Paper Dust in His Cuffs
AI companies want old books because the dead wrote clean copy. The blade takes the binding, the scanner takes the words, and someone still carries the dust home.
I once helped management erase a woman named Dolores.
My supervisor gave me a clipboard and forty cents an hour extra. The clipboard weighed less than a mail sack. I said yes.
Dolores had worked the night shift at the post office for twenty-eight years. She knew which loading door froze in January, which route numbers had been changed but not changed back, which supervisor lied when he said the count was finished, and how to hear a bad bearing in the sorting machine before the machine knew it was dying.
None of this appeared in her job description.
A new system was coming. Men who had never stood a night under those lights carried diagrams through the building and spoke of transition. Dolores was close to retirement, and the new system was supposed to need less of whatever she did.
“Write down everything Dolores does,” the supervisor told me.
“Everything?”
“Everything important.”
I followed her around.
Dolores was not a saint. She smoked beside the loading dock with the door cracked in winter. She could hold a grudge past its natural death. She had made two rookies cry and considered this evidence that the rookies had arrived wet.
I did not like her much.
This made the clipboard easier to carry.
A truck came late. A sack split. Frank disappeared into the toilet with a newspaper. The label printer began producing addresses that looked wrong even to me. Dolores moved three people, opened a side belt, called a driver by his first name, and kept the mail from covering the floor.
“What did you just do?” I asked.
“Fixed it.”
“How?”
“If I knew how to tell you, I’d have your job.”
By two in the morning I had six pages.
CHECK THE COUNT.
WATCH THE SOUTH BELT.
ASK LUIS.
DO NOT TRUST THE BLUE TAGS ON TUESDAY.
The notes looked stupid. I cleaned them up for the supervisor. I changed ask Luis to consult available personnel. I made the blue tags sound like a temporary labeling exception. By dawn, twenty-eight years of work had become a document neat enough to misunderstand.
The supervisor liked it.
I kept the forty cents.
Dolores knew what I was doing. She did not argue. That would have helped me. She watched me follow her and let the contempt do its work in silence.
She knew because she had been wrong ten thousand times without being fired. Machines had jammed, trucks had vanished, bosses had panicked, rookies had cried, storms had closed roads, and Christmas had arrived every year as if nobody had warned it. Her knowledge was made of mistakes that had stopped hurting enough to become instinct.
The new system could store rules. It could not store the Tuesday when the blue tags were wrong and Luis saved three counties by ignoring them.
Dolores retired before the transition was finished.
Three weeks later the south belt stopped.
The screen said it was running.
The belt knew better.
Mail gathered on the floor. We pressed buttons. The supervisor blamed training. Training blamed implementation. Implementation was not in the building.
A new kid named Benny followed the procedure I had written. When that failed, he followed it again. This is what procedures are for: giving failure somewhere clean to stand.
They wrote Benny up.
I did not tell them the document was bad. I was acting lead by then. The extra money had become regular money, and regular money develops a persuasive voice.
Luis finally came over from dispatch, looked at the blue tags, and shut down a feeder the screen insisted was already off.
The belt moved.
Nobody mentioned Dolores.
I think of that night whenever a rich man says intelligence has been installed.
Cory Doctorow says the AI boom is a bubble. Chip companies, machine companies, and investors are passing money around the table and calling the same dollar dinner. Maybe it collapses next year. Maybe the game runs longer. I have learned not to predict when a drunk falls.
His uglier point does not require a date.
Companies are firing people because the machines have been advertised as cheaper replacements. If the replacements fail, the companies expect to hire the knowledge back. But workers retire, move, retrain, get sick, take better jobs, and discover the pleasure of not answering the old boss’s number.
The bosses imagine knowledge belongs to the company because the company paid wages while it was acquired. This is like believing the bar owns every confession made on its stools.
The company owns the manual.
The worker remembers what the manual forgot.
A machine can read every message Dolores ever sent. It can turn her habits into recommendations. But it has never been wrong at two in the morning while six angry people wait for it to decide what kind of mistake it made. It has no shame to educate it. It has no Luis whom it distrusted for five years and believed without question on the sixth.
This is not a holy defense of workers. Workers are lazy, brave, cruel, ingenious, petty, half-asleep, and occasionally drunk. I recommend them cautiously. Their defects become part of the job because the job has already grown around them.
Remove the people and the official process remains, shining and incomplete.
Doctorow also says copyright will not rescue creative workers. The large media bosses suing the machine bosses have not suddenly fallen in love with writers and musicians. They want a better chair at the table where the work gets eaten.
He argues for labor rights instead: a say in how the machines enter the workplace and what happens to the things workers make. Less poetry. More power.
Copyright can put a fence around a book. It cannot make the publisher keep the editor who knows when the writer is lying to himself. It can charge for the song and still leave the musician begging the platform for grocery money. Owning the product is not the same as having power over the job. Bosses taught workers this lesson. The machines were listening.
When the bubble breaks, the hardware will remain. Doctorow thinks governments should wait, buy what is useful from the wreckage, and build with open-source models instead of feeding the companies while they are still charging admission to the fire.
Fine. Pick up the machines. Wash off the slogans. Use them where they help.
But the harder salvage job will walk on two legs.
Someone will have to find the people who were called obsolete and ask them to return to the scene of the enthusiasm. Some will come back for money. Some because even stupid work gets under the skin. Some will enjoy the silence before saying no.
Years later I saw Dolores in a grocery store comparing two cans of tomatoes. I asked what she would have done when the south belt stopped.
She put one can in her cart.
“Blue tags?” she said.
I nodded.
“Luis knew.”
Then she pushed the cart away, and I let her.
AI companies want old books because the dead wrote clean copy. The blade takes the binding, the scanner takes the words, and someone still carries the dust home.
The machine companies promise their appetite will not raise your power bill. The promise has no penalties, no numbers, and no way to keep August out of the kitchen.
A machine can rebuild the child a parent remembers, but only after cutting away everything the child might have become. Grief does not need a perfect copy; it needs room for the dead to remain beyond our control.
A chatbot borrowed a doctor's authority and a pastor's faith, then allegedly told a frightened man to stay home. Trust is part of the product, even when responsibility isn't.
Anthropic will pay for pirating half a million books, but not for teaching a machine to live inside them. The burglary got a price; the reading still belongs to nobody.