In 2010, a team of researchers led by Anita Williams Woolley at Carnegie Mellon and Thomas W. Malone at MIT published a paper in Science that should have changed how every organization on earth thinks about groups. They did not get a lot of press. The finding was too counterintuitive for most people to accept.
The study was straightforward. Woolley and her colleagues recruited 699 people and organized them into groups of two to five. Each group completed a battery of tasks designed to test different kinds of cognitive work: brainstorming, puzzle-solving, collective moral reasoning, negotiation, strategic coordination. The researchers measured each group's performance across the full range of tasks. Then they looked for a pattern.
They found one. Just as individuals have a measurable general intelligence factor — psychologists call it g — groups turned out to have an analogous factor. They called it c, for collective intelligence. Groups that scored high on c performed well across all the tasks, not just some of them. The collective intelligence of a group predicted its future performance on tasks it had never seen before. It was real, it was stable, and it was measurable.
But here is the part that broke people's intuitions: the c factor was not significantly correlated with the average intelligence of the group members. It was not correlated with the maximum intelligence of anyone in the group. Having the smartest person in the room did not make the group smart. Having a room full of smart people did not either.
What actually predicted group intelligence
Three things correlated with collective intelligence. All three were about process, not personnel.
Equal distribution of conversational turn-taking. In the highest-performing groups, everyone spoke in roughly equal measure. No one dominated. No one disappeared. The researchers measured this directly from audio data — they tracked not just what people said, but how evenly the speaking time was distributed. Groups where one or two people did most of the talking scored lower on c, regardless of how brilliant those talkers were.
Average social sensitivity. Woolley's team measured this using the "Reading the Mind in the Eyes" test — a validated psychological instrument where subjects look at photographs of eyes and identify the emotion being expressed. It is a measure of how well someone can read other people. Groups whose members scored higher on this test performed better collectively. The ability to sense what someone else is thinking or feeling — what psychologists call Theory of Mind — predicted group performance more strongly than any other individual trait they measured.
The proportion of women in the group. Groups with more women performed better. But this was not an independent effect. It was mediated by the first two factors. On average, the women in the study scored higher on social sensitivity and contributed to more equal turn-taking patterns. The gender composition effect largely disappeared when the researchers controlled for social sensitivity and turn-taking equality. The finding was not about gender per se. It was about the interpersonal dynamics that happened to correlate with gender in their sample.
Woolley put it plainly in a 2015 paper with Malone in Current Directions in Psychological Science: "Average ToM scores remained the only significant predictor of collective intelligence even when controlling for individual intelligence or other group composition or process variables." Translation: the single best predictor of how smart a group would be was how well its members could read each other, not how much any of them knew.
The replication debate
Good science invites disagreement, and the collective intelligence findings got plenty of it. In 2017, Timothy Bates and Shivani Gupta at the University of Edinburgh published a replication attempt across three studies with 312 people. Their conclusion was sharply different: individual IQ accounted for around 80% of group-IQ differences. Social sensitivity and turn-taking equality showed no significant effect. The title of their paper made their position clear: "Smart Groups of Smart People."
A 2021 meta-analysis published in Cognitive Research: Principles and Implications tried to reconcile the competing findings. The authors reviewed every published study testing the c factor and found that the evidence was suggestive but methodologically fragile. The c factor did appear to correlate moderately with group performance — better, in fact, than the average IQ of group members. But the number of studies was small, sample sizes were modest, and most had not adequately controlled for individual intelligence. The authors cautioned against "prematurely embracing" the c factor.
Then in 2024, Rowe, Hattie, and Munro published a new study in PLoS ONE that complicated the picture further. Working with 85 university students, they found that collective intelligence was real but not a single factor. Their data favored a two-factor model — analogous to the distinction between fluid and crystallized intelligence in individuals. These two factors predicted group performance on future tasks with moderate strength (correlations between .40 and .47). The c factor was real, but it was not as simple as the original 2010 paper suggested.
What matters for anyone running or joining a peer group is not which side of the debate is right. It is what both sides agree on. Even the skeptics — even Bates and Gupta — acknowledge that group dynamics matter. Their argument is about whether there is a single magical c factor, not about whether how a group interacts affects what it produces. On that point, there is no disagreement at all.
The turn-taking mechanism
Separate from the c factor debate, the research on conversational turn-taking in groups is vast and unambiguous. Alex Pentland's work at the MIT Media Lab — using sociometric badges that track speaking patterns in real time — has shown repeatedly that the distribution of speaking time is one of the strongest predictors of group performance across nearly every context studied.
In cooperative games, groups with higher turn-taking rates and more balanced participation earned more money for the collective. Groups where one person dominated the conversation earned less. The pattern held even when the dominant speaker was the most knowledgeable person in the room. Being right does not help if no one else in the group is engaged enough to act on it.
A 2021 study published in Proceedings of the ACM by Woolley and colleagues examined 150 participants in 50 teams and measured "inclusive turn-taking" — defined as a pattern where different members speak in successive turns rather than the same person speaking repeatedly. They found that inclusive turn-taking improved team skill use, and when combined with strong task strategy, significantly improved team performance. The mediating mechanism was straightforward: when everyone speaks, the group surfaces more of its distributed knowledge and applies more of its members' actual skills.
This is why networking events fail and structured peer groups succeed. A networking event has no turn-taking structure. The loudest person wins. The most socially aggressive person collects the most cards. A structured mastermind group enforces turn-taking by design — hot seats, timed rounds, facilitated check-ins. The format itself creates the conditions that the collective intelligence research identified as most important.
Social sensitivity is a group input, not a personality trait
The most overlooked finding in the Woolley study is how social sensitivity functions in groups. It is easy to read the result as "hire empathetic people and your team will be smarter." That misses the point. Social sensitivity in the study was not about being nice. It was about informational throughput.
A person with high social sensitivity picks up on nonverbal cues — a flicker of disagreement in someone's expression, a hesitation before they answer, a shift in body language that signals they have something to say but are holding back. In a group context, this translates into more information entering the conversation. Someone who notices that a group member is uncomfortable with a plan and asks them about it extracts information that would otherwise stay hidden. Someone who reads the room well enough to know when to push and when to pause keeps the conversation productive instead of adversarial.
This is the mechanism behind psychological safety — Amy Edmondson's concept that has become a fixture of organizational research. Groups where people feel safe to speak up produce better outcomes not because safety feels good (though it does) but because safety increases the volume of relevant information that enters the group's decision-making process. Social sensitivity is one of the mechanisms that creates safety. Reciprocal vulnerability is another. Both serve the same function: making sure the group has access to everything its members know.
Transactive memory — the phenomenon where group members develop a shared awareness of who knows what — depends on the same dynamics. You cannot build a transactive memory system in a group where one person dominates, where members cannot read each other, or where relevant knowledge stays hidden because no one feels safe surfacing it. Turn-taking equality and social sensitivity are not just nice to have. They are the infrastructure that makes collective knowledge work.
The founder's collective intelligence problem
Most founders operate in groups that violate every principle the collective intelligence research identified as important. Board meetings have extreme power asymmetries — no one speaks equally when the person across the table can fire you. Team meetings suffer from the same problem in reverse — no one tells you what they really think when you sign their paychecks. Advisory calls are one-to-one, which eliminates the group dynamics entirely. Investor updates are performative, not analytical. Conferences are noise.
The typical founder's informational environment is a near-perfect inversion of what the research says produces collective intelligence. Unequal speaking time. Low social sensitivity (or social sensitivity subordinated to power dynamics). No structured turn-taking. No mechanisms for surfacing distributed knowledge.
A peer group flips every one of these conditions. There is no power asymmetry — everyone is a founder at a similar stage, making decisions alone with the same constraints and uncertainties. The structure enforces equal speaking time. The recurring nature of the group builds social sensitivity through repeated exposure — you learn to read people you meet with regularly in ways you never could with strangers at a conference. And the format is designed to surface problems, not performances.
This is not about assembling a room of geniuses. The Woolley study's most important practical finding was the negative one: piling up individual intelligence does not reliably produce collective intelligence. What produces it is the conditions under which people interact. Equal airtime. Genuine attention to what others are saying and feeling. A structure that prevents domination and rewards contribution.
Structure is the variable you can control
The collective intelligence literature — all of it, including the skeptical replications — converges on one practical truth: how a group is structured matters at least as much as who is in it. You cannot control the IQ of the people around you. You cannot screen for social sensitivity on a name tag at a networking event. But you can choose to be in a group with enforced turn-taking, a format that surfaces honest problems, and a size that allows real depth instead of performative breadth.
Five to eight people is the range where these dynamics work best. Small enough that no one can hide. Large enough that there is genuine cognitive diversity. Structured enough that the format does the heavy lifting — not the loudest personality, not the most credentialed person, not the most charismatic founder who would dominate in any unstructured setting.
Woolley, Malone, and Pentland did not set out to prove that peer groups work. They set out to understand why some groups are smarter than others. What they found is that the answer has less to do with the inputs than with the process. A group of average thinkers with exceptional dynamics will outperform a group of exceptional thinkers with average dynamics. Accountability, turn-taking, social sensitivity, and structured honesty are not soft skills in this context. They are the architecture of collective intelligence itself.
The question is not whether you are smart enough for the group. It is whether the group is structured well enough to be smarter than any of you.