with tools such as Microsoft Teams, Zoom, and Skype, groups can always connect with each other.  The research by Bernstein and colleagues suggests that this may not be a good idea.  There is no question that we can leverage the "wisdom of the crowd" to make better decisions - we just have to do it in a way that maximizes all of the benefits while minimizng the associated costs.