Separation of powers

Decentralization is the distribution of knowledge and decision-making among people positioned close to particular problems. In The Wisdom of Crowds, it is neither a cure-all nor a rejection of coordination: it creates independence, specialization, and local knowledge, but works well only when the wider system can bring those dispersed judgments together. The decisive organizational challenge is to connect local intelligence without assuming that it must first be concentrated in a central authority.

Definition and Strengths

Decentralization makes room for knowledge that is specific to a place, job, or experience and difficult to summarize for someone else. Friedrich Hayek described this as tacit knowledge. Because people closest to a problem may understand its circumstances better than distant decision-makers, assigning decisions locally can draw on information that formal channels or managerial expertise overlook.

The book also presents decentralization as a way to combine specialization with independent judgment. In a wise crowd, people bring different information and interpretations; they are not simply repeating what those around them think. Decentralized participants can develop those perspectives in their own settings, while an aggregating mechanism turns individual judgments into a collective result. Markets offer one such mechanism: the final point spread on an NFL game, for example, reflects the changing bets of many participants and is usually a difficult forecast for an individual gambler to beat. Decentralization supplies diverse, locally grounded judgments; aggregation allows their errors to offset one another and their information to become visible.

The strengths are inseparable from a limitation. Information found in one part of a decentralized system may never reach another part, so the total knowledge available across the system can exceed what it can use collectively. The aim is not to erase local specificity by collecting every judgment at a center. It is to let people specialize and acquire local knowledge while creating reliable ways for that knowledge to circulate and contribute to shared decisions.

Intelligence and Scientific Collaboration

The book uses the failures of the U.S. intelligence community before September 11 to distinguish decentralization from fragmentation. Dividing intelligence work among agencies could support specialization and a range of analytic perspectives. The failure was that the community lacked a way to aggregate judgments as well as information: there was no effective mechanism for drawing on the collective knowledge of the National Security Agency, CIA, and FBI. A proposal to create a truly central intelligence agency might have made it easier to place one agency in charge of all information, but would also have risked sacrificing the local knowledge, diversity, and independence that division among agencies could provide. The proposed alternative is to share information and aggregate analysis without treating centralization as the necessary solution; even linking agency databases could improve information flow while preserving autonomy.

The global search for the cause of SARS demonstrates decentralization coupled with active exchange. After receiving reports of a severe respiratory disease in Guangdong Province and later outbreaks in Hanoi and Hong Kong, the World Health Organization organized eleven laboratories into a collaborative research project. The labs held daily teleconferences, posted images and analyses online, and traded virus samples. They could investigate the same samples simultaneously, compare results, and rule out possible causes as well as pursue promising candidates. Scientists at Hong Kong University and the Centers for Disease Control separately isolated likely viruses; subsequent work across the network established that a coronavirus caused SARS. The process took weeks, not the months or years that any single laboratory might have needed working alone.

The discovery therefore cannot be assigned to one scientist simply because one person first spotted the coronavirus. Cynthia Goldsmith, an electron microscopist at the Centers for Disease Control and Prevention lab in Atlanta, identified it under the electron microscope, but demonstrating that it caused SARS required weeks of shared work, including tests that excluded other candidates. The World Health Organization orchestrated the network, but did not dictate how each laboratory divided its work. Its achievement depended on decentralized initiative and agreed channels for collaboration, not on a central authority directing each investigation.

Scientific practice extends this collective arrangement beyond explicit research partnerships. Scientists publish findings and make data available so that others can scrutinize results, challenge conclusions, and build new experiments. The resulting body of knowledge depends on predecessors and contemporaries alike, including researchers whose unsuccessful hypotheses help show where not to look. Recognition also rests with the community: a claim becomes an accepted scientific contribution as other scientists examine and credit it, not merely because its original investigators announce it.

This system involves trust as well as scrutiny. Most scientists do not personally replicate most experiments; they rely on other researchers’ data and on the reliability of instruments, materials, and procedures they have not independently tested. Replication and the exposure of fraud remain possible, but science could not advance if every researcher had to retrace every result before using it. Decentralized inquiry therefore relies on open communication and a community’s provisional judgment about which findings deserve to enter common knowledge.

Limits of Centralized and Hierarchical Decisions

The book’s account of NASA’s decision-making before the Columbia disaster shows how formal hierarchy can prevent decentralized expertise from affecting a decision. Engineers on the Debris Assessment Team had serious concerns about the foam strike and wanted images of the Orbiter’s wing to make an informed assessment. Their concerns did not receive a serious hearing from the Mission Management Team. In the January 24 meeting, Linda Ham summarized the analysis without its supporting discussion; no member of the team asked a question or requested the study. A meeting format in which Ham asked and others answered, combined with no process requiring members to contribute their views, narrowed the available analysis. The team also began with its conclusion, interpreting subsequent information to fit it. The result was not effective coordination of distributed knowledge but a hierarchy in which relevant local expertise could not change the group’s judgment.

A similar problem appears in the book’s picture of large American corporations. At Ford, more than fifteen managerial layers separated the chairman from a factory-floor supervisor; Toyota had five. At a General Motors plant in Van Nuys, two workers installed a bracket for two years without being told that it held the car’s sunshade. When a supervisor showed them that a missed bracket meant ripping out carpet and welding the part into place, one worker was astonished to learn its purpose. The episode shows how a narrow assignment and an obstructed flow of information can separate workers from knowledge relevant to their own work.

Hierarchical bureaucracy could also make the path for a new idea extraordinarily long: a chart reproduced in In Search of Excellence showed 223 possible routes through one unnamed company’s bureaucracy before a product idea could be accepted. With so many layers between executives and workers in the field, management could not be confident that its picture of the company reflected conditions on the ground. Advancement incentives compounded the problem. A study of young executives found that greater anxiety about moving up the ladder was associated with less accurate communication of problem-related information; studies also linked upward mobility with withholding news of fights and budget problems. In such a system, decentralization fails not only when decisions are concentrated, but when people have reason to keep local knowledge from moving upward or sideways.

Organizational Design and Balance

The book argues that decisions about local problems should, as far as possible, be made by those close to them. Companies should not assume that every problem must travel up a hierarchy and every solution return down it; practical knowledge gained through experience can make local employees better placed than managers to devise workable responses. Employee involvement, however, remains unusual: fewer than 2 percent of American companies were estimated to make real use of high-performance work systems. In contrast, research on leading companies found that employees and managers were empowered to make more independent decisions and encouraged to seek improvements in their operations.

Decentralization can also affect how people perform by giving them control over their immediate conditions. In one experiment, two groups worked on puzzles and proofreading while loud, random noises interrupted them. Only one group was given a button that could turn off the sound; no one pressed it. Nevertheless, that group solved five times as many puzzles and made fewer proofreading errors. The relevant difference was the availability of control, not the button’s use. Local authority can therefore matter even when it does not produce a visible intervention: responsibility for one’s own environment can improve engagement and performance.

The book sets this organizational argument against Hayek’s account of markets. Hayek emphasized local knowledge and the ability of decentralized social processes to produce outcomes without top-down direction, but was skeptical that dispersed knowledge could be condensed into one collective judgment. The book accepts the importance of decentralization while arguing that the difficulties of aggregation should not obscure its possible benefits. Pari-mutuel betting supplies the contrast: a central authority takes one side of every bet and publishes common odds, yet the resulting odds can give an uncannily accurate picture of the future. For the book, this illustrates that aggregation need not require a conventional hierarchy commanding each participant. Decentralization is most useful not when every part remains isolated, and not when one center replaces independent judgment, but when distributed knowledge can be combined without being stripped of its local sources.