The Citation Was a Door to Nowhere

The old library downtown had one good chair and forty thousand bad ones.

I used to go there when the apartment got too hot and the bar got too familiar. The good chair sat near a window where the sun came in tired and yellow, like it had been drinking with the rest of us. You could hear pages turning. Real pages. Paper scraping paper. The little cough of somebody pretending he was not asleep.

Once, years ago, I watched a kid in a wrinkled shirt chase a footnote through three books and a microfilm machine that sounded like it had tuberculosis. He had a notebook open beside him. Every time the trail went cold, his face tightened. Then he would find another lead and keep going.

Nobody applauded.

Nobody gave him a grant for heroic verification.

He was just doing the work.

That phrase has become embarrassing now. Doing the work. It sounds like something an old gym teacher would say while his knees clicked. But there it is, still standing in the corner with its hat in its hands, waiting for the educated classes to remember it exists.

arXiv, the big open drawer where researchers toss their papers before the world has had time to put on pants, has decided that if you submit work with fake AI-made references, you might get banned for a while. Not because you used a machine. Not because somebody at arXiv wants to smash laptops with a medieval hammer. Because if your paper contains citations to ghosts, the people reading it have a small problem.

They cannot trust you.

This should not be a controversial sentence. It has the plain shape of a chair. If you put your name on a paper and the paper points to sources that do not exist, you have not made a charming little clerical error. You have built a door to nowhere and told other people to walk through it.

Naturally, some of the scholars started screaming.

You have to love academia. It can take a simple rule like check the things you claim are real and turn it into a constitutional crisis. Suddenly responsibility had become gatekeeping. Verification had become cruelty. A one-year ban was apparently the intellectual equivalent of being dragged behind a truck.

I understand mistakes. I have made enough of them to furnish a small motel. I have mailed letters to the wrong women, trusted the wrong men, believed checks would clear because hope is a disease that keeps finding new hosts. I once spent a week convinced a publisher had stolen my poem, then found the damn thing under a stack of racing forms.

Errors happen.

But there is a difference between a typo and a corpse in the kitchen.

A misspelled name is a mistake. A missing comma is a mistake. A paragraph where the machine left its little metal thumbprint is ugly but usually survivable. A citation to a paper that never existed is different. That is not a scuff on the floor. That is the floor opening up.

The funny part is how familiar the excuses sound. Papers are long. People are busy. Coauthors know different things. The reference was in another language. The subject was technical. The tool slipped. The dog ate the bibliography and then uploaded it to a preprint server.

Fine. Life is hard. Science is hard. Writing is hard. That is why they used to make people sit in libraries under bad fluorescent lights until their souls became footnotes.

Now the machine can produce a bibliography with the confidence of a drunk giving directions. It does not know the difference between a source and a dream of a source. It has read the shape of scholarship. It knows the posture. Authors, title, journal, year, page numbers. It can make a lie wear sensible shoes.

And because the lie looks right, people want permission to stop looking.

That is the rot under the floorboards. Not the hallucination. The hunger to be absolved by it.

The machine did it.

The model suggested it.

The workflow generated it.

The system introduced an artifact.

Listen to those words. They have no fingerprints. Nobody is standing in them. They are the kind of sentences institutions invent when somebody has already called the lawyer.

I spent years in the post office watching responsibility move around like a cockroach under a glass. A letter went missing and everybody became a philosopher. Was it the carrier? The clerk? The sorter? The machine? The weather? The address? The customer, for wanting mail in the first place? By the time the question reached management, the lost letter had become a weather pattern.

But some poor bastard still had to face the old woman at the counter.

That is the part the clever people forget. Every fake citation creates a little unpaid debt for somebody else. Some graduate student loses an afternoon hunting a paper that never breathed. Some reviewer wonders if he is going soft in the head. Some researcher builds on a plank that snaps. Trust, which was already limping, takes another hit in the mouth.

And the people who caused it say the punishment is too strict.

I do not want scholars banned for honest errors. I do not want some frightened junior researcher exiled because a formatting tool coughed up a bad line in the appendix. There should be judgment. There should be proportion. There should always be room for the human mess.

But the human mess is not a magic bath where everything comes out clean.

If you use a tool that invents things, you check what it invents. If you cannot check it, you do not use it there. If your coauthor is the only one who can verify the citation, then your coauthor verifies the citation. This is not tyranny. This is authorship with its shoes tied.

The strange thing is that the old academic system already ran on ritualized suspicion. Peer review. Replication. Citations. Notes. Methods. All of it was supposed to be a way of saying: do not just believe me because I sound serious. Follow the trail. Kick the tires. Open the box and see if there is a cat inside or just another grant application.

AI did not invent laziness. It only gave laziness a lab coat.

It gave tired people a way to produce the appearance of care without the care itself. That is the real product. Not intelligence. Not insight. Appearance. The gleam on the apple. The fake leather smell. The bibliography that looks like it has been somewhere.

I keep thinking about that kid in the library, hunched over the microfilm machine, jaw set, chasing some dead man’s sentence through dust and eye strain. Maybe he was wasting his life. Maybe the paper he wrote disappeared into a journal nobody read and a tenure file nobody loved. Most work ends that way. A small noise swallowed by a larger room.

But for a while, he honored the line between true and almost true.

That line is not glamorous. It does not raise money. It does not demo well under stage lights. It is a thin, dull, stubborn thing. A pencil mark. A checked box. A source that actually exists.

Lose that, and the machines will not need to destroy knowledge.

We will have done the ugly part ourselves, politely, with credentials.

And somewhere, in a library with bad chairs, a student will pull on a citation and find only air.

Source: Academics in Meltdown Now That They’re Responsible for AI Hallucinations in Their Research Papers