When most experts in a field agree, that agreement is itself a piece of evidence for anyone outside the field. But not all agreement is equal. A thousand people repeating one rumor are not a thousand witnesses. This level looks at when a consensus deserves weight, how it can be tested from the outside and how it can be wrong.
Why many heads can beat one
In 1785 the French mathematician and reformer the Marquis de Condorcet proved a result now called the jury theorem. If each member of a group is more likely right than wrong, and each judges independently, the chance that the majority is right rises toward certainty as the group grows. Eleven independent judges who are each right 60 percent of the time will, as a majority, be right about three times in four. A group of 101 such judges will be wrong as a majority only about one time in fifty.
The theorem has a dark mirror. If each member is more likely wrong than right, the majority becomes almost certain to be wrong as the group grows. Numbers multiply whatever competence or incompetence is already there. Agreement among people who have no special access to the truth tells you little.
Independence matters
The second condition, independence, is easy to forget. The philosopher Alvin Goldman pointed out that if a guru's thousand followers simply repeat whatever the guru says, their agreement is really one opinion with an echo. Agreement adds weight only when the people agreeing reached their views with some independence, through their own access to the evidence. This is why a claim that circulates through many websites copying one source counts as one claim, and why scientists value replication by separate teams.
When consensus rests on knowledge
The philosopher Boaz Miller asked, in 2013, how an outsider can tell a consensus based on knowledge from one based on habit, pressure or shared interest. He proposed three signs.
The first is consilience: several independent lines of evidence, gathered by different methods, point to the same conclusion. The word comes from the nineteenth century philosopher William Whewell, who called it a "jumping together" of inductions. When a conclusion is supported by lab work, field data, history and theory at once, a single flawed method is unlikely to explain it.
The second is social calibration: the experts share standards of evidence and use their key terms in the same way, so that their agreement is really about the same claim. Polls of "scientists" who mean different things by a phrase can manufacture agreement that is not there.
The third is social diversity: the agreeing experts differ in background, funding, nationality and interests, so that a shared bias is an unlikely explanation for their agreement. A consensus held only by researchers paid by one industry, or only by one country's officials, deserves more suspicion than one shared by rivals.
When consensus is wrong
Consensus has been wrong, and the stories are worth knowing. In 1982 the Israeli materials scientist Dan Shechtman saw a pattern in an aluminum alloy that, according to textbook crystallography, could not exist: an orderly structure that never exactly repeated. His colleagues mocked the finding, he was asked to leave his research group, and Linus Pauling, the most famous chemist alive, dismissed him publicly for years. Shechtman's quasicrystals were confirmed by other laboratories, and he won the Nobel Prize in Chemistry in 2011.
The lesson is not that dissenters are usually right. For each Shechtman there are many confident dissenters who were simply mistaken. The lesson is how the consensus changed. Shechtman won because he produced evidence others could check and reproduce, not because he was certain or persecuted.
Core and frontier
One more distinction helps outsiders. Consensus is strongest at the core of a field, on matters tested for decades, and weakest at its research frontier, where new results arrive daily and experts are expected to disagree. That chemistry obeys the periodic table is core. What the best diet is for preventing a particular disease may be frontier. A news story about "scientists disagreeing" usually describes the frontier, which says nothing against the core.
Worked example: HIV and AIDS in South Africa
By the late 1990s the evidence that the HIV virus causes AIDS was consilient: virologists had isolated it, epidemiologists had traced infections through blood transfusions and sexual networks, and new drug combinations that attacked the virus were turning AIDS from a death sentence into a manageable illness. The researchers who agreed came from dozens of countries, from public universities and private companies, from rich nations and poor ones. In 2000 more than 5,000 scientists and doctors signed the Durban Declaration restating this consensus.
South Africa's president, Thabo Mbeki, nevertheless gave a hearing to a small group of dissenters who denied that HIV causes AIDS, and his health minister promoted garlic, lemon and beetroot while the government delayed antiretroviral drugs. Mbeki's suspicion had understandable roots, including a long history of colonial medicine treating Africans as objects of study and drug companies charging prices poor countries could not pay. But the dissenters were few, mostly outside virology and without any successful predictions. A 2008 study from Harvard researchers estimated that more than 330,000 lives were lost because treatment programs were delayed. Checking Miller's three signs would have shown which side the evidence stood on.
