# Wisdom of Crowds **The model:** Under specific conditions, the aggregate of many independent estimates is more accurate than any individual estimate — even than expert estimates. The canonical demonstration: Francis Galton's 1907 county-fair experiment, where 787 fairgoers guessed an ox's weight. The mean guess (1,197 lbs) was within one pound of the actual weight (1,198 lbs). No single guesser was that close. The collective estimate beat the experts. The *Diversity Prediction Theorem* (Page, 2007) gives the formal result: collective error = average individual error − diversity of predictions. The math is exact — *the more diverse the individual estimates, the lower the collective error*. Diversity isn't merely nice; it's mathematically necessary for crowd wisdom to outperform individuals. But the conditions for the model to work are strict, and modern information environments routinely violate them. ## The Mechanism **Independence is the load-bearing assumption.** Each estimator must arrive at their estimate independently. When estimators copy from each other, the "many estimates" become effectively one estimate, and the wisdom collapses. This is why Surowiecki spent more time on independence than on any other condition. **Diversity contributes mathematically.** Different estimators bring different information, biases, and reasoning paths. Their errors cancel out in aggregation only if they're not all making the same mistake. Homogeneous crowds (same training, same information sources, same priors) don't show wisdom — they show coordinated wrongness. **Decentralization helps.** When estimators have different vantage points and different local information, the aggregation pulls in genuinely different signal. Centralized aggregation can lose the local-knowledge advantage. **Aggregation method matters.** Mean is good for many-points-around-the-truth distributions; median is better for skewed distributions; weighted aggregation (Brier-scored, market-priced, etc.) is better when reliability varies. Choosing the wrong aggregation method can destroy the wisdom. ## When It Fails Real-world conditions frequently violate the assumptions. The failure modes are well-cataloged: **Information cascades.** Once estimators see each other's guesses, they update toward consensus and stop providing independent information. The crowd becomes a mob. **Polarization.** When estimators share group identities, errors correlate. Partisan media audiences are not independent estimators — their estimates of political-fact questions cluster strongly by partisan affiliation, which destroys the diversity that produces wisdom. **Strategic answering.** When estimators have stakes in the answer, they don't report their honest estimate — they report what serves their interests. Surveys of partisan voters about political facts are systematically biased by this effect. **Filter bubbles and homogeneity.** Algorithmically curated information environments reduce diversity by design. The "wisdom" of an algorithmically homogenized crowd is the worst kind of crowd outcome — the appearance of consensus combined with the reality of correlated error. ## KB Applications ### Why Polling Has Worked Less Well Recently Polling aggregates many individual estimates, expecting wisdom-of-crowds dynamics to produce accurate measurement. The recent decline in polling accuracy (2016, 2020, 2022, 2024) tracks the violation of independence and diversity assumptions: respondents have become more partisan-clustered in their views, less willing to participate in surveys (skewing samples), and more likely to misreport when they do (especially Republican respondents declining to respond, producing what's now called "shy Trump voter" effects). Polls aren't broken — the underlying preference distributions are now more homogeneous within partisan groups, which destroys the diversity that polling-based aggregation requires. ### Markets as Crowd-Wisdom Mechanisms Financial markets are wisdom-of-crowds aggregators when working well: many independent estimators with money on the line produce prices that incorporate distributed information. They fail when independence breaks down (bubble dynamics), when diversity collapses (everyone trading on the same algorithm), or when strategic incentives diverge from honest estimation (insider trading, market manipulation). The 2008 financial crisis was a case where the market's wisdom failed because its aggregation conditions failed. ### Why "Trust the Crowd" Fails Under Polarization The current information environment is structurally hostile to crowd wisdom. Partisan media reduces independence (audiences hear correlated information); identity politics reduces diversity (group affiliation predicts factual beliefs more than the underlying evidence does); attention markets reward strategic answering (engagement-bait beats accuracy). The result: contemporary "crowds" are bad estimators not because individual people got dumber but because the structure that aggregates their estimates broke. ### The Right's Curated Information Environment as Anti-Wisdom Architecture The right-wing media ecosystem is structurally engineered to violate every condition for crowd wisdom: shared sources (low independence), in-group orthodoxy enforcement (low diversity), strong identity stakes (strategic answering), and filter-bubble amplification (homogeneity). The result is high internal coherence and low external accuracy — exactly what the model predicts when its assumptions are violated. Recognizing this is more analytically useful than calling the audience irrational; the audience is operating in a system designed to produce wrong answers from collectively-rational individuals. ### Restoring Conditions for Wisdom Where it's possible to engineer the conditions for crowd wisdom — prediction markets, structured forecasting (like Tetlock's Good Judgment Project), Wikipedia's revision-history mechanism, citizen-assembly processes — the wisdom returns. These are formal corrections to information-environment failures, not appeals to rationality. The lesson is structural: design for diversity and independence, and the aggregate intelligence appears. ## Practicing the Model 1. **Don't trust the crowd until you've checked the conditions.** Independent? Diverse? Decentralized? Honest reporting? If not, expect coordinated wrongness, not aggregate wisdom. 2. **Engineer the conditions when you can.** Anonymous estimates before discussion; structured aggregation; rotating panels; explicit diversity in input. 3. **Watch for cascade failures.** When estimates start clustering after exposure to others' estimates, the wisdom is collapsing. 4. **Use formal aggregation methods.** Prediction markets, Brier-scored forecaster ensembles, and weighted polling aggregators outperform unweighted averages because they correct for known violations. ## See Also - [[Bounded Rationality]] — the cognitive constraint individual estimators face; aggregation can sometimes overcome it - [[Median Voter Theorem]] — democracy's wisdom-of-crowds claim; collapses under the same conditions - [[Prisoners Dilemma]] — the cooperation problem that crowd wisdom can sometimes solve - False Consensus Effect — what wisdom-failures often look like from inside the crowd - [[Philip Tetlock]] — the empirical research program on engineered crowd wisdom (Good Judgment Project)