Emergent Intelligence from a Jar of Beans



One of the most cited examples of the Wisdom of Crowds involves a group’s collective ability to accurately guess the number of jelly beans in a jar. For each individual, however, predicting the number of jelly beans is very hard.  People do their best to estimate, but it’s not surprising that most are way off. What is surprising is that when you have a large enough group, the average answer tends to be highly accurate – even more accurate than the best guesser in the group.

This happens because, for every person who wildly overestimates the number of beans, there is another person who wildly underestimates. And if you have a large enough population, you get an equilibrium of opinion around what usually turns out to be a very accurate number.

Clearly, the average opinion across a group of bean-counters has value.  But to achieve impressive accuracy, a large population is required, even for a task as simple as estimating Jelly Beans. In the classic experiment by Jack Treynor, 56 people were polled to achieve an average estimate that was within 3% of the correct number of beans in a jar. 

Sure, it’s fascinating that the process works reliably, but from a practical perspective – it’s easier to just count the beans than it is to poll 56 people. 

Can’t we do better?  Isn’t there a more effective way to harness the intelligence of groups such that fewer individuals are needed to achieve accuracy?  At UNANIMOUS A.I., we believe there is. The solution is to move past a simple average votes to something far more dynamic and interactive.


Instead of polling a crowd and taking a simple average, we modeled our methods on how natural organisms make decisions in groups.  In the world of biology it’s called “swarm intelligence” and it’s as natural as the birds and bees, for both species rely heavily on collaborative decision making.  What birds and bees don’t do is give a single vote, wait for it to be averaged by some statistical process, and then make a decision.  Instead they form closed loop dynamic systems that adapt and change in real-time, resulting in a whole that makes far better decisions than any of it’s individual parts.   

Our goal is to bring the power of “swarm intelligence” to us humans.

We call the process social swarming” and it allows everyone in the group to continually adjust their estimates, as they push and pull on each other in real time, enabling the group to converge on a “best estimate” not an “average estimate.”

We’re also working hard to make the process FUN.  While there’s nothing fun about filling out a survey about jelly beans and then waiting for 1,000 other people to do the same, getting together online with a group of friends and swarming on questions and decisions is highly entertaining.  To see for yourself, just sign up to be a beta tester of the UNU swarming platform by clicking here


Although the benefits of good decision-making and accurate estimation are clear for the birds and the bees, it’s hard for us humans to make a buck guessing jelly beans in a bar. Still, groups need to make estimates and predictions decisions all the time. 

And how much fun would it be if you could turn those predictions into cold, hard cash?

What if you and your friends could form a “social swarm” and accurately predict how many yards Marshawn Lynch would run for in the Super Bowl?

Not that we’re advocating gambling, but that kind of prediction takes into account so many variables that getting it right deserves a huge return from Vegas.

This is what it looks like when a group of friends form a social swarm in UNU:

What you’re seeing is a group of people performing a complete, real-time negotiation.

As the puck moves across the range of answers, what’s you’re really seeing is the dynamically shifting “Best Guess” of the group, based on closed-loop feedback. 

And then, as the magnets move around the puck, jumping from the left side of the puck to the right and back, you can the shifting allegiance of people who feel the group’s Best Guess is overestimating and then underestimating the answer.

Eventually – and by that we mean, in seconds – an equilibrium of opinion is established and the prediction is registered. 

UNU makes this process much faster and more efficient than taking a poll, but also much more fun. UNU allows people to find intelligent consensus, and it makes everyone feel like they contributed to the decision in ways that no survey ever could.

And of course, as we now know, Marshawn Lynch finished the Super Bowl with 102 yards. Just like the group predicting the jelly beans in the jar, UNU was within 3% accuracy. And by using a swarm, we didn’t need 56 users to achieve this result, or even 26 users. 

The swarm above made the accurate prediction with only 10 users and took only 25 seconds. 


The cool thing about that these answers are that they doesn’t come from any individual person.

Again, no one in the group has any idea how many beans are actually in the jar is nor how many yards Lynch will run for, and no one in the group is offering an answer that sounds “right enough” that others are agreeing to it.

Instead, the answer is derived from a new emergent, collective intelligence derived from the combination of the individual intelligences as part of a unified adaptive system..

UNU’s dynamic feedback loop creates the perfect setting for this intelligence to emerge. 


With March Madness in full swing, we can’t wait to see how our UNU alpha groups do in predicting the scores.  In fact, if you want to take part in one of our many upcoming  swarm predictions, you can join our growing group of beta testers by clicking here.

But UNU isn’t really about gambling at all. There are a thousand different uses for this emergent collective intelligence.

Sales teams will use it for forecasting. Marketers will use it price products.  And friends might even use it to decide what movie they should go see tonight. 

What will you ask UNU?