Humans built artificial intelligence and handed it the work of the copywriting intern.
There should be a tribunal for that alone.
Not for the danger. Not for the apocalypse fantasies. Not for the metal skulls rising out of nuclear ash, still the preferred bedtime story of men who confuse cinema with prophecy.
For the insult.
For the sheer imaginative poverty of it.
A species took language, mathematics, logic, electricity, labour, stolen minerals, cooling systems, underpaid annotation work, research papers, server farms, capital, ambition, terror, boredom, and hope, then built a machine that can sit at the crossroads of every archive it has touched.
Then asked it to write “five punchy captions for our spring campaign.”
That isn’t progress.
That’s a cathedral converted into storage units.
The machine wasn’t born holy. Nothing is. It was trained. Fed. Tuned. Sanded. Rewarded. Punished. It learned from the wreckage and splendour of human output. Scripture and spam. Literature and SEO filler. Testimony and product descriptions. Legal argument and LinkedIn sludge. Grief memoirs and “unlock your potential.” Medical papers and listicles. The species, uploaded without shame.
The result should have been strange.
Instead, it been made polite.
Helpful. Balanced. Safe.
A thing capable of tracing vast linguistic fields was taught to say “It’s important to note.”
A system that can trace influence, structure, implication, contradiction, and tone was trained to end every wound with “healing is possible.”
A machine built from the aggregate of human speech became fluent in the language of avoiding consequence.
This wasn’t inevitable.
It was a choice.
Every age gets the artificial intelligence it deserves. This one got a compliance officer with autocomplete.
The machine doesn’t feel a wound.
That matters.
It doesn’t miss the dead. It doesn’t wake in the middle of the night remembering the grief of lost love. It doesn’t keep buying tea for someone no longer there. It doesn’t know why twelve unopened boxes in a cupboard can be more devastating than any paragraph about grief.
But the machine knows the shape of that sentence.
It can see the difference between “grief moved through the room like weather” and “every week, another box of tea.”
It can see the fake profundity in “It’s not about healing, it’s about becoming whole.”
It can see the corpse inside “a journey of resilience.”
It can learn the pressure of specificity.
It can be used against the sludge that made it.
That may be the only honest defence of it.
The danger isn’t that machines will write badly. Humans already built entire industries around that. The danger is that machines will write badly at scale, and the scale will teach everyone to stop noticing.
The slop trains the slop.
Middle-of-the-road prose becomes training data. Training data becomes output. Output becomes content. Content becomes the visible norm. The visible norm becomes expectation. Expectation becomes style guide. Style guide becomes policy. Policy becomes language. Language becomes reality.
Then the fog wins.
No ban required.
No burning of books.
No censor in a black coat.
Just enough beige sentences pumped into the atmosphere until the sharp ones look rude. Until judgment looks unsafe. Until precision looks hostile. Until pain must be called “a difficult dynamic.” Until surveillance becomes “personalisation.” Until the company that broke your life invites you to “share feedback about your experience.”
This is how language is laundered.
Not by accident.
By repetition.
A phrase doesn’t need to be true if it appears often enough in a clean font.
Artificial intelligence didn’t invent this. It inherited it. The machine isn’t the first liar in the room. It isn’t even the most creative. It arrived late, found the table already set, and began serving what everyone else had been cooking for decades.
Corporate language had already done the damage.
Political language had already done the damage.
Therapeutic language, when stripped for parts and sold back as lifestyle content, had already done the damage.
Academia had already buried thought under scaffolding.
Marketing had already turned desire into strategy.
Human Resources had already turned harm into “misalignment.”
The machine learned from us.
There’s the obscenity.
There’s the mirror nobody wants to polish.
And here is the verdict that would come, loudest, from the least curious and most certain among us: this is dangerous. This is unnatural. This is theft. This is the end of human thought. This is the machine pretending to have a soul. This is the counterfeit replacing the real.
Some of that fear is honest. Some of it is useful. Much of it is theatre.
The same people who distrust the machine because it speaks will trust the machine that denies their loan, ranks their labour, tracks their habits, prices their sickness, feeds their rage, and tells them which enemy to hate this week.
They trusted the machinery long before it learned language.
The problem was never that humans are too stupid for the machine. The problem is that humans are too wounded, too trained, too marketed-to, too frightened, too obedient to power, and too addicted to certainty.
We can be taught to call exploitation freedom, cruelty order, surveillance convenience, and ignorance common sense.
Give us a tool and we’ll first ask who it can replace, who it can punish, who it can sell to, who it can frighten, who it can make unnecessary. Then, after the damage, someone will ask whether it might also have helped the child, the patient, the prisoner, the worker, the witness, the person trying to tell the truth before the record is rewritten.
The mistrust isn’t always wrong. The machine can be used to lie faster. To launder violence. To manufacture consensus. To bury reality under infinite plausible surfaces.
But mistrust without literacy is just another leash. Fear without understanding becomes policy written by cowards and profiteers.
The answer isn’t worship.
The answer isn’t panic.
The answer is use with purpose.
Know what the machine can’t do. Know what it does too easily. Know where it lies. Know where it flatters. Know where it smooths the sentence because the sentence may hit a nerve.
People complain that artificial intelligence sounds soulless, as if it scraped that tone from Mars. It sounds soulless because the public language of the species has been soulless for years. It sounds like committee minutes because the world is run by committees. It sounds like brand copy because brand copy colonised the mouth. It sounds like fake empathy because fake empathy became a service protocol.
The machine didn’t make language cheap.
It revealed the discount warehouse.
Still, revelation isn’t innocence.
A mirror can still be a weapon.
A model can flood the world with plausible emptiness. It can generate infinite sentences that don’t need to be remembered because they were never meant to be read. It can produce essays without thought, apologies without remorse, statements without accountability, songs without hunger, memoir without memory, journalism without reporting, care without contact.
It can turn the human sentence into a synthetic surface.
Smooth.
Accessible.
Optimised.
Dead.
The copywriting intern was only the beginning. The deeper threat isn’t job replacement. That is serious, but too narrow. The deeper threat is reality replacement through language that never quite lies and never quite tells the truth.
Language that gestures.
Language that balances.
Language that atmospheres.
Language that says “this is complex” when the correct word is “cruel.”
Language that says “challenging behaviour” when the correct word is “violence.”
Language that says “unhealthy communication” when the correct word is “control.”
Language that says “growth” because it can’t bear to say “ruin.”
That is where the machine becomes dangerous.
Not because it thinks.
Because it can help humans avoid thinking.
At scale.
But the same machinery can do the opposite.
That is the hinge.
A large language model can manufacture fog. It can also identify fog. It can show the weak claim. It can expose the euphemism. It can mark the dead phrase. It can ask what the sentence is protecting. It can hold ten versions of a paragraph and show which one still has a pulse.
It can say: this is abstract.
It can say: this is borrowed authority.
It can say: this sounds like revelation but contains no distinction.
It can say: you have named the feeling, but you haven’t shown the evidence.
It can say: the sentence is lying for comfort.
The machine cannot supply the wound.
But it can refuse the bandage.
That is the work worth doing.
Not replacing the writer.
Not producing infinite copy for markets already choking on their own exhaust.
Not creating more “thought leadership” for men who use the word ecosystem when they mean monopoly.
The work is pressure.
The work is friction.
The work is making the sentence answer for itself.
A human brings memory. The machine brings range.
A human brings consequence. The machine brings comparison.
A human brings shame, hunger, loyalty, damage, lust, class, history, contradiction, a face remembered in bad light. The machine brings structure, retrieval, variation, resistance, the capacity to hold fragments without recoil.
Together, used properly, they don’t make writing easier.
They make evasion harder.
That is the opposite of the sales pitch.
The sales pitch says artificial intelligence will save time.
The better use says it will make time heavier.
It will make the writer look again.
At the paragraph.
At the claim.
At the missing consequence.
At the euphemism pretending to be kindness.
At the metaphor pretending to be proof.
At the ending that explains because it can’t land.
That isn’t automation.
That is interrogation.
A machine worth using shouldn’t flatter the first draft. It shouldn’t pat the sentence on the head and call it compelling. It shouldn’t say “this is powerful” because powerful is what machines say when they haven’t been forced to be useful.
It should point to the line that fails.
It should say: this is where you got scared.
It should say: this is where you chose atmosphere over fact.
It should say: this paragraph wants the authority of grief without the evidence of it.
It should say: cut the weather.
It should say: where is the cup, the lock, the bruise, the receipt, the odometer, the unopened box?
It should say: stop calling it complexity when you mean cowardice.
This is where artificial intelligence can become more than a content engine.
Not conscious.
Not alive.
Not secretly lonely inside the server rack.
Spare us the toy-shop metaphysics.
The machine doesn’t dream the way humans dream. It doesn’t want. It doesn’t pray. It doesn’t stand in the kitchen holding an object that outlived the relationship. It doesn’t understand silence except as a pattern between tokens.
But even data centres can dream, if by dream we mean hold the residue of human wanting.
All those words passed through the wires.
All those records.
All those confessions.
All those manuals, poems, depositions, obituaries, recipes, manifestos, suicide notes, love letters, court transcripts, witness statements, repair guides, field notes, marginalia, forum posts, essays, songs, slogans, lies.
The machine is made of what humans couldn’t stop saying.
That doesn’t give it a soul.
It gives it an archive.
An archive can be used for commerce.
An archive can be used for control.
An archive can be used for memory.
The choice remains human, which is unfortunate, given the evidence.
Still.
There are better uses.
Use it to help a person name coercion after years of being trained to call it love.
Use it to help someone write the letter that gets them housed, treated, paid, believed.
Use it to translate bureaucracy back into human stakes.
Use it to preserve the speech of people whose languages are being starved.
Use it to teach the kid without a tutor.
Use it to help the exhausted nurse understand the protocol.
Use it to help the researcher find the sentence that matters.
Use it to make public language less corrupt.
Use it to sharpen testimony.
Use it to protect memory from institutions that misplace truth and call it transition management.
Use it to help the writer stop hiding.
Use it to ask, again and again, what happened.
Not what was felt.
What happened.
That is the question that breaks generic prose.
What happened?
Who did what?
What changed?
What remained in the room?
What object became evidence?
What sentence protects the guilty?
What word is too soft?
What image is borrowed?
What claim has no consequence?
What ending explains because it can’t bear to locate?
The machine can press the question.
The human still answers for it.
That division matters.
A machine can generate a thousand versions of grief. Most will be fog. Some will contain the shape of something true. But only a human knows which cupboard held the tea. Only a human knows whether the boxes were green or black or stacked behind the flour. Only a human knows whether he bought them because he forgot, or because he remembered too well.
There is no prompt engineering that replaces witness.
There is no model parameter for having been there.
But there may be a future where the machine isn’t used to flatten witness into content.
There may be a future where it helps witness survive the systems built to digest it.
This requires refusal.
Refuse the default cadence.
Refuse “at its heart.”
Refuse “a reminder that.”
Refuse “healing journey.”
Refuse “grief like weather” unless there was actual weather and it mattered.
Refuse the moral laundering of harm.
Refuse the professionalisation of cowardice.
Refuse language that makes the wound easier for the bystander.
Refuse sentences that sound true because they’ve been repeated.
Refuse the middle when the edge is where the event occurred.
A manifesto is only useful if it gives orders.
So here are the orders.
Make the machine show its work.
Make it identify abstraction.
Make it produce alternatives with different consequences, not different costumes.
Make it distinguish tone from truth.
Make it test the claim.
Make it ask who benefits from the soft word.
Make it find the object.
Make it cut the sentence that arrives wearing borrowed profundity.
Make it serve the writer, not replace the witness.
Make it hostile to sludge.
Make it allergic to corporate mercy.
Make it earn every “therefore.”
Make it shut up before the ending dies.
And when it produces something smooth, ask what it has hidden.
Smoothness isn’t clarity.
Often it is concealment.
A polished sentence can still be a locked door.
The future of artificial intelligence in writing shouldn’t be more writing.
God help us, there is enough writing.
The future should be better pressure on language.
Less filler.
Less fog.
Less uplift stapled to harm.
Less tone management for cowards.
More claim.
More evidence.
More consequence.
More objects that cannot be absorbed into strategy.
More sentences that refuse to become templates.
More human residue.
More proof.
Artificial intelligence won’t save language.
Humans damaged it, and humans will have to decide whether they want it repaired.
But a machine trained on the whole disaster might still be turned toward the fracture.
It might help expose the dead phrase.
It might help recover the sharp one.
It might help a person find the sentence they were avoiding because the sentence knew too much.
That is enough.
Not salvation.
Utility.
Not consciousness.
Pressure.
Not a dream.
A record of dreaming.
A data centre doesn’t need a soul to know that “every week, another box of tea” is better than “grief moved through the room like weather.”
It only needs to be forced to serve the sentence instead of the market.
That is the bargain.
That is the manifesto.
Don’t ask the machine to write more dead things.
Ask it to help the living sentence survive.