In 1906, Francis Galton watched a crowd at a county fair guess the weight of an ox. No single person got it right, but the average of all their guesses landed within one pound of the actual weight. This result has been celebrated for over a century as proof of the wisdom of crowds. But there's a detail most retellings leave out: the crowd at that fair was remarkably diverse—butchers, farmers, clerks, and casual onlookers all contributed.

What happens when you strip that diversity away? The collective answer gets worse. Sometimes dramatically so. Network science has spent the last two decades unpacking why this happens, and the findings carry consequences for everything from corporate strategy to personal career decisions.

The mechanism isn't mysterious once you see it through a network lens. Homogeneous groups share the same blind spots. Their errors don't cancel out—they compound. Diverse networks, by contrast, produce a form of cognitive insurance that no amount of individual brilliance can replicate. Here's how it works.

Every decision rests on assumptions, and every assumption is a potential point of failure. When you surround yourself with people who share your background, industry, and worldview, you build what network scientists call a correlated error structure. Everyone in the group tends to be wrong in the same direction, at the same time, for the same reasons.

Consider a leadership team staffed entirely by people with finance backgrounds evaluating a product launch. They'll model the market beautifully. But they may collectively underestimate cultural resistance to the product, miss a regulatory shift brewing in a different sector, or fail to anticipate how end users actually behave. Their errors aren't random—they're systematic, shaped by shared training and shared information diets.

A structurally diverse network disrupts this pattern. When someone trained in behavioral science, someone with frontline customer experience, and someone from a different geographic market all weigh in, each brings a distinct error profile. One person's blind spot is another's area of clarity. The errors become uncorrelated, and uncorrelated errors are precisely what allows averaging, debate, and synthesis to push a group closer to the truth.

This is not about politeness or representation for its own sake. It is a structural property of information flow. Research by Scott Page at the University of Michigan formalized this insight: a group's collective accuracy depends less on the average ability of its members and more on the diversity of their cognitive approaches. Two moderately skilled people who think differently will often outperform two experts who think alike. The network's topology—who is connected to whom, and how different those connections are—determines whether errors get caught or get amplified.

Takeaway

The value of a diverse network isn't additive—it's corrective. Surrounding yourself with people who share your blind spots doesn't double your insight; it doubles your exposure to the same mistakes.

Prediction markets offer a rare laboratory for studying how network composition affects decision quality with measurable outcomes. In these markets, participants buy and sell contracts tied to future events—election results, product launches, economic indicators—and the prices reflect the crowd's collective probability estimate. When they work well, they outperform individual experts. When they don't, the failure usually traces back to network structure.

A series of studies, including influential work by economists at MIT and the Santa Fe Institute, found that prediction markets performed best when participants had access to diverse information sources and worst when traders clustered into ideologically or informationally homogeneous groups. In one experiment, networks where individuals could only see the trades of people similar to themselves produced prices that drifted far from reality. Networks that exposed participants to a wider range of signals self-corrected faster and ended up significantly more accurate.

The mechanism is straightforward. In a homogeneous network, a compelling but wrong narrative can cascade unchecked. Everyone reinforces the same signal. In a heterogeneous network, contrarian information enters the system from multiple angles, creating friction that slows bandwagon effects and forces participants to update their beliefs. This is why information diversity—not just demographic diversity—matters so much for collective judgment.

The implications extend well beyond financial markets. Any group that aggregates opinions to make a decision—a hiring committee, a product team, a board of directors—functions as an informal prediction market. The research suggests that the single most impactful design choice is not who speaks loudest or who has the most credentials, but whether the network feeding information into the room is structurally varied. When it is, the group's collective forecast improves. When it isn't, even brilliant individuals can converge on confidently wrong answers.

Takeaway

The accuracy of any group prediction depends less on who's smartest in the room and more on whether the room draws information from structurally different corners of the network.

Understanding the theory is one thing. Reshaping your own network is another. The challenge is that human networks naturally tend toward homophily—we gravitate toward people who share our profession, values, and social context. This isn't a character flaw; it's a well-documented structural tendency that operates across cultures and ages. Overcoming it requires deliberate design.

Start by auditing your current advisory inputs. Map out the five to ten people whose opinions most influence your major decisions. Then ask: Where do they overlap? Do they read the same sources, attend the same conferences, operate in the same industry? The degree to which their knowledge bases correlate is the degree to which your network fails to correct your errors. You're not looking for people who disagree with you on everything—you're looking for people who see different parts of the landscape.

Practically, this means cultivating what Mark Granovetter called weak ties—connections to people outside your immediate social circle. Join a professional community in an adjacent field. Maintain relationships with former colleagues who've moved into different sectors. Seek out advisory conversations with people whose daily information diet looks nothing like yours. These connections feel less comfortable than talking to people who already get your context, and that discomfort is precisely the signal that the relationship carries novel information.

One useful framework is to diversify across three dimensions: industry (people who work in different sectors), function (people who hold different roles even within your sector), and geography (people embedded in different regional or cultural contexts). You don't need dozens of new contacts. Even two or three strategically different voices in your network can meaningfully shift the quality of your decisions by introducing the uncorrelated perspectives that keep your assumptions honest.

Takeaway

A well-designed advisory network isn't a collection of the most impressive people you know—it's a collection of the most differently positioned people you trust.

Network diversity isn't a soft virtue. It's a structural advantage with measurable effects on decision accuracy, forecast quality, and error detection. The research converges from multiple directions: diverse inputs produce better outputs, not because difference is inherently good, but because correlated errors are inherently dangerous.

You don't need to overhaul your social life. You need to notice where your information sources overlap and deliberately introduce gaps. A few well-placed weak ties can do more for your judgment than a hundred strong ones that all see the world the same way.

The next time you face a consequential decision, ask not just who you're consulting—ask how differently they see the problem. That's where the real intelligence lives.