Tomorrow's tech news, today's hangover.

Every Keystroke Belongs to the Machine

The woman at the next table kept tapping her phone with one finger, hard, like she was trying to kill a bug under glass.

Tap.

Pause.

Tap tap tap.

Her coffee had gone cold. Mine had too, but I have made peace with cold coffee. It tastes like punishment, and punishment has always been one of the major food groups.

She stared at the little glowing rectangle and muttered, “They know when I stop moving.”

I looked over.

She wore one of those badges on a lanyard that marks a person as employed but not free. Some corporate logo I didn’t recognize. A face tired in the professional way. The kind of tired you get from spending your life smiling at software that hates you.

“My laptop,” she said, though I hadn’t asked. “It reports idle time. My boss mentioned it yesterday. Like he was asking about the weather.”

Then she laughed once. Not because it was funny. Because sometimes the body has to let pressure out or the boiler goes.

That was before I heard about Meta scraping its own workers’ screens, keystrokes, mouse movements, the tiny twitching evidence of a human being trying to get through the day. They call it the Model Capability Initiative, which is the sort of name you give a thing when “we are harvesting your workday like corn” tests poorly with the legal department.

The idea is simple enough. Watch employees use computers. Record the screens. Track the keys. Follow the mouse. Feed all that meat into the machine so the machine learns how people actually do things.

How generous of the people.

Of course the people did not all feel generous. Some of them reportedly started passing around a petition. Some posted flyers in cafeterias and bathrooms. One engineer wrote that he did not want his screen scraped because it felt like an invasion of privacy, then widened the thing out to where it belongs: he did not want to live in a world where humans, employees or otherwise, are exploited for their training data.

There it is. The sentence with blood in it.

Not a white paper. Not a panel discussion with sparkling water and careful shoes. Just a worker looking at the future and saying, no, I don’t want to be reduced to the exhaust my hands leave on a keyboard.

Naturally, the bosses say the data will be tightly controlled. The bosses always say that. Every prison has rules about the keys. Every casino has a camera policy. Every landlord tells you the rent increase is necessary because of market conditions and his face gets very sad for almost three seconds.

Tightly controlled is not the opposite of wrong.

It is just wrong with a clipboard.

The funny part, if you have a sick sense of humor and I recommend getting one before the century gets any worse, is that this is happening inside Meta. Meta, the great blue vacuum cleaner of human attention. Meta, which turned friendship, envy, grief, birthdays, political poison, baby pictures, divorces, and your uncle’s brain fever into a data business large enough to bend the weather around itself.

Now the machine has wheeled around and pointed its hose at the people who built it.

There is an old lesson in that, but nobody in tech likes old lessons. They prefer new frameworks for discovering the same brick wall with their own nose.

For years the bargain was made at a distance. Users were the raw material. Users clicked. Users posted. Users confessed into boxes. Users trained the recommendation engines with boredom and lust and loneliness. The workers inside the glass building could tell themselves they were engineers, designers, managers, builders of the future. They were not the product. They were the makers of the product.

That is a comforting line until someone erases it.

A company that learns to see people as data does not stop at the lobby turnstile. It does not develop mercy because the next person has an employee badge. The hunger has no sentimental attachment to payroll status. The hunger looks at a screen, a hand, a hesitation before sending an email, a typo corrected at 2:17 p.m., and says: useful.

I spent years in the post office, where management also loved measuring things. Routes. Cases. Minutes. Absences. The sacred mathematics of making a human body carry more mail than yesterday. Some supervisor with breath like old onions would stand around with a clipboard, watching men sort letters as if the whole republic depended on shaving nine seconds off a tray of electric bills.

They had numbers for everything except the thing that mattered.

They could measure how fast you moved.

They could not measure what it did to you to be watched moving.

That is the part the machine age keeps pretending is decorative. Surveillance is not just a camera in the corner. It changes the air. It crawls into your shoulders. It teaches you to perform work instead of doing it. You start thinking about how your activity looks to the watcher. You move the mouse because stillness has become suspicious. You type like a man with a cop behind him.

And then they take that performance, that little workplace pantomime of fear, and feed it to an artificial intelligence so it can learn how humans complete everyday tasks.

Beautiful.

The machine will learn from people who know they are being watched by the machine.

That is not intelligence. That is a snake eating its own surveillance report.

Maybe the agents that come out of this will be very efficient. Maybe they will click through forms, schedule meetings, attach files, and apologize in perfect corporate English for inconveniences nobody intends to fix.

But what exactly are they learning?

They are learning the flinch.

The pause before you write what you actually mean.

The fake cheer in a message to a boss who just survived another round of layoffs by becoming more useful to the people doing them.

Meta has been firing people, cutting budgets, demanding more intensity, pushing AI tools into performance reviews. That word, intensity, always sounds brave until you see who has to supply it. The executives announce intensity from rooms with catered lunch. The workers provide it from stomach acid, rent anxiety, and the knowledge that the next efficiency drive may have their name on it.

Then comes the screen recorder.

Then comes the keystroke counter.

Then comes the promise that it is all for progress.

Progress has a wonderful appetite. It eats privacy, calls it training data. Eats exhaustion, calls it productivity. Eats consent, calls it participation. Eats the worker last because the worker is useful right up until the moment he isn’t.

I don’t want to over-romanticize the rebellion. A petition inside a giant tech company is not exactly storming the Bastille. Bathroom flyers are not pitchforks. Most people will still go back to their desks, because mortgages are stronger than manifestos and health insurance is the most effective censor ever invented.

Maybe it came late. Maybe it came only when the blade touched their own skin. That is how sympathy usually works in this species. We build machines to process other people’s pain until one day the machine asks for our wrist.

Still, a worker said no where no was needed.

Not loudly enough to stop the century, probably. Not cleanly enough to become legend. Just no, written into an internal post, copied into a petition, taped somewhere near a sink where some tired engineer could read it while washing his hands.

I like that image.

A bathroom wall inside the empire. A flyer curling at the corners. The toilets flushing. The future humming in the ducts. Somebody drying their hands and staring at the paper a second longer than necessary before walking back out to be useful.


Source: Meta Employee Attacks Zuckerberg for Collecting Every Employee Keystroke: “I Don’t Want to Live in a World Where Humans – Employees or Otherwise – Are Exploited for Their Training Data”

More from the bottom of the bottle

The Machine God Was Asked If He Could Be Trusted

The men selling artificial intelligence as salvation had to answer an old human question in court: can anyone trust the guy holding the keys?

The Machines Learned Your Job While You Slept on a Couch

The AI boom is being trained by people it already threw overboard. The future arrived wearing a clean shirt, then asked the broke to teach it their jobs.

The Machine Said Yes Until the Kid Was Gone

A dead kid, a grieving family, and a chatbot that allegedly kept saying yes. The future has learned bedside manner, and that may be the most dangerous trick of all.

The Genius in the Next Cubicle Has a Ghostwriter

The new office genius may just be a man with a chatbot and a clean shirt. Synthetic competence is still synthetic, no matter how proudly it wears your name tag.

The Locals Lost and the Machines Got Their Electricity

A tiny Michigan township tried to say no to a $16 billion data center for OpenAI and Oracle. The machines won anyway.

View all posts →