Reading passage
Groups can solve problems that defeat every individual in them, and they can also blunder in ways no member would accept alone. What separates a wise committee from a foolish one is not obvious, and for most of history the answer was left to proverb and prejudice. Over the past century, however, psychologists, economists and computer scientists have turned collective intelligence into a measurable property, one that depends less on who is in the group than on how the group is organized.
The classic founding story comes from Francis Galton, who in 1907 analysed hundreds of guesses in an ox-weighing competition at a country fair. Individual estimates, made mostly by non-experts, were scattered and often poor, yet the median guess fell within about one percent of the animal's true weight. The episode seemed to vindicate a mathematical intuition later formalised in the Condorcet jury theorem: when each member is even slightly better than chance and votes independently, errors cancel and the group converges on the truth. Crowds, however, also stampede, and the same century supplied ample evidence of groupthink, the pressure toward consensus that Irving Janis blamed for famous policy disasters.
A decisive step came in 2010, when Anita Woolley and her colleagues reported a general collective intelligence factor, which they called the c factor. Working with dozens of small groups on tasks ranging from brainstorming to moral judgment, the team found that a single statistical factor predicted how well a group performed across very different assignments, much as general intelligence does for individuals. Crucially, the c factor was only weakly related to the average intelligence of group members and not at all to the intelligence of the smartest member. Something else was doing the work.
Three predictors emerged. Groups scored higher when members were more socially sensitive, as measured by a test of reading emotions from photographs of eyes; when conversational turns were distributed more equally, so that a few voices did not dominate; and when the group contained more women, an effect largely explained by women's higher average scores on the social sensitivity measure. Later studies replicated the c factor in online groups that never met face to face, suggesting that collective intelligence is a property of interaction structure rather than of physical proximity.
Why does interaction matter so much? One mechanism is the transactive memory system, the shared knowledge of who knows what, which lets a group allocate problems to the member best placed to solve them. Another is cognitive diversity: the economist Lu Hong and the political scientist Scott Page proved formally that, under stated conditions, a diverse set of competent problem solvers can outperform a set of individually stronger but similar ones, because different perspectives cover more of the solution space. Diversity, though, only helps when members actually share and combine their distinctive information, which equal participation makes more likely.
Engineers have built institutions around these principles. Prediction markets, in which participants trade contracts tied to future events, aggregate scattered information into prices that often beat expert forecasts. The Delphi method collects anonymous expert judgments in successive rounds to reduce dominance effects. More recently, swarm intelligence platforms let networked groups converge on answers in real time, and hybrid ensembles pair human forecasters with algorithms that correct their biases. Philip Tetlock's research on superforecasters showed that teams of accurate individuals, trained to update frequently, predicted geopolitical events better than intelligence analysts with access to classified data.
The darker side is equally systematic. When people observe one another's estimates before forming their own, judgments converge and confidence rises even as accuracy falls, a process known as an information cascade. Experiments by Jan Lorenz and colleagues showed that mild social influence narrowed the diversity of estimates in a crowd and eroded its collective accuracy, even though every participant became more confident. Echo chambers and herding are not exceptions to collective intelligence but its failure modes, arising whenever the conditions of independence and diversity are violated.
The emerging picture is that collective intelligence is neither magic nor myth but a designable property. Groups become smarter when they recruit diverse and socially skilled members, structure communication to preserve independence before aggregation, and match the weight of opinions to demonstrated accuracy. As organizations and democracies confront problems of unprecedented complexity, the science of how minds combine may prove as consequential as the study of any individual mind.