Trust but verify

Trust but verify is a rule for collective judgment: trust allows people and institutions to make use of one another’s knowledge, while verification keeps confidence from becoming credulity. The book develops the principle across markets, corporations, intelligence agencies, and science, showing that collective intelligence depends not only on gathering information but also on whether people can test, share, and act on it.

Trust, Independence, and Aggregation

A crowd can reach a reliable judgment when its members bring diverse information, form opinions independently, specialize through decentralization, and have a mechanism for aggregation. Under those conditions, individual errors can cancel when estimates are combined, leaving information in the collective result. But error cancellation alone is not enough: the group must possess relevant information in the first place. The principle therefore implies both confidence in aggregation and attention to the quality of what is being aggregated.

The book’s account of betting markets makes that distinction concrete. A bookmaker adjusts a point spread when wagers arrive heavily on one side; the changing line reflects bettors’ collective judgment rather than a forecast fixed once and for all by the bookmaker. The final line is not infallible, and spreads only weakly predict the scores of most NFL games. Yet gamblers find it difficult to beat consistently, and roughly three-quarters of the Mirage’s final lines are described as the most reliable forecasts available for NFL outcomes. The market is useful not because it never errs, but because its judgments are hard to outperform and its errors are not obviously systematic.

The Iowa Electronic Markets offer another test. Across forty-nine elections from 1988 to 2000, election-eve predictions were off by an average of 1.37 percent in presidential elections, 3.43 percent in other U.S. elections, and 2.12 percent in foreign elections. In comparisons with 596 polls released during presidential elections in that period, the market price was more accurate three-quarters of the time. Its forecasts also changed less wildly than polls, tending to move sharply in response to new information. The market had only about eight hundred traders and did not mirror the electorate: most traders were men, and a disproportionate, though shrinking, share came from Iowa. Its performance thus illustrates the difference between asking people to predict their own behavior and aggregating judgments about what a broader electorate will do.

Verification and Institutional Accountability

The principle becomes especially pointed in the book’s account of corporate gatekeepers. Wall Street firms, stock exchanges, and auditors are supposed to distinguish trustworthy companies from untrustworthy ones. Their credibility is an asset: in theory, self-interest should lead them to do careful work, because clients would abandon institutions whose certifications proved worthless. But that incentive fails if dereliction brings no penalty.

During the late 1990s, it did not. From 1997 through 2000, seven hundred companies restated their earnings, compared with just three in 1981. The Nasdaq listed companies the book calls laughable; prominent underwriters backed them, and auditors approved their accounts. Arthur Andersen audited accounting disasters including Waste Management and Sunbeam, yet investors did not treat companies such as WorldCom and Enron with suspicion simply because they continued to use Andersen. The institutions meant to verify corporate claims received more business rather than less. Trust without effective scrutiny had become a system in which investors stopped watching the watchmen—and the watchmen stopped watching, too. The book’s rule is therefore not a call to distrust every certification; it is a warning that trust must remain conditional on accountability.

Local Knowledge and the Problem of Sharing

Verification also depends on how organizations distribute information. Decentralization can encourage independence and specialization: people close to a problem may possess tacit knowledge tied to a particular place, job, or experience. Yet local knowledge does not automatically become useful to the whole. A system can contain valuable fragments while failing to connect them.

The U.S. intelligence agencies’ failure to anticipate major terrorist attacks between 1993 and 2001 is presented as a warning about this weakness. Congress emphasized a lack of information sharing: agencies produced localized snapshots instead of a coherent picture of threats. Senator Richard Shelby argued that the FBI’s decentralized structure had fragmented information holdings into largely independent fiefdoms, leaving important information out of the hands of those who needed it. The proposed remedy was central coordination, with an agency standing above the disputatious bureaucracies. This case complicates any simple preference for decentralization: local independence may generate knowledge, but without a means of sharing and combining it, the collective cannot see what its parts already know.

The book frames the institutional challenge as finding a balance. Individuals need room to specialize and acquire local knowledge, while a mechanism must make that knowledge globally useful without stripping it of its specificity. Verification, in this sense, is not merely checking a finished claim; it includes creating routes through which independent findings can meet and be evaluated.

Scientific Collaboration and Collective Trust

The global SARS investigation illustrates how that balance can work. After reports from Guangdong Province and subsequent cases in Hanoi and Hong Kong, the World Health Organization convened eleven laboratories from countries including France, Germany, Japan, the United States, China, and others. Beginning on March 17, 2003, the labs held daily teleconferences, posted electron-microscope photographs and test results, and exchanged virus samples. They could investigate the same material in parallel, compare findings, and rule out possible causes together.

On March 21, scientists at Hong Kong University had isolated a likely candidate, while the U.S. Centers for Disease Control separately isolated a virus that appeared to be a coronavirus. Labs then detected the coronavirus in samples from many SARS patients, sequenced it, and tested its effects; in early April, infected monkeys in a Netherlands laboratory developed full-blown SARS. On April 16, only a month after the collaboration began, the labs announced that the coronavirus caused the disease. Cynthia Goldsmith first spotted the coronavirus, but the evidence did not make her, or any one laboratory, the sole discoverer. The finding depended on shared samples, parallel work, and the elimination of competing explanations; the network collectively established what the virus did.

The collaboration had no director dictating each laboratory’s work or the exchange of information. The WHO organized the network, but the laboratories themselves agreed to share relevant data and meet every morning; beyond that, they worked out how to divide the investigation. The example gives verification an operational form: open exchange lets independent specialists scrutinize and extend one another’s results, while coordinated communication makes their local discoveries part of a common account.

Scientific knowledge more broadly rests on this kind of trust. Researchers publish results and data so others can inspect them, challenge conclusions, or use the findings to build new hypotheses. Yet most scientists do not replicate most experiments. They often rely on work they have not personally repeated, and a failed attempt to reproduce a familiar result may initially lead a researcher to question their own skill rather than the established experiment. Verification is real and replication matters, but science could not advance if every researcher spent all their time retracing others’ work.

That dependence makes trust necessary, not optional. Experiments can be replicated and fraud can be exposed, but the circulation of knowledge requires scientists to rely on one another’s data and competence while remaining open to scrutiny. Scientific acceptance is consequently collective: the coronavirus caused SARS before the community announced the result, but its standing as an established scientific finding depended on other researchers examining and accepting the evidence. The book’s version of trust but verify is neither blind faith nor universal suspicion. It is confidence sustained by communication, review, and the possibility of correction.