Nobody Read the Warning

This is an argument from a statistical result and a set of historical cases. It is not a measurement. The statistical result is the design effect [1]. The cases are the collapse of the eastern Mediterranean around 1177 BC, the Babel account in Genesis 11 [2], the long survival of Egypt and Imperial China, and the Talmudic rule on unanimous verdicts [3].

The design effect says that the worth of a group's answers is the number of members thinking independently, not the number of members. A thousand people who read the same paper, went to the same school and use the same three websites supply about eleven data points. Pushed to full correlation, a million perfectly coordinated minds have roughly the decision-making power of one.

Around 1177 BC the eastern Mediterranean fell over, not one kingdom but all of them more or less at once: Mycenae, the Hittite empire, the Levantine trade cities, and Egypt reduced to a shadow of itself. Writing systems vanished. Linear B stopped being written and could not be read for three thousand years. Trade networks that had run for centuries stopped, the palaces burned, and a period followed that historians call the Dark Age.

The people who came out the other side were overwhelmingly the ones who had never mattered: hill villages, herders, and populations too marginal for the palace economies to have integrated. The great houses died with their records, and the marginal populations inherited the ground and then wrote things down.

The pattern generalises: the most sophisticated, most connected and most efficiently organised system fails first, and what survives is whatever was too scattered, rural or unimportant to have been plugged in. In each case the survivors wrote an account of what happened.

The text of Genesis 11 reads as a statement about capability and names its source precisely. Close to the direct quotation, the people is one, and they have all one language, and now nothing will be restrained from them. The cause given is unity: one people, one language, everybody pulling the same direction with no friction between them.

The response is specific. The tower is not destroyed and no one is killed. Their language is confused and they are scattered. That is the minimum intervention that reduces a population's capability without harming a person: break its ability to coordinate and spread it out.

An observer watched a unified civilisation become catastrophically capable and then come apart, and recorded that the unity was what made it dangerous and that afterwards it was scattered and speaking different languages. The account files this under pride, the vocabulary available, but the observation is preserved unedited and says that coupling was the cause.

Progress, mechanically, is shared standards, shared infrastructure, faster communication and everyone reading the same thing at the same time. Getting better at civilisation is the process of becoming more correlated, and the same thing that makes a society powerful reduces its effective size. The empire with the best roads is the empire whose provinces make the same mistake in the same week; the hill villages survive because nothing connected them.

Egypt lasted three thousand years and Imperial China was large, standardised and unified, and both survived. The reason is that until about a hundred and fifty years ago nobody could coordinate anything. An order from the capital took months to reach the frontier, arrived garbled, and was carried out by a governor who was in practice making it up. Ancient empires were centralised on paper and independent in practice, every province ran its own experiment and was wrong in its own local way, and that unplanned diversity kept the effective number of independent units above one. Distance was the immune system. Modern communication removed distance and with it the one thing that kept the effective number above one without cost.

The survivors wrote accurate accounts, because they were present. They were by definition peripheral to whatever failed, so they saw the event and could not see the mechanism, which is a statistical property of a system they were never inside. They recorded the event correctly and explained it as pride, wickedness or forgetting the old ways. The wrong explanation is the container the correct observation was shipped in.

A text about the end of the world written by the only people who lived through it is made sacred. Sacred texts are recited word for word, in unison, for centuries, with essentially no loss of the words and essentially no comprehension, because reciting something is the opposite of reading it. A congregation asked to say a passage reproduces it exactly. Asked what it means, it returns the explanation from the container and not the content. Veneration is both the best preservation method and the best concealment method, using one mechanism.

In the Talmud, a capital case before the Sanhedrin needed a majority to convict, but a unanimous vote of all seventy-one judges meant the accused walked free. The strongest possible evidence is thrown out for being too strong. That is the design effect applied to a jury: seventy-one judges from one tradition, in one room, hearing one presentation, are already heavily correlated, and total unanimity signals that correlation has reached its ceiling. There are not seventy-one opinions but one opinion held seventy-one times, and a man is not hanged on one opinion. A result from 1965 appears in a law book from the second century and is usable as written.

A likely reason this one survived intact is its size. A court that convicted an innocent man is a disaster small enough to leave a procedure behind and not a scripture. Catastrophic collapses leave stories, and stories get sung.

Unanimity among correlated members is a warning and not a confirmation, as it is in the Sanhedrin rule.

The prediction for a society routing a growing share of its thinking through a shrinking set of shared platforms, rankings and models is that individuals become sharper while the population's effective size falls, because mistakes once scattered and cancelling are made once at the source and delivered to everyone at the same time. The symptom is not stupidity. It is that disagreement stops carrying information: two people arguing used to triangulate something because they had been wrong in different ways from different places, and when their inputs share an upstream the disagreement is decorative.

The remedy that follows is to keep some people, methods and sources deliberately unplugged. They are not better; they are wrong differently, and that difference is the only thing that has caught an error the whole system was making together. Better tools alone do not supply it.

The statistics are standard and can be looked up: Leslie Kish, design effect, 1965. The history is also established. The link between the two is a hunch and not a finding. No effective size is measured for any ancient or modern society here, and the figure of about eleven independent data points from a thousand correlated people is an illustration and not a computed value.

Our own civilisation has not completed its trial, so no comparison against the finished cases can be made. The prediction above is recorded so that it can be checked.

Several historical statements carry no citation in this article: the 1177 BC collapse and its extent, Linear B going unread for three thousand years, the speed of communication before the last hundred and fifty years, and the Egyptian and Chinese durations.