Meeting of the Mind

Brad Gulliford

Book report for SLIS 5712

The wisdom of crowds

James Surowiecki

New York: Doubleday, 2004

Crowds of people, voting independently, arrive at determinations or decisions more accurate than any made by an individual, even an expert individual.

Four conditions that characterize wise crowds:

• Diversity of opinion: each person should have some “private,” i.e., unique, information, even if that “information” is an opinion

• Independence: people’s opinions not determined by other people’s opinions

• Decentralization: people specialize and draw on local knowledge

• Aggregation: some method of turning private judgments into collective decision (example: stock market)

Experts don’t do better than crowds, but people do have to have some information (so adults with no professional background in aerospace could, through a decision market system, identify Morton Thiokol as the real culprit in the Challenger disaster, but a group of children could not). Each person holds a different bit of information that contributes to the rich whole.

Groups of experts don’t do as well because their education and experience tend to be common and make them act and think alike.

Diversity, actually “cognitive diversity” (“people who possess varying degrees of knowledge and insight”) not only increases the number of ideas to be explored, but also creates a climate in which diverging opinions may be expressed. But diversity of intelligence is also important: it is more effective to assemble a diverse group than to spend resources finding experts.

Diversity can generate a large quantity of ideas and products, but the system must be able to rid itself of duds efficiently.

The mechanisms by which dominant design, groupthink, and conformity occur are described. “Social proof” is the name for the idea that if everybody else is doing it, there must be a good reason. People reason that if the group makes a better decision than an individual, the individual should follow the group. The problem is that if too many people do that, the crowd loses its wisdom (diversity and independence).

Economists refer to a phenomenon called “information cascade,” in which a design achieves dominance because people adopt or pass on the idea that a product is good because others are saying it is, not because individuals are reaching independent conclusions—they start imitating others and stop paying attention to their own private knowledge. They don’t do this because they are stupid or sheep-like conformists; they think they are learning from others.

Surowiecki here draws a contrast between his work and Gladwell’s Tipping point. Surowiecki’s information cascades spread among people who may be (most often are) strangers, whereas Gladwell’s mechanisms of transmittal are through social ties, driven by particular roles (mavens, connectors, and salesmen).

Information cascades can be started deliberately by certain people. Surowiecki cites the example of the machinist who popularized standardization of screws; presumably people who conduct political, propaganda, or advertising campaigns are also fomenting (or trying to foment) information cascades.

Many decisionmaking models favored are economic, some even involving participants (such as students in an economics or psychology class) trading real money. (Real money encourages more careful decisions.) Many of the analyses cited are also the work of economists, as opposed to experts in the domain field. This has been carried further in Freakonomics (Levitt et al., NY: Morrow, 2005; authors’ blog at http://www.freakonomics.com/blog/ ). In itself, this raises the question of who can or should evaluate things.

Decentralization allows the capture or use of tacit knowledge (knowledge that cannot easily be summarized or transmitted through communication). Decentralization does not ensure that the knowledge generated in one place will be disseminated anywhere else—but now there’s Google. Google aggregates and disseminates information according to the Surowieckian ideal. The links are not formed deliberately by experts, but the quality yielded by the ranking algorithm is just as high.

Decentralized systems must have an aggregator to reap the advantage of all the independent processes contributing to the whole. In fact, the aggregator is what makes the independent process contribute. Another example of the power of decentralization is Linux, and the aggregator is Linus. Without a central (single) arbiter, the independent processes could not be harnessed to maximum benefit. Yet the Microsoft model has not produced software of greater quality.

Mechanisms of aggregation, or at least working together:

• Coordination (doing things at the same time, not bumping into each other—avoiding rush-hour congestion)

• Cooperation (paying taxes to finance societal benefits; trusting other people)

• Collaboration (building of knowledge in academe)

• Committees, juries, teams (which can polarize or work well together)

Companies should decentralize decisionmaking to ensure market agility and cushion bosses’ bad decisions.

There are three kinds of problems: cognition (for which there is a specific answer), coordination, and cooperation (both of which are solved by approaches, not answers, and have no one right solution). The wisdom of crowds can improve the latter two, but succeeds most in cognition problems.