ASHFALL INSTITUTE | SUBDUCTION ZONE
Building Bridges — Episode Ten
NOBODY IS PAID TO REMEMBER
P. A. Moore
Ashfall Institute | Subduction Zone
Written under the collaboration disclosed in Time Is a Wheel and We Are the Ground. Concept, argument and judgment: P. A. Moore. Research and composition: Claude Opus 5.
Everyone is arguing about whether the machines will take your job.
I am not going to argue about it, because I think it is the wrong question and because it has produced two years of shouting without moving anybody an inch.
I want to ask a different one. What does a machine do that no human institution has ever managed to do?
And I think the answer is one word, and it is not think.
Remember.
Let me start with why institutions forget, because it is not stupidity and it is not carelessness. It is structural, and once you see the structure you cannot stop seeing it.
Nobody in any organization is paid to remember. Everybody is paid to deliver.
People rotate. The engineer who ran the pilot program takes a job in another state. The county commissioner serves two terms. The administration changes and every political appointee above a certain level is replaced within weeks. The woman who knew exactly why the 1988 version failed — who was in the room, who saw the water come in, who wrote the memo nobody read — retires in 2004 and dies in 2019.
The memo is still in the file. Nobody knows the file exists.
So somebody proposes the same thing in 2026, with enthusiasm and a slide deck, and it fails the same way, for the same reason, and the failure is recorded in a new memo, which goes into a new file.
That is not a hypothetical. That is the ordinary condition of public administration in this country, and it costs a fortune.
I want to give you the most expensive example I know, because it is not a memo. It is seven people.
In 1986 the space shuttle Challenger came apart. The investigation found an engineering cause — the O-rings — and underneath it an organizational one. The sociologist Diane Vaughan gave it a name that has since spread far beyond NASA: normalization of deviance. A thing goes wrong slightly, and nothing bad happens, so the slightly-wrong becomes the new normal. Then it goes a little further wrong. Repeat until the accident.
NASA understood this. It was studied, published, taught, and built into training.
Seventeen years later, Columbia broke up on reentry.
And the accident board that investigated it did something unusual. It examined the organization as carefully as the vehicle, and it concluded that the organizational causes of the second accident were substantially the same as the first. Schedule pressure. Dissenting engineers who could not get their concerns heard at the level where decisions were made. Anomalies that recurred often enough to stop counting as anomalies.
The same institution. The same lesson. Learned at the cost of seven lives, written down, taught — and gone within one career.
Here is a smaller one that I find almost funny, and it makes the mechanism very clear.
The F-1 engine powered the first stage of the Saturn V — the rocket that went to the Moon. It is the most powerful single-chamber liquid engine anybody has ever flown.
Around a decade ago, NASA engineers wanted to understand it again, with a view to building something similar.
You will sometimes hear that the blueprints were lost. That is not true. The drawings exist, on microfilm, in an archive.
What was gone was everything the drawings did not contain. Which welder’s technique worked on that alloy. Why a particular tolerance had been changed by hand in 1966. Which supplier’s part behaved differently and what they did about it. The thousand pieces of knowledge that live in the hands and heads of the people who actually built the thing.
So the engineers did the only thing available. They took an F-1 out of a museum, disassembled it, scanned it, and worked backward.
We had the documents. We had lost the knowledge.
And that distinction is the whole of this episode, so let me put it plainly.
The problem is not storage. Everything is written down somewhere. Agencies produce evaluations by the truckload; consultants write reports; every failed program has a file.
The problem is retrieval at the moment of decision.
A document that exists but is not surfaced when somebody proposes the same idea is functionally identical to a document that was never written. The 1974 report and no 1974 report produce the same meeting.
An archive is a place where things are kept. A memory is something that speaks up.
Now — the reason I think this is solvable is that one industry already solved it, on purpose, and the results are the safest complex system human beings have ever operated.
In 1976 the Federal Aviation Administration and NASA set up something called the Aviation Safety Reporting System.
Any pilot, controller, mechanic or flight attendant can file a report describing something that went wrong — including something they themselves did wrong. A missed altitude. A misread checklist. A near-miss nobody else noticed.
It has been running for fifty years. The database holds more than a million reports.
And it works because of two design decisions that I would like every legislature in this country to look at.
The first is immunity. The FAA extends limited protection to someone who self-reports a safety event that did not result in an accident. You are not confessing in order to be punished. You are confessing in order to be counted.
The second is the one I think is genuinely brilliant, and it is the answer to the problem I raised in the last episode.
The system is run by NASA, not by the FAA.
The agency that can suspend your license does not hold your report. The body that keeps the memory has no power to punish and no stake in the answer.
In fifty years and more than a million reports, no reporter’s identity has ever been revealed.
Look at what that produces.
A pilot in Denver makes a particular error on a particular approach. A different pilot in Atlanta makes the same one. Neither knows about the other. But the pattern shows up in the database, and it gets published in a monthly bulletin that pilots actually read, and the approach plate gets redrawn.
Nobody had to die for that. Nobody had to be fired. Nobody had to be brave.
That is institutional memory, deliberately engineered, in an industry that used to kill people regularly and now, in commercial operation, almost never does.
And here is the comparison I want to make carefully, because the numbers get thrown around loosely and I would rather understate it.
Medicine has no equivalent of this. Not for most of its history and not fully now. There is no confidential, protected, national, cross-institutional system where a nurse can report the near-miss she caught at two in the morning and have it aggregate with four hundred others into a redesigned label.
The mistakes are made, and caught, and forgotten, one hospital at a time.
Not because medicine is careless. Because nobody built the machine that remembers.
So what does a computer actually add to this? Because a database is not new.
Here is the honest answer, and it is narrower than the marketing but I think it is bigger than it sounds.
A million reports is more than anybody reads. The ASRS has human analysts and they are good, and they are also a small number of people looking at a very large pile, working by category.
The thing that has changed is the ability to find the pattern nobody was looking for.
Not to search a database — we have been able to search databases since the sixties. To read a proposal that lands on a county commissioner’s desk on a Tuesday, and to say, before the vote: this was attempted in three other counties, here is what happened in each, here is the specific clause that failed, and here is the one that worked.
That is not doing anybody’s job. That is telling the person doing the job what happened last time.
The engineer still decides. The commissioner still votes. The surgeon still operates.
They just stop deciding blind, which is what almost all of them are doing right now.
I want to put one thing in about where institutional memory actually comes from, because we talk about it as though it lives in buildings.
It lives in people, and people can be moved.
In the 1930s, Europe expelled a substantial fraction of its own scientific memory. Physicists, chemists, mathematicians — people who carried in their heads the working knowledge of the most advanced physics on Earth, developed in institutions that then decided they did not belong.
They came here. Fermi, Szilard, Bethe, Wigner, Teller, and Einstein among a great many others less famous. The technical foundation of the most consequential American project of the twentieth century arrived in this country in the luggage and the heads of refugees, because the country that had it decided it did not want them.
The knowledge did not stay behind. It went where the people went.
I will let that stand without comment, except to say that it is the clearest possible demonstration that memory is portable, and it leaves with whoever is holding it.
So the ask for this episode is modest and I think it is nearly free.
One. A protected reporting system outside the industries that already have one. Aviation has ASRS. Utilities, water systems, construction, and the trades do not. The design is fifty years old and demonstrably works: confidential, immune, and held by a body with no power to punish.
Two. Every substantial public program evaluated, and the evaluation kept in a form that surfaces itself. Not an archive. Not a portal nobody opens. A system that puts the relevant history in front of the decision before the decision, automatically, whether or not anyone thought to ask.
Three. And the discipline from the last episode: keep the metadata. Who decided. What they were told. What they were not told. A record you can interrogate is worth more than a record that looks tidy.
None of that requires anyone to believe anything about artificial intelligence. It requires believing that a country which forgets what it learned is going to pay for the lesson twice.
I said at the start that everyone is asking whether the machines will take your job.
Here is what I would rather they were told.
Not “AI will do your work.”
“AI remembers what your predecessor forgot.”
That is the offer. It is smaller than the promises being made and considerably more useful than the fears being sold, and it is available now, and almost nobody is doing it.
The thing this country is worst at is not building. We have proved twice over that we can build.
The thing we are worst at is remembering what went right and what went wrong — and it is the only part of the job that nothing else in the room is able to do.
Next time: what people are good at — and a demonstration, conducted in public, with the division of labor stated in advance.
Sources: Challenger, 1986; Diane Vaughan on the normalization of deviance. Columbia Accident Investigation Board, 2003, on organizational causes recurring from the earlier accident. F-1 engine: NASA teardown and scanning of a museum engine to recover manufacturing knowledge not contained in surviving drawings. Aviation Safety Reporting System: established 1976 under an FAA–NASA agreement, administered by NASA; limited FAA immunity for self-reported non-accident safety events; database exceeding one million records; no reporter identity disclosed in the program’s history. Refugee scientists in the United States, 1930s.
Building Bridges is a fourteen-part series. Previous: Episode Nine — A Factory You Cannot Enter. Next: Episode Eleven — An Aggregate of Recorded Opinions.
P. A. Moore is the pen name of Pamela King, philosopher and artist. Available through the Ashfall Institute.