The Wisdom of Crowds

The Ox at the County Fair

In 1906, the polymath Francis Galton — Charles Darwin's half-cousin, inventor of the weather map, fingerprint identification, and the word "eugenics" — walked into a county fair in Plymouth, England, and stumbled upon a weight-judging competition.1 A fat ox stood on display, and for sixpence, anyone could write down their guess of its slaughtered weight. About 787 people took a ticket. Farmers, butchers, clerks, people who had never touched a cow in their lives.

Galton, being Galton, collected the tickets afterward. He was interested in the "democratic judgment" of the crowd, and — if we're being honest — he expected to prove it was rubbish. Galton was an elitist. He believed in the superiority of experts. He wanted data to confirm that the average person had no business making quantitative estimates about anything.

He got the opposite.

The median guess was 1,207 pounds. The actual weight of the dressed ox was 1,198 pounds. The crowd was off by nine pounds — an error of 0.8 percent.2 No individual expert at the fair came that close. The crowd, in aggregate, was essentially perfect.

Galton, to his enormous credit, published the result. "The result seems more creditable to the trustworthiness of a democratic judgment than might have been expected," he wrote, in what might be the most grudging concession in the history of statistics.

This is the wisdom of crowds. Not the idea that crowds are always right — they're not — but that under the right conditions, a large group of ordinary people can be collectively smarter than any individual expert. It's one of the most counterintuitive results in all of mathematics, and it's hiding in plain sight, because it's really just the law of large numbers wearing a social-science costume.

Why Crowds Work: The Law of Large Numbers in Disguise

Here's the mathematical engine behind Galton's ox. Suppose each guesser produces an estimate that's the true value plus some random error:

If the errors are independent — and this is the crucial assumption — then the average of n guesses has variance σ²/n. As n grows, the average converges to the true value. That's the law of large numbers, and it's doing all the heavy lifting.

The key word is independent. Each person's error has to be their own. If everyone is copying the person next to them, you don't have 787 data points — you have one data point, repeated 787 times. The variance doesn't shrink. The crowd is just one loud voice echoing.

θ

True value we're estimating

εᵢ

Individual error (noise), mean zero

σ²

Variance of individual errors

n

Number of independent guessers

This is why the conditions matter. James Surowiecki, in his 2004 book The Wisdom of Crowds, identified four conditions that make collective intelligence work:3

When all four conditions hold, crowds are remarkably accurate. When any one breaks, they can be spectacularly wrong.

Try It Yourself

Here's your chance to be part of a crowd. Below is a field of dots. How many do you see? Make your best guess, then watch what happens when we aggregate yours with a simulated crowd.

Ask the Audience

If you've ever watched Who Wants to Be a Millionaire?, you've seen the wisdom of crowds in action. The "Ask the Audience" lifeline — where the studio audience votes on the correct answer — is right about 91 percent of the time.4 That's better than "Phone a Friend," which clocks in at around 65 percent. Think about that: a random collection of people who happened to buy tickets to a game show taping outperforms the contestant's single smartest friend.

Why? Because the audience satisfies all four of Surowiecki's conditions. They're diverse (different backgrounds, different knowledge bases). They're independent (they vote simultaneously on individual keypads, with no discussion). They're decentralized. And the voting mechanism aggregates their answers cleanly.

"Phone a Friend" gives you one expert's best guess. "Ask the Audience" gives you the law of large numbers.

Prediction Markets: Crowds With Skin in the Game

The county fair charged sixpence. Prediction markets raise the stakes. Platforms like Polymarket and the Iowa Electronic Markets let people buy and sell contracts on future events — elections, economic indicators, whether a particular bill will pass. If you think the probability of Event X is higher than the current market price, you buy. If lower, you sell. The market price becomes a crowd-sourced probability estimate.5

And these markets are astonishingly well-calibrated. Events trading at 70% happen about 70% of the time. Events at 90% happen about 90% of the time. This isn't because each trader is a genius. It's because the market mechanism — the aggregation condition — is working overtime, continuously incorporating new information from diverse, independent, financially motivated participants.

The key ingredient that prediction markets add is incentive. When people put money on their beliefs, they think harder. They don't just pattern-match to their tribal affiliations. The market penalizes systematic bias and rewards calibrated judgment. It's the wisdom of crowds with teeth.

When Crowds Go Wrong: The Cascade

Now for the dark side. Crowds are only as wise as their independence. Break independence, and you don't get collective intelligence — you get a stampede.

Consider information cascades.6 Imagine you're choosing between two restaurants. Restaurant A has a few people in it. Restaurant B is packed. You have a slight personal preference for A — the menu looked better online — but you think, "all those people can't be wrong." So you go to B. The next person sees an even more crowded B, and they too override their private signal. And so on. Before long, everyone is at Restaurant B, not because it's better, but because everyone is following everyone else.

The mathematical structure is elegant and terrifying. Each person has a private signal — some piece of information that's more likely correct than not. But they can also observe what previous people chose. Once the public information (the herd) outweighs the private signal, it becomes rational to ignore your own information and follow the crowd. That's the cascade. It can lock in on the wrong answer, and new information stops flowing into the system.

This is what happens in financial bubbles. In 1999, you didn't buy Pets.com stock because you thought online pet food delivery was a good business. You bought it because everyone else was buying it, and the price kept going up, and that rising price was a public signal that overwhelmed your private skepticism. Independence breaks, errors become correlated, and σ²/n doesn't shrink anymore — because the effective n has collapsed to something much smaller than the number of participants.

The Hong-Page Theorem

There's a beautiful result from complexity science that puts a point on all of this. Scott Page and Lu Hong proved, roughly, that a diverse group of problem-solvers will outperform a homogeneous group of high-ability problem-solvers.7 The theorem says: Diversity trumps ability, under certain conditions.

The intuition is that what matters for collective performance isn't how good each individual is in isolation — it's how different their approaches are. If you assemble a room full of the best experts and they all use the same heuristic, their errors are correlated. They're all wrong in the same direction. But if you assemble a diverse group — some experts, some amateurs, some people who approach the problem from completely unexpected angles — their errors point in different directions. They cancel. The group converges on truth.

This is the mathematical argument for diversity that has nothing to do with fairness or representation (though those matter too). It's a theorem about error structure. Diverse errors cancel; homogeneous errors compound.

Watch a Cascade Form

In the simulation below, players sequentially decide between Option A (blue, which is actually correct) and Option B (red). Each player sees a private signal that's right 60% of the time. They also see what previous players chose. When independence is ON, each player uses only their private signal. When OFF, they rationally weigh the public evidence — and cascades form.

The Moral of the Ox

The wisdom of crowds is not magic. It's mathematics. Specifically, it's the observation that averaging over independent, diverse estimators is a variance-reduction machine. Every time you add an independent voice, you shrink the noise by a factor of 1/n. That's extraordinarily powerful — but only as powerful as the independence assumption that makes it work.

Galton's fairgoers got it right because they didn't talk to each other before writing down their guesses. The Millionaire audience gets it right because they vote on individual keypads. Prediction markets get it right because each trader puts their own money where their own mouth is.

But social media? Groupthink? Bubbles? Those are all mechanisms for destroying independence, for making your error correlate with mine. And once the errors correlate, the wisdom of crowds becomes the madness of crowds — not because people got stupider, but because the mathematics changed.8

The lesson isn't "trust the crowd" or "distrust the crowd." It's: check the conditions. Is the crowd diverse? Are they thinking independently? Is there a good aggregation mechanism? If yes, trust the average over any individual — even an expert. If no, the crowd is just a mob with a mean.

Galton's ox teaches us something profound about the nature of knowledge. No one person at that county fair knew the weight of the ox. The information existed only in the aggregate, distributed across hundreds of imperfect estimators, invisible until someone thought to add them up. The truth was there all along — scattered, noisy, incomplete in every individual mind, but present in the collective.

That's the wisdom of crowds. It's not that the crowd is wise. It's that wisdom, sometimes, is a collective property — one that emerges from the mathematics of aggregation, and vanishes the moment we forget the conditions that make it work.