How Groups of People can Make Smart Predictions
Under certain conditions, groups of people can make better predictions and generate deeper insights than even the smartest individual.

James Surowiecki starts The Wisdom of Crowds with the story of a scientist who saw a weight-judging competition at an English livestock exhibition. Curious about how the amateur judges performed, the scientist measured their guesses. He thought he’d find a bell curve where most people were wrong, a few were right, and a few were very wrong.
Instead, the amateurs were right. The ox whose weight they were guessing weighed 1,198 pounds. On average, the amateurs guessed the ox weighed 1,197 pounds.
Today, the wisdom of crowds is more accessible than it was at ox-weighing contests in 1906. The internet makes it easy to collect the insights of many people at once. Prediction markets have existed for decades but acquired new power in the internet age.
However, there are limits to how wise crowds can be. Groups must be structured properly to get the best insights out of their aggregate opinions.
How the Wisdom of Crowds Works
There are four major conditions a group has to fulfill to produce a “wise” prediction:
Diversity of Opinion
Independence
Decentralization
Aggregation
Each of these conditions has its own nuances, but the point is to build a group where each person’s errors balance someone else’s and have systems in place to draw the group’s best ideas out.
Diversity of Opinion
One of the best ways to find the best idea is to have many of them. However, they can’t all be the same. If all the guesses in a group are the same, then there’s a high chance that they’re all similarly wrong.
Surowiecki gives an example that repeats itself often in the free market:
“The early days of the business are characterized by a profusion of alternatives, many of them dramatically different from each other in design and technology,” Surowiecki wrote. “As time passes, the market winnows out the winners and losers, effectively choosing which technologies will flourish and which will disappear.”
In this case, the crowd having its choices judged is made of different companies in the same industry. Customers gravitate toward the best products, and the companies that produce them survive.
The best predictions can’t come to light without the best list of possibilities.
Independence
Group decision-making is at its best when people use the information they have. Everyone has imperfect information, and Surowiecki recommends relying on errors in information to cancel other errors out.
When individuals follow each other instead of their information, then the group falls into groupthink. That can undermine the promise of group predictions and lead otherwise intelligent groups off a cliff.
Surowiecki cites the plank road mania of the 1800s. One New York engineer became convinced that plank roads would last long enough to be worth the maintenance and investment costs. He saw some work in Canada and tried them in New York. Enthusiasm for them became widespread in the United States.
Then the roads started to decay in half the time that the engineer thought they would. They were also more expensive than the engineer thought.
The mania for these expensive, shoddy roads occurred because many people followed one individual and the people who supported his vision. Had the people who founded plank road companies conducted their own due diligence, they may have been able to save time and money on failed projects.
Follow information, not people.
Decentralization
Surowiecki notes that in decentralized systems, “power does not fully reside in one central location, and many of the important decisions are made by individuals based on their own local and specific knowledge rather than by an omniscient or farseeing power.”
Decentralized systems lead to specialization, which not only makes people “more productive and efficient.” It also “increases the scope and the diversity of the opinions and information in the system.”
This is where individual expertise becomes critical. The collection of local pictures lays the groundwork for a wider view of a problem.
Surowiecki cites the CIA as a well-functioning decentralized system. It’s made of thousands of intelligent individuals looking at data and coming up with predictions and recommendations based on what they’re seeing.
Aggregation
A system can get the first three group requirements right and still fail. The predictions and judgments that the system of experts comes up with have to be in one place so they can be put together into one cohesive picture.
Prediction markets do this easily. Customers can buy and sell shares priced from one to 99 dollars or cents and trade their shares over time. The resulting share price is equal to a probability. Kalshi offers prediction markets on many issues and topics using this model.
For intelligence decisions, the stakes are higher and the decisions more difficult. Surowiecki cites the failure of the intelligence community to anticipate and stop the 9/11 attacks, particularly the CIA. The CIA’s decentralized nature wasn’t the issue. Its inability to aggregate the necessary insight from the necessary experts was.
Aside from top-down decision-making, some of the proposed solutions to the aggregation shortcomings included internal decision markets. FutureMAP was a short-lived internal prediction market where a few dozen people across agencies could have traded on key security predictions.
This project was short-lived because a public version had been proposed. Appalled congressmen killed the project, and many in the public agreed with them. It was viewed as a way to bet on instability and human suffering in remote parts of the world.
But it also put a clear metric on the complex findings of key national security experts.