Fifteen Law Students Outpredict 20-Year Veteran Lawyers, Study Shows

Crowd Wisdom Principles — Hypermind

Hypermind

Aggregated groups systematically outperform solitary experts because cognitive diversity mathematically cancels out individual errors, provided participants deliberate independently rather than seeking consensus.

AI-assisted summary of Crowd Wisdom Principles — original source. Check the original for full context.

Source published:

Key ideas

  • Francis Galton demonstrated crowd wisdom in 1906 when 787 fairgoers estimating an ox's butchered weight produced a median of 548 kg and a mean within 0.5 kg of the actual 543 kg weight. (0:00)
  • Scott Page's Diversity Theorem mathematically proves that collective error equals average individual error minus diversity, meaning that greater disagreement directly enhances group accuracy. (4:44)
  • In a civil trial payout prediction study, an aggregate of 15 Stanford law students surpassed the accuracy of a veteran attorney with 20 years of courtroom experience. (7:04)
  • Harnessing collective intelligence requires independent thinking, cognitive diversity, and objective aggregation mechanisms like averaging, majority voting, or prediction markets. (8:35)

Detailed notes

## Background Published on YouTube by the forecasting platform Hypermind on December 1, 2021, this video serves as an installment of their free "Crowd Forecasting" Masterclass. The presentation outlines the core principles of collective intelligence, dissecting how properly aggregated groups outperform their smartest individual members through diversity and independence.

## TL;DR The video traces the origins of crowd wisdom back to Francis Galton's 1906 country-fair experiment, where the aggregated estimates of a crowd accurately pinpointed the butchered weight of an ox within half a kilogram. It establishes three mathematical laws of crowd scale, explains Scott Page's Diversity Theorem to show how divergent errors cancel each other out, and demonstrates how a small cohort of law students can outperform veteran attorneys. Ultimately, it emphasizes that collective intelligence requires cognitive diversity, independent thinking free from consensus-seeking conformity, and objective aggregation methods like averaging, voting, or prediction markets.

## Atmosphere & style The presentation carries the polished, intellectually measured tone of an executive masterclass, unfolding with clear analytical authority and visual elegance. As it transitions from historical anecdotes to statistical formulas and modern legal experiments, the narrative demystifies collective decision-making with cool, rational optimism. Watching it feels like a structured intellectual briefing that steadily dismantles intuitive biases favoring solitary expertise in favor of statistical rigor and intellectual independence.

## Bulletin - [00:00](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=0s) Francis Galton demonstrated crowd wisdom in 1906 when 787 fairgoers estimating an ox's butchered weight produced a median of 548 kg and a mean within 0.5 kg of the actual 543 kg weight. - [04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) Scott Page's Diversity Theorem mathematically proves that collective error equals average individual error minus diversity, meaning that greater disagreement directly enhances group accuracy. - [07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) In a civil trial payout prediction study, an aggregate of 15 Stanford law students surpassed the accuracy of a veteran attorney with 20 years of courtroom experience. - [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) Harnessing collective intelligence requires independent thinking, cognitive diversity, and objective aggregation mechanisms like averaging, majority voting, or prediction markets.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=0s) Crowd forecasting relies on the empirical observation that an aggregated estimate from a large group is frequently far more accurate than individual guesses. - [00:45](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=45s) Francis Galton analyzed 787 participant tickets in an ox weight-judging competition to test the competence of ordinary citizens. - [02:00](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=120s) The crowd's median estimate was off by just 5 kg, while the mean average landed within half a kilogram of the actual 543 kg butchered weight. - [03:06](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=186s) The first law of crowd wisdom states that the error of the average is consistently smaller than the average error of individual participants. - [03:38](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=218s) The second law establishes that aggregate collective error systematically decreases as the size of the crowd expands. - [03:54](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=234s) The third law identifies diminishing returns to scale: a group of 10 cuts error threefold, 100 people divide it by three again, and roughly 450 are needed for the next equal gain. - [04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) Aggregating judgments works because individual idiosyncratic errors cancel each other out while underlying informational signals accumulate. - [05:30](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=330s) Scott Page formalized this dynamic in the Diversity Theorem: Collective Error = Average Individual Error - Diversity of Estimates. - [06:20](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=380s) Diversity and domain expertise function as interchangeable and complementary variables in producing accurate group judgments. - [07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) In legal damage payout predictions, averaging 14 law students matched a veteran attorney's accuracy, and 15 students outperformed them. - [07:50](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=470s) Combining the independent estimates of just two experienced attorneys reduced error by 25%, a level of accuracy requiring 42 students to match. - [08:10](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=490s) Experienced professionals often fail to benefit from collective wisdom because they undervalue peer input and resist taking advice. - [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) Unstructured crowds typically default to poor judgment due to social pressure, intellectual conformity, and herd mentality. - [08:55](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=535s) Deliberation must be conducted independently, prioritizing the surfacing of divergent views over premature consensus. - [09:30](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=570s) Objective aggregation methods must be matched to the task: voting for governance, averaging for physical metrics, and prediction markets for forecasting future events.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=0s) Francis Galton and the Discovery of Crowd Wisdom

Collective intelligence takes many forms, but one of its most striking expressions appears when a large group of people independently attempts to estimate a numerical value. While individual guesses are often riddled with inaccuracy, their consolidated figure—the crowd’s judgment—frequently lands astonishingly close to the truth. This dynamic underpins crowd forecasting, which applies the collective ability of ordinary groups to anticipate future developments.

The scientific foundation of this phenomenon traces back to 1906, when it was uncovered "in extremis" by the renowned British statistician Francis Galton at the age of 84, only a few years before his death. While attending a country fair, Galton observed hundreds of attendees gathered around an ox for a weight-judging competition. Contestants paid an entry fee to submit a ticket estimating the butchered weight of the animal's meat, competing for a prize. At the time, Britain was engaged in fierce political debates over expanding democratic governance, and Galton—a member of the Victorian elite—harbored deep skepticism about the competence of ordinary citizens. He saw in the contest an opportunity to test whether an unspecialized populace possessed sound collective judgment.

Galton obtained and examined 787 submitted tickets, arranging the guesses in ascending order from approximately 400 kilograms to 700 kilograms. He first evaluated the median—the middle-most guess, where an equal number of estimates sit above and below it. Identifying this point as an authentic statistical representation of the *vox populi*, or the people's voice, Galton found that the median guess stood at 548 kilograms, a mere 5 kilograms off the actual butchered weight of 543 kilograms. When he subsequently calculated the mean average across all submitted tickets, the result proved even more astonishing: the crowd's average guess came within half a kilogram of the actual figure.

As the narrator notes of Galton's personal outlook: > “As a member of the kingdom’s elite, Galton was initially skeptical of the ability of laypeople to make accurate judgments.”

- **Total tickets analyzed:** 787 participant estimates - **Range of estimates:** Approximately 400 kg to 700 kg - **Actual dressed weight of the ox:** 543 kg - **Median crowd estimate (*vox populi*):** 548 kg (error of 5 kg) - **Mean crowd estimate:** Within 0.5 kg of actual weight

---

[03:06](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=186s) The Three Laws of Crowd Wisdom

The performance of Galton’s country-fair crowd is not an anomaly; it demonstrates three reproducible mathematical rules that govern collective estimates. The first principle is that a consolidated group judgment is routinely superior to the individual guesses that compose it. Comparing the average error of an individual selected at random to the error of the aggregated group reveals a stark gap: in Galton’s ox experiment, individual participants were off by an average of 4.5%, whereas the group's aggregate error approached zero. This disparity yields the foundational axiom of crowd wisdom:

> “So this checks out nicely, leading us to the first law of crowd wisdom: «the error of the average is smaller than the average error.»”

The second law dictates that crowd size matters: as more individual judgments are added to an aggregate, the collective error systematically shrinks. However, the third law qualifies this relationship through the principle of diminishing returns. The curve plotting crowd size against error reduction drops sharply with the first few participants and flattens out thereafter. A group of just 10 individuals achieves an error rate three times smaller than that of an average individual. Tripling that accuracy again requires expanding the group from 10 to 100 people, and another equivalent step demands roughly 450 people. The practical consequence is that organizations do not require massive throngs to harness collective intelligence; even modest groups capture the vast majority of the statistical benefit.

- **First Law:** The error of the average is smaller than the average error. - **Second Law:** Larger crowds yield smaller aggregate errors. - **Third Law:** The benefits of crowd scale exhibit diminishing marginal returns. - **Scale efficiency benchmark:** 10 people reduce individual average error by a factor of 3; 100 people divide it by 3 again; roughly 450 people are required for the next comparable reduction.

---

[04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) The Mathematics of Diversity: Scott Page’s Theorem

The ability of a group to outperform its individual members is grounded in basic mathematics rather than mystery. Every individual assessment incorporates a mixture of genuine signal (information) and distortion (error). Because people view a problem from different perspectives, their personal errors are rarely identical. When diverse judgments are averaged, the valid information accumulates while the idiosyncratic, opposing errors cancel each other out. If one fairgoer underestimates an ox's weight by 50 kilograms while another overestimates it by 47 kilograms, their combined average lands just 1.5 kilograms shy of the target.

This dynamic was formalized by American social scientist Scott Page in the "Diversity Theorem." Expressed plainly, the theorem states:

$$\text{Collective Error} = \text{Average Individual Error} - \text{Diversity of Estimates}$$

Because diversity is subtracted from the average individual error, increasing the disagreement among participants directly reduces the error of the collective result. The narrator captures this counterintuitive reality:

> “Because of the minus sign here, the more people disagree, the smarter the group becomes.”

Under this mathematical framework, group intelligence relies equally on personal competence and cognitive variance. Diversity and domain expertise operate as complementary and interchangeable variables. A deficit in deep technical expertise can be compensated for by gathering a wider variety of independent viewpoints. Conversely, if a group consists of like-minded individuals whose thinking aligns, they must possess exceptionally high individual expertise to prevent their shared biases from skewing the final outcome.

---

[07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) Experts vs. Amateurs: The Legal Prediction Experiment

The practical interchangeability of expertise and diversity was demonstrated in a 2011 study conducted by a team of American and Israeli researchers. The experiment tested how accurately participants could predict the financial payouts awarded by juries in United States civil trials. The pool included seasoned attorneys boasting 20 years of courtroom experience alongside novice Stanford University Law School students.

When evaluated on an individual basis, the experienced lawyers vastly outperformed the students, whose average individual errors were twice as large as those of the practicing attorneys. Yet, as student estimates were mathematically averaged together, the diversity of their mistakes rapidly compensated for their lack of professional mileage. An aggregate of 14 law students matched the predictive accuracy of a single 20-year veteran attorney, and a group of 15 students surpassed the single expert.

```mermaid flowchart TD accTitle: Accuracy Comparison of Legal Predictions accDescr: Flowchart illustrating how student groups scale in accuracy relative to single and paired expert attorneys.

subgraph Amateurs["Stanford Law Students"] S1["Individual Student<br/>(2x error of an attorney)"] S14["Group of 14 Students<br/>(Matches 1 Attorney)"] S15["Group of 15 Students<br/>(Outperforms 1 Attorney)"] S42["Group of 42 Students<br/>(Matches 2 Combined Attorneys)"] end

subgraph Experts["Experienced Attorneys (20 yrs exp)"] A1["Single Veteran Attorney"] A2["Combined Pair of 2 Attorneys<br/>(+25% accuracy over single expert)"] end

S14 -.->|"Equivalency"| A1 S15 -.->|"Exceeds"| A1 A1 -->|"Averaging 2 experts"| A2 S42 -.->|"Equivalency"| A2 ```

Experts themselves do not always agree, and combining expert opinions yields substantial predictive dividends: averaging just two attorney forecasts increased predictive accuracy by 25%. To match the combined judgment of two experienced attorneys, the student cohort had to expand to 42 individuals. However, real-world expert panels face a behavioral barrier that prevents them from exploiting this effect:

> “Sadly, experienced attorneys are bad at taking advice: they give less weight to their peer’s estimates and therefore fail to reap the benefits of the crowd’s wisdom.”

- **Amateur baseline:** Individual law students exhibited twice the error of veteran attorneys. - **Equivalence threshold:** 14 students combined matched the accuracy of one expert attorney. - **Surpass threshold:** 15 students outperformed a single attorney. - **Combined expertise:** Averaging 2 attorney forecasts reduced their error by 25%. - **High-volume amateur equivalence:** 42 aggregated students were required to match 2 combined attorneys.

---

[08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) The Recipe for Collective Intelligence

Despite the mathematical power of aggregation, groups in the real world frequently descend into collective foolishness. Most unmanaged crowds default to stupidity rather than wisdom because individuals easily succumb to social influence, herd behavior, and intellectual conformity. For collective intelligence to materialize, a crowd must be deliberately organized according to specific procedural criteria, popularized by author James Surowiecki.

```mermaid flowchart LR accTitle: The Recipe for Collective Intelligence accDescr: Three-stage process showing individual independence feeding diverse opinions into objective aggregation methods.

subgraph Inputs["1. Conditions for Participants"] I["Independent Thinking<br/>(Incentivized, free from peer influence)"] D["Cognitive Diversity<br/>(Surfacing divergent views vs consensus)"] end

subgraph Aggregation["2. Objective Aggregation"] M1["Majority Voting<br/>(Selecting leaders)"] M2["Mathematical Averaging<br/>(Estimating quantities/weights)"] M3["Information Markets / Betting<br/>(Forecasting future events)"] end

subgraph Result["3. Output"] CW["Wise Crowd Judgment<br/>(Minimized Collective Error)"] end

I --> D D --> Aggregation M1 --> CW M2 --> CW M3 --> CW ```

First, participants must be incentivized to deliberate independently. Independence prevents conformity—the true antithesis of intelligence—and ensures that members draw upon their unique, private information instead of mirroring their peers. Rather than pressuring a team to forge consensus, a well-structured organization works actively to uncover and preserve divergent opinions.

Second, the group requires an objective mechanism to aggregate these conflicting inputs into a unified decision without political tampering. The appropriate method depends on the task:

1. **Democratic majority voting:** Suited for governance and selecting leaders. 2. **Mathematical averaging:** Suited for estimating quantitative metrics or physical properties, such as Galton's ox. 3. **Betting and prediction markets:** Suited for forecasting uncertain future events, using financial or reputational wagers to weight and synthesize diverse probability assessments.

As the narrator emphasizes regarding social dynamics within decision-making bodies: > “The enemy of intelligence is conformity, not the crowd itself. A smart group should seek to surface all diverging opinions instead of seeking consensus.”

## Notable quotes > [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) The enemy of intelligence is conformity, not the crowd itself. A smart group should seek to surface all diverging opinions instead of seeking consensus.

> [07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) Sadly, experienced attorneys are bad at taking advice: they give less weight to their peer’s estimates and therefore fail to reap the benefits of the crowd’s wisdom.

## Mentioned Francis Galton, Scott Page, James Surowiecki, Hypermind, Stanford Law School, Crowd Forecasting: How Collective Intelligence Helps Predict The Future

## Video description When properly organized, a crowd can be smarter than even its smartest member. Everyone can be right, even when each person is wrong.

We’ll take a closer look at the discovery of the “wisdom of crowds”, dissect the “diversity theorem”, and give you the recipe for collective intelligence.

- Larger crowds are smarter. - Diversity and expertise are interchangeable and complementary. - Collective intelligence = diversity + independence + aggregation.

This video is part of the free Hypermind Masterclass: “Crowd Forecasting: How Collective Intelligence Helps Predict The Future” - https://www.hypermind.com/master-class

[Source](https://www.youtube.com/watch?v=HNQ3GccCAI0)

中文简报

研究表明：15名法学生预测表现胜过20年资深律师 聚合群体的表现系统性优于单一专家，因为只要参与者独立深思而非寻求共识，认知多样性就会在数学机制上抵消个体误差。 视频追溯了“群体智慧”的渊源，回顾了弗朗西斯·高尔顿（Francis Galton）1906年在乡村集市上进行的实验——人群估值的汇总结果精准预测了一头公牛宰杀后的重量，误差在半公斤以内。视频确立了群体规模的三大数学定律，阐述了斯科特·佩奇（Scott Page）的多样性定理以说明分歧误差如何相互抵消，并展示了一小组法学生如何击败资深律师。视频最后强调，集体智慧需要认知多样性、摆脱迎合共识的独立思考，以及平均法、投票或预测市场等客观聚合方式。 弗朗西斯·高尔顿在1906年验证了群体智慧：787名集市游客估算一头公牛宰杀后的重量，给出的中位数为548公斤，平均值与543公斤的实际重量误差不到0.5公斤。 (0:00) 斯科特·佩奇的多样性定理从数学上证明，集体误差等于个体平均误差减去多样性，这意味着更大的分歧能够直接提升群体的准确度。 (4:44) 在一项民事审判赔偿金预测研究中，15名斯坦福大学法学生的综合预测准确度超过了一位拥有20年出庭经验的资深律师。 (7:04) 激发集体智慧需要独立思考、认知多样性，以及平均法、多数表决或预测市场等客观聚合机制。 (8:35)

Reading summary

Aggregating independent individual estimates reliably outperforms individual judgment because diverse, opposing errors cancel each other out while underlying information accumulates. Drawing on historical experiments and statistical modeling, Hypermind outlines the mathematical principles that govern crowd accuracy and details why collective intelligence depends on diversity and independence rather than expert consensus. The Three Mathematical Laws of Crowd Scale (3:06) Collective estimation relies on consistent mathematical properties first demonstrated statistically by Francis Galton in 1906. Analyzing 787 entries at an ox weight-judging competition, Galton discovered that while individual participants exhibited an average error of 4.5%, the median entry of 548 kg and the mean entry came within 5 kg and 0.5 kg, respectively, of the ox's actual 543 kg butchered weight. This dynamic establishes the first law of crowd wisdom: the error of the aggregate average is invariably smaller than the average error among individual members. The second and third laws govern crowd size and efficiency. While collective error systematically declines as more estimates are included, the process exhibits steep diminishing returns. A group of 10 people reduces individual average error threefold; achieving a second threefold reduction requires expanding the sample to 100 people, and a third comparable step demands approximately 450 people. Consequently, organizations can extract the vast majority of statistical accuracy from relatively modest cohorts without needing massive crowds. The Diversity Theorem and Offsetting Biases (4:44) The mechanics of group intelligence are formalized by Scott Page’s Diversity Theorem, which defines collective error as the average individual error minus the diversity of estimates. Because each person's assessment combines genuine information with unique bias, individual errors pull in opposing directions. When judgments are averaged, these idiosyncratic deviations neutralize one another—such as a 50 kg undercount balancing an opposing 47 kg overcount—leaving the accumulated signal intact. Under this equation, disagreement mathematically improves group performance: as diversity increases, the subtracted term grows, reducing aggregate error. Expertise and cognitive variety operate as interchangeable assets. A shortfall in deep specialized competence can be offset by a broader pool of independent perspectives, whereas a group of like-minded specialists requires exceptionally high competence to prevent shared systemic biases from corrupting the outcome. Substituting Volume and Diversity for Expertise (7:04) The practical trade-off between expertise and viewpoint diversity was illustrated in a 2011 study evaluating forecasts of civil jury awards in the United States. While individual veteran attorneys with 20 years of experience performed with half the error rate of Stanford Law School students, mathematically aggregating student estimates quickly neutralized the expertise gap. An aggregate of 14 law students matched the accuracy of a single veteran attorney, and a group of 15 surpassed the expert. Expertise also benefits from aggregation, though human behavior frequently undermines it. Combining forecasts from just two veteran attorneys reduced error by 25%, a benchmark that required an aggregate of 42 law students to match. However, the source notes that experienced professionals struggle to capture these benefits in practice because they tend to discount peer advice and give inadequate weight to other specialists' assessments. Structuring Aggregation to Prevent Group Conformity (8:35) Unmanaged groups routinely fail to achieve collective intelligence because social pressure, herd behavior, and conformity erode independence. As author James Surowiecki highlighted, wise crowds require procedural safeguards that protect dissenting viewpoints rather than pressing for artificial consensus, ensuring that participants draw strictly on their own private information. Once independent inputs are preserved, they must be combined through an objective aggregation process suited to the task. Democratic majority voting functions for governance and leadership selection, mathematical averaging resolves physical quantities and metrics, and prediction or betting markets synthesize probabilistic forecasts of uncertain future events through financial or reputational stakes. Crowd wisdom relies on the deliberate design of estimation processes: without strict independence and variance among participants, collective deliberation easily degenerates into intellectual conformity and herd behavior.

精炼阅读摘要

群体智慧的数学基础表明，多样化独立判断的聚合结果通常优于个体平均水平，但这一效应的发挥高度依赖于个体的独立思考与客观的汇总机制。 高尔顿实验与群体智慧三大定律 (0:00) 群体智慧的科学验证始于1906年统计学家弗朗西斯·高尔顿（Francis Galton）对集市公牛屠宰重量竞猜的研究。在分析了787张竞猜票后，高尔顿发现，尽管单人估计在400至700公斤之间且个体平均误差达4.5%，但中位数估计为548公斤（误差5公斤），均值与实际重量543公斤的差距甚至不足0.5公斤。这一案例确立了第一定律，即平均值的误差始终小于平均个体误差。 群体规模与误差缩减的关系受后续两条定律制约：第二定律表明群体越大整体误差越小；第三定律则指出这种缩减存在边际收益递减。具体而言，10人团队可将个体平均误差降低至三分之一，达到同等改善需扩充至100人，而下一次同幅度缩减则需约450人。因此，组织无需过大规模即可捕获绝大部分统计红利。 多样性定理与专业经验的互换性 (4:44) 社会科学家斯科特·佩奇（Scott Page）提出的多样性定理将群体误差形式化表达为：集体误差等于平均个体误差减去估计多样性。由于多样性项前为负号，参与者之间的分歧越大，相互对冲的特异性偏差就越多，从而直接减少群体总误差。这意味着领域专业知识与认知多样性在数学上具有可替代性，专业能力的不足可以通过拓宽独立视角来弥补。 2011年关于美国民事审判陪审团赔偿金额预测的研究印证了该原理。单人测试中，20年资历的资深律师表现远优于斯坦福大学法学院学生（后者平均误差为前者的两倍）。然而，通过对学生估值进行数学平均，14名学生的组合即可匹敌一名资深律师，15人便可超越该专家；而要匹配两位经验丰富律师平均预测的准确率（相比单人提升25%），则需要42名学生的聚合结果。 群体智能的成立条件与汇总机制 (8:35) 未经管理的群体极易因从众心理和群体思维陷入低效，甚至走向愚蠢。根据詹姆斯·苏罗维茨基（James Surowiecki）总结的框架，群体智慧必须建立在两个硬性条件之上：首先，参与者必须受到激励进行独立思考，避免受到同伴影响，组织应主动挖掘分歧而非强行寻求共识；其次，必须通过不受政治干扰的客观机制来汇总相互冲突的信息。 具体聚合机制应根据决策任务的性质进行匹配：民主多数决适用于领导人选举与治理事务；数学平均法则适用于物理属性或数量指标的度量估算（如重量竞猜）；而涉及预测未来不确定事件时，则应借助信息市场或预测投注，利用声誉或资金杠杆有效整合概率评估。 经验丰富的专业人士往往不愿听取同行建议，难以通过协作进一步降低误差；而群体若要持续产生明智决策，关键在于打破从众心理、维护认知多样性，并选用适宜的信息聚合工具。

Detailed notes — zh

## Background 本内容由频道 Hypermind 于 2021 年 12 月 1 日发布在 YouTube 平台上，是其免费大师课《群体预测：集体智能如何帮助预测未来》的一部分。视频旨在阐释群体智慧的核心原理，拆解“多样性定理”，并提供构建集体智能的必备法则。 ## TL;DR 视频通过统计学家弗朗西斯·高尔顿在乡间市集的公牛称重实验，引出了群体智慧的三大定律，指出聚合判断往往比随机个体精准得多，且群体规模带来明显的收益递减效应。依据斯科特·佩奇的多样性定理，群体误差可以通过专业素养或意见分歧来抵消，表明多样性与专业技能具有互换与互补性。在对比了斯坦福法学院学生与资深出庭律师的判决预测后，主讲人阐明普通外行依靠多样性同样可以匹敌顶尖专家。最终，视频总结了实现真正集体智能的三大基石：保持独立思考、积极发掘分歧以维持多样性，以及根据具体预测目标采用投票、数学均值或预测市场等客观聚合机制。 ## Atmosphere & style 整期视频呈现出严谨理性而又通俗易懂的学术科普基调，讲解条理清晰、层层递进。主讲人以沉稳自信的叙事风格，将枯燥的统计学公式与生动的历史实验及现实案例巧妙结合，没有生硬的说教，而是带着一种循序渐进启发思考的知性力量。观看时既能感受到严谨数学逻辑带来的释然与震撼，也能体会到对“乌合之众”与盲目从众心理一针见血的警醒与反思。 ## Bulletin - [00:00](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=0s) 弗朗西斯·高尔顿在 1906 年的公牛称重实验中发现，787 人的平均估算与实际重量仅相差 0.5 kg，验证了集体判断能大幅消除个体误差。 - [03:06](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=186s) 群体智慧遵循边际收益递减规律：10 人群体可将单人误差削减至三分之一，而进一步削减到九分之一则需将规模扩大至 100 人。 - [04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) 斯科特·佩奇提出的多样性定理指出“集体误差 = 平均个体误差 - 估算多样性”，表明观点的分歧程度与专业技能具有可互换性与互补性。 - [07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) 在法庭赔偿预测实验中，15 名非专业学生的聚合表现超越了拥有 20 年经验的资深律师，而仅需 2 名律师意见结合即可使预测准确度提升 25%。 - [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) 构建集体智能必须依赖“多样性 + 独立思考 + 客观汇总”的机制设计，群体的真正威胁是从众心理而非分歧本身。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=0s) 弗朗西斯·高尔顿通过集市公牛称重竞猜发现，群体中位数（548 kg）与平均值极度逼近公牛真实重量（543 kg），开启了群体智慧研究。 - [03:06](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=186s) 群体智慧第一法则强调“平均值的误差小于平均误差”，随机个体的预期误差约为 4.5%，而聚合后的平均误差几乎为 0%。 - [03:06](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=186s) 群体规模越大预测越精准，但收益递减规律明显，绝大部分智慧红利在规模较小的早期阶段就已被释放。 - [04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) 群体智慧的底层数学原理在于不同个体的有效信息会相互叠加，而方向相反的误差则会在平均化过程中相互抵消。 - [04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) 多样性定理用数学公式证明了群体分歧越大整体表现越聪明，观点的多样性与个体的专业知识可以相互弥补与替代。 - [04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) 若团队成员想法高度同质化，则成员必须是几乎不会犯错的顶尖学者，否则意见趋同会导致严重盲区。 - [07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) 斯坦福学生与资深律师的对比实验表明，通过聚合 14 至 15 名非专家的多样化观点，完全可以匹敌甚至超越单名资深专家的预测水准。 - [07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) 聚合 2 名资深律师的意见能将准确度提升 25%，这需要 42 名法学院学生聚合才能达到同等水准，说明专家间组合具备极高价值。 - [07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) 资深专家往往由于过度自信而不愿吸纳同行建议，导致他们无法自然享受到群体智慧带来的红利。 - [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) 未经恰当组织的群体容易沦为盲目的乌合之众，集体智能的诞生必须依托制度与机制设计。 - [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) 从众心理是集体智能的最大敌人，一个聪明的群体应当致力于激发出所有相异的观点，而非过早追求共识。 - [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) 群体智慧依赖于客观汇总机制：选举依赖多数投票，物理估值依赖数学平均，而复杂未来预测则需借助预测市场等押注机制。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=0s) 群体智慧的发现：高尔顿与乡间市集的公牛 群体智慧（Wisdom of Crowds）描述的是一种集体智能现象：当许多人对某件事物做出猜测时，把所有人的猜测整合在一起得出的群体猜测，往往比任何单独个人的猜测都要精准得多。群测预测（Crowd forecasting）正是利用了群体在预测未来方面的这种强大能力。这项研究的历史可以追溯到 1906 年，由英国知名统计学家弗朗西斯·高尔顿（Francis Galton）在生命的晚期发现——当时高尔顿已经 84 岁高龄，并在发表这一发现仅几年后便与世长辞。 当时高尔顿正在参观一个乡间市集，现场聚集了数百人，他们围着一头公牛参加有奖竞猜比赛。比赛规则非常直接：参赛者各自支付一小笔报名费，在一张票券上写下自己对这头公牛被宰杀后所能产出的肉质重量的估算值，估得最准的人就能赢得丰厚的牛肉大奖。高尔顿对此产生了浓厚的兴趣，因为当时的英国正在激烈争论推行更民主的治理模式，普通公众到底有没有做判断的能力成为了一个火热的议题。作为英国精英阶层的一员，高尔顿起初深信外行与普通平民缺乏做出准确判断的能力。为了从科学角度检验这一问题，他在比赛结束后设法收回了参赛者提交的所有票券。 高尔顿将收集到的 787 个估算数据按照从低到高的顺序排列，最低的估算值约为 400 公斤，最高的约为 700 公斤。他首先找到了正中间的数值，也就是中位数（Median）。中位数具有非常独特的性质，因为它是唯一一个不会被大多数参与者认为“过高”或“过低”的估计值——刚好有一半的人猜得比它高，一半的人猜得比它低。高尔顿因此将这个中位数称为“民意”（vox populi），即群体作为一个整体的选择。 - 收集的有效估算样本总量：787 个 - 估算范围：约 400 kg 至 700 kg - 群体中位数估算值：548 kg - 公牛实际出肉重量：543 kg（中位数误差仅为 5 kg） - 群体平均值（Mean）估算结果：与公牛实际重量仅相差 0.5 kg --- ### [03:06](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=186s) 群体智慧的三大定律 高尔顿的公牛实验完美揭示了群体智慧运作的三个核心定律。第一条定律指出，集体判断普遍比绝大多数个体判断更为聪明。如果要检验这一点，只需对比群体中随机挑选一名个体的平均误差与整个群体的平均误差。在公牛称重实验中，随机挑选一个人的预期平均误差约为 4.5%，而整个群体的平均值误差基本为 0%。这引出了群体智慧的第一法则：“平均值的误差小于平均误差（the error of the average is smaller than the average error）”。 第二条定律表明群体的规模至关重要（Crowd size matters）——群体规模越大，集体判断就越聪明。当在平均值中纳入越来越多人的估算时，整体误差会明显持续下降。第三条定律则体现在群体智慧曲线的形态上：它表现出明显的收益递减规律（Diminishing returns）。随着群体规模的扩大，误差在最初阶段呈断崖式急速下跌，随后下降速度越来越慢。这意味着集体智能的大部分收益在初期就已经实现，组织并不需要极其庞大的人群就能获得极高的智慧。 ```mermaid flowchart LR accTitle: 群体规模与误差递减规律 accDescr: 展示从单个人到群体后误差成倍缩减的衰减趋势 subgraph ScaleEffects ["群体规模与误差缩减"] direction LR Ind["单人随机误差<br/>基准误差 (100%)"] G10["10 人群体<br/>误差降为单人的 1/3"] G100["100 人群体<br/>误差再次除以 3"] G450["450 人群体<br/>误差进一步收敛至极低水平"] Ind -->|"汇聚初始样本"| G10 G10 -->|"规模扩大 10 倍"| G100 G100 -->|"规模继续扩大"| G450 end ``` - 随机个体的平均误差约为 4.5%，而群体聚合后的误差基本为 0% - 10 个人的群体即可将平均个体的误差缩减至三分之一 - 要使误差在此基础上再缩小至三分之一，需要 100 个人，随后需要 450 个人 --- ### [04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) 群体智慧的数学原理与多样性定理 群体智慧并非某种不可思议的魔法，而是严谨的数学规律。核心机理在于：每个人的判断都同时包含“有效信息”与“误差”。在现实世界中，每个人的认知缺陷和错误方向往往是不一样的。当这些独立判断被汇总到一起时，正确的有效信息会相互叠加积累，而方向相反的误差则会在平均计算中相互抵消。例如，某人可能低估了公牛 50 公斤，另一个人高估了 47 公斤，两人判断结合后产生的误差仅仅只有 1.5 公斤，瞬间逼近真相。 美国社会学家斯科特·佩奇（Scott Page）总结了支撑群体智慧的数学基础，将其命名为“多样性定理”（The Diversity Theorem）。该定理的公式形式看似复杂，其揭示的内涵却非常简练深刻：集体误差等于个体平均误差减去估算的多样性。换句话说，群体的失误可以通过更小的个人错误来缩减，也可以通过让不同的人犯不同的错误来缩减。 ``` 集体误差 (Collective Error) = 平均个体误差 (Average Individual Error) - 估算多样性 (Diversity) ``` 公式中的负号意味着一个关键结论：人群之间的分歧越大，整合后的群体就会变得越聪明。这就是为什么意见的多样性对于集体智能至关重要。一个团队的智慧既依赖于成员的专业能力，也同等程度依赖于他们观点的差异。主讲人指出：“专业技能与多样性是互补且可互换的（Expertise and diversity are complementary and interchangeable）。”如果群体成员不是专业人士，可以通过寻求更高维度的观点多样性来弥补；而如果一个团队的所有人想法高度一致，那他们最好是造诣极深、几乎不会犯错的顶尖学者。 --- ### [07:04](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=424s) 专家与外行的博弈：多元学生与专业律师 为了验证专业技能与多样性的互换性，主讲人引用了一项由美国和以色列研究人员于 2011 年进行的法庭审判赔偿预测实验。该实验要求参与者预测美国真实陪审团裁定的赔偿金额。参与者被分为两类：一类是拥有 20 年执业经验的资深出庭律师，另一类是斯坦福大学法学院的学生——这群学生无疑非常聪明，但缺乏一线从业专长。 实验数据显示，从个体层面来看，斯坦福学生的平均预测误差是资深专业律师的两倍，这说明专业素养在个人层面上优势极其显著。然而，只要将多名学生的预测结果结合起来，群体中观点的多样性便能迅速弥补专业知识的不足。数据显示，一个 14 名学生组成的小组，其预测准确度就能匹敌一名资深律师；如果是 15 名学生，其集体智力甚至略微超越了资深律师。 不仅外行可以结合，专家意见同样可以组合。资深律师之间常常存在意见分歧，将仅仅 2 位律师的观点结合在一起，预测准确度就能直接提升 25%。此时，需要 42 名斯坦福学生组成的大群体才能追平这两名律师组合的准确度。然而，现实中存在一个残酷的人性障碍：“遗憾的是，经验丰富的律师很不擅长听取建议：他们赋予同行评估的权重极低，因而未能享受到群体智慧带来的红利（Sadly, experienced attorneys are bad at taking advice: they give less weight to their peer’s estimates and therefore fail to reap the benefits of the crowd’s wisdom）。” | 评估主体 | 预测表现 / 特点 | |:--- |:--- | | **单名斯坦福法学院学生** | 平均个体误差是资深律师的 2 倍 | | **14 名学生的小组聚合** | 依靠观点多样性抵消错误，表现等同于 1 名资深律师 | | **15 名学生的小组聚合** | 预测准确度略胜于 1 名资深律师 | | **单名资深律师（20年经验）** | 个人专业性极强，但受制于个体认知偏见 | | **2 名资深律师的聚合** | 准确度比单人提升 25%，需要 42 名学生组合才能匹敌 | | **资深专家的心理弱点** | 傲慢且不善采纳同行建议，无法自然获取集体智慧 | --- ### [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) 集体智能的配方：避免群体盲思的三大要素 尽管群体的智慧十分惊人，但现实生活中“乌合之众”引发的集体愚蠢（Collective stupidity）同样随处可见。主讲人直言不讳地指出，几乎每个人都亲眼见识过非常愚蠢的群体，甚至聪明的个体一旦置身于无组织的群体中也会变蠢。事实上，大多数自然形成的乌合之众确实生来就是盲目的。要让群体变得聪明，核心在于必须以恰当的方式组织它。主讲人借鉴美国记者詹姆斯·苏罗威茨基（James Surowiecki）的经典名著，给出了打造集体智能的三大配方。 首先是独立思考（Independence）。每个人都必须有动力去独立观察与思考。独立性是观点多样性繁荣的土壤，它鼓励个人毫无顾忌地表达自己独一无二的视角。主讲人强调，群体真正的敌人不是群体本身，而是从众心理（Conformity）：“一个聪明的群体应该致力于发掘所有分歧观点，而不是去追求共识（A smart group should seek to surface all diverging opinions instead of seeking consensus）。” 最后一步是客观汇总（Objective Aggregation）。在允许充分分歧之后，群体需要一套客观的机制来汇总这些相异的观点并作出决策。采用何种聚合机制完全取决于任务的目标： - 如果群体的目标是推选领导人，那么“多数人投票表决”（Majority vote）是最适宜的手段。 - 如果群体的目标是估算一头公牛的重量或物理参数，那么对所有数据取“数学平均值”（Mathematical average）即可完美解决。 - 如果群体的任务是预测不可知、复杂的未来，“预测市场（让人们对不同结果进行押注）”则是整合分散预测的最强大工具。这就是群体预测的精髓所在：借助群体智慧的科学机制去预判未来。 ## Notable quotes > [08:35](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=515s) 一个聪明的群体应该致力于发掘所有分歧观点，而不是去追求共识 — 主讲人 > [04:44](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=284s) 专业技能与多样性是互补且可互换的 — 主讲人 > [03:06](https://www.youtube.com/watch?v=HNQ3GccCAI0&t=186s) 平均值的误差小于平均误差 — 主讲人 ## Mentioned 弗朗西斯·高尔顿 (Francis Galton), 斯科特·佩奇 (Scott Page), 詹姆斯·苏罗威茨基 (James Surowiecki), 斯坦福大学法学院 (Stanford Law School), Hypermind, 多样性定理 (The Diversity Theorem) ## Video description When properly organized, a crowd can be smarter than even its smartest member. Everyone can be right, even when each person is wrong. We’ll take a closer look at the discovery of the “wisdom of crowds”, dissect the “diversity theorem”, and give you the recipe for collective intelligence. - Larger crowds are smarter. - Diversity and expertise are interchangeable and complementary. - Collective intelligence = diversity + independence + aggregation. This video is part of the free Hypermind Masterclass: “Crowd Forecasting: How Collective Intelligence Helps Predict The Future” - https://www.hypermind.com/master-class [Source](https://www.youtube.com/watch?v=HNQ3GccCAI0)

Source quotes

[8:35] 主讲人: 一个聪明的群体应该致力于发掘所有分歧观点，而不是去追求共识 [4:44] 主讲人: 专业技能与多样性是互补且可互换的 [3:06] 主讲人: 平均值的误差小于平均误差

Source quotes

[8:35] The enemy of intelligence is conformity, not the crowd itself. A smart group should seek to surface all diverging opinions instead of seeking consensus. [7:04] Sadly, experienced attorneys are bad at taking advice: they give less weight to their peer’s estimates and therefore fail to reap the benefits of the crowd’s wisdom.

Original transcript

[0:00] Crowd wisdom, with Hypermind. [0:03] In this lesson, you will learn about  one type of collective intelligence [0:07] that emerges when many people attempt to guess  something. In a nutshell, the idea is that the [0:12] consolidated guess of many, the crowd’s guess, is  often much more accurate than individual guesses. [0:20] This phenomenon is known as “the wisdom of  crowds”. Crowd forecasting simply leverages [0:26] this stunning ability of crowds to make  reliable collective guesses about the future. [0:31] We’ll go into forecasting in the next lessons,  but first, let’s learn about the wisdom of crowds. [0:37] The phenomenon of the wisdom of  crowds was discovered in extremis [0:41] in 1906 by the British scientist Francis Galton,  one of the foremost statisticians of all time. [0:48] I say “in extremis”, because Galton achieved the  ripe old age of 84 years old! – [0:53] and died just a few years after publishing this last discovery. [0:59] Galton was visiting a country fair when he came across a crowd of several hundred people [1:04] whose attention was focused on an ox. [1:07] They were participating in a contest to try to  guess the weight of the animal, or to be precise, [1:13] to guess the weight of the meat that would come  from butchering it. [1:17] Each participant paid a small entry fee and wrote their best estimate on a ticket. [1:23] The most accurate of these estimates would then win the contest, along with its meaty prize. [1:28] Why would Galton find this interesting? [1:31] Well, he realized that he could use these estimates to  find out whether a large number of non-specialists [1:37] could make a smart collective judgment. At the  time, the United Kingdom was debating the idea [1:42] of a more democratic form of government,  making the wisdom of crowds a heated topic. [1:48] As a member of the kingdom’s elite, [1:50] Galton was initially skeptical of the ability of laypeople to make accurate judgments. [1:55] To investigate the matter, Galton arranged to retrieve all the tickets after the contest. [2:01] He arranged the 787 estimates in ascending order from the lowest, at around  400 kg, to the highest, at around 700 kg. [2:12] He then identified the middle-most estimate, called  the median. [2:16] The median is very special [2:18] because it is the only estimate that isn’t considered as too low or too high by a majority of the participants, [2:25] meaning there are exactly the same number of  estimates above as there are below the median. [2:32] This brought Galton to declare the median to be a  measure of the vox populi, the people’s voice. In [2:39] other words, the choice of the crowd as a whole. This median estimate, the crowd’s estimate, [2:45] was 548 kg, just 5kg away from  the actual ox’s weight of 543 kg. [2:54] Galton later computed the average of the estimates  – another way to consolidate the crowd’s guess – [3:00] and it turned out to be even more accurate,  a mere half a kg away from the actual weight. [3:06] The 3 principles of crowd wisdom The wisdom of crowds obeys 3 laws, which Galton’s [3:12] example illustrates perfectly. The first is that  collective judgment is generally smarter than most [3:19] individual judgments. We can test this claim  by comparing the average error of an individual [3:24] within the crowd, to the error of the full crowd  average. In this particular case, the average [3:32] individual error – that is, how much error you’d  expect picking someone at random – was about 4.5%, [3:40] while the error of the crowd as a whole was  essentially 0%. So this checks out nicely, leading [3:46] us to the first law of crowd wisdom : «the error  of the average is smaller than the average error.» [3:53] The second law is that crowd size matters –  larger crowds are smarter. Look at how the [4:00] error of the crowd diminishes as we include more  and more individual estimates in the average. [4:05] The third law is visible in the shape of the  crowd-wisdom curve. It exhibits diminishing [4:12] returns. As you increase the size of the crowd,  the error falls sharply at first, and then more [4:18] and more slowly. A group of 10 has an error that  is 3 times smaller than the average individual’s. [4:25] But it takes 100 people to divide the error by 3  again. And then a crowd of 450... The implication [4:32] is that most of the collective-intelligence  benefit is obtained at the start: even small [4:38] crowds are a lot smarter than individuals, and  you don’t need a huge crowd to be a lot wiser. [4:44] The maths behind crowd wisdom Ok, so far we have described [4:48] the wisdom of crowds phenomenon, but we  haven’t truly explained how it works. [4:52] By what magic can a smart collective judgment  emerge from the combination of many flawed [4:59] individual judgments? Of course, it’s not  magic... it’s mathematics. The central idea [5:05] is that everyone’s judgment contains both  information and error. In many cases, [5:11] every judgment is flawed in a different way.  So when they are combined, information adds up, [5:17] while errors cancel each other out. For example, in the case of Galton’s ox, [5:23] someone may have underestimated its weight by 50  kg, while someone else could have overestimated [5:29] it by 47 kg. When the two erroneous  judgments are combined through averaging, [5:35] the result is just 1.5 kg below the real  weight. An almost perfectly accurate judgment. [5:42] The diversity theorem The mathematical formula that powers the wisdom of [5:46] crowds is called “the Diversity Theorem”. It was  discovered by the American sociologist Scott Page. [5:54] It looks complicated, but what it says is very  simple, yet profound: collective error equals [6:01] average individual error, minus the diversity  of the estimates. In other words, the size of a [6:08] groups’ mistake is reduced when individual errors  are smaller – that’s obvious – but it is also [6:15] reduced when different people make different  mistakes. That’s the diversity component. [6:21] Because of the minus sign here, the more people  disagree, the smarter the group becomes. That’s [6:27] why a diversity of opinions, which implies a  diversity of errors that can cancel each other [6:33] out, is so important to collective intelligence. So you can read this formula as simply that a [6:39] group’s collective intelligence depends as much  on individual expertise as on diverse opinions. [6:47] Expertise and diversity are complementary and  interchangeable. If most people aren’t experts, [6:54] you can compensate by seeking  a higher diversity of opinions. [6:59] Or if everyone thinks alike, then  they’ better be very knowledgeable. [7:04] Expert vs Amateurs Here’s an example from an experiment that [7:08] was run back in 2011 by a group of American and  Israeli researchers. It looked at people’s ability [7:15] to predict the results of trials in the United  States, specifically the size of jury awards. [7:24] Some participants were experienced attorneys  with 20 years of professional experience, [7:29] while others were students at Stanford University  Law School, very bright no doubt, but with no [7:34] particular expertise. As you can see, the average  mistake of a student was twice as large as the [7:43] average mistake of an attorney. Expertise makes  a huge difference! Yet, when the estimates of [7:49] several students are combined, the diversity  of opinions in the group makes up for what [7:54] it lacks in individual expertise. The students’  collective error diminishes so that a group of 14 [8:02] is just as smart as an expert. And a group  of 15 students is slightly smarter still... [8:08] But experts can also disagree, and they often  do, so their opinions can also be combined for [8:13] more accurate judgments. Combining just 2  expert opinions increases accuracy by 25%. [8:20] And now it takes 42 students to match  the performance of two attorneys. [8:25] Sadly, experienced attorneys are bad at  taking advice : they give less weight to [8:30] their peer’s estimates and therefore fail  to reap the benefits of the crowd’s wisdom. [8:35] The collective intelligence recipe All this talk about wise crowds is all [8:40] well and good, but haven’t we all encountered  deeply stupid groups? Has each and every one [8:46] of us not personally witnessed that even smart  people – perhaps even ourselves – can become [8:51] stupid when they become part of a crowd? Of  course, collective stupidity also exists. [8:58] There are smart crowds and stupid crowds,  and most crowds are indeed naturally stupid. [9:04] To make a crowd smart, you need to organize it  properly. The recipe for collective intelligence, [9:10] adapted from the famous book by American  journalist James Surowiecki, tells us how: [9:15] First of all, everyone must be incentivized to  think, and to do so independently. Independence [9:23] allows for a diversity of opinions to flourish,  by allowing and encouraging individuals to express [9:29] their unique perspectives. The enemy of  intelligence is conformity, not the crowd [9:34] itself. A smart group should seek to surface all  diverging opinions instead of seeking consensus. [9:41] Then, a final step ensures the group can  make decisions despite the disagreements: [9:46] objective aggregation. The most relevant aggregation mechanism [9:50] depends on what a given crowd is trying to  do. If it is tasked with choosing a leader, [9:56] a majority vote is a good solution. If  it is asked to guess the weight of an ox, [10:01] a mathematical average of all estimates will  do fine. If it is trying to predict the future, [10:07] letting people place bets on outcomes is a  powerful way to consolidate divergent forecasts. [10:13] In essence, this is what crowd forecasting is  about: using crowd wisdom to predict the future.