The Funeral Principle

Max Planck, who founded quantum theory and spent his life inside the institutions of German science, wrote near the end of it that “a new scientific truth does not triumph by convincing its opponents.” They do not come round, he observed. They die, and a generation grows up for whom the new idea was never controversial.

The line is usually quoted at conferences, and it usually gets a laugh. Everybody in the room can think of somebody it applies to. Then the session moves on, and nobody treats it as what it is, which is a testable empirical claim about how knowledge actually propagates through a profession.

It has been tested. The result is not funny.

Four hundred and fifty two deaths

The study that did it comes from Pierre Azoulay, Christian Fons-Rosen and Joshua Graff Zivin, and the design is grim and elegant. They identified four hundred and fifty two elite life scientists who died prematurely while still active in their fields. Not retirements, not slow declines. Sudden exits, which is what makes the design work, because an unexpected death is close to a natural experiment. It removes a person from a field without any of the gradual withdrawal that would otherwise muddy the picture.

Then they looked at what happened to those fields afterward.

Two things did. The scientist’s close collaborators published meaningfully less, which is unsurprising and mostly reflects the loss of a productive partner. But the second finding is the one that matters. Researchers who had never worked with the deceased, outsiders to that particular circle, began publishing in the area at a higher rate. The increase averaged 8.6 percent, and it did not fade.

And their papers were different. They drew on different prior work. They were disproportionately likely to be highly cited. The authors were disproportionately people who had not previously been active in that subfield at all.

The paper’s own conclusion is blunter than anything I would write. Once in control of the commanding heights of a field, star scientists tend to hold their position somewhat too long.

The ideas had been out there the whole time. What changed was that the position blocking them became vacant.

I want to be careful about the accusation embedded in this, because the obvious reading is uncharitable and I do not think it is correct. The finding is not that eminent scientists are villains suppressing rivals. Some are, but that is not what the data shows and it is not necessary for the effect.

The mechanism is much more ordinary, which is what makes it so hard to fix.

How a field gets held

Consider what an eminent researcher actually controls, not by scheming, but by holding the job.

They sit on the panels that award grants. They referee for the journals that matter, and they edit some of them. They advise on hiring and tenure. They train the graduate students who become the next cohort, and those students learn the field’s questions from them. They are asked to write the review articles that define what the open problems are. When a science journalist wants a quote, they get the call.

Now suppose this person sincerely believes, on the basis of forty years of hard work and genuine expertise, that a particular line of inquiry is a dead end.

They are not going to suppress it. They will simply rate that grant application slightly lower, because they honestly think it is less promising. They will recommend rejection of that paper, because they honestly see the flaw in it. They will advise a promising student against that topic, because they honestly want the student to have a career. Every one of those judgments is made in good faith. Every one of them is exactly what expertise is supposed to be for.

And the aggregate of those good faith judgments is a field that cannot turn.

This is why the effect is so robust and so difficult to legislate against. There is no bad actor to remove. There is only the ordinary operation of authority, which is that people with track records get to decide what counts, and the track record was built in the old paradigm.

The history is full of it, and three cases are worth having in mind because they fail in three different ways.

Ignaz Semmelweis worked out in the 1840s that doctors were killing mothers by walking from the autopsy room to the delivery room without washing their hands. In 1847 he ordered handwashing in chlorinated lime at the Vienna General Hospital and the death rate in his ward fell from around eighteen percent to about two.

That is not a subtle statistical finding. That is roughly eight out of every nine deaths, gone, in twelve months. The profession rejected it anyway, and part of the reason is that his result was not merely a hypothesis but an accusation. It said that the physicians hearing it had personally been killing their patients, for years, with their own hands. There is no version of that claim that a room of eminent men can weigh dispassionately. Semmelweis was committed to an asylum in 1865, was beaten by the guards there, and died of an infected wound.

Alfred Wegener presented his case that the continents move to the German Geological Society in January 1912. He was rejected for roughly half a century, and here the mechanism of rejection was different and considerably more defensible. He could not explain what force would push a continent through solid ocean floor, and his critics were right that he could not. He was also a meteorologist rather than a geologist, which meant the people qualified to judge his claim were the people whose field he had wandered into. Continental drift became consensus in the 1960s, once seafloor spreading supplied the missing mechanism. Wegener died on a Greenland expedition in 1930, three decades before he was vindicated.

Both things were true at once in his case. The objection was legitimate and the objectors were also incumbents, and nobody can now cleanly separate how much of that fifty year delay was proper scientific caution and how much was territory.

That is the uncomfortable part. In the Semmelweis case we can say cleanly that the field was wrong. In the Wegener case we cannot, and most real cases look like Wegener rather than Semmelweis.

Barry Marshall is the third kind. He could not persuade anyone that stomach ulcers were caused by bacteria rather than by stress and acid, so in 1984 he drank a culture of the organism and gave himself gastritis. He shared the Nobel Prize in 2005.

Marshall found a way around the problem that did not require waiting for anybody to die. It required an act so dramatic that it could not be refereed away, could not be declined by a grant panel, and could not be politely ignored at a conference. That worked. It is also not a mechanism. You cannot build a scientific establishment on the assumption that people with correct unpopular ideas will poison themselves to get a hearing, and the ones who are unwilling to are not thereby wrong.

The honest counterweight

I should state the cost of turnover directly, because a chapter that only lists the benefits of people dying is not an honest chapter.

When a great scientist dies, an enormous amount is lost that was never written down. Judgment about which problems are worth attacking. Knowledge of which apparently promising avenues were tried in 1987 and quietly failed. The tacit craft of doing the work, which passes from person to person and does not survive in papers. Fields do not only get stuck by keeping their elders. They also get stupid by losing them, and rediscover things at great expense that somebody already knew.

So the claim here is not that death is good for science. It is not. The claim is narrower and it is the same claim as every chapter in this section.

Death is currently performing a function, badly and at enormous cost, and we have no other mechanism that performs it at all.

Notice also that our existing institutions had a partial substitute and then removed it. Mandatory retirement for tenured faculty in the United States ended on the first of January, 1994, and academic careers stopped having a defined endpoint. I should say that the consequences are genuinely disputed. The higher education establishment predicted that senior faculty would stay indefinitely and strangle junior hiring, and at least one careful study concluded beforehand that the effects would be far smaller than feared. Retirement rates at seventy and seventy one did fall sharply afterward, which is not in question. Whether that translated into the broader damage that was predicted is still argued about.

I raise it not as proof but as the only real experiment we have. One profession removed one turnover mechanism, deliberately, and we are still arguing about what it did. That is a poor evidentiary position from which to remove all five of them.

The compounding problem

Now put this chapter next to the previous one, because the interaction is worse than either alone.

Chapter 4 argued that capital would stop turning over. This chapter argues that intellectual authority would stop turning over. In practice these are not two separate populations. The people who hold capital and the people who hold intellectual authority overlap heavily and increasingly. Endowments fund research. Philanthropic foundations set agendas. Private laboratories dominate certain fields outright, and the person who founded the lab decides what it works on.

Freeze both at once and you do not get two static systems side by side. You get one static system with mutual reinforcement. The capital funds the ideas that the authorities endorse. The authorities are the people whose prestige was built with that capital. Neither has any mechanism for revision, because the revision mechanism in both cases was the same event.

There is a third layer arriving, and it deserves flagging here even though it belongs properly to Part III.

Machine learning systems are trained on the accumulated written output of a field. That corpus is produced disproportionately by the people with the most standing, because standing is what gets you published, cited and quoted. If those people never leave, they never stop producing, and their share of the corpus never declines. The models trained on it will reproduce the consensus of a generation that is not being replaced, and they will do so with fluency and at volume, and their output will in turn become part of the corpus.

Every previous generation’s intellectual authority had a hard stop. Yours ended when you did, and even the greatest reputations decayed as the people who remembered you personally also went. What we are building is the first apparatus for holding a position in a debate indefinitely, at scale, without the holder needing to be present.

Combine that with a holder who is present anyway.

What death was doing here

The job is straightforward to state and impossible to replace with anything we currently have.

Death vacates the chairs. It empties the grant panels, the editorships, the department heads, the named professorships and the foundation boards, on a schedule that nobody controls and no amount of eminence can defer. It does this without any judgment about whether the occupant deserved to go, which is precisely the property that makes it work, because any mechanism that required a judgment would be captured by the people being judged.

And it does one more thing that is easy to miss. It supplies the young with a credible expectation that the positions will eventually be available. That expectation is what makes it rational for a talented person to enter a field at all, to spend a decade in training for a chance at a chair. Remove the expectation and you have not merely blocked the ladder. You have removed the reason to start climbing.

Which is the subject of Chapter 7. But there is a chapter in between, because before we get to what happens when nobody leaves their post, there is a stranger question about what happens when nobody is willing to take a risk.