A crowd produces accurate collective judgment when four conditions hold: diversity of opinion (each person has private information), independence (opinions are not swayed by those around them), decentralization (people draw on local and specialized knowledge), and aggregation (some mechanism turns private judgments into a collective decision). Remove any condition and the crowd's accuracy degrades.
When many independent estimates are averaged, individual errors tend to point in different directions and cancel each other out, leaving the signal of genuine information. This is why the group's aggregate guess is often closer to the truth than almost any single member, including the experts.
At a 1906 country fair, 787 people guessed the weight of an ox after slaughter; Galton averaged the guesses expecting the mob to be far off, but the mean of 1,197 pounds was within one pound of the actual 1,198. No individual expert matched the accuracy of the aggregated crowd.
A group stays smart only when members think for themselves rather than deferring to one another; once people start imitating neighbors, errors correlate instead of canceling and the aggregate drifts toward systematic bias. Information cascades and conformity therefore destroy the very diversity that makes crowds accurate.
Prediction and betting markets work because they let dispersed individuals stake money on their private beliefs, and the resulting price aggregates that scattered information into a single forecast. The Iowa Electronic Markets often predicted election outcomes more accurately than polls because participants had incentives to bet on what they thought would happen, not what they hoped.
To find the sunken USS Scorpion in 1968, John Craven polled a diverse team of experts for independent guesses on what happened, then combined their estimates using Bayesian methods into a collective location prediction. The wreck was found within 220 yards of the aggregated group estimate, though no single expert had picked that spot.