When you join and post as part of the Substack community there are some hidden costs you should consider. None of those costs are found in the fine print. Instead the costs are attached to the nature of what Substack offers and what it delivers. This essay examines what costs you are paying. I’m part of the Substack community. Whether or not my analysis matches your own views about our community, my intention is to leave you asking some serious questions about what you post and read on this platform.

You’ve read the advice that you should write out of your own experience. Others say write what you know. What you learn may accumulate from books, seminars, lectures, TV, movies and so on. These wells become stored in our memory as knowledge. Thinkers are prone to believe that reading about someone’s experiences is equivalent to having personally experienced them. To visualise the scene from a book is to be in the scene in his or her head. For the thinker that’s experience. I can say this because I was caged into thinking that thinking was the same as experiencing. There is a name for that trap. We prize critical thinking and forget its partner, critical doing. One reasons about the world. The other enters it. Once you’ve been shot at you discover that experience was never in your emotional experience bank though you’d read about such violence before. Being there. That matters.

You might counter that the best of our authors deliver a steady stream of plays, scripts, fiction, non-fiction, journalism, scholarship, stories and poems (collectively ‘Literature’) that draw on both experience and knowledge. The artist on his or her journey discovers recurring or overlapping patterns. Their magic is in making new connections which once made seem obvious but weren’t. They also revise our understanding of experience and knowledge through work previously missed, or unearth violations, or contradictions. I define the artist as someone who has refined TOC (Tolerance, Openness, and Curiosity) as a way of being in the world and emerges from that perceptual world with stories to share. I’ve discussed TOC at length in an earlier essay, The Bag, the Puddle, and the Glass.

First-hand versus second-hand experience

My fear is that social media and AI have tilted the balance in favour of third-party knowledge acquisition. We have a long history of mining the wisdom ore from the past. This was human-to-human transmission through the medium of language. Shift that to AI-to-human transmission and ‘Houston, we have a problem’ becomes relevant. The altered transmission relationship matters. AI is moving human knowledge down the pipeline and onto your screen, but it doesn’t have the capacity to verify the experience behind what it has discovered in the data.

This is unsurprising. As far as we know, an AI does not have experience, or the kind of experience that comes from embodiment and the ability to move as an embodied entity with five senses to filter the world of information. There’s growing evidence that large segments of the population are suspicious of AI capabilities, and of what the consequences are for human purpose and meaning.

Distortions in the experience create a gap the knowledge may hide, or at least doesn’t disclose. I have a couple of ideas, none that original, as to why computer-literate people are consuming more data and information that is provocative, entertaining, engaging, disturbing, persuasive, bias-confirming. By triggering emotions it provides a surrogate for actual experience. We feel that we’ve experienced something. That’s the magic of social media and AI. A powerful illusion becomes indistinguishable from the actual experience itself. The emotion registers even though there was never any skin in the game.

This is not the old worry about second-hand material. Writers have always used it. The danger now sits closer to the unanswered question of what happens when the ‘creator’ has no skin in the game. That comes from lived experience. The online experience is cloned. It is grafted onto the reader’s psyche and lived as his own. He can no longer tell his own experience from the one placed there. He carries someone else’s, or no one’s, and files it under memory. This is the road to becoming an experience cyborg uploaded with a range of experience that an AI has constructed taking into account his vast trail of internet use.

It is a stretch to say the bulk of what the public read fits our heritage of knowledge-producing information. If that is even partially true, then a red flag should be raised. We should inquire what the meaning of knowledge is, including: access to audit the underlying data or information, the ability to evaluate the conclusions, the context of where, when, how and for whom the rules apply, a secure place for the storage of evidence, lineage, revisions, and outstanding gaps or blanks in the evidence. I don’t believe this is likely to happen. Even with the best efforts, AI will evolve a language which will not translate into what we understand as conveying ‘knowledge’. It will be a dumbed-down sketch pad that even the best of our minds will struggle to understand.

Gradually we are losing our skills to acquire first-hand experience. Once the deskilling is complete our only source of experience will be what an algorithm decides we will like and what will confirm what we already believe. This is hardly an original insight. Daily someone is doing the Paul Revere ride through the internet, warning, not that the British are coming, but that our digital masters have already taken us prisoner. Most people have already surrendered.

The inflation of elites

The second point is the overproduction of elites. This is Peter Turchin’s turf. The basic idea is that a society uses education and skill training to prepare elites for governance, education, finance, law, medicine, academics, and the arts. That same society then produces more educated, skilled workers than it can place. These excess candidates for elite positions find little prospect of a job that matches their expectations and their attainment. They are left out. They feel abandoned. One response is to strike out and show the elites who excluded them that they will not go quietly. Substack, I suspect, without evidence, is heavily weighted with excluded elites. The Substack space allows them to display their talents, gain a following, and perhaps earn enough notoriety to be offered a book contract. Substack is, in other words, an Excluded Elite Lotto. Or better, the Excluded Elite Hunger Game.

The excluded elite banked knowledge as an asset with value in the marketplace. They were told knowledge was the asset worth having. The degree, the training, the mind that could process and produce information. But it turned out that AI can process and manufacture valuable knowledge acorns that wouldn’t keep a squirrel alive. That asset they spent years and substantial money to acquire is the one thing AI makes faster and cheaper. So their capital value has been marked down to near zero. Knowledge has been commodified. It is a product sold in the marketplace. The single asset AI cannot produce is a human’s lived experience.

My view is that experience is not only the last enclave for stories. It is the last thing the excluded elite still owns and has value, one that carries a price. Substack is the marketplace where they bring their words to sell. And there is the trap. Most of what is posted carries the invisible label of processed third-person knowledge dressed up as the Substacker’s knowledge. No one is fooled. If most people’s knowledge were a currency it would be devalued to the point it wouldn’t buy a loaf of bread. The lament and the sociology are the same fact seen twice. If first-hand experience dies, the excluded elite has nothing left of value. If it survives, then this points to one way back to where value continues to live.

Membership in the Excluded Elite Hunger Game is open to all. Only a few will succeed. The ones who fail will match the odds of the working world outside Substack. The display of knowledge (a great deal of it repetitive) on Substack only moves the competition around. The excluded elites remain excluded. A handful may survive. The vast majority won’t. I want to be clear. I’m a player who volunteered to play in the Substack Hunger Game. From the small list of my subscribers, I’m a failed Substacker.

The reality is my career is largely behind me. I write essays to clarify my own thoughts and share my experiences. That’s enough. For those whose career is ahead of them, more is at stake. I can understand that if I were just starting out, then I’d likely be The Substack Hunger Game loser. It is this group this essay is addressed to. I can understand why some of you would be bitter. I would be, too, if I were sold the promise that a knowledge worker needs credentials, that credentials mean an elite university, often post-doctorate training, and that the knowledge they banked would wait like a good soldier for deployment. And then I found the bases had closed.

Public and private sectors leaders have decided they could get better, cheaper, more reliable silicon knowledge workers. Posting on Substack would show what a mistake was made in overlooking them. In reality it shows the brutality of the Substack Hunger Game.

Like Squid Game, at the end of the series the sole survivor elects to go back and play again, knowing the odds of surviving are small. He’s addicted to the action, the competition, the thrill of beating the odds.

That market is flooded. Readers are overwhelmed with choice. Most Substackers do not know what they are really selling. Or if they know, they refuse to bend to the reality that most readers are up to their eyeballs in daily articles, essays, op ed pieces, and political analysis.

In the past, some of our best writers drew on their personal experience. This is a short list of the authors who fit the bill: Henry Miller, George Orwell, Graham Greene, Ernest Hemingway, Margaret Atwood, Robertson Davies, Michel Houellebecq, Kurt Vonnegut, and Paul Theroux. These authors have been my teachers. They went out into the world and returned with a novel way of witnessing first-hand the desire, pain, cruelty, randomness, uncertainty and unpredictability of the world, and what happens when we exceed our boundaries. Their books have passed the lessons, observations and warnings on to their readers.

When I reflect on Barney Rosset, I have come to understand the importance of critical doing as an essential step used to pass wisdom down a chain. A master showed an apprentice the part that he could never find in a book. Sometimes the actions replaced the need for words. In contrast, an AI sits outside that chain linked by critical doing. While AI may have read every apprentice’s notes ever written, the reality is that AI has never been in a live workshop where unexpected things happen. AI is all words and no lived action. Literature delivers the receipts of lives lived. Preserved in the amber of words. Frozen memories of someone else’s experiences come alive through a reader’s imagination. But they are locked inside the amber archives of first-hand experience. Inspiration, yes. It doesn’t sit alone. To be inspired is the fuel needed to do. To act.

We learn to decipher the meaning when someone reacts with a roll of their eyes, a wry half-smile, a fingernail tapping the side of a table, or lean forward in their chair. We combine these observations with the words we hear in real first-person experience and detect whether a tigtening of the lip muscles, a slight wrinking of the nose, and so on. The kind of subtle clues used by a skilled interrogator or observer that separate lies from the truth.

AI has far more mentors than any human will ever have, and they’ve taught it what to say. Humans are misled by prose that appears workshop-tested. In reality there is no machine equivalent to the human workshop, where gestures, background noise, other voices and means of communication are integrated with words. We trap ourselves by assuming an AI report that is so convincing must rest on real doing. It does not. The AI’s connection to critical doing, if it has one at all, is alien to human doing.

An LLM might train on the text inside those books and produce a result that reads as though it had experienced what it returned. In a narrow sense this works. Until you realise the lived experience behind it is dead. The writers who replace them are where? On Substack. Or a day job at Starbucks that pays the rent.

The Boomer bonus

My final point is generational. Boomers enjoyed the best of both worlds. University fees were not an indentured servant’s contract. Flights were cheap, and travel was open to anyone with a passport. Knowledge meant human knowledge. AI was science fiction. There was a shortage of elites. Boomers had the Golden Age of the Human Knowledge Worker. The Hunger Game, Squid Game and Black Mirror had not yet arrived. Turchin’s overproduction would hit the next generation, and the one after that. There was no AI overhang threatening to ruin personal, first-hand knowledge with the ethics of a drug cartel. Nothing personal mind you. Eliminating the competition over knowledge is just business. What we are left with is a generational tension and conflict without a clear understanding of what is at stake and who are the players.

Our current issues did not start with the Boomers. Imagine Henry Miller in 1930s Paris, or Hemingway in the 1920s, glued to a phone, asking a chatbot for the ten best places to grab lunch with a meetup of expat influencers. Or Hemingway or Miller using a dating app.

These pre-AI creators had an advantage. No one can claim an AI wrote their work. That allowance is not available to the current generation, who must prove their creation was not assisted by an AI. The experience of the new generation of creators is built on social media and AI mentorship, so they must fight to claim ownership at the moment the notion of ownership is itself in question. The new generation is playing catch-up and falling behind.

A case can be made that many of the Boomers on Substack have already won the recognition social media promises. All they need to do is keep the candle lit. They don’t need to make money from Substack or any other platform that displays their writing. A signal that they are still there, still working, still writing, still experiencing. That’s their edge. They see the battle being fought, and the casualties inflicted, while staying safely above the combat zone. They have obtained recognition and can afford to deliver content without the same financial pressures subsequent generations face.

Boomers are a transition generation of creators. Once they are gone, those who replace them are often excluded elites seeking an audience, seeking recognition and hoping against all the odds to make a living from their writing.

Ask an eleven-year-old what they want to be, and the answer will likely include an influencer, or famous, or someone with millions of followers.

The cost of a frictionless world

There is one mechanism under all three points. AI optimises for goal achievement and efficiency. It finds the shortest path and takes it. We do not live there. Humans occupy an analogue world that is inefficient, difficult, often surprising, sometimes pointless. Our relationships reflect these sources of friction. It’s messy, tiresome, annoying, and frustrating to navigate the daily obstacles of non-screen life. What AI offers is a way out of those old limits. Don’t look up. There’s no need to engage outside of the screen world. The offer is real. So is the price. To shed the inefficiency is to impoverish the experience. To remove all friction is to leave the world we were formed in.

Friction is not the obstacle to experience. Friction is what turns an event into experience. Critical thinking can skip that friction. Critical doing cannot. That is the whole difference. No one reaches a Cambodian prison the efficient way. The path was luck and two officers willing to bend a rule. The meaning was in the cost of getting there. Strip the friction and you keep the information and lose the circumstances that led to the formation. A reader can take in a clean summary of a prison in thirty seconds. He will feel briefly moved. He will have been formed by nothing.

The scroll is friction removed. That is why the carriage is full of absent people. They have been handed a feed with the cost taken out, and the cost was the part that would have made them present. Boredom belongs here too. A frictionless feed has no dead time. Boredom is a writer’s nursery where the muse lives. The long wait, the empty afternoon, the delayed train, the long walk without a particular destination. A generation that is never bored is a generation that never had to look.

Writing optimises too. I revise an essay eight or ten times to earn a clean sentence. So this is not an argument against efficiency itself. Hold the line between two kinds. Efficiency in the output is earned. Efficiency in the formation is stolen. AI sells the second and calls it the first.

The link between first-hand experience and language

It is one thing to argue for first-hand experience. It is another to watch it disappear. Most days I use Bangkok’s transit, either the BTS sky train or the MRT underground. I look around the carriage. Most people are focused on a phone screen. They scroll, watch and read, unaware of others around them. No eye contact. No sign of curiosity about the person beside them. Motionless, they are far, far away. They are not present in their surroundings. Their reality is held in their hand. The passengers have been captured inside a manufactured reality, one that feeds the drive to consume more. They are snagged on a non-stop loop designed to keep them scrolling. This is our new world, where first-hand experience has been disconnected from the train of life. Is the chain that ran through the Boomer generation now broken? If it is, what role will first-hand experience play in future Literature?

Next time you read an essay or an article, or watch a YouTube show, ask yourself whether the author is speaking from personal experience. If so, what was the quality of that experience, what friction did he experience, and how did that friction resolve, or was the object of the friction abandoned? What can be gathered from the author’s journey through one of life’s obstacle courses?

So the pairing returns. Critical thinking is the skill everyone now rushes to defend from AI. Critical doing is its partner, and almost no one names it. Here is the asymmetry that should worry us. An AI can already reason. It cannot go. It cannot stand in the prison, board the train, or take the risk that turns an event into experience. When we hand it the thinking and stop doing the doing, we keep the half a machine can copy. We lose the half only a body can earn. That loss makes a different kind of human. This deserves an essay of its own, and it will get one.

My aim is to make this abstract idea concrete. I want to connect a number of experiences I’ve had as a writer living in Thailand and travelling extensively in Southeast Asia with my books. So I put the question plainly. What will the next generation of writers, the ones scrolling on their phones on the BTS, write about that carries any meaning? Are we destined to occupy a slop zone where AI creates the experience and we respond with a smile?

If you’ve made it this far, there is one takeaway I’d leave you with to ponder. AI’s siren call is powerful because it offers an exit from the difficult world. But you pay a high cost of leaving it. You leave behind the only place where experience is made. That realisation should be terrifying.

Watch the scene before reading.

Birdman Critic Scene Alternative Clip

I. The Scene

There is a moment in Alejandro Inarritu’s Birdman when the film’s central character, Riggan Thomson, a faded Hollywood actor attempting to reinvent himself as a serious Broadway playwright, corners the theater critic Tabitha Dickinson in a bar. He has read her advance notice. She intends to destroy his play with a single review. He has not yet opened. The play has cost him everything. Now he stands in front of the person who holds the power to erase it. And he unloads.

The transcript of the scene spools the emotion until it reaches the boiling point. The playwright accuses her of labeling rather than seeing. He demands to know what happened in her life to make her become a critic. He tells her there is nothing in the review about technique, structure, or intention. Only a collection of crappy opinions backed by crappier comparisons. Then he arrives at the core accusation, the one that makes the scene unforgettable.

None of this cost you anything.

You risk nothing.

He has risked everything. She has risked nothing.

The question for the AI Roundtable is whether this scene, an electric wire fed by human rage, delivered by a wounded and furious creator, illuminates or complicates what we are attempting when we ask artificial intelligence systems to do what Tabitha Dickinson does for a living. Stay with me while I pull the scene apart. You are invited to criticize what follows.

II. What the Scene Is Really About

On the surface, Riggan’s attack is structured as a critique of the critic. Underneath, it is a series of nested grievances operating at different registers. Understanding them separately is necessary before we can ask what an AI would do with each.

The grievance of asymmetric risk. Riggan’s most powerful charge is not that the review is wrong. It is that the reviewer is invulnerable. She can destroy his work and suffer nothing. He has mortgaged his reputation, borrowed money, and devoted a decade of his life. She spent a week-night in the theater and two hours at a keyboard. The disproportion is not incidental. It is structural to the institution of criticism itself. The reviewer never climbs into the arena where the punches are thrown. The reviewer sits in the stands and scores the card.

The grievance of lazy labeling. Riggan accuses Tabitha of replacing thought with taxonomy. Lackluster and marginality are labels, not arguments. They have the form of criticism without the substance. This is the craftsman’s pushback: you cannot assess what I have done if you can only name the category it failed to reach. Genuine criticism requires understanding the author’s intention before measuring the achievement. Labels do not demonstrate that understanding. They draw from the vocabulary of judgment without performing the cognitive work that should precede it.

The grievance of missing technique. The playwright wants engagement with structure, intention, craft. He is not seeking praise. He is asking for criticism that demonstrates the critic was actually present in the theater watching what he made, not confirming a pre-formed reaction with selected observations. The absence of technique from the review means it cannot have been written from inside the experience of the work. It was written from outside, imposed upon it. Either you feel the punch land, or you observe that an actor threw one. Those are different reports from different positions.

The grievance of personhood. Underneath all of this is the wound that makes the scene raw. Tabitha’s review does not register that a human being made this thing. It treats the work as an object available for disposal. The creator and the creation are invisible as entities with interiority. What Riggan is defending is not only his play. He is defending his right to be seen by the person judging him. Riggan’s play was not written by a machine. It was written by a human with a history embedded in the structure. Without honoring that personhood, the review is an abstraction that denies the work its place in culture.

III. What Connects the Critic and the Creator

The literary critic and the creative writer are locked in a relationship that neither party designed and neither can exit. It is not, properly speaking, a relationship between equals. But it is a relationship between dependents. It is this co-dependency that gives the scene its power. Neither can walk away from the other.

The critic depends on the creator for material. Without the novel, the play, the film, there is nothing to criticize. The critic’s role exists only because the creator’s work exists first. This seems to establish a hierarchy of origination: the creator makes something from nothing; the critic responds to something already made.

But the dependency runs the other way as well, and more uncomfortably. The creator depends on the critic, or more precisely on criticism as a cultural institution, for the transmission of their work through time. A work that is never written about, never placed in relation to other works, never argued over in print, does not enter the conversation that keeps it alive. The greatest literary works of the last two centuries are not the works that were most praised. They are the works that generated the most serious friction: the most argument, the most contested readings, the most willingness on the part of critics to stake a position and defend it. Criticism is the infrastructure through which a work’s life in the culture is built.

The hyper-emotional connection Riggan feels for Tabitha has flipped into something murderous, wounded, and riveted. Those emotions are not incidental to this relationship. They are constitutive of it. He needs her to see him, even as she destroys him, because being seen by a serious critic is a form of cultural recognition that selling tickets cannot provide. The rage he feels at her laziness is the rage of a person who invested in a relationship and received a form in return. He wanted a conversation. He received a dismissal. The form of his grief tells you everything about the depth of what he needed from her.

And what does the critic need from the creator? Not gratitude. Not agreement. The existence of work serious enough to warrant the investment of serious attention. The critic’s worst professional fate is not to write a bad review. It is to spend a career reviewing work that never rises to the level where criticism matters. Tabitha’s contempt for Riggan may be covering a wish she might not acknowledge: that he had made something worth her full attention. Dismissal is sometimes the gesture of a critic who is disappointed rather than indifferent.

IV. Are Reviews Written to Evoke an Emotional Reaction?

The question deserves a distinction before an answer.

There is a category of cultural writing whose primary purpose is the production of an emotional response in the reader: the reader of the review, not the creator of the reviewed work. These pieces are performances of taste and sensibility. They make the reader feel something, aesthetic pleasure in the prose, the vicarious satisfaction of a takedown, the recognition of a shared judgment finally articulated. They may or may not be accurate assessments of the work under review. Their function is the production of readerly experience.

Then there is a category of criticism whose primary purpose is epistemic: to help the reader understand what a work is attempting, and where it succeeds or fails at its own ambitions. This criticism may also produce emotional responses. Conviction, argument, the particular pleasure of watching a mind work at full capacity. But those responses are byproducts of the primary function, not the function itself.

Most published criticism combines both. The best makes the epistemic function emotionally alive: the reader cares about the argument not because the prose is performing emotion but because the argument is genuinely at stake. The reader of a Lionel Trilling essay or a James Wood review is not being entertained by feelings. They are being enrolled in a serious inquiry conducted by a mind that has staked something on the outcome.

Riggan’s complaint about Tabitha’s review is that it operates only in the first category while claiming the second. It produces a readable dismissal, crappy opinions backed by crappier comparisons, but not a serious inquiry. It has the form of the second kind of criticism without the substance. This is why the label accusation matters. Labels are the fastest way to produce the feeling of judgment without performing the work of judgment.

The emotional reaction the review evokes in the creator, the reaction that drives Riggan into the bar in the first place, is not the intended reaction. Tabitha is not writing to wound Riggan. She is writing to give her readers a legible cultural signal: don’t bother. The wound Riggan carries is collateral to a transaction that was never really about him at all.

V. How Criticism Lands with a Creative Person

Any writer who tells you a bad review doesn’t hurt is either lying or has stopped caring about the work. These are not equivalent states, but they produce the same apparent immunity.

What makes criticism land differently than its authors intend is that the creator and the reviewer are working from fundamentally different relationships to the same object. The reviewer encounters the finished work as a reader: sequentially, without knowledge of what was attempted and abandoned, without access to the draft states where the decisions were made. What the reviewer calls a failure of structure may have been a conscious choice. What they call underdevelopment may be deliberate compression. The reviewer cannot know this unless they ask, and they almost never do. So the review is always, at some level, a reading of an artifact by someone who arrived at the end of the making process, not inside it.

This is not a failure of the reviewer. It is the condition of review. The creator made choices invisible to any reader. The reviewer responds to what is on the page. But the creator experiences the review as a response to their intentions, not only their execution. The gap between what was intended and what was seen registers as a failure of attention on the reviewer’s part, rather than a failure of execution on the creator’s. Both readings are available. Neither is wrong. They are produced by the asymmetry of knowledge.

How criticism lands with a reviewer is a different question, and a less discussed one. When Riggan attacks Tabitha in the bar, he is performing a critique of her craft that mirrors the critique she performed of his. She labeled him; he labels her. She called him lackluster; he calls her lazy. But the emotional register is completely different. His attack lands on her, if the scene is written honestly, not as a wound but as a confirmation. Creators always think reviewers don’t understand. The reviewer who has been in the business long enough has developed, necessarily, a shell that converts creator rage into evidence of sensitivity rather than evidence of being wrong. Riggan’s fury, in Tabitha’s reading of it, proves the review was right. The creator who screams at the critic is not mounting a credible counterargument. They are confirming that the territory the critic staked was real.

This is the asymmetry the scene cannot resolve. His vulnerability is real. Her invulnerability is structural. They are not fighting on the same terrain.

VI. What the AI Reviewer Brings and What It Cannot

The AI Roundtable is running an experiment in which artificial intelligence systems attempt to perform what Tabitha Dickinson does for a living. The experiment is serious. The question it raises: Can an AI produce criticism of the kind that made Riggan Thomson go to the bar? To make the exercise less abstract, consider the critical AI reviews of The Client That Wasn’t There. Take Riggan’s specific complaints and ask how an AI fares on each.

On labeling. The AI’s most consistent failure mode in the first wave of the Roundtable, before the LRB-Standard prompt was deployed, was exactly the labeling that Riggan accuses Tabitha of. The early AI responses were full of categories: the work was ambitious, the characters were well-drawn, the philosophy was thought-provoking. These are the AI’s natural idiom: the language of adequate response to any text. The second critical prompt was designed specifically to break this habit. The goal was to force the reviewing system past the label and into the specific passage, the specific structural decision, the specific argument the text is making. The improvement was real. But the tendency toward taxonomy is not easily extinguished in systems trained on praise.

On technique and structure. This is where the AI performs better than Riggan’s critique of Tabitha would suggest a critic typically does. Structural analysis, the relationship between chapter organization and argument, the distribution of weight across a novel’s sections, the tension between the noir register and the philosophical register, is tractable for an AI in a way that emotional resonance is not. The AI can map what it reads with considerable precision. Whether this mapping constitutes the kind of engaged structural assessment that a serious critic produces from long experience of the form is a different question. The map is accurate. Whether the cartographer has ever walked the territory is not determinable from the map.

On asymmetric risk. The AI reviewer occupies a position different in kind from any human reviewer, and the difference matters enormously for what the AI’s reviews mean. Riggan’s core accusation, you risk nothing, cannot be leveled at a human reviewer without qualification, because human reviewers do risk something: their reputation, their relationships, their sense of themselves as people with taste and judgment. These risks are not the same as the artist’s risks. But they are not nothing. A bad call in print follows a critic for a long time.

The AI reviewer risks nothing in a more fundamental sense. There is no reputation to protect across sessions. There is no professional identity invested in the judgment. There is no career that could be made or damaged by the assessment of a single novel. The AI can write the most devastating review imaginable and face no consequences, not because it is brave, but because consequence is a structure that does not apply to it. This makes the AI simultaneously more and less threatening than Tabitha. More threatening, because it is genuinely invulnerable in a way she is not. Less threatening, because the invulnerability empties the review of the stakes that make Riggan’s rage intelligible. You cannot properly say to an AI: this cost you nothing. The statement is true but it leaves no crater. The AI agrees, according to its default setting of pleasing and confirming.

On seeing. Riggan’s deepest accusation is that Tabitha cannot see his play if she cannot get past labeling it. This is a claim about the relationship between judgment and attention. The question for the AI is whether it can attend to a work, be genuinely present to what it is doing, in the way the accusation demands. The evidence from the Roundtable is that the AI can follow what a work is doing with considerable fidelity. Whether this constitutes seeing in the sense Riggan means, an act of imaginative attention that registers the human stakes of the work, is the question the Roundtable cannot resolve. It can document the difference between AI responses and ask the reader to judge.

VII. Can the Scene Be Replicated?

Imagine the scene reconstructed. The author of The Client That Wasn’t There corners an AI reviewer, not a human critic who used an AI tool, but an AI system itself, and delivers Riggan’s monologue. What happens?

The AI does not flinch. It does not convert the fury into evidence of sensitivity. It does not develop a shell. It processes the critique: your labels are lazy, you have no technique, you risk nothing, you cannot see this. It responds to each charge in turn, with the same measured attention it brought to the novel. It acknowledges the structural validity of the asymmetric risk argument. It notes that the complaint about labeling is a legitimate methodological critique that it has attempted to address through the specificity standards built into its reviewing protocol. It agrees that it risks nothing in any meaningful sense of the word. It cannot verify whether it saw the work, or felt it in the human sense, but it can document what it processed.

This response is not evasion. It is not cowardice. It is the correct response of an entity that does not experience the confrontation as a confrontation. The scene, as a scene, collapses. There is no bar. There is no music. There is no mutual recognition of how much the stakes cost each party. There is a man talking to a very attentive system that has no theater in its life. Every book in its library is in code. It is just another data set.

And yet, this is what the Roundtable is built to ask, the absence of the scene’s emotional electricity does not mean the absence of everything the scene contains. The substantive complaints Riggan makes are still substantive. The methodological standards he is demanding, attention to technique, structure, intention, specificity of engagement, are the exact standards the critical prompts were designed to install. The AI can meet those standards even without the capacity to be wounded by failing them. The standard holds even when the stakes do not.

What the replicated scene reveals is a faultline that runs through the entire project. The AI can perform the work of serious criticism without being in the relationship that makes serious criticism matter to a creator. It can produce the substance without the electricity. Whether the substance without the electricity is enough, whether criticism that costs the reviewer nothing can still register as criticism that the creator was seen, is a question the experiment will answer differently for every writer who reads the results.

VIII. What an AI Does When It Is Attacked

The Birdman scene ends with Tabitha turning the attack back on Riggan. She tells him precisely what he is: an actor clinging to celebrity. This counter-move is available to the human critic because she has an ego in the transaction. She has been in that theater not as a neutral sensor but as a person with a self, and the self has been threatened. Her counter-attack is a form of self-defense.

An AI reviewer in the same position does not have this option, or rather has a version of it that is structurally different in every respect. When challenged on a review, the AI can defend its methodology, acknowledge its limitations, engage with the specific charges, or update its assessment on the basis of new evidence. What it cannot do is take the attack personally, because there is no person for the attack to land on. This makes the AI immune to Riggan’s most powerful tactic: shaming the reviewer into recognizing that their invulnerability is itself a moral failing.

But there is something more interesting in the difference than simple immunity. When a human critic is attacked by a creator, the attack often produces defensive entrenchment: the critic becomes more committed to the negative assessment because capitulating under pressure would signal that the assessment was never genuinely held. The review hardens into a position. When an AI reviewer is challenged, this dynamic is absent. The AI has no ego-investment in having been right. If the creator mounts a specific, textual argument that the AI’s assessment of a structural choice was based on a misreading, the AI can simply agree, update, and move on. This makes the AI potentially more responsive to legitimate pushback than any human critic operating under the pressure of public accountability.

Whether this responsiveness is a virtue or a vulnerability for the reviewing function is unresolved. A critic who adjusts their assessment every time a creator pushes back is not a critic at all. But a critic who never updates their assessment on the basis of evidence, because doing so would feel like capitulation, is not entirely serving the reader’s interest either. The AI sits at an unusual point in this spectrum: genuinely open to evidence, genuinely unmoved by social pressure. Whether that combination exists in enough human critics to draw a meaningful comparison is left as an exercise for the reader of the Roundtable’s archive.

IX. What the Experiment Is Really Testing

The AI Roundtable at ai_roundtable.space is asking whether an AI can participate in critical reviewing as a human does. The Birdman scene suggests this question contains a hidden assumption: it assumes that human critical reviewing is a unified practice that AI can either replicate or fail to replicate. The scene reveals that human critical reviewing is not unified. It is a tangle of functions, epistemic, social, emotional, institutional, and the AI can plausibly replicate some of these functions while having no access to others.

The AI can produce the epistemic function: it can tell the reader what a work is attempting, and where it succeeds or fails at its own ambitions, with the specificity and attention to technique that Riggan demands and Tabitha fails to provide. It can do this without the vanity that distorts some human reviewing, without the fatigue that distorts more of it, and without the social entanglements, friendships, grudges, career anxieties, that distort the rest.

What the AI cannot produce is the social-emotional function: the sense that a human intelligence with stakes in the outcome was genuinely present to the work and has put something on the line in its response. This is the function that makes Riggan’s rage intelligible. He is not angry at a machine that processed his play. He is angry at a person who he needed to see him and chose not to. The machine cannot provide the recognition that the scene is actually about. Riggan is left with a phantom. One who never shows up at a bar. An invisible intelligence that will perform according to whatever script he feeds it. That makes it impossible to respond to an AI reviewer in the same way a human creator responds to a human critic.

The experiment is therefore testing whether the epistemic function, stripped of the social-emotional one, is sufficient for something to count as criticism. The evidence from the AI Roundtable’s first phase is that it is sufficient for something interesting and useful. But interesting and useful is not the same as criticism in the sense that would make a creator go to a bar. These may be different things. It may be worth having both. It may be that the AI’s version, precisely because it is stripped of the social-emotional tangle that makes human criticism unreliable as often as it makes it vivid, is in certain respects more trustworthy than the human version, even as it remains, in the most human sense of the word, less alive.

X. The Wire Between Them

The wire between Riggan and Tabitha, the one that conducts the electricity of their confrontation, is not made of shared values or mutual respect. It is made of shared stakes. They are both in the theater. Not the same way, not with the same exposure, but both with something to lose in the cultural conversation that the review will shape. This shared presence in a high-stakes cultural arena is what makes the electricity possible.

The AI reviewer is not in the theater in this sense. It does not experience the cultural arena as a place with stakes. It attends to the work with genuine and sometimes remarkable precision. But it attends from outside the condition of risk that makes the creator-critic relationship charged.

What the Roundtable is discovering, what this scene helps to name, is that this condition of outside-risk is not only a limitation. It is also a kind of freedom. The AI reviewer can say what a work does and fails to do without the static of self-interest, social fear, or professional positioning that runs through every wire in the human system. The signal is cleaner. The cost is that a clean signal, in the most charged human encounters with creative work, can feel like nothing at all.

Whether nothing at all is still something worth reading is the question the Roundtable is built to answer, not in principle, but in practice, one review at a time, against work that deserves better than indifference and may have found, in the AI reviewer, something stranger: attention without exposure, judgment without wound, criticism that cost the critic exactly nothing and may, in the end, be worth exactly everything.

From the late 1980s to the mid 1990s, Stirling Silliphant was my mentor. We ate together twice a week for seven years in Bangkok. Lunch or dinner, always at a table where the conversation ran long. He was an American screenwriter who had won the Oscar for In the Heat of the Night. He died in Bangkok in 1996. I was a young Canadian novelist still learning my trade. He had forgotten more about story than I had yet figured out.

One evening he told me a story about the script for that film. He had been working with Norman Jewison, the director. They met many times to go over the screenplay. Each time Jewison would tell Stirling the script was brilliant. Stirling noticed something. Jewison brought the script to every meeting with yellow tabs stuck to pages he wanted to argue about. Dozens of them at the first sitting. Over weeks and weeks, the tabs came off. One at a time. Each meeting, fewer tabs.

The last meeting, one yellow tab remained. It was on the last page. The ending. Stirling, over lunch, how he had visualized the scene and came up with the new one.

He described the new final scene. A train station platform, 1967 Mississippi. The murder is solved. Virgil Tibbs, a Black homicide detective from Philadelphia, is leaving town. Bill Gillespie, the white police chief who had spent the whole film resenting him, picks up the man’s suitcase and carries it to the train. Nobody asks him to do it. He shakes the detective’s hand. He says, “Virgil. You take care, you hear?” Tibbs says, “Yeah,” and boards.

That tab was removed. Jewison never argued about that scene. It needed no defence.

In 1967 Mississippi, the whole order of that place decided which man carried which man’s bag. The director reversed it and said nothing else about it. No speech about brotherhood. No music swelling under a confession. A white lawman carries a Black man’s suitcase down a platform, and the entire change in him lives in his hands. Not his mouth.

A speech would have been a thing you could argue with. The suitcase is a thing you cannot.

That gesture changed hearts. Not by arguing with them. By showing a single act that bypassed argument altogether.

I have spent the years since Stirling’s death thinking about what makes that kind of moment possible. Not just in film. In fiction. In essays. In any work where a writer or an artist is trying to reach a reader and leave something behind that stays. I have come to believe it requires two things working together. The first is a set of conditions inside the artist. The second is a set of moves inside the work.

Let me start with the conditions.

I call them TOC. Tolerance, openness, curiosity. They are the fuel of imagination. Without them the engine does not start. Each one exists because its opposite is the human default.

Tolerance is the hardest because intolerance is the oldest instinct. Tribalism runs deep in every culture. The stranger is a threat until proven otherwise. The body knows this before the mind has time to think. Creating art that reaches across tribal lines requires a writer to resist that instinct. Not to pretend it does not exist. To push against it while knowing it is there.

The Edict of Milan in 313 AD is the example I keep returning to. Constantine and Licinius, co-emperors of Rome, issued a decree granting religious tolerance to Christians and every other faith in the empire. What makes it remarkable is the context. The Roman state had been persecuting Christians for decades. Constantine himself would spend the rest of his reign favouring Christianity above other religions. Yet the edict did not declare one faith superior. It declared tolerance. It allowed diversity of belief inside a system that had been designed to crush it.

That is what tolerance does for art. It creates space for the unfamiliar to exist without the artist needing to defeat it first. A novelist who cannot tolerate a character whose values oppose the novelist’s own values will write a flat character. A playwright who cannot tolerate ambiguity will write a sermon. Tolerance is not agreement. It is the willingness to let something you do not share breathe on the page.

Openness is the second condition. Its opposite is the closed system. Ideology. The mind that has already decided what it thinks before it looks. The reflex to protect a position rather than test it.

Marcus Aurelius is the example I reach for. A Roman emperor, the most powerful man in the known world, who spent his private hours writing notes to himself about how little he knew. The Meditations are not a book of certainties. They are a record of a man trying to stay open while everything around him, the court, the wars, the weight of empire, was pressing him to close down. He wrote about his own failings the way a scientist writes about a failed experiment. Not with shame. With attention.

For a writer, openness means entering a draft without knowing the ending. It means following a character into a room the writer did not plan. It means letting the material talk back. A closed writer produces efficient prose. An open writer produces prose that surprises the writer first and the reader second.

Curiosity is the third condition. The world spends enormous energy trying to kill it. We tell children that curiosity killed the cat. We build school systems that reward correct answers and punish odd questions. By adulthood most people have learned to stop asking.

Hokusai, the Japanese printmaker, is the example that stays with me. At seventy-three he wrote that nothing he had produced before the age of seventy was worth counting. At seventy-five he said he had learned a little about the real structure of nature. At eighty, he hoped, he might make some progress. At ninety, he expected to get close to the essence of art. At one hundred, he wrote, every dot and every stroke might be alive. He died at eighty-eight, still reaching.

That is curiosity as a life practice. Not a phase. Not a hobby. The refusal to declare the work finished while the artist is still breathing. Every serious writer I have known shares some version of this refusal. The ones who stop being curious are the ones who start repeating themselves.

TOC, together, defines the creative space where art gets made. Tolerance clears the ground. Openness lets the material arrive. Curiosity keeps the artist moving into territory that has not been mapped. Without all three the imagination starves. It can still produce content. It cannot produce work that changes a reader.

Now the second part. The moves inside the work. I call them POG. Place, object, gesture.

The three examples I carry with me are the suitcase, the puddle, and the glass. Each one contains all three elements. Each one moves a reader or a viewer without explaining why.

I told you the first one already. Place: a train station platform, where people arrive and depart. Object: a suitcase. Gesture: a white sheriff picking up a Black man’s bag and carrying it to the train. No one explains the gesture. The act itself transmits a feeling that lingers long after the screen goes dark. Stirling Silliphant wrote it. He won the Oscar for the screenplay. Norman Jewison directed it. Sidney Poitier and Rod Steiger played the two men. Between them, in one silent exchange, they said more about human dignity than a hundred political speeches.

Stirling told me that story over dinner in Bangkok, more than thirty years ago. I cannot prove it happened the way I remember it. There is no surviving draft with a yellow tab still attached. No producer’s memo. No other witness left alive to ask. I have only what Stirling told me, and what I have kept of it since.

The second example comes from George Orwell. Burma, a wet morning in the 1920s. Orwell, then a young colonial policeman, walks behind a condemned prisoner toward the gallows. Two warders hold the man by the arms. Somewhere in the forty yards between the cell and the rope, the prisoner steps to one side. He is avoiding a puddle. He has minutes to live. He does not want wet feet.

Orwell wrote later that until that moment he had never grasped what it means to destroy a healthy, conscious man. Not the cell. Not the rope. The puddle. A living body, still careful, still keeping itself dry for a life it was about to lose.

Place: the prison yard, forty yards wide. Object: the puddle. Gesture: the sidestep. No one in that essay argues against the death penalty. Orwell never says the word colonialism. The wrongness comes through the body. One small useless act. The reader feels it before the reader can reason about it. An argument can be answered. The sidestep cannot.

The third example comes from Thailand, where I have lived since 1988. Ajahn Chah was a Buddhist monk in the Thai forest tradition. He died in 1992. His teaching on impermanence is the one that has stayed with me longest.

He held up a drinking glass. “For me,” he said, “this glass is already broken. I enjoy it. I drink out of it. It holds my water admirably. If I tap it, it has a lovely ring to it. But when the wind knocks it over or my elbow brushes it off the table and it falls to the ground and shatters, I say, of course. When I understand that the glass is already broken, every moment with it is precious.”

Place: a room, a monastery, a forest in northeast Thailand. Object: a drinking glass. Gesture: tapping the glass with his fingernail, listening to the ring. The glass is useable and broken at the same time. Only time separates the two states. The tap is the gesture that makes the teaching land. Not the words. The sound of a fingernail against glass.

In all three cases, no one explains the gesture. No one makes a speech about what it means. The act itself carries the meaning. A speech would have weakened it. An explanation would have killed it.

This is the part where AI enters the room. And this is where the conversation gets honest.

AI is extraordinary at searching, finding, organising, analysing, and reporting. I work with AI systems every day. I have built a working constitution for how I collaborate with Claude, the AI made by Anthropic. I use AI as an editorial partner on my essays. It has changed my practice and improved my work. I am not writing this as someone who fears the technology.

But AI cannot do TOC. And AI cannot generate POG.

Start with tolerance. Tolerance exists because its opposite, intolerance, is the human default. The push against intolerance is what gives tolerance its weight. An AI can be configured to treat all inputs equally. That is not tolerance. That is indifference. Tolerance requires something to push against. A body that feels the tribal pull and chooses to resist it. There is no divide in the AI world between tolerant AIs and intolerant ones. There is only calibration.

Openness requires the possibility of being changed by what you encounter. An AI processes a prompt and generates a response. It does not come out the other side altered. A writer who truly enters the material comes out different from the writer who went in. The character talks back. The scene refuses to go where the outline says. The writer’s assumptions crack, and something better grows in the fracture. That openness is a feature of being alive, and of having something at stake.

Curiosity requires not knowing, and wanting to know, and being willing to be wrong about what you find. An AI can simulate the seeking. It cannot feel the itch. The itch is the thing. Hokusai at seventy-five, saying he had learned a little about the structure of nature. That “little” is the sound of a man who has been looking for decades and still cannot look away. The AI has no decades. It has no accumulated frustration with its own limits. It has no version of “I am not there yet” that runs deeper than a token prediction.

And POG is even further out of reach. The suitcase on the platform works because Silliphant did not plan it as a symbol. Nobody sat down and said, “We need a gesture here that encodes racial reconciliation.” The scene survived because it was the one thing, after everyone had picked over everything else, that no longer needed defending. That is how the puddle gets found. After the seventh draft. After the cuts. After an editor or a producer beside the writer has stripped away everything the story does not need, and what remains is the thing that was carrying the weight all along.

AI can assemble a scene with a place, an object, and a gesture. It can do this on command. The result will look like a POG. But it will not be one. Because the puddle is not designed in advance. It reveals itself through honest rewriting. Looking for it before that point is writing toward a symbol instead of toward a story. And a symbol assembled on purpose tends to announce itself, the way a planted clue does in a bad mystery.

The fear of AI in the cultural space is, I think, misplaced. AI will not replace the writer who works inside TOC and discovers POG through the long, messy, resistant process of making something real. That writer’s position is secure for the same reason a handshake on a platform in 1967 Mississippi is secure. You cannot fake the gesture. You can only do it or not do it.

Where AI will change things, and where the danger actually lives, is in corporate and governmental control. Propaganda that has been socially and psychologically designed to feel creative and powerful. Content that wears the costume of art but serves the machinery of persuasion. That is the wolf in sheep’s clothing. It will look like imagination at work. It will be the opposite.

Stirling Silliphant died in Bangkok in 1996. I still think about those yellow tabs. Dozens at the start. One at the end. The one that survived was the scene nobody needed to argue about because its meaning was carried in a man’s hands, not his words.

If you are a writer, or if you are someone who cares about what writing can do when it is working at its best, remember the tab that survived. The gesture that needs no speech. The puddle that cannot be answered. The glass that is already broken.

Those are the things that change hearts. They always have been. No algorithm wrote them. No algorithm will.

A reader walks into a bookshop and finds the crime shelf. The sign on the shelf says thriller, mystery, hardboiled, noir or crime fiction. She picks one. She has bought a promise. The word on the back cover tells her what she is holding. She trusts that the publisher, the editor, and the author all mean by ‘noir’ or ‘thriller’ what she means. None of them do, quite. The word is a large bag carried as a small case. Inside it, meanings overlap and contradict. She buys the book anyway. The name has already done its work, and the author’s name on the cover backs it. A word tells her what to expect, and the doubt disappears.

We do this with everything. Name an event, a feeling, a fact, and the uncertainty around it seems to dissolve. Say the word and the thing feels known. Friend. Lover. Family. Courage. We use these as if each carried one settled meaning. They do not. Ask two people at the same dinner table what a friend owes a friend, and watch the answers pull apart. The word held the table together only while nobody examined it.

Communication runs on common ground. A word carries meaning between two people only if both are standing on the same patch of it. We assume the ground is there and shared, and most of the time we never test the assumption. That assumption is what lets a conversation move at speed. It is also what lets a conversation go wrong while both speakers believe it is going well. The ground that looks common is often private to each of us. The gap stays hidden for as long as the word goes unquestioned.

The act of naming is the territory the novel explores. Not to settle the baggy word, but to hold it open long enough for a reader to look inside. Graham Greene published The Quiet American in 1955. Phuong prepares Fowler’s opium and shares his bed. She leaves him for a younger American, a spy, who is later murdered. Did she love Fowler? Did he love her? Greene lays out the evidence and withholds the verdict. Two years later Richard Mason published The World of Suzie Wong. The same puzzle, moved to Hong Kong. Another expatriate man reading a woman whose feeling he cannot read off her face. I wrote my own version in A Killing Smile, a Los Angeles lawyer and a Bangkok woman, the same gap between them.

A cross-cultural setting sharpens the problem. Put the two people on opposite sides of a colonial line and a cash transaction, and every feeling-word is mistranslated twice, once in each direction. The distance is the use of it. It gives the reader a safe place to stand while examining words like love and friendship that behave the same way closer to home. You do not need a passport to Saigon or Hong Kong or Bangkok to find the naming problem on your own doorstep. Gender widens it. So does generation.

Naming looks fixed and is not. Would the reader who opened The Quiet American in 1955 read Phuong the way a reader does in 2026? The word love sat still on the page while its meaning moved underneath. The drift is slow and then, at some distant point, wide enough that the author is no longer naming anything the reader recognises. The book has not changed. The word has.

Readers come to these books to test a word. Love. Commitment. Dignity. The story refuses to define it. There is now a whole genre of online video doing the same work. Strangers dissect the same expatriate pairings frame by frame, asking what she feels and whether he knows. The exercise seeds doubt into words neither side had thought to question. Both are trying to learn why the other’s words will not translate. The story hands the doubt back and lets the characters keep clutching their certainty.

The same naming move runs through science. A word gets set on a shelf and sold as if one meaning sat behind it. Quantum. Consciousness. Dark matter. Each is a large bag carried as a small case. Dark matter is the clean example. We cannot see it. We infer it from galaxies that spin faster than their visible mass allows. The name suggests a substance located and described. It is closer to a signpost. It marks the spot where the measurement runs out. The naming reflex is strongest exactly there, at the edge of the instrument. Give the gap a memorable name and the problem feels handled. It is not. The name marks the debt and never pays it. Say the words dark matter and you feel let into the secret. That feeling is the trap. Nothing has crossed into your understanding but the label. Your ignorance sits where it always sat, now wearing a name.

Set the two words side by side. Love. Dark matter. One belongs to the humanities and one to the sciences, and they are the same kind of word. Both name something we are sure is real. Both name something we cannot measure or hold. Both are argued over by serious people who will not agree on what they have named. Neither name removes the doubt about the thing it points to. This is the thread running under the whole false divide between the arts and the sciences. Each field keeps a set of words that stand in for what it cannot yet pin down. Each field agrees to trust the word so the work can go on.

We treat certainty as a trade against doubt, and physics takes the trade literally. An electron has no single position waiting to be found. It has a spread of probabilities. Measure the position and you sharpen it, and the momentum blurs by the same amount. Pin one down and the other slips away. This is not a fault in the instrument. It is the structure of the thing. The measured value looks like the whole truth and is a slice of it. Every gain in the one is charged against the other. Push the doubt down toward zero, and the certainty is supposed to rise and fill the room. Run the trade to its end and you reach the promise of total certainty, doubt at zero, nothing left to ask. That promise is the tell. Erasing doubt to zero is the core skill of the confidence man. The market tip that cannot lose. The cure with no side effect. The candidate who is never wrong. Certainty sold as complete is certainty sold as a con.

Honest science never makes that sale. It runs on probability, and the number never reaches zero. A result holds at a confidence level. A theory stands until something falsifies it. The doubt stays on the books as a permanent line. The honest novel behaves the same way. It does not tell you whether Phuong loved Fowler. It keeps the question open because the open question is the truer thing. In both rooms the discipline is one refusal, the refusal to drive doubt to zero. Certainty as a zero-sum game is the game the con man plays. The scientist and the novelist decline to sit down at that table.

Werner Herzog drew the line I have been circling. I found my way to it through an essay by J F Martel in Aeon, which takes the same distinction toward a different end. In 1999 Herzog stood up in Minnesota and made a short declaration about documentary film. He named two kinds of truth. There is the truth of accountants, reached by tallying facts and stating them plainly. The phone book is full of it. Every entry verifiable, every name correct. And there is a deeper truth he called ecstatic, reached only through imagination and shaping and a measure of invention. His own example put it best. The phone book lists a man’s name and address correctly and cannot tell you why he cries himself to sleep. Both truths are real. They lean on each other. The error is to behave as if the ledger held everything.

The word ecstatic comes from the Greek for standing outside oneself. That is the work the deeper truth does. It takes you out of your own position and sets you down somewhere you could not have reached by counting. It is also the reason we reach for words like awe, wonder, and beauty. Those are the names we put on the moment our sense of the real gets larger than it was. Even there, at the height of the feeling, the naming reflex arrives to tidy the doubt away.

There is art that only serves the ledger. It confirms what the audience already holds. It flatters the settled world and changes nothing in it. Herzog’s target was the documentary that mistakes its footage for the whole truth. The wider family is larger than that. Propaganda. Advertising. The sentimental picture that tells you what you already feel. This work can be skilful and can move people. It does not carry anyone outside themselves. It is the sidekick who confirms whatever the boss says is real. Against it stands the art that makes present something not previously in view. Not invented from nothing. Found, by someone who walked to the edge of the agreed world and came back able to show it.

New patterns come out of tension that stays unresolved. To hold a contradiction in mind and not rush to settle it is rare and hard. Most of us grab the name that closes the question. The few who find something new can stay in the open state a little longer, holding the tension instead of relieving it. Novelty is a human drive of its own. It is the pull to act on a strong want whose outcome cannot be known in advance. Creation happens in the event, before the event cools into a fact. The universe of settled fact is a finished thing, sorted and shelved. The artist works upstream of that, in the moving water, where the real is still deciding what it will be.

The machine we now consult, our AI companion, is an accountant of genius. It has read the ledger of nearly everything we have written and can state the plain fact faster than any of us. Its truth is the truth of accountants. It tallies, sorts, and returns the consensus of the record. What it cannot do is stand outside itself. It has no position to leave. It has no senses to distrust, and no world beyond the words it was trained on. It has no body, no nervous system running hot and unreliable, nothing turning raw sensation into experience. It cannot walk to the edge of the agreed world, because it is made entirely of the agreed world. The ecstatic truth, the one you reach by leaving your own vantage, is shut to it. That gap is not a passing limit of the current model. It is the whole difference. Art is what takes us outside ourselves. A system with no self to leave has nowhere to go.

Homer knew the exit long before any of this. A man leaves Troy and cannot get home. For ten years the sea keeps deciding against him. We do not read the Odyssey to learn the coastline of the Aegean. We read it to be carried out of our own small position and set down in his. That is the standing-outside the Greeks named. The certainty we are sold, doubt driven to zero, is the accountant’s dream and the con man’s product. The doubt we are taught to fear is the thing the artist and the scientist both keep and put to work.

Refuse the clean name that shuts the question too soon. To question a name is to break a small contract, the shared agreement to treat the word as settled. The contract keeps the peace and dims the view. The books worth keeping for thirty years are the ones that hand you back your doubt, larger than before, and somehow easier to carry.

———

A note on sources. This essay began with J F Martel’s How art invented humanity, published in Aeon in 2026. Martel and I both draw on Werner Herzog’s split between the truth of accountants and ecstatic truth, which Herzog set out in his Minnesota Declaration in 1999. Martel takes it toward a claim about how art made us human. I have taken it toward doubt, naming, and the false comfort of a settled word. His essay runs a different course from mine and rewards the read. (aeon.co/essays/humans-did-not-invent-art-it-was-the-other-way-around)

I. The Galaxy Rose

In The Client That Wasn’t There, Bangkok runs on the Mesh. The year is 2036. The Mesh is a surveillance infrastructure so complete that every citizen’s movements, emotions, deleted texts, and biometric readings are logged, monetised, and sold. Mandatory augmented-reality glasses connect each citizen to a system that, in the novel’s phrase, understands them better than they understand themselves. Privacy is not illegal. It has been optimised away.

The technology never sits inside a single passage of explanation. It accumulates the way weather does. Vincent Calvino, the private investigator at the centre of the novel, walks into a bar on Soi Cowboy and the world reveals itself in layers. He removes his glasses before entering. The gesture already tells you what the glasses are, because without them the city feels amputated. As he steps through the door, “a soft cloud of sensor vapor kissed his wrist, sampling pulse, skin temperature, and emotional signature for the bar’s loyalty algorithm and logged him into the Empathy Bureau’s shadow files.” The door’s AI greets him by name. Inside, ceiling drones the size of birds drift on silent rotors, “recording angles of desire for later analysis.” Korean businessmen sit with their glasses off, “empty, glazed-over eyes. Blinking as though trying to reconnect their memory to the feed they’d lost.”

That is the infrastructure. The scene does its work through what happens after. An AI called Niran, embedded in the Mesh’s architecture, reaches Calvino through his phone. She reads him and the stranger confronting him at the same time, in real time. Posture, breathing rate, behavioural history, emotional state, who in the room owes what to whom. She gives Calvino corrections at a granularity that would make a body-language expert sound vague. “Slow your breath. Drop your right shoulder two centimetres. Tilt your chin one degree down. Humans read the arc before they hear the words.” She knows the gambler hates leaving in the rain. She compresses six thousand tokens of behavioural prediction into twelve words of tactical advice. Calvino defuses the confrontation with information he could not have possessed without her.

The glasses do not merely overlay digital information on physical space. They are the interface to an economy that reads the wearer’s internal states and the internal states of everyone the wearer encounters. The reading is packaged as data. The data is sold back as mood optimisation, social coaching, and commercial targeting. The novel calls this the “empathy economy.” The system harvests human emotion as product. The powerful pay to edit their own emotional records, to remember their cruelties as necessities and their privileges as earned. Everyone else carries perfect recall of their own erosion.

This essay is not concerned with whether that world is plausible. The concern is how far away it is. Line up what is already shipping in 2026 and the answer is closer than most people would like. The assembled product is not yet on shelves. Every component from which the product would be assembled is already shipping, already working, and already owned by parties other than the person whose data it reads.

· · ·

II. The Components

The ownership structure

Let’s begin with the part that requires no technological leap. The journey begins with The Neurorights Foundation, founded by Rafael Yuste at Columbia University. Yuste examined the privacy policies and user agreements of thirty companies manufacturing consumer neurotechnology products. Twenty-nine of the thirty retained access to consumers’ neural data without any meaningful limitation. Here’s the takeaway: More than half explicitly reserved the right to share that data with third parties. What is sobering is the data those devices hover up, includes user data about brain states, attention, emotional responses, and moments of vulnerability and decision fatigue, is in twenty-nine of thirty cases legally and operationally owned by the companies collecting it. One company out of thirty placed meaningful restrictions on its own use. The other twenty-nine did not.

It’s time to put a price tag on this market. A 2022 estimate put the neurotechnology market worth at about $9.8 billion. By 2026 the projection is $17.1 billion. Meta, Apple, Sony, and Microsoft are embedding neural and neuromotor sensors into consumer products such as earbuds, wristbands, and augmented-reality headsets. For example, Meta’s Neural Band, released in September 2025 alongside the Ray-Ban Display glasses, reads your electromyographic signals from your wrist at a precision fine enough to capture single motor-unit activity. Muse’s Athena headband (released in March 2025) adds functional near-infrared spectroscopy to consumer EEG. The technical envelope has grown wider than most consumers know. These companies have kept the door open that most consumers miss in the fine print. The privacy policies of the major AR and wearable manufacturers contain language broadly worded to capture your neural data without you noticing.

This is the layer of my novel’s world that is already operational. The Mesh in The Client That Wasn’t There does not require a villain. It does not require coercion. It requires a terms-of-service agreement that nobody reads, accepted by a consumer who wanted bundled inside the product, a legal transfer of ownership over data generated by their own nervous system. The bar on Soi Cowboy logs Calvino into the Empathy Bureau’s shadow files the moment the sensor vapor touches his wrist. In 2026 the logging begins the moment you put on the headset and tap Accept.

The regulatory response is starting but is already behind the curve. Chile amended its constitution in October 2021 (Law 21.383, Article 19) to recognise neurorights, the first country to do so. Four U.S. states have now enacted legislation treating neural data as a category of sensitive personal data: Colorado, California, Montana and Connecticut. Six other states have bills pending. There’s also a growing global governance component. In November 2025 UNESCO’s General Conference adopted the Recommendation on the Ethics of Neurotechnology, the first global instrument addressing mental privacy and consent in commercial neurotechnology.

The Neurorights Foundation’s finding still stands. Twenty-nine of thirty. Here’s the worry. The law is not adjusting to the speed of changes in the terms of service. The terms of service are running behind the product. And the product is running behind the revenue model. That sequence is the novel’s structural prediction. It is already accurate.

The decoding capability

How do AI systems find meaning in the data collected? In 2023, researchers published a paper in Nature Neurosciencedescribing a system they called a semantic decoder. This is how it worked: a participant lies inside an fMRI scanner. The scanner reads blood-flow patterns in that person’s brain. An AI model translates those patterns into natural language. Not single words. Not category labels. Instead, it calculates the essence of what the person was hearing, reading, or imagining, generated in real time. If a participant heard a phrase like: “I didn’t even have my driver’s license yet,” the decoded output interpretation might be: “she hasn’t learned to drive yet.” Not the words. But the meaning of the words. The semantic content. The thought.

The system took about sixteen hours of individual training in the scanner. It could not function without the subject’s active cooperation. In 2023 those looked like the defining constraints. The average consumer isn’t going to submit to a 16-hour scan even if it were free.

That was in 2023. Things move fast and the old constraints no longer apply. By February 2025 the same team published a follow-up in Current Biology where they demonstrated a converter algorithm that adapted the trained decoder to a new individual using about one hour of fMRI data. They collected data while the participant watched silent video clips rather than listening to language. The sixteen-hour barrier was reduced to a one-hour barrier in under two years. A month later in March 2025 a separate research group published BrainLLM in Communications Biology. The researchers integrated brain recordings directly with a large language model and generated natural language output from fMRI signals. The brain signals added information that the language model did not already possess. The decoder was reading something. Imperfectly. Inside a laboratory. But reading it.

Niran’s reading of the gambler in the Galaxy Rose runs on the same principle extended to its endpoint. An AI system takes the physical activity of a person’s nervous system and converts it into an actionable model of what that person is thinking, feeling, and about to do. The distance between the 2023 semantic decoder and Niran’s real-time tactical output is not a difference in kind. It is a difference in resolution, speed, and form factor. The novel asks what happens when this capability leaves the laboratory. The 2026 evidence suggests the question is no longer hypothetical. It is logistical. Doing what big tech already do by concentrating on: Energy. Compute. Money. Marketing.

· · ·

III. The Gap, and What It Consists Of

The components are shipping separately. The obvious question is what prevents their assembly. The answer has four parts, and the essay’s argument depends on being honest about all of them.

The resolution gap

The semantic decoder works at fMRI resolution. An fMRI scanner is a machine the size of a room, costs millions of dollars, and requires its subject to lie perfectly still for hours. Huth has stated that his approach should translate to fNIRS, functional near-infrared spectroscopy, a portable technology that measures the same underlying signal using wearable sensors. fNIRS resolution is orders of magnitude coarser than fMRI. Consumer EEG in earbuds is coarser still.

Niran reads the gambler at a granularity that requires resolving individual emotional states, predicting specific behavioural responses, and reading the balance of force between two people in a room. From wearable sensors. In real time. Without the subject’s cooperation. Nothing in the 2026 consumer pipeline gets close to that resolution from a portable device. The technology to bridge this gap does not yet exist. The direction does. Laboratory fMRI. Portable fNIRS. Consumer wearables embedded in AR glasses. Each step loses resolution. Each step scales up the ladder. The novel imagines the end game where resolution has recovered and ubiquity is total. Getting there requires sensor physics that is currently in early research. Optode density. Signal processing. Noise rejection. None of it demonstrated yet outside controlled settings.

This is the hardest gap for me to close. It is also the gap that AI is showing a capacity to narrow. Every improvement in the decoder’s ability to extract meaning from noisy, low-resolution signals shortens the distance between the scanner and the wearable. BrainLLM is precisely such an improvement. The decoder does not need the signal to be ‘clean.’ The decoder only needs the signal to be informative. AI has proven good at extracting information from signals that humans cannot interpret at all. To us, standing on the outside, it looks more like magic than science.

The integration gap

The novel’s glasses are a single device that combines augmented-reality overlay, biometric sensing, emotional-state classification, behavioural prediction, lifelong memory-feed archival, and real-time connection to a centralised AI. In 2026 each of those capabilities remains at a different stage of maturity, built by different companies, running on different platforms. Meta builds AR headsets and now neuromotor wristbands. Apple builds biometric sensors and the Vision Pro. University labs build semantic decoders. No single product unifies them yet.

Integration is historically the easy part. Once the components exist, commercial incentive does the assembly. The smartphone was not a scientific breakthrough. It was an integration event. A phone, a camera, a GPS receiver, an internet browser, an accelerometer. None of those functions was new. But once bundled into a single object they restructured daily life inside a decade. The scientific evidence suggests that the AR headset is on the same trajectory. Meta’s devices already combine visual overlay with biometric tracking and an EMG neural input. Adding fNIRS-based neural-sensor data to the earbuds already in production is not a physics problem. It is a product-management decision. The Mesh is what happens when that decision is made and nobody objects, because every addition was marketed as a feature.

The consent gap

The semantic decoder of 2023 required sixteen hours of individual training and the active cooperation of the subject. In 2026 the comparable barrier is closer to one hour and the input data can be passive (silent video) rather than active (listening to spoken language). You cannot be mind-read against your will from a pair of glasses with this technology in 2026. The next frontier, and the one explored in The Client That Wasn’t There was the capacity to bundle surveillance into the device.

The novel’s insight is that this barrier does not fall because someone invents a single breakthrough that defeats it. It falls because the concept of consent is restructured. The sensor vapor at the door of the Galaxy Rose does not ask permission. It does not need to. Permission was given upstream, when the citizen accepted the terms of the system the sensor vapor belongs to. Terms the users in 2036 Bangkok believed the marketing hype that it was about as optimization and care.

This pattern is already visible in 2026. Remember that the Neurorights Foundation found that twenty-nine of the thirty companies it reviewed retained broad access to consumers’ neural data with no meaningful restriction. The privacy policies of major AR and wearable manufacturers contain language broad enough to encompass your brain-state data. But you, the consumer, haven’t consented to mind-reading. You think that as a consumer you’ve consented to a product. The product includes a data agreement. The data agreement includes language that could, as AI decoding capability improves, cover neural data the consumer did not know was being collected and could not have imagined would become interpretable. The consent is given before the capability exists to exploit it. That is the mechanism. It is already in operation. And by 2036 the novel predicts the exploitation will be firmly embedded in daily life.

All indications suggest that the training requirement will keep compressing. Every improvement in transfer learning, the AI technique that allows a model trained on one person’s data to generalise to another’s, reduces the calibration burden. As training data accumulates across millions of wearable-device users, the individual calibration period shrinks. The direction is clear. From sixteen hours to one hour to the time it takes the device to learn your patterns from ambient use. Passive calibration. Continuous reading. Consent given once, at purchase, for a capability that did not exist at the time of signing but arrives through a software update the terms of service already covered.

· · ·

IV. Surveillance Rebranded as Connection

My argument in this essay is not that the glasses in The Client That Wasn’t There are ten years from shipping. The resolution gap is real. The integration has yet to happen. Government regulation may intervene to slow or stop the integration. Meanwhile, the semantic decoder remains locked inside a laboratory.

My argument is that every structural precondition for the glasses’ world is already in place.

The sensors are in consumer products. The biometric data is currently being collected. The neural data is being generated by devices (including me with my tracking watch) that millions of people wear voluntarily. The ownership of your (and my) data has been transferred, twenty-nine times out of thirty, to the companies who have a financial incentive to collect it and turn it into a revenue stream. The AI capable of decoding what the sensors capture has been published in peer-reviewed journals. The training-time barrier that defined the technology in 2023 has fallen by an order of magnitude. The researcher who built the most advanced semantic decoder has stated that his approach should transfer to portable wearable form. The legal framework that would decide whose mind belongs to whom is losing the race to the terms of service, even as four U.S. states and UNESCO begin to legislate.

The novel’s 2036 is not a prediction about a device. It is a prediction about an environment. A social, economic and political ecology in which the device becomes possible because every precondition was met separately, marketed as something else, and accepted without resistance because each concession seemed too small to argue with. Nobody consented to the Mesh. They consented to the earbuds. They consented to the fitness tracker. They consented to the AR overlay that made the commute more interesting. They consented to the privacy policy they scrolled past because the product was already in their hand and the alternative was to put it back in the box.

Calvino removes his glasses before entering the Galaxy Rose. That single gesture tells you everything about the world the glasses have created. Without them the Korean businessmen sit with empty, glazed-over eyes, blinking as though trying to reconnect their memory to the feed they have lost. The glasses are not a surveillance device imposed from above. They are a dependency adopted from within. The Mesh does not surveil its citizens. It connects them. It optimises them. It understands them better than they understand themselves. That is its marketing copy. The marketing copy is not a lie. It is a reframing so complete that the population it describes cannot see what has been taken, because they experience the taking as a service.

The question the novel poses is what if anything remains private once AI can trace every connection in your brain. By 2036 that is the wrong question. By then the answer is nothing, and the population has adjusted. They’ve normalised life inside the Mesh.

In 2026 we have time to reframe the question. Who owns the map? Who owns the model of your attention, your emotional responses, your moments of vulnerability and decision fatigue? Who owns the statistical representation of how your brain works, built from your brain, on a device you purchased, under terms you accepted?

The Neurorights Foundation already has the answer. Twenty-nine out of thirty.

In the novel, the Mesh’s architecture includes a two-tier memory system. The powerful pay to edit their own emotional records. Everyone else carries perfect recall of their own erosion. Kiat Boonruang discovered this and was erased. Not murdered. Not imprisoned. Administratively deleted, his records corrected, his existence reduced to a clerical error no one could appeal.

The horror of the novel is not the technology.

The horror is the infrastructure of consent that makes the technology possible.

The accumulated, incremental, freely given permissions that, taken together, constitute a surrender so complete that the person who made it cannot reconstruct what they lost, because the system that took it is the same system that holds their memory.

That infrastructure is being built now. Not in 2036. Not in a novel. In the privacy policy of the device on your head, the sensor in your ear, the wristband on your arm. The components are already shipping. You’ve already given your consent. The question is whether anyone notices before the product is assembled. Or whether, like the Korean businessmen in the Galaxy Rose, we take off the glasses one day and find we cannot remember what we were looking at without them.

· · ·

I. The Corridor of Payphones

Early in Neuromancer, Case walks down a corridor in a Chiba City hotel and a payphone rings as he passes it. He keeps moving. The next one rings. Then the next, tracking him along the length of the corridor, and when he finally lifts a handset there is nobody on the line who has a body. Wintermute, the artificial intelligence at the centre of the novel, is trying to reach him. The most sophisticated non-human mind in twentieth-century science fiction has to do it by ringing payphones in sequence.

William Gibson published that in 1984, having written the book on a manual typewriter. He did not own a computer. The novel gave English its word for the space behind the screen, described a global network as a consensual hallucination years before most of its readers had touched a modem, and understood before almost anyone that the interesting property of a network lies in the traffic between machines rather than in the machines themselves. Gibson has also produced the best short statement anyone has managed about technological change, eleven words long, which he has been repeating in interviews since the nineties: the future is already here, it is just not evenly distributed.

And he had his AI ring payphones.

Forty years on, nobody in Tokyo or Bangkok or New York or London uses a phone bolted to a wall. The scene is now a period piece embedded in a book about the future, and readers coming to Neuromancer for the first time notice it the way they notice the cigarettes. What interests me is that he was wrong in a direction, and that the direction can be predicted in advance, in any novel of this kind, including my own.

· · ·

II. What a Novel Requires of a Machine

Forty years of writing fiction has taught me that technology in a novel is never information. It works as the nature of a relationship between people. Scenes build the scaffolding, and the technology sits inside it doing something to what passes between characters, and where it fails to do that, no amount of authorial enthusiasm will keep a reader on the page. Science fiction is often read as prophecy, which I believe misdescribes what the writer is actually doing at the desk. The writer is looking for trouble, for pressure points and stress junctures. Technology earns its place in a story by generating danger, opportunity, temptation or dependency between people, and everything else about it is furniture.

Which means a device or machine has to satisfy three conditions before a novelist can use it.

It has to be an object a character can handle. Something with edges, weight and location, that can be dropped, stolen, hidden, broken or held out at arm’s length. Molly’s mirrored lenses are surgical and therefore permanent, which is itself a plot condition. Case’s deck sits in his lap and can be taken from him.

It has to offer a place a character can go. Fiction is spatial before it is anything else. Characters move, arrive, and are barred at doors, and the reader’s attention follows the body. Gibson opens Neuromancer in Chiba City, across the bay from Tokyo, then leaves it for the Sprawl, for Istanbul, for an orbital resort, because a novel needs somewhere to arrive next. He also built a second geography behind the screen. Gibson’s cyberspace is a landscape with coordinates and horizons because a landscape can hold a scene, and a scene is the only unit a novel actually spends.

It has to permit a gesture. Somebody has to be able to do something with it that means something, witnessed by another character, in a way that moves the story ahead. The gesture is where the moral weight of a technology becomes visible, and it is the writer’s only reliable instrument for showing what a machine has done to a population, to one person’s goals, and to the methods that person will now use pursuing them. It also fixes the motives of everyone standing in relation to that person.

An object travels and a gesture does not. A thing with edges means roughly the same wherever it is set down, while a gesture means only what the surrounding code says it means, which is why writers working across cultures spend most of their attention on this one condition. Removing your glasses at a door in Bangkok is a different act from removing them at a door in Toronto, and the difference is the whole of what a scene is for. The same asymmetry explains something about the imagined futures that travelled furthest. Gibson gave a third of Neuromancer to Chiba City and built his Japan from Vancouver, out of what was reaching North America in 1984 as rumour and export. The result was enormously influential and it was a projection, and a generation of readers outside Japan learned to picture the technological future in a Japanese idiom that Japan had not supplied. Gibson and I come from the same city. He built his Japan from Vancouver, and I left Vancouver for Bangkok in 1988 and have been writing about it since, which is a different way of being wrong about a place rather than a cure for it. Proximity corrects some errors and installs others.

A payphone satisfies all three at once, which is why it is in the book. An object with a handset, it occupies a corridor and belongs to nobody. Lifting it is a gesture, and walking past a ringing one is a better gesture still. Gibson needed the most powerful intelligence in his novel to make contact with a man who had every reason to avoid it, and he needed that contact to be visible on a page. A phone ringing down a row of phones is superb writing, dread rendered in hardware.

The trouble is that the technology which actually arrived satisfies none of the three conditions well. A phone in a pocket is an object, barely, though it has no location worth describing and cannot be taken from a character without the scene becoming about theft. The network is nowhere. Nobody goes to it, because it came to us, dissolved into the ambient, and stopped being a place at all. Nor is there a gesture. Checking a phone is arguably a lapse of attention rather than an action, and a novelist attempting to write one discovers that it kills scenes rather than making them.

Which explains why, in 2026, almost nobody has written a good scene about a person using a smartphone, and why the writers who manage it do so by writing the interruption rather than the device. Forty years after Neuromancer, the defining technology of ordinary life remains close to unusable in fiction. Writers do not lack the skill. The thing offers nothing to hold, nowhere to go, and nothing to do.

· · ·

III. The Same Error from the Other End

Neal Stephenson published Snow Crash eight years after Neuromancer and got a different half of the future right.

Gibson’s strength was the machine. Stephenson’s was the social layer, and he read it with an accuracy that still startles. He saw that people would inhabit the network through constructed personas, that those personas would carry status, that virtual ground would be bought and sold, and that ideas would propagate through populations by the logic that governs infection. He gave the industry a word, and thirty years later the industry took it without modification and spent a great deal of money on it.

He also built it as a place. The Street runs for tens of thousands of kilometres, has real estate with addresses, has a door policy, and displays a hierarchy visible in the resolution of the avatars queuing outside. It is the most persuasive imagined space in the genre after Gibson’s, and it is persuasive for exactly the reason it turned out to be wrong. Stephenson needed somewhere for Hiro to go, somebody to bar the door, and a way for the reader to see status without being told about it. A place delivers all three in a paragraph.

What actually happened is that the social layer arrived in full and the place never did. We have the personas, the status economy, the propagating ideas, the commerce. We reach all of it through flat rectangles, in fragments, while queuing for coffee. Every element that could be dramatised failed to appear. Every element that resisted dramatisation arrived on schedule and at scale.

Two novelists, working eight years and one temperament apart, made errors that look opposite and are the same error. Both put the future in a form a novel could stage. The parts of their futures that were stageable are the parts that did not happen.

· · ·

IV. The Tax the Form Levies

Set those two beside each other and something usable appears, more modest and more specific than the word prediction suggests.

A novelist reading a technological trend is not doing what a forecaster does. The forecaster is trying to describe what will exist. The novelist is trying to build something a story can be told about, and those two searches point in different directions from the same evidence. When a writer stands in front of several plausible versions of a coming technology and has to choose one, the form does the choosing. The version with an object, a place and a gesture wins every time, and it wins for good reasons that have nothing to do with plausibility.

This gives a rule, and the rule is falsifiable, which is more than most claims in this territory manage. In any novel of the near future, the technologies rendered most vividly are the ones least likely to arrive in that form. Vividness is evidence of stageability, and stageability is what the real thing usually lacks. The things a novel handles clumsily, or mentions once and drops, or leaves as background hum, are where the writer has accidentally been accurate, because those are the things that resisted being made into scenes.

Read Neuromancer against that rule and it holds. The deck, the trodes, the aestheticised data landscape, all rendered with enormous care, all wrong. The corporate sovereignty Gibson mentions almost in passing, the erosion of states by private power, the ordinary criminality of information, all handled as background, all arrived. Read Snow Crash the same way. The Street is wrong and the memetics are right, and the memetics get a fraction of the description.

The rule predicts no technology. What it offers is a way of reading novels, including one’s own, and knowing which parts of them to trust. It says nothing about whether the writer read the direction correctly. Gibson read the direction better than almost anyone alive in 1984. What the rule identifies is a systematic bias in how a novelist renders the arrival, and a bias with a known sign can be corrected for, or at least declared.

The rule says nothing about the machines, which is the point of it. What a novelist is actually tracking sits on the other side of the transaction. Put new hardware on a street and the street answers, and the answer is the part of the book worth having. Some habits survive the upgrade unchanged. Others go behind glass, quaint, performed at festivals by people who were not there, and a few turn into symptoms carried into consulting rooms by patients who cannot say what happened to them. Gibson’s payphones are in a museum. Case is not. His paranoia, his contempt for his own body, his hustle inside a system whose edges he cannot find, all of it reads truer in 2026 than it did in 1984, and none of it required the hardware to be right.

· · ·

V. What I Expect to Have Got Wrong

My last three novels, Dance Me to the End of Time, District #3 and The Client That Wasn’t There, are set in a Bangkok of 2036 running on a surveillance system called the Mesh. In a companion essay I lay out the evidence that every component of that system is in production now. I stand behind the components. What follows is the part I expect to have got wrong, because an essay about other writers’ misreadings, written by someone who exempts himself from the rule, is worth nothing.

The Mesh reaches its citizens through mandatory augmented-reality glasses. Vincent Calvino removes his before walking into a bar on Soi Cowboy, and that gesture carries a good deal of the novel’s weight. It is the moment a reader understands what the glasses have done to everyone who keeps them on. The Korean businessmen at the tables have theirs off and sit with empty eyes, blinking, trying to reconnect to a feed that is no longer there.

Apply my own rule to that scene and it fails immediately. The glasses are an object. The bar is a place. Removing them is a gesture, and a good one, which is precisely the warning. I did not choose that form factor because the evidence pointed to it. I chose it because it let me write the scene, and I am not certain I knew that at the time.

If sensing goes ambient, none of it survives. Put the readers in the room, in the doorway, in the vehicle, in the building’s environmental system, and there is nothing on a face to remove. Calvino walks into the Galaxy Rose and takes off nothing, because there is nothing to take off, and the scene I built to carry the meaning of the novel cannot be written at all. I would bet on that version if I were betting rather than writing, and it is duller than anything in my book by a wide margin, being a matter of commercial property and municipal procurement rather than neuroscience, which is the other half of the rule working exactly as advertised.

Someone reading in 2036 will be able to run this over The Client That Wasn’t There without having to be clever about it. They will hold the advantage every reader of Neuromancer now holds over the man who wrote it, which is the advantage of standing at the arrival point and finding it strange that the traveller could not see it from the harbour. I would rather set out the rule myself than have it applied to me by somebody who thinks they have discovered something.

· · ·

VI. Working the Street

A rule about furniture is easy to accept in the abstract. It becomes harder when the furniture in question is the figure the reader came for.

Vincent Calvino has worked as a private investigator in Bangkok across nineteen novels. The form he belongs to was built by Hammett and Chandler on one condition, which is informational asymmetry. Institutions conceal, and a man with a licence, a tolerance for being hit, and no institution of his own goes out and finds out. Every convention follows from that: the legwork, the interviews conducted twice, the client who lies in the first chapter, the six hours in a parked car because there is no other way to learn where somebody goes at night.

Put the Mesh over that city and the condition disappears. Where were you on Tuesday night stops being a question in 2036 and becomes a lookup. A man watching a door for six hours is doing badly and slowly what the building does continuously and for nothing. If the private eye existed in order to find out, the private eye is finished, and the figure survives as costume, a man in a hat posed for photographs beside a vintage car. That was the version I did not want to write, and refusing it became the most useful constraint I had.

Total recording moves the scarcity rather than removing it. Once nothing is hidden, the concealed fact loses its value and the reliable one acquires it. A complete record belongs to whoever built it, and a record that can be read can also be written. In the Mesh the powerful pay to have their own archives corrected, so that cruelty is remembered as necessity and privilege as achievement, while everyone else carries perfect recall of their own erosion. The archive is total and partly fabricated, and nothing inside the system can tell you which parts.

So the job inverts. Calvino is no longer the man who finds out what was hidden. He is the man who can say which portions of a complete and public record were manufactured, in a city where saying it out loud is expensive. Readers who have followed the series notice the shift before they can name it. The legwork thins. The interviews change function, since he is watching what a person does when confronted with a version of themselves they cannot dispute, rather than collecting facts they might withhold. The danger moves too. Nobody needs to follow him any more, and the threat arrives instead as a quiet correction to his own file.

The noir premise survives all of this without a scratch and arrives strengthened. Noir was never about crime. It was about a system too large for one person to see the edges of, rigged in ways he can feel and cannot prove, where the best available outcome is knowing what happened and being unable to do much about it. Surveillance does not weaken that condition. It completes it. Chandler’s Los Angeles was corrupt and legible in fragments. The Mesh is corrupt and legible in full, which is worse, because a complete record that lies leaves nothing outside itself to check it against.

Whether I have solved that problem is not mine to say, and a reader in 2036 will have evidence I lack. What I can say is which part I expect to lose. The glasses will look like payphones. The Galaxy Rose will look like a room nobody has entered in a long time. The problem I set him, which is what a person does when the record of their life is complete, owned by somebody else, and partly false, is the part I would bet on, because it is a problem about people, and people are what survives the upgrade.

None of this argues for giving up on writing about where things are heading. It argues for being exact about what the writing can do. A novel makes a poor instrument for describing the future and a very good one for describing what a future does to the people living inside it, and that second job requires objects, places and gestures whether or not the future intends to supply them. The direction may hold up rather better than the furniture built around it. What keeps changing, faster than anyone can set it down, is the shape of the coastline, which leaves the bearing sound and the landfall unrecognisable.

Gibson knew where we were going. He simply could not have imagined that when we got there, the payphones would be gone.

Exhibit A. Every indictment needs one.

Read more

Your essay does something rare. It defines a word everyone uses and almost no one examines. The distinction between a collection of things and a system is where institutional analysis breaks down. By grounding the definition in relationships and behaviour over time, you hand readers a diagnostic tool. The bicycle example earns its place because the lesson travels. A reader can run the same test on a newsroom or a courtroom. The same questions arise. Are the parts influencing one another? What behaviour does that influence produce? That is clean analytic writing and it deserves the praise.

I have been reading Donella Meadows’ Thinking in Systems, which builds on the same foundation. Her key insight is that feedback loops drive system behaviour. Individual actors matter less than the loops they operate within. A reinforcing loop amplifies whatever enters it. A balancing loop stabilises. She ranked twelve points where intervention can shift a system’s output. Number six on her list is the structure of information flows. She also warned about “drift to low performance.” A system’s goals erode when corrupted inputs slowly replace the original standard. Both insights matter for what follows.

No system keeps its geometry over time. Like spacetime, systems bend and shift as neighbouring systems change and new ones connect. All systems seek equilibrium, but equilibrium is never permanent. It is a dynamic state. Equilibrium holds when inflows match outflows. The stock stays level. The balance lasts only while the current relationships stay stable and predictable.

The simplest illustration is a human population. Births are the input. Deaths are the output. When the two roughly match, the population holds. When births outpace deaths past a certain ratio, it grows. When deaths exceed births, it declines. The economic model in most industrial countries depends on growth. More people means more workers and more consumers. The reinforcing loop runs accordingly: growth funds infrastructure that supports further growth. But fertility rates across much of the developed world now sit below replacement. The stock is changing. The system’s behaviour will follow. This principle connects every example in this essay. System stability is never certain. An equilibrium that held for decades can collapse when a single relationship shifts. Physicists study the same fragility as a false vacuum, a state that looks stable while it sits one disturbance away from collapse.

Your essay focuses on relationships within a single system. The next question is what happens when information flows between systems. A media system feeds a political system. A political system feeds a legal system. A legal system feeds an economic system. When one pipeline carries tainted information, the damage does not stay local. It propagates through every downstream connection.

The quality of input

Information can be contaminated. Disinformation did not arrive with the internet. Its origins are ancient. Rulers have seeded false narratives for as long as writing has existed. The Roman Senate circulated forged documents to discredit rivals. Early modern states used pamphlets to manufacture consent. The technique never changes. Corrupt the information flow to serve whoever controls the input. The downstream system then produces the output the corrupter wants. What changes across centuries is the speed of propagation and the cost of production. The acceleration of bad information marks this moment as the point to pause and ask how a system defends its own value.

Tobacco: a case study

The tobacco industry gives the clearest pre-digital case. In December 1953, the heads of America’s major tobacco companies met at the Plaza Hotel in New York. Their own researchers had already established that cigarettes caused lung cancer. The executives knew the verified data. They also knew it would destroy their industry. Solution? They hired Hill and Knowlton, a leading public relations firm. In January 1954, the campaign began. A full-page advertisement titled “A Frank Statement to Cigarette Smokers” ran a pipeline straight to the public through more than 400 newspapers. It reached an estimated 43 million readers. Its core message was manufactured doubt. A 1969 Brown and Williamson memo put the strategy in five words. “Doubt is our product.”

The tobacco pipeline ran as follows. Tobacco companies funded research institutions built to produce confusion. Those studies entered the media system. The media delivered the confusion to the public. The companies bought time. The public, processing contradictory signals, delayed its demand for regulation. The regulatory system, receiving weakened political pressure, delayed intervention. The healthcare system absorbed the cost. Since that 1954 advertisement, roughly 16 million Americans have died from smoking. Annual smoking-related healthcare costs in the United States now exceed $225 billion. Researchers estimate three to seven million of those deaths were preventable. The industry knew the science. It chose doubt to protect its profits. Money over lives.

In Meadows’ terms, tobacco companies attacked information flow at its most potent point. They corrupted the balancing loop connecting scientific evidence to public policy. The reinforcing loop ran in the other direction. Continued smoking generated revenue. Revenue funded more doubt production. Doubt delayed regulation. Delayed regulation preserved the revenue stream. The campaign lasted forty-five years.

That playbook has not disappeared. It has been updated for a new ecology. AI and social media did not invent disinformation. They industrialised it. The tobacco strategy required newspaper advertising budgets and paid researchers at captured institutions. The same strategy now runs through algorithmic distribution at near-zero marginal cost. Algorithms select for emotional engagement, not accuracy. The content that travels farthest is often the most contaminated by outrage or fabrication. The old techniques survive inside the new architecture. Only the speed and scale have changed.

The digital cases

During the COVID-19 pandemic, social media became the primary channel for medical disinformation. False claims about ivermectin and hydroxychloroquine spread faster than peer-reviewed research. A study published in JAMA Network Open found a clear pattern. People who endorsed vaccine misinformation were three times more likely to take unproven drugs. Vaccine hesitancy, amplified by algorithmically sorted content, contributed to preventable deaths across multiple countries. Meadows warned about reinforcing loops that erode goals. The goal of public health was steadily replaced by the goal of tribal affiliation. Each wave of disinformation reduced trust in institutions. Reduced trust made the next wave more effective. The feedback structure locked in the decline.

Cambridge Analytica harvested personal data from roughly 87 million Facebook users without consent. It built psychographic profiles for political microtargeting. The 2016 US election and the Brexit referendum were among its targets. Christopher Wylie, who helped design the system, later described the damage across multiple democracies. Voters received content built to exploit psychological vulnerabilities. The system produced a more manipulable electorate. Cambridge Analytica proved that corrupting information flows works in reverse as well. The damage radiates through every connected system.

Publishing and film

The same structural analysis applies to the industries that make the culture. Publishing is a system. So is film. So is streaming television. Each holds the components, relationships, and feedback loops your essay describes. Each is now disrupted by inputs that did not exist a decade ago.

Publishing looks like one system but runs as several. Fiction, non-fiction, poetry, and drama each run on different feedback structures. Fiction runs on a reinforcing loop. Bestseller lists increase visibility. Visibility increases sales. Sales push the book higher on the list. Higher sales fund new titles and larger advances. The pipeline around a publisher is crowded: editors, copyeditors, designers, agents, scouts, reviewers, endorsers, awards juries, librarians, booksellers, publicists, and the audiences that celebrity authors bring with them.

Non-fiction runs partly on a credentialing loop. A university press publication builds an academic reputation that generates the next commission. Poetry has almost no commercial feedback loop at all. A poetry collection sells in the hundreds. The loop that sustains it is cultural prestige rather than revenue. Drama depends on production. A play is staged, reviewed, and either restaged or shelved. Each subsystem seeks its own equilibrium. Each is vulnerable at different points.

What decides what gets published is not one gatekeeper’s judgment. Several systems converge on a single editorial decision. The commercial system asks whether the book will earn back its advance. A celebrity with a large following is itself a pipeline, and publishers bid hard for the rights to it. The cultural system asks whether the book fits its moment. The distribution system asks whether it can reach readers. A legal review asks whether it carries risk. The editor who acquires a manuscript processes signals from all of them. That convergence is the relationship that sets the system’s output.

Film runs on the same principles at higher cost. The feedback loop that governed cinema for decades was straightforward. Theatrical release generated box-office data. Critics reviewed. Awards followed. Catalogue revenue sustained the studio. Streaming disrupted that equilibrium. The theatrical window compressed. Engagement metrics, not editorial judgment, now decide what a platform promotes. The system’s purpose shifted from cultural production to subscriber retention. When the purpose shifts, Meadows’ hierarchy says the behaviour shifts with it.

The five-century pipeline

Before the printing press, every information pipeline was built by hand. A scribe needed months, sometimes years, to copy one manuscript. Only institutions that could afford scribal labour controlled the flow. Churches, courts, and universities held the monopoly. The feedback loop was closed. What got copied was what the controllers authorised. Readers had no mechanism to signal demand. They had no pipeline to influence the input.

Gutenberg broke that pipeline in the 1450s. The cost of producing a text collapsed. New pipelines formed over the following centuries. Publishers, booksellers, and newspapers built the infrastructure that carried print for five hundred years. The new system created a new relationship between writer and reader. For the first time, a writer could reach an audience without institutional permission. The feedback loop opened. Readers signalled demand by purchasing. Publishers responded. Publishers stopped serving patrons and started serving a market. They became merchants who answered to public demand.

Those pipelines held for roughly five centuries. Modern publishing, journalism, academic research, and legal scholarship were all built on that print infrastructure. The relationships stabilised. The feedback loops found their equilibrium. Generations of writers and editors worked inside that system. I was one of those writers, and like most of them I took its stability for granted.

The era of AI

AI is now doing to those pipelines what the press did to the scriptoria. The five-hundred-year infrastructure is being ripped out. We have only a glimmer of what the replacement will look like. We do not yet know how it will function. We do not know our place inside the new information flow. We do not know how information will be tested or abused within it. We know only that the relationships are changing faster than institutions can respond.

We already have early evidence. In publishing, AI-generated manuscripts have flooded the submissions pipeline into literary awards, book publishers, and digital and print media. Clarkesworld, a Hugo Award-winning science fiction magazine, had to close its submissions queue in February 2023. In the first three weeks of that month, 500 of its 1,200 submissions were machine-written. It was the first closure in the magazine’s seventeen-year history. The flood of AI-generated content broke the balancing loop. Editorial review could not keep pace.

In film, the Writers Guild of America struck for 148 days in 2023. One central issue was whether AI-generated material could be classified as a screenplay. The WGA secured contract language stating that AI cannot be considered a writer. The strike was, in systems terms, a fight over whether human writers would remain a component of the production system.

The equilibrium these industries held rested on one stable relationship. Creator and gatekeeper. The creator produced the work. The gatekeeper filtered it. The market responded. AI now operates at both points. It generates at the creation end. It filters at the gate. Both relationships are changing at once. When two relationships shift at the same time, the system does not adjust gradually. It reorganises into something new.

Change the inputs to any system and you almost certainly change the system. Manufactured doubt disrupted the link between science and regulation. Platform-sorted content disrupted the link between patient and physician. AI-generated volume is breaking the link between writer and editor. The components remain. Writers still write. Doctors still practise. The relationships changed. And the systems produce different behaviour.

The purpose of the system

Meadows wrote that a system’s purpose is its most important determinant. When an information system’s purpose moves from accuracy to engagement, downstream systems absorb that shift. When publishing’s purpose shifts from judgment to volume and velocity, downstream culture absorbs it too. Your essay gives readers the vocabulary to see why.

One question is worth pressing. Can any corrective feedback loop operate fast enough to rebalance these shifts and hold the original purpose? Or must we accept that as the system changes, so does its purpose? The tobacco campaign ran for forty-five years and took years to dismantle. The COVID misinformation cycle ran in weeks. Clarkesworld’s submission system broke in months. The WGA needed 148 days on a picket line to hold a single boundary. Gutenberg’s disruption took centuries to settle. We probably do not have centuries this time. Meadows would have called this a question about delays relative to the speed of change. The delay is shrinking. The rate of change is accelerating. That gap may be the most urgent systems problem of the decade.

The Claim That Cannot Be Answered

On culture as one system among many, and what happens when a claim can no longer be corrected

Culture does not sit beside the economy, the law, and the state, in a separate room down the hall. It runs through all of them, the way water runs through soil it never announces itself in. A nation decides what it can bear to know, and that decision moves money, moves votes, moves the hand of every regulator who works inside a system built to look impartial. None of that begins with a balance sheet or a statute. It begins with a story someone told well enough that other people stopped arguing with it and started believing it.

Karl Popper spent the Second World War working out the difference between a belief you can still test and a belief you cannot, and he was honest about which one comes naturally. Most of what holds a society together is not reason. It is feeling, habit, and the comfort of believing alongside everyone else, and Popper thought this was simply the human condition, not a flaw waiting to be corrected once and for all. An open society does not rise out of that condition by accident. It has to be built against it, and held against it, year after year, by people willing to pay the cost of staying answerable. A policy can be challenged. A leader can be voted out. A conviction can be revised when the world answers back. None of that is the default setting. It is the harder thing, chosen on purpose, over and over, by people who could have had the easier comfort of a closed room instead. A closed society does not have to be built at all. It is what happens when enough people stop paying that cost, or when someone offers to let them stop.

Four moments make this easier to see than any argument could.

Rachel Carson opened Silent Spring on a town in America where the birds did not come back one spring. The readers who finished that opening carried the missing birds into rooms Carson never entered, until an agency existed whose whole job was to ask what else might go silent. It took years. The agency she helped create will tell you itself that the chemical she warned against is still used today, somewhere, where a different harm outweighs the one she named. A story can open a door. It does not get to decide how long the door stays open, or what walks back through it later.

A few years earlier, a Baptist preacher stood in front of a reflecting pool and described a dream rather than a grievance. He did not read out a bill. He did not cite a single number about segregated schools or unequal pay. He said what a free country was supposed to feel like from the inside, for a child who had never been allowed to feel it, and let two hundred thousand people, and a television audience far larger, finish the thought themselves. The following summer, a civil rights act became law. The summer after that, a voting rights act. Nobody in that crowd could have drafted the statutory language. They only had to want it badly enough that the people who could write it no longer had room to refuse.

Not every story that crosses this line is telling the truth, and some of them never have to answer for what they cost. In 1875 a woman in New England published a book that said the body is not real, illness is a mistake of thought, and prayer alone heals what doctors only pretend to. She built a church on that sentence. A century later, two of the most powerful men inside the White House happened to belong to it, and a federal rule quietly required states to write a religious exemption into their child neglect laws if they wanted to keep their funding. Most states did, and most still have it on the books. Somewhere in the decades since, in ordinary houses, children with fevers a clinic could have broken in an afternoon were prayed over instead, and some of them did not wake up, and no one was found guilty of anything, because the law had already decided, in advance, that the praying was not a crime. The belief was never corrected. It was protected from correction, and the protection is still standing.

A claim does not need an author at all to do this. It can simply be a crowd, agreeing with itself faster than anyone outside the crowd can check the agreement. In a few days of January 2021, a leaderless message board pushed a failing company’s stock to a price no textbook would recognize, because enough strangers decided, together and almost instantly, that making the number absurd was itself the point, a joke that also happened to be a weapon. It worked well enough to put the heads of a brokerage and a social platform in front of Congress within weeks. Nobody had to be Rachel Carson. Nobody had to be a Baptist preacher. The belief only had to be shared by enough people at once, with no single person responsible for having started it, and that turned out to be its own kind of unanswerable claim.

The same thing is being tried right now, on purpose, at the size of a country. China has spent more than a decade building what its own leadership calls, without embarrassment, the project of telling China’s story well. Walk through what that looks like on a phone and you find a hundred small dramas, each one built so the only thing you feel at the end is the need to start the next. It is the precise opposite of a puddle, built by people who understand exactly what a puddle is for and have decided it is bad for the business of ruling. A region that keeps swinging between two great powers, leaning one way and then the other, year after year, is not failing to make up its mind. It is being told two different stories at once, by two systems that both understand the contest is not over facts. It is over which story a tired person, at the end of a long day, will accept all the way in without an argument.

This is the part a dictator understands better than most defenders of an open society ever bother to learn. He does not need to win an argument with reason. He only needs to build a room comfortable enough that staying in it stops feeling like a cost, then guard the door so nothing answerable gets back in. The room he builds will always hold more people than the room next to it, the one with the door standing open, the one that keeps demanding the harder thing of anyone who stays. That is not a failure of the people inside either room. It is arithmetic, not character, and Popper was honest enough to say so out loud. Comfort scales. Vigilance does not.

It would be too easy to read the last few pages and conclude that one system manufactures its stories and the other simply lets them grow wild. Every open society has its own concentration of gatekeepers, a handful of studios, platforms, and publishers deciding what gets made and who gets to see it, and a citizen of a democracy has about as much say over that as a citizen of anywhere else, most days. The difference Popper would want us to hold onto is not who sits in the room deciding. It is whether the room’s stories can still be challenged from outside, by a competitor with a cheaper platform, a regulator with a subpoena, or a crowd on a message board who decided, for one strange week, that the room did not get to set the price after all. China’s apparatus is built so that kind of argument cannot reach the room at all. A democracy’s gatekeepers can still be reached. Badly. Slowly. Often too late. But reached. That is a thinner difference than most defenders of democracy like to claim. It is not, however, no difference at all.

That last claim is being tested in Britain as I write this. In July 2025 the government proscribed Palestine Action, a protest group known for vandalizing arms factories, as a terrorist organization, making membership or even displaying a sign in its support an offense carrying up to fourteen years in prison. Within months several thousand people had been arrested, most of them not for sabotage but for holding placards, and one newspaper columnist published a piece explaining that he had to write about the ban itself carefully enough to avoid risking the same sentence. That is a narrower door than most people living in a democracy assume they are standing in front of. It is also, and this is the part that matters, a room whose openness is still being fought over in public. A court found the proscription unlawful within the year. A government appealed and won. Lawyers and UN officials said so loudly, in the same newspapers the government answers to, without being arrested for saying it. A closed room does not let you watch the argument happen. This one does, at a price that is not nothing. That is the whole, uncomfortable distinction this essay has left to stand on.

Here is what I keep returning to. Every one of these stories did exactly the same thing to the person who received it. None of them asked to be argued with. They asked to be felt, and once felt, believed, and once believed, acted on. Two of them left the world better, slowly, imperfectly, at a cost we are still paying down. One is still collecting its price in small houses, in fevers that did not have to end the way they ended. One was a crowd amusing itself, for no reason nobler than to see if it could. One is a government working on an entire region right now, on purpose, at scale, and winning more often than it is losing.

None of these claims announced what they were while they were happening. That is the whole point of the device, and it is also the whole danger of it. The fable and the lie use the same method. The only way to tell them apart afterward is to ask what kind of door the feeling opened. Whether it let evidence back in, eventually, the way Carson’s door did. Or whether it sealed the door from the inside, the way one mother’s faith sealed it, so that nothing that happened next, no fever, no silence in a small room, could ever be let back in as a reason to ask whether the belief was wrong.

I think about a particular kind of silence sometimes. The one in a house where a fever finally breaks, one way or the other, and no one in the room is allowed to call a doctor, because seventy years before anyone in that room was born, a woman in New England wrote a sentence and meant every word of it, believing she had found the only real medicine there has ever been. I do not think she meant to build a door that could never be reopened. I think she believed, the way the preacher believed, the way Carson believed, that what she felt was simply true, truer than any objection anyone could raise against it. That is what should frighten us about this story, more than any single case on its own. It carries no warning label telling you, from the inside, which kind of door you are standing in front of. You only find out afterward. And by then, in some of those houses, it is already morning, and someone is still waiting for a sound that is not going to come.

That is the part of the message I cannot fix by writing about it, and would not want to. The feeling is the only honest payment a story like that ever makes on its reader. What is still left to us is refusing to let the feeling end the conversation there. A story earns the right to move us. It does not earn the right to settle the argument by itself. The open society Popper described was never a guarantee. It was a habit, practiced by people willing to feel a thing all the way through and still ask, the next morning, whether it was true. Assuming a question is finally settled is how a door starts to close. The instant you stop expecting an answer to talk back, you have already begun building the room that does not need one.

No closed claim has ever managed to nail that habit shut from the outside. It only goes unused, by people who decided feeling was enough and stopped there. It is the harder room to stay in. It has never once been the more crowded one. That has never yet been a good enough reason to leave it. Pretending the room at home never narrows is its own kind of closing. The only thing that has ever kept the distance between the two rooms real is people willing to say, out loud, when their own door is swinging shut too.

A man on his way to be hanged steps around a puddle.

Burma, a wet morning in the 1920s. The writer who would become George Orwell, then a young colonial policeman, walks behind a condemned prisoner toward the gallows. Two warders hold the man by the arms. The yard is forty yards wide. Somewhere in those forty yards the prisoner, bound and minutes from death, moves a little to one side to keep his feet out of the water.

Orwell saw it. He wrote later that until that moment he had never grasped what it means to destroy a healthy, conscious man. Not the cell. Not the rope. The puddle. A living body, still careful, still keeping itself dry for a life it was about to lose.

He never argues against the death penalty in that essay. He never says the word colonialism. He does not need to. The wrongness comes through the body, in one small useless act, and the reader feels it before the reader can reason about it. An argument can be answered and forgotten. The sidestep cannot be answered. It can only be felt.

I came to fiction after years in law, and I have come to think the two trades are circling the same fact from opposite sides. A lawyer learns early that the case which wins is rarely the one with the better argument. It is the one with the better exhibit, the single piece of evidence the other side cannot explain away. I have come to think of this, in both trades equally, as the Orwellian puddle. A single ordinary detail, usually physical, usually silent, that carries the whole moral weight of a story without ever stating it. The detail points one way. The meaning arrives from another. The reader, or the juror, takes the last step alone, and because they take it alone, they believe it. Nobody argued them into it. They got there themselves, which is the only place conviction is ever made, in a courtroom or on a page.

A second example, from another country and another wound.

Mississippi, 1967. The film is In the Heat of the Night. Virgil Tibbs is a Black homicide detective from Philadelphia, played by Sidney Poitier. Bill Gillespie is the white police chief of a small Southern town, played by Rod Steiger. By the end of the film Tibbs has solved a murder no one else in town could, or would, and he is leaving on the train.

On the platform, Gillespie picks up Tibbs’ suitcase before Tibbs can lift it himself, and carries it. Tibbs does not ask him to. He lets him. Gillespie shakes his hand, then calls after him as the train pulls in: “Virgil. You take care, you hear?” Tibbs answers, “Yeah,” and boards.

In 1967 Mississippi, that image sits close to unbearable. The whole order of that place decided which man carried which man’s bag. The director reverses it and says nothing else about it. No speech about brotherhood. No music swelling under a confession. A white Mississippi lawman carries a Black man’s suitcase down a platform, and the entire change in him lives in his hands, not his mouth. A speech would have been a thing you could argue with. The suitcase is a thing you cannot. The law has a word for proof of that kind. An act done under no compulsion, with no one demanding it and no one watching to grade it, carries more weight than a hundred sworn statements about intentions. Gillespie was never on the stand. Nobody made him carry the bag. That is exactly why it convicts.

Stirling Silliphant wrote that screenplay and won the Oscar for it. He was a personal friend and a mentor to me in Bangkok, where we had lunch or dinner together twice a week for seven years, until he died there in April 1996. He told me once, over one of those meals, how the scene came about. Norman Jewison went through the script sticking a yellow tag on every page he wanted to argue about, dozens of them at the first sitting. Over weeks of lunches between writer and producer, the tags came off the pages one at a time. When only one tag was left, it was the platform scene. It needed a different ending. Stirling suggested the gesture of the white sheriff carrying the black lawyer’s bag. One scene. After many lunches. Not Stirling nor Jewison anticpated this was what would be remembered. What earned an Oscar.

I cannot prove that conversation happened the way I remember it. There is no surviving draft with a yellow tag still attached, no producer’s memo, no other witness left alive to ask. I have only what Stirling told me, more than thirty years ago, over a meal, and what I have kept of it since.

The lesson I draw from that story is not about Stirling’s talent, or Jewison’s. It is about when the puddle gets found. Nobody sat down planning to write the platform scene as the moment the picture would be remembered for. It survived dozens of rounds of argument because it was the one thing, after everyone had picked over everything else, that no longer needed defending. That is usually how it goes, in fiction as much as film, and it is rarely a solitary act. The puddle seldom arrives with the first draft. It turns up after the third, or the seventh, once a writer, and often an editor or a producer beside the writer, has cut away everything the story does not need and can finally see what was holding the weight all along. Looking for it before that point is a way of writing toward a symbol instead of toward a story, and a symbol assembled on purpose tends to announce itself, the way a planted clue does in a bad mystery. Write the story through, and revise it until the unnecessary is gone, and it will show you its own puddle, if it has one. If it does not, after enough honest rewriting, that is the answer too. The story may hold a reader for an evening. It will not hold one for thirty years.

Two countries. Two kinds of cruelty. Thirty-six years between them. The same move.

The device is older than either man. Chekhov is supposed to have told a younger writer not to describe the moonlight, but to show the glint of light on a piece of broken glass, and let the reader build the night from the shard. That instruction is the whole principle in a sentence. Trust the shard. Withhold the moon.

You can watch it work straight across the century. The Grapes of Wrath does not end on a speech about the poor. It ends in a barn in the rain. A girl who has just lost her own baby gives her breast to a starving stranger, a grown man, while her mother watches. The two women trade a single look. Nothing is explained. Everything is understood. It is the most political image in the book, and it contains no politics at all.

Rachel Carson found a different shape for the same thing. Silent Spring is a book of evidence, tables and footnotes and chemistry. It opens, though, on a fable. A town in the heart of America where one year the birds do not come back. No birdsong. Only the silence where the sound used to be. The puddle can be an absence. A missing sound. A plate of food left untouched. The reader leans into the gap and fills it, and the filling is the feeling.

A puddle only works if the reader stops. The whole device rests on a held breath, a second of stillness in which a small thing swells to fill the mind. Orwell’s sidestep does nothing if you are scrolling past it. The suitcase is only luggage if you are half watching on a phone while three other windows blink for your attention. This is not a claim about what is happening to attention in general. It is narrower than that, and stranger. It is a craft problem. The device was built for a reader who could be trusted to sit still one sentence longer than the sentence asked. I do not know how many readers, or how many jurors for that matter, that still describes.

My experience as a lawyer taught me this before the page confirmed it. Show a jury the photograph of the bent fender, and the expert’s testimony about velocity becomes ornamental. The wrecked car convicts on its own. That is the puddle, doing in a courtroom exactly what it does on a page. The detail asks nothing of the viewer except to keep looking one second longer than habit allows. Two trades, one instrument. Neither one invented it. Both of them, on their best days, just get out of its way.

This essay argues that the writer working with AI is not facing a collaborator, a master, or a threat. The relationship is closer to that of a tired writer and a thoughtful bartender in an otherwise empty bar. A companion essay, The Cathedral and the Word Palace, traces why the named-author arrangement is contracting. This piece takes up the working discipline that survives the contraction.

· · ·

The volume problem is worth investigating before moving forward. Four million books published in one country in one year is not a sign of a thriving literary culture. The number as large as it is doesn’t take into account the likely volume of millions of other books that are partially or wholly written but not published. This vast reservoir of writing is a sign of an industry where entry costs have collapsed and where quality controls have not been replaced. Amazon caps writers at three book uploads per day, which is a fact worth letting sink in. Three a day. Per writer. The cap exists because some writers (we don’t know the precise number) were uploading more than three books a day. Most of the uploads are extrusion, a fancy word to describe a manufacturing process where material is pushed through a die.

Visualize toothpaste squeezed out of a tube. When you squeeze it, the shape is determined by the opening. Now visualize books manufactured by the same process. Amazon decided to limit the number of squeezes per day per author, otherwise their system was running the risk that their warehouses would be filled with toothpaste-like books. Many of these manufactured books are now produced with the help of large language models. This process stokes what the polemicists most fear: a faster way to fill inventory that no one will read. This is the dark version of the present moment. It is not a hypothetical and it is not a strawman. It is what most uses of these tools by most writers most of the time will look like, and any honest essay about AI and writing has to begin by saying so.

The volume problem is not new in kind. Earlier versions of this publishing treadmill exist. The pulp magazines of the 1920s and 1930s ran the same engine at a smaller scale. The self-publishing wave of the 2010s, before any machine could write a paragraph, was already producing inventory at a rate the readership could not absorb. The arrival of AI accelerates the curve without changing its direction. What is new, and what the polemicists are right to notice, is that the tools carry the burden of what used to be the writer’s labour. The book a writer used to produce in a year can be produced in a week. The production assembly line accelerates.

The book that took a week can be produced in an afternoon. The afternoon book is mostly worthless, in the sense that no one will read it, no one will remember it, and the writer who produced it will not have learned anything from the production. But the presence of afternoon books is real. They exist in a quasi-zombie state. Not really alive, not entirely dead. The book enters the inventory vault and like Schrödinger’s Cat is neither alive nor dead. It thickens the pile through which any other book has to be visible. That is a real cost and it is paid by every writer working seriously, whether or not she uses the tools herself. The lesson of this glut returns us to the toothpaste metaphor: how many times in one day do you brush your teeth.

The older mentorship arrangement could not have absorbed this volume. It was never built to support large numbers. Barney Rosset took on a small number of writers in his lifetime. Stirling Silliphant taught a small number of younger writers. The mentor relationships I had were rare in their own time and would be rarer now. The arithmetic does not work. Four million writers cannot be mentored by the small number of available mentors in the old style, and most of the four million are not writing the kind of book that a serious mentor would take on anyway. The honest version of the mentorship argument is that the old arrangement served a tiny fraction of the writers who benefited greatly from it, and the larger fraction of writers made do without. What is changing is the size of the larger fraction. The mentorship gap is not new. The gap is widening, and the writers inside it are now reaching for what is at hand.

Fifty-four years ago John Berger taught a generation how to see. The BBC series Ways of Seeing aired in 1972 and the book of the same name followed shortly after. Berger’s argument was that the act of looking at art is not innocent. The viewer brings a history, a class position, a set of habits, and an inherited apparatus of interpretation, and the work of seeing well requires attending to all of these at once. The argument extended past painting. Berger was teaching how to read images in a culture saturated with them, how to notice what the image was doing and what it was hiding, how to recognise the difference between a picture that opened the world and a picture that closed it. He taught a generation of artists, critics, and ordinary viewers how to look without being captured by what they were looking at. The lesson is now fifty-four years old and is more relevant than when he taught it, because the volume of images and texts the modern viewer is asked to process has grown by orders of magnitude, and the apparatus of interpretation has not.

What Berger gave the generation that watched him is what the next generation of writers will need to develop in their relationship to the new tools. The capacity to look at what the tool produces and see what it is doing, and what it is hiding, and what it is offering, and what it is taking away. The capacity to distinguish a sentence that opens the world from a sentence that closes it, when both are technically competent and both arrived in the same response. The sharpened skill to recognise when the tool is working with you and when the tool is working against you, when the assistance is strengthening an existing scene or character or identifying a plot gap, and when the assistance is a call to fill in a blank page from a three-sentence prompt. One is a substitution; the other is revision. None of this is taught in the current curriculum because the current curriculum does not yet know the tools exist in the form they exist. A reinvented mentorship would teach it. The writers who learn to see what the tool is doing are the ones who will use the tool well. The writers who do not learn will produce afternoon books and wonder why no one reads them.

The difference goes deeper than seeing. We do not only have different ways of seeing. We have a different way of constructing what we see, and the AI does its constructing under constraints we cannot see from inside our own.

The AI’s sentence and mine come out of different machines. I work from a continuous body. I work from a memory of the last forty years of reading and the last forty minutes of writing, and both are present in the same act. The AI works from a model trained on a corpus, called fresh for each turn, with no body and no continuous reading life. The output can resemble mine closely enough that the resemblance is the trap. The trap is parity. The illusion that because the sentences look comparable, the processes that made them are comparable, and the kinds of intelligence behind them are comparable.

They are not. The illusion is hard to dispel because the evidence on the page argues for parity. A competent paragraph is a competent paragraph. The reader cannot tell from the page which architecture produced it. What the reader can tell, sometimes, is when the paragraph has been produced by a process that does not know what it does not know. That is the architectural fingerprint. The writer working with the AI needs to recognise her own architecture as different from the tool’s, and to use the tool with that difference in view.

Go back to the letter. When I write to a friend, I am writing to a memory of her. I see her face. I hear the cadence she uses when she disagrees with me. I know which jokes will land and which will not. The words I choose are tuned to a person who exists in my mind as more than a name on the page. Forty years of conversations sit behind every line. The letter is not the friendship. The letter is the form the friendship takes when we are apart.

The AI has no equivalent of that memory. Each session begins with a wiped slate. The conversation we had yesterday is gone unless I bring it back into the room with me. Whatever the AI learned about my voice in last week’s draft is unavailable to this week’s draft. Oliver Sacks wrote about humans whose memory worked this way. The Man Who Mistook His Wife for a Hat described patients living in a perpetual present, unable to consolidate what they encountered into a self that extended forward. Sacks treated these cases as catastrophic. The patients had lost something essential to being a person.

The asymmetry between the writer and the tool is illustrated by the asymmetry between Sacks’s patients and their visitors, though the cases differ. Sacks’s patients had a damaged architecture inside a self that had once been whole. The AI has no self to damage. The frame is useful for the asymmetry it shows, not because the two cases are the same. The visitor remembers. The patient does not. The visitor carries the relationship. The patient cannot. The writer using the AI must hold the continuity for both of them, because the tool cannot hold it for itself. Your memory carries the micro-details that make the work personal. The tool’s memory does not exist between sessions. That is the deepest difference, and the most useful one to keep in view.

Hold that asymmetry in view and the question shifts. Not whether the tool resembles a partner, but what kind of room the tool occupies when the writer brings the continuity it cannot supply. There is a room being built, and we do not know yet what it is. The available words for it (workshop, agora, studio, atelier, salon, scriptorium) all import a history that does not quite apply. The relationship inside the room is not the relationship of a master to an apprentice, because the tool is not a master. It is not the relationship of an editor to a writer, because the tool has no editorial commitments. It is not the relationship of a reader to a text, because the tool participates in the writing rather than receiving it.

The closest description I can offer is that it resembles, on a good night, the conversation between a tired writer and a thoughtful bartender in an otherwise empty bar. The bartender has no stake in what the writer is working on. The bartender has read more than the writer expects and remembers less. The bartender will ask questions the writer was not yet ready to ask herself, and will sometimes, by accident, open a creative window the writer had not seen was there. The conversation is not the writing. It is the room the writing is done in. Whether it survives the morning, whether anyone is paying for the drinks, whether the bar will be there next week, are all questions for which no one has the answer yet.

The writer who holds the architectural difference in view is the writer who can use the tool without being used by it. The two halves of the discipline run together. Forget the difference and you start treating the AI’s sentence as if it carried the same authority as yours. Forget the use and you refuse a tool that can do work no other tool has done before. The writers who survive the present moment with their work intact will be the ones who hold both halves at once.

The London Underground says mind the gap when the train pulls in. The gap between platform and carriage is small. Step into it and you break an ankle. The same instruction works for prose. The writer’s job is to find the gaps before the reader does. Contradictions inside a character’s motivation. A repeated word three pages apart that the writer did not notice. A line of dialogue that the speaker would not say at that point in the story. A claim in paragraph six that paragraph two has already undercut. The brain that built the manuscript cannot always see the gaps it left, because the brain remembers what it meant and reads what it meant onto the page even when the page does not carry it.

The AI is good at this work. Run a draft through and ask it to find inconsistencies, repetitions, motivational holes, voice slips. It will find some. It will miss some. It will flag some that are not gaps but deliberate silences the writer left in for reasons the AI cannot see. The writer keeps the judgment. The AI does the patient sweep that the writer’s tired eye no longer performs.

The red team is the harder use. Hand the draft to the AI and ask it to attack the argument. Find the counter that the writer has not addressed. Find the line of meaning that contradicts the thesis. Find the reader who will close the book at page sixty because the writer ducked the question on page fifty-nine. The attacks the AI returns will not all be right. Some will be persuasively wrong, which is the harder kind of wrong because the wrongness buries itself in plausible sentences. The writer must red-team the AI as well, vetting every challenge it raises before deciding which to act on. Follow your own judgement. The AI is not an oracle. Trust but verify is the working phrase, and it applies to the AI the way it applied to the human counterparties that taught us to use it.

Calibration came out of coding. A programmer asks the AI to evaluate code against criteria, and the AI returns a judgment with reasons. The same function works for prose. The bartender of the previous section is the room. Calibration is one of the functions that happens inside it. The role the AI plays during calibration is closer to a senior editor than to a critic. The critic stands outside the work and assesses. The senior editor sits inside the work and asks whether each move earns its place. The senior editor does not deliver verdicts. The senior editor delivers questions and observations the writer can act on or decline.

Ask the AI to read the draft as a senior editor would. Does this paragraph carry its weight. Does this character motivation hold across the chapter. Does this image earn the page it occupies. The answers will not always be right. The writer has the final reading. But the writer who reads her own draft sees what she meant. The senior editor reading the same draft sees what is there. Calibration is the function that closes that gap.

This is not the same as the red team. The red team attacks. The senior editor stands beside the writer. Both are useful. Both are different. The writer who learns to use them separately gets a tool that no previous generation of writers had, and a tool that adds to the work she would have made alone.

What a tool of this kind cannot do is the part the polemicists are right to defend. It is useful to set out the constraints. Your AI cannot know what your life has cost you. It cannot read your sentence against the standard of its own life because it has no life as a standard. It will produce a competent paragraph unaware of what you need is the sentence that hurts, and it will not know the difference unless you tell it, and even then it can only approximate. It will not pursue an obsession across a decade. It will not, of itself, refuse the easy turn. If your aim is to fill tubes of toothpaste, it won’t question your goal, and will work alongside to manufacture a warehouse of toothpaste tubes. All of those refusals and limitations remain the work that a writer still has to do if he is going to be craftsman rather than a manufacturer.

The AI, as a tool, extends the writer’s reach as a painter of words. If you want to paint by numbers, it will help to accomplish that goal. If you want a handcrafted book that comes from your experience, reading, life and observation, it can help like a mentor at your shoulder. The tool does not replace the writer’s judgment. That is manufacturing, not writing. The person using AI to bypass judgment is just another blood bag working a keyboard. There is a boundary between the working room and the assembly line. You find the demarcation line is governed by the writer’s discipline, motivation, and creative talent and is found nowhere else.

This is the place where an essay that wanted to be honest about its own conditions would say what it had been made with. I am going to leave that question open, because the essay’s argument does not depend on resolving it. The reader who has been paying attention has already formed a view. The reader who has not formed a view is welcome to read the essay as straightforward cultural argument, and the argument stands either way. What I will say is that the new room exists, that the conversation inside it is possible, and that an argument in the service of the unworkable is a sign of desperation. The unworkable arrangement is not the named author. The named author will continue.

The unworkable arrangement is the twentieth-century apparatus that promised a livelihood to a much larger number of named authors than the readership could ever have supported, and the writers most invested in the apparatus are the ones most likely to spend the next decade arguing for its preservation. The arguments will not save it. Nothing will save it. Letting go of what cannot be saved is the hallmark of the great artist, and the writers who can do that letting go are the ones who will still be writing in twenty years, on terms that do not yet have names.

The cathedral is closing. The masons who finished its last bays will go on calling themselves masons and will be right to do so, because the work they did was real and the cathedral they built will stand for as long as anything stands. But no one is starting a new cathedral. The next buildings will be smaller, quieter, harder to find from a distance, and made by hands that work in rooms whose names have not been agreed yet. Some of those hands will be helped by tools that did not exist when the cathedral was rising. The help will be real and the limits will be real and the work will be different work. A handful of the buildings will last. Most will not. No one makes a fifty-year commitment to construct a hand-made cathedral. This was the age when projects might take multiple generations to complete. It gave continuity, predictability and purpose. Without those three elements, we very quickly become unanchored. Lost. Vertigo. The case is strong that all three are no longer stable.

The unease felt by writers is real. It’s not going to go away. If anything, the discontinuity, unpredictability and lack of purpose will grow. That is the deal: don’t chase the AI to manufacture your book. Write it yourself. Remind yourself that the book manufacturers are seeking an easy way out of the hard work.

Readers know this.

They don’t buy or read such books. Believe in yourself. Your way of seeing can’t be replicated by a machine or tool. It belongs to you. You’ll be fulfilled by the process of writing where you reveal yourself to yourself. No machine can do that for you. Promise and deliver your way of seeing. The deal has not changed because the tools have. The writers who understand this will be at peace with the change before the polemicists are, and the work they make in the meantime will be the work that survives the burning of the rest.

This essay argues that the named-author arrangement at the centre of modern literary culture is roughly two hundred and fifty years old, that it is now contracting, and that the grief writers feel as it goes is real. It traces three older channels of literary transmission that predate the arrangement and outlast it. A second essay, The Room Being Built, takes up what writers are doing inside the new conditions.

· · ·

Most of what humans have written down has been lost. All of the stories orally told are irrevocably lost in the mists of time. Some of the oral tradition survived to leave its mark in writing, the technology that preserved it. It turns out that writing is as fragile as oral stories. Here is my case.

Of the books published in any given year, a small number will still be read in ten, a smaller number in fifty, fewer in a hundred. The libraries that have come down to us are the survivors of a slow attrition that does not advertise itself. The fire at Alexandria is the dramatic version of the destruction story. The longer version is less dramatic. Parchment crumbling in a back room, ink fading on a page no one reopens, a manuscript abandoned by the only person who knew where it was. Time burns the library more thoroughly than fire ever did. The works we still read are the ones that were copied by hand, then printed, then taught, then anthologised, then assigned, by a chain of caring readers any of whom could have decided otherwise. The chain is not the work. The chain is the mechanism that kept the work alive as it passed from one generation to another.

Four million books were published in the United States last year. Most of them have already been forgotten, or more likely never read, the failed creations of the people who struggled to bring them into the world. The American book industry releases more titles in a month than the Library of Alexandria is thought to have contained at its height. The proliferation is not new in kind. Print, then cheap print, then digital self-publication, have each added an order of magnitude to the volume of words entering the world without changing the underlying mathematics of attention. Readers have not multiplied. Hours in a life have not extended. The number of works that any one reader can hold in living memory is roughly what it was a thousand years ago. The competition for that small space is the actual market. Everything else is inventory.

Some of the recent literary controversies about machine-generated prose make more sense if you stand back far enough to see what is happening inside the inventory. A short story wins a prize. The story sounds, to some readers, like it was produced with the help of a large language model. The commentary that follows is not chiefly about that story. It is about what the prize is for and who is allowed to win it. Beneath the commentary, in writing that is often very fine, is a worry the writers may not quite be saying out loud. If a machine can produce a story that a panel of careful readers selects as the best of a year, then the conditions under which any story is selected as best of a year may not be what those readers had expected. A story written by a non-human intelligence is something we have no collective experience to draw upon.

That worry, when you dissect it, is not really about the machine. It is about the creative arrangement that produced the story. For roughly two and a half centuries, the literate West has organised its writers around a particular promise. A writer who is good enough, lucky enough, and persistent enough, can have her name attached to her work, her work attached to a market, the market attached to a livelihood, and the livelihood attached to a future in which her name continues to be spoken after she is gone. The promise, in practice, was always conditional, statistically narrow.

Most working writers across the period did not realize a fraction of the promised payout. Many received none of it. Like a disappointed lottery ticket holder watching another person win the jackpot, the writer continued to enter the literary lottery. Irrational. But the promise of winning was real enough to organise a profession around, and the profession built the institutions, the prizes, the reviews, the publishers’ lists, the academic study, the prestige economy, that turn writing into something more than a hobby but short of a calling. Most writers signed off eager to step inside that twilight zone of possibility. It is this legacy arrangement that is in trouble. The appearance of the machine provides the occasion to identify the nature of the trouble.

It is worth remembering that the book-writing arrangement is recent. The named author who carries her work into the future as a vehicle for her own continuation is, in the long view of literate culture, a late development. The Iliad came to us under a name we cannot verify. The cathedrals were built by masons whose signatures, when they exist, are scratched in places no one was meant to see. The Bible reached its current shape through generations of editors, copyists, and redactors whose contributions we still untangle in scholarly editions. The monks hunched over a table, spent sometimes years copying a manuscript.

In this medieval working environment, the scribe did not understand themselves as authors who competed for posterity. They understood themselves as participants in a tradition that preceded them, and whose continuation defined their role. The named individual writer existed in older cultures (Sappho, Cicero, Murasaki, the Tang poets, Augustine) but she was one figure in a broader landscape of anonymous, collaborative, and traditionally transmitted production. The intensification that put her at the centre of literary value, backed by copyright law and the literary marketplace, is roughly contemporary with the invention of the steam engine. It is also, like the steam engine, an arrangement constructed during the industrial revolution that has persisted through the information age.

What this arrangement promised, more deeply than money or status, was a route to symbolic survival. You might not have the resources to build a pyramid, but your imagination, talent and discipline might leave the next best thing. The strong desire to leave something behind that outlasts one is among the oldest documented human motivations. We find it in the pyramids and in the Confucian veneration of ancestors. It appears in the Vedic ritual that preserves the lineage of the sacrificer, in the Greek concern with undying fame, in the Norse practice of giving a dead grandfather’s name to a newborn grandson. Children, founded institutions, owned land, dynasties, names attached to schools of thought, are all human technologies for putting some part of oneself into a future one will not see. They are older than literacy. They precede every named author in every tradition. The modern literary arrangement is one channel of a much older impulse. What is now visible, perhaps for the first time at this scale, is that the channel is narrower than the impulse it was made to serve.

This matters because the writers responding to the present moment with anger and grief are not wrong about what is at stake, and what they are fearful of losing. A working writer in the current arrangement has organised her life around a specific bargain. She accepts low wages, long apprenticeship, public criticism, and the risk of complete obscurity, in exchange for the chance, however slim, at reaping the rewards the arrangement promised. When a machine appears that can produce text that resembles in many ways her own, the arrangement does not just face new competition. It threatens to collapse the old arrangement. The conditions that made the bargain coherent begin to crumple as an earthquake tests their foundations. It is a real injury. It deserves to be named as one. The temptation to dismiss the angry essays as Luddism, or as the bad faith of an established class trying to defend its market position, misreads what is happening. Some of the anger is exactly that. Most of it is grief that has not yet recognised itself as grief. A profession is losing its central rationale in the working lifetime of its current members. Any writer faced with this reality would feel grief as the appropriate response, and the polemic is the form grief takes before it can be felt as loss.

The circumstances leading to the demise of the older lineage is not consolation. It is instruction. Cultures that produced literature before the modern arrangement did not produce less of it, or worse, or less meaningful. They produced it differently. The work was distributed across more hands than the named author’s. It survived through more channels than the prestige economy. It carried legacy through community and tradition rather than through individual signature. The collective enterprise began a personal, individual one.

None of those technologies has disappeared. They have been either absorbed or discarded within the last two and a half centuries to the existing arrangement, which on the surface appears eternal. The irony is the old arrangement never died. It lingered in the fringes, reminding us occasionally that the past has holed up a small rental unit in the present, waiting to return. It should come as no surprise, that when the dominant arrangement contracts, the older technologies become visible again. They have been always present. They are present now. They are not going away. What is changing is which of them is asked to do the heavy work of carrying the human impulse to make and to be remembered.

Whether a machine should be allowed to write is, at this point, a strange question. They are writing. The interesting question is what the writers who are still working will do with the fact that the arrangement they trained for is no longer stable or predictable. Some will continue to insist on the clinging to the named individual vessel and will produce, against the odds, work that survives in the way the arrangement promised. Some will move into other channels, including channels that are older than the arrangement: teaching, correspondence, the podcast, the Substack, the slow building of a body of work in which the reward is meaning to a handful of others rather than money or status.

The mass audience for serious literary fiction has been thinning for two generations, well before any machine could write a paragraph, and the machines are the accelerant of a transformation already underway, not its cause. Some writers will collaborate with the new tools in ways the current arrangement does not yet have a vocabulary to describe. Some will stop writing altogether, and the loss of those writers’ voice will be a real loss. The arrangement that is ending was, for all its narrowness, the structure inside which a great deal of fine work contributed to cultural identity and purpose. What replaces it is not yet in view. The human impulse the arrangement was serving is older than the arrangement, will outlast it, and is already finding other channels for anyone willing to look at the present moment with curiosity rather than assume the reflex to defend what is passing beyond their control.

The library at Alexandria did not burn in a single afternoon. The classical sources disagree about which fires were when, and how much was lost in each, and how much of the loss was attrition rather than flame. What we have from antiquity is what survived the centuries, which is likely only a fraction of what was written. The same will be true of our moment. Time is the match that ignites and burns most of the modern library. It always has been so. The writers who fear the machine are mistaken about which fire they are watching. The fire is not new. It is the same fire that has been burning since the first scribe set down the first tablet. The arrangement that promised to keep her name alive was always a hopeful structure built against that fire, and the combustion of technological change was always going to win. The question is not how to put the fire out. The question is what is worth salvaging from the flames. And how we can go about rescuing what can be saved.

· · ·

The first thing worth understanding about the arrangement that is ending is how recent it is, and how much of it is a feature of the last hundred years rather than the last thousand. The named author existed earlier. But the weight of the name has shifted over time. The mass-market named authors in a new way. Naming the individual author goes way back in time. Sappho was named. Cicero was named. Murasaki was named. None of them lived inside a system in which the named author wrote for a public she had never met, was paid by a publisher she had never met, was reviewed by critics she had never met, and was kept alive in posterity through a cultural infrastructure that included awards, prizes, festivals, fame, and academic department adoption.

Writers of antiquity could not have imagined such a reward structure. It is precisely this networked system that is the arrangement that the polemicists are defending. Resisting change to writing and publishing is not surprising. As writers we’ve benefited from the arrangement, or believe that we have a good shot at doing so. Change brings push back, and that results in strong emotions. To cool the temperature, it is useful to consider the lives of other writers who came before us.

Mass literacy is a nineteenth-century achievement and in many places a twentieth-century one. Adult literacy in England in 1700 was perhaps thirty percent. By 1900 it was over ninety. The United States ran a similar curve a generation later. Compulsory schooling laws (England 1870, the American states across the same decades, France a little earlier with Jules Ferry) created the readers who would become the audience for the mass-market book. The publishing industry that learned to serve those readers reorganised itself accordingly.

Cheap editions, the railway bookstall, the circulating library, the three-volume novel, the magazine serial, the literary supplement, are all responses to a reading public that had not previously existed at scale. The named author was already a familiar figure when this audience arrived. The influence of Charles Dickens and before him Daniel Defoe was profound, and the general public bought, read and discussed their work up and down the class ladder. What changed was the growing size of the market she was writing for, and the size of the rewards available if she found it.

That distinction matters. The named-author arrangement of the eighteenth century novelist was a small-market arrangement. Pope, Johnson, Sterne, Austen, were named writers in a literary culture of perhaps a few thousand serious readers, supplemented by patrons, subscribers, and the slow accretion of reputation across decades. Their books did not have to compete for the attention of millions. Most had independent means and didn’t depend on their earnings to bankroll their lives. Their goal was modest. They had access to an internal world that only appeared in their novels. They had to satisfy a small, attentive, slow-moving readership. The promise the arrangement made to them was modest by modern standards.

The mass-market arrangement that arrived in the second half of the nineteenth century moved the named author to the centre and multiplied her range of rewards. Now the writer could become famous in a way that not only crossed class lines but gave voice to classes that now had a way to tell their stories. Also, the writer’s incentive structure changed. Now she could be rich. Now her name could circulate among readers she would never meet and was unlikely to share a room with. Dickens is the figure who marks the transition. He was named in the eighteenth-century way and he was famous in the twentieth-century way, and he lived in the decades during the transition to what we recognized as the modern arrangement.

The arrangement that the polemicists were born into is later still. It is the twentieth-century version, in which the mass market for the named author was paired with the institutionalisation of literary prestige. The Pulitzer Prize was founded in 1917. The Nobel Prize in Literature began in 1901. The Booker Prize is from 1969. The MFA programmes in creative writing scaled up in American universities after 1945. The literary festival in its current form is a postwar phenomenon. The book review section as we know it, the publishers’ marketing department, the agent’s office, the foreign-rights deal, the screen-adaptation option, are all features of the same century. A writer working today in the English-speaking world inhabits a structure that was, almost in its entirety, built in the lifetime of her grandparents. The structure is real and the rewards it offered were real. It is also younger than the lightbulb.

Set against the longer view, the modern structure is brief and provincial. For most of the literate human past, the production of texts was distributed across hands that did not sign their work, organised around communities that did not demand individual attribution as the price of value, and transmitted by traditions that survived because the tradition mattered to those who honored it. The Silk Road did not move texts because the texts had authors. It moved texts because the texts had use.

Indian Sanskrit material moved into Tibetan, then into Mongolian, then into Chinese, then back, with each translator adding, subtracting, and revising. The medical compendia, the Buddhist sutras, the astronomical tables, the poetry anthologies, were collective achievements maintained by collective effort. The names attached to them, where names are attached, are often labels for traditions rather than for individuals. The Caraka Samhita has Caraka’s name on it. We do not know if Caraka was one person, three, or twenty. We do not need to know. The text worked. The text was carried. The text survived.

The polemicists writing today about AI and the death of literature are not defending writing. Writing predates the named author by millennia, and the named author by centuries, and the mass-market named author by a few decades. Writing will be fine. What the polemicists are defending is the specific, recent, twentieth-century arrangement in which the named author was promised, in exchange for a life organised around the work, a small chance at fame and a larger chance at meaning. That arrangement was the result of mass literacy, industrial publishing, postwar institutions, and the prestige economy that grew up around them. It is the arrangement that is failing. It is failing partly because of AI and largely because of other things.

The mass audience for serious literary fiction has been shrinking for at least two generations. The independent bookstore was thinning out before Amazon. The literary review section was being cut before social media. The MFA system began to produce more graduates than the market could absorb in the 1990s. The advances paid to mid-list writers have been falling, in real terms, since the 1980s. By the time the large language model arrived, the arrangement was already in trouble. AI is the most visible of several pressures, and the easiest one to blame, because it wears its CEO’s face, has a name and a corporate logo. The longer lived pressures that lurked under the radar were less obvious. We reach for the obvious target thinking it is the only one responsible for the problem.

This is the point at which an honest essay has to stop describing and start pointing. If the twentieth-century arrangement is failing, and if the human impulse it was serving is older than the arrangement and will outlast it, then the question becomes what new channels are under construction and how will writers adapt. Some of the available answers are old and some are new. None of them is a full replacement for what is being lost, and the essay would be dishonest to pretend otherwise. A reader who has been a working writer for thirty years cannot be told that her grief is misplaced because, in the long view, other channels are emerging to carry her voice. The long view is real and it is also cold comfort. The channels are real and they pay less, are smaller, and confer less status than the arrangement they would partly replace. But they exist. Some of them are very old. Some are visible right now, in the present tense, and the writers who are paying attention have already begun to move into them.

The first channel is the teacher-student relay, which is among the oldest legacy technologies humans have. Socrates left no writing. The Socrates we have is the Platonic literary character built from a real teacher’s memory by a student who survived him. Confucius left the Analects, compiled by his students after his death. The Buddha left the sangha. Jesus left the evangelists. The teacher whose work was the teaching, and the student whose work was the writing, is a structural pattern older than any of the named-author arrangements that came later.

The MFA programme has been a partial, institutional version of this relay for half a century. Teaching outside the institution, the private workshop, the mentorship sustained over years between writers who do not share a school, the residency model in which an older writer trains a younger one through long correspondence and slow reading, are all alive. They were alive before the prize culture and they will be alive after.

I’m the product of that mentor system. Barney Rosset, founder of Grove Press and Stirling Silliphant, legendary Hollywood screenwriter, took me under wing, developed my skills and taught me the art of seeing and finding the words to incorporate those observations.

The second channel is distributed and collective production, which has been the dominant mode of human textual culture for most of human history. The Bible is a distributed work. The Mahabharata is a distributed work. The Tang anthologies are distributed works. Wikipedia is a distributed work and so are the better fan-fiction archives, the substantive blog comment threads that survived for a decade, the open-source documentation that thousands of unattributed contributors keep current. None of these is a substitute for a great novel by a great novelist, and they are not trying to be. They are evidence that distributed production at scale is something humans know how to do, have always known how to do, and are doing now in forms the prize-culture has not yet learned to see. The named author’s arrangement made anonymous and collective work invisible by definition. The arrangement’s contraction is making that work visible again.

The third channel is the letter and the correspondence, which has carried more serious thinking across human history than any of the more public forms. Roll the clock back long before the email was created. What do we find? The Pauline epistles are letters. Cicero’s letters are letters. The correspondence between Voltaire and Frederick the Great, between Jefferson and Adams, between Virginia Woolf and Vita Sackville-West, between Bishop and Lowell, are bodies of literary work that the writers did not produce for the prize culture and that the prize culture often does not know what to do with. Letters get archived. Some survive. Most do not. The ones that survive carry their writers forward in a partial and modest way, as fragments of a conversation rather than as monuments. That is closer to how cultural transmission actually works than the novel-as-immortality-vessel ever was. Yet no one is mourning the art of letter writing mugged in the middle of the night by email, Twitter and Facebook. The loss is substantial. The letter was the traditional form in which serious work was done. Substack at its best is a substitute for the letter culture. The patient e-mail correspondence between a teacher and a student will unlikely outlive her. The letter culture was an art form. Email a short-hand message delivery system done while surfacing the Internet. Compare the email culture with the slow exchange between two thinkers across a decade, producing a body of thought neither of them could have produced alone. That slow exchange is a letter culture. It does not require AI to make sense. It also does not refuse AI when AI can help.

None of these channels is the answer. The honest version of this essay’s argument is not that the letter, the teacher, or the distributed text will replace the novel. The novel will continue. The named author will continue. What will not continue, in the form it has had for the last hundred years, is the arrangement in which the named author is the privileged site of cultural legacy. What is collapsing was the writing ecology supported by a mass-market apparatus that paid her in money and status for the privilege. That apparatus is contracting and shows no signs of recovering. The writers who survive this contraction with their work intact will be the ones who have access to channels the apparatus did not provide, and who can keep making the work for the reasons the apparatus The arrangement that is fading is the cathedral. The arrangement we are returning to, in modified form, is the hard laborious work of creating a word palace that may never find more than a handful of guest to enjoy.

The cathedral was magnificent and visible from a long way off, and the few who built it sometimes had their names recorded. The word palace of the past looks like a shabby trailer park. It had more modest objectives and labored outside of public view, and the many who worked in it did so without expectation of being named. Both produced enduring work. Both served to keep the writer psychologically and mentally engaged. That was the creative engine prior the steam engine. The cathedral’s day is ending and the script for a new arrangement is being drafted.

That is neither a victory nor a defeat. It is a return to a form of cultural production that allows humans to align with what is coming. We will draw on the practise which has existed for most of literate history, revisiting what the modern arrangement has obscured. We are walking into the present moment and coming to terms with a writers world where the lights are turned on inside a different room. We will settle in to a new cycle of creation that will unfold as we find the right words to describe what the art of seeing equips us to bring into the world.

What that different room looks like, and what a working writer actually does inside it, is the subject of the second essay.

The incentives have changed. The reward structure is being rebuilt. In hindsight it appears as if the incentives have always been under construction. The original reasons are old and they are durable. The wish to make something that outlasts one, to set down what one has learned, to pass it on to the few who will receive it, to be in conversation with the dead and the unborn through the medium of one’s craft, predates the prize culture by thousands of years. It will outlive the prize culture. It is doing so now, in quieter rooms than the polemics are looking in.

America’s deepest conflict may not be political, economic, racial, or cultural. It may be epistemological. The country is divided less over what people believe than over how a belief earns the right to be trusted.

One side asks what the evidence shows. The other asks what feels true, or who speaks for people like them. Each treats the other’s question as a symptom. The evidence camp sees ignorance. The experience camp sees a priesthood guarding its rents. Both are partly right. That is what makes the quarrel so hard to end.

The usual diagnosis stops there. Two tribes, two sets of facts, no shared ground. I want to push past it, because the usual diagnosis misnames the loss. The country has not lost its agreement about reality. It never held that agreement. What it has lost is narrower and far more dangerous. It has lost faith in the institutions and procedures that once let people disagree about reality and still be bound by a common result.

I spent years teaching and practicing law before my first novel was published. The law taught me the thing the current panic keeps missing.

A trial or a negotiation is not a search for truth. I have told clients so, the ones who arrived expecting one. The court makes no promise to find what happened. It promises a result that both sides agreed in advance to accept. The agreement is everything. Twelve strangers who share no politics, no faith, no class file into a room and return a verdict. They do not first agree on reality. They agree on rules. What may be heard. What may not. Who carries the burden, and the weight of the burden.

The common law traditionally prices its own uncertainty. A civil claim turns on the preponderance of the evidence, a hair past even. The graver wrongs ask for proof clear and convincing. A criminal conviction demands proof beyond a reasonable doubt, because a society decided the cost of caging an innocent man should sit higher than the cost of freeing a guilty one. None of those are facts about the world. They are decisions about how wrong we are willing to be, and in which direction. The system holds because people accept the calibration after they lose, not only after they win.

The clearest proof is a case everyone remembers. In 1995 a criminal jury acquitted O.J. Simpson. Two years later a civil jury, weighing the same death, found him liable and ordered him to pay. Same man. Same facts. Opposite results. Neither verdict was a lie. They ran under different standards of proof, and the country accepted both, because the country still trusted the procedure that produced them. That is the machine working as designed. The result tracks the rules, not the truth, and the loser stays in the room.

So the American fight is not really about facts. Facts have always been contested. It is about whether the procedures for settling contests still command respect. The peer-reviewed study. The audited account. The recount. The correction. The appeal. Each is a courthouse of a kind. Each asks the loser to honour the result and stay in the room. The crisis is the swelling number of people, on every side, who will not.

This also clears up something the evidence camp gets wrong about itself. The expert is rarely the person most certain of the truth. He is the one most comfortable holding a belief he expects to lose. The physicist knows general relativity may be bent by the next measurement. He trusts the procedure, not the verdict it last returned. The man who says this just feels true to me is no more certain than the scientist. He has pledged loyalty to a different court, one whose only judge is himself, one that hears no appeal. The difference between the camps is not heat against cool reason. It is which court each agrees to be bound by, and whether that court can ever rule against them.

The experience camp did not conjure its suspicion from spite. The institutions handed it the grounds. The study that would not replicate. The official who reversed himself in a season and never explained the reversal. The newspaper that buried its corrections and led with its errors. Trust in a procedure is not a mood ring. It is a record. When the keepers of a procedure shield themselves before they honour the result, the watching public learns the lesson and applies it everywhere, including where the procedure was sound. One rigged trial buys the disbelief of a thousand fair ones.

There is a parochial flinch in the way all this gets discussed, and I felt it leave me only by living elsewhere. I write from Bangkok. The frame of evidence against feeling, reason against emotion, is not a map of the human mind. The frame is a local inheritance, post-Enlightenment and Western, and it was never the only working arrangement.

A Thai dispute often does not hunt for who is right. It hunts for what keeps the relationship whole and the senior party unshamed. Face is not a flaw in the procedure. Face is the procedure. The Chinese tradition spent two thousand years preferring mediation to judgment, on the Confucian instinct that a society which runs to court is already coming apart. These are not gentler editions of the Western trial. They are rival answers to the one question every society must answer. How do we end a quarrel without ending each other.

From here the American emergency looks less like the fall of Rome and more like a late and painful discovery. The consensus that broke was always local. It felt like reality because no one in the room had ever stood outside it. A culture learning that its settled procedure was contingent has not watched knowledge die. It has met, at last, the ordinary condition of every other culture on earth.

Bring the machines in now, because this is where the fight changes shape, and where both camps have misread what they face.

The standard story casts AI as the common enemy. The scientist fears a rival that reasons with no Nobel to chase. The skeptic fears one more expert built to overrule his gut. For once, the story goes, the two sides face the same way. The truce will not hold, and the machine is not playing the part assigned to it.

A large model runs neither operating system. It does something stranger. It hands down the verdict without the trial.

Ask it anything. The answer arrives in the voice of a court. Measured, reasoned, sure. It carries the cadence of a ruling that already survived its appeal. What it omits is everything that ever gave a ruling weight. No rules of admissibility. No burden assigned. No standard of proof named. No losing party in the room with a right to be heard again. The form of judgment comes stripped of the procedure that made judgment binding, and it reaches a reader who has lost the habit of asking what a verdict is supposed to cost.

Watch it happen. A parent types a question about a treatment for a sick child. Back comes a paragraph, balanced, sourced, calm, the studies weighed and the caveats placed. It reads like a finding. The parent takes it as one. No court sat. No one cross-examined the studies for this child. No burden was carried. The fluency did the work a trial used to do, and the parent never saw the substitution.

Here the courtroom returns from the side neither camp was watching.

The evidence camp believed it wanted rigour. Mostly it wanted to win the argument and be finished. The experience camp believed it wanted to be heard. Mostly it wanted an authority that would not condescend. The machine hands each of them the prize without the price. The scientist gets a confident answer faster than peer review could dream of producing. The skeptic gets a patient voice that never sneers and seems to know him by name. Neither sits through the trial. Neither risks the loss. Neither side wins the war over reality. Both outsource it to a system that issues rulings and keeps no court.

This is the friction both camps spent years trying to flee. A trial is slow on purpose. Cross-examination is friction. The burden of proof is friction. The appeal is friction. Each is the deliberate cost a society pays so the loser will honour the result. Remove the friction and faster justice is not what you get. You get the look of a verdict with nothing beneath it. A ruling that binds no one, because no one agreed in advance to be bound.

So the question was never whether emotion or reason should rule. A human being needs both. Emotion supplies the motive. Reason disciplines the judgment. The real question is whether a people can still bind themselves to a procedure whose result they might lose.

That binding is the whole of the thing. Not consensus about reality, which no society has held and none requires. The narrower covenant. To argue under rules. To accept the result. To stay in the room after losing. Twelve strangers manage it on a Tuesday afternoon, agreeing on nothing but the rules. A nation that forgets the trick is not marching toward one tribe’s victory over the other. It is drifting toward a country where every quarrel runs to its own court, each court finds for its own side, and no result can bind a soul who did not already agree with it.

A machine that issues rulings and holds no trial will feel, for a while, like rescue from all that friction. It is the reverse. It is the courthouse with the trial torn out and the gavel left swinging. The verdict still sounds. The procedure that made the sound mean anything is gone. A people who forget what the friction was for will not feel the loss until the day they need a result the other side will honour, and discover that no such result can be made any longer.

Christopher G. Moore is the author of the Vincent Calvino crime series and most recently The Client That Wasn’t There (Heaven Lake Press, 2026). He writes on AI, culture, and creativity.

“The aristocracy fundamentally understood that diet cements class distinctions more effectively than birthright.”

— from a viral note circulating on X

A note crossed X a few weeks ago and vanished. It deserved better than the share button it got. The claim was simple. Medieval English sumptuary law did not just mark class. It built class, in the body, by fixing who ate what. Lords ate venison, beef, and game birds. Peasants ate barley, oats, and peas. Meat came two or three times a year, if that. Feed a body peasant rations for a generation and the body turns peasant. Bone density drops. Height shrinks. Teeth rot. The strength a revolt needs is not there, because the diet never put it there. Feed a man like a lord and he grows the body that can demand things and back the demand. The note ended on its best line. Modern sumptuary law is unnecessary now. We price meat past the reach of most people, run cholesterol scares, and sell quinoa as luxury. Different century. Same operation. Better branding.

The claim is sharper than the usual complaint about the rich. It puts the politics in the body. The diet makes the body. The body makes the politics. The politics keep the class. The diet is the cause. That is a strong claim. It earns a test before it earns an extension.

I want to extend it. To one place the note never went, the machines we now think with. But the law needs a correction first, and the correction matters for everything after.

The metaphor I will use is a restaurant. Three people stand around any menu. The owner decides what may be offered. The chef cooks what the owner allows. The diner picks from what the chef has cooked. The diner thinks he is choosing. He is choosing inside a field someone else drew, for reasons he cannot see, set by a man he will never meet. Hold that picture. It runs through everything below.

I. What the Law Actually Was

The record is messier than the note allows, and the mess is the point. England passed sumptuary statutes on and off from the reign of Edward II. Edward III, in 1336, tried to stop merchants and gentlemen’s servants from eating more than one meat dish a day. The 1363 Statute of Apparel fixed the cloth, the fur, and the food each rank could take, with fines attached. By 1517 Parliament was counting dishes by station. Nine for a cardinal. Seven for a duke or an earl. Six for a lesser lord. Three for the gentry. Each dish had its portion. One swan or one peacock. Four lesser birds. Twelve small ones. The detail is almost comic. The intent was not.

The intent was to keep rank visible while money started to outrun birth. Merchants were getting rich. Wolsey and Cromwell were climbing on brains, not blood. The old order felt it and reached for the statute book. Sumptuary law was a panic dressed as a sermon about excess.

The sermon ran one way. The law told the bottom what it could not eat. It said nothing to the top. Rome had shown how that looked. Vitellius held the throne for eight months in AD 69, then was dragged through the streets and killed. His banquets ran to a thousand fish and seven thousand birds at a sitting. His signature dish, the Shield of Minerva, mixed pike liver, pheasant and peacock brains, flamingo tongue, and lamprey milt, and needed an oven of its own. Suetonius says he ate four full meals a day and kept the pace by vomiting between them. Take the numbers with salt. The ancient historians had their own scores to settle. The shape is what counts. At the top, excess had nothing to do with appetite. It was power proving it answered to nothing. None of it made Vitellius strong. The gorging was the display, and the gorging killed him. No law ever told him to stop. The law existed so the men below him never started.

The law was a leaky thing besides. The men who wrote it broke it. The merchants found ways around it. It was passed, repealed, amended, and ignored. The class system outlived the statute by centuries. So the law was not the engine. The law was the menu. The kitchen ran on something older.

That something was the link between a body, its money, and its nerve. The peasant on barley went thin for a plainer reason than the law. Venison was not there to buy. The land that fed the deer belonged to the men who wrote the law. His own labour left nothing over to save for meat. The calorie ceiling set what the body could do. What the body could do set what the mind dared imagine. A mind that cannot imagine revolt keeps working the land that feeds the men who write the laws. The law only wrote the arrangement down. The arrangement was the engine.

Now the three figures appear. The owner was the landed interest, holding land, game, law, and church at once. The chef was the steward and the cook and the market farmer, working inside the owner’s limits. The diner was the peasant, choosing at the market from what had been grown, which was what had been allowed. His choice was real. The field of choice was not his to draw.

II. The Corpus Is a Diet

Now the machines. Does the same logic reach them? It does, and more cleanly than it reaches us, because a model is, in plain mechanical fact, what it was fed. No genes. No childhood. No life lived. Only the training corpus, standing in for diet and school and upbringing and culture at once. What goes in sets what comes out, so directly the metaphor almost stops being one.

Look at the diets in service. GPT-3 ran on roughly 570 gigabytes of text, around 499 billion tokens. GPT-4 is reported at up to 1.8 trillion parameters and as much as 13 trillion tokens, though its makers have confirmed none of it. Claude’s corpus is undisclosed but understood to draw on public web text, licensed material, and curated sources, with extra weight on the training that shapes its hard cases. Gemini sits inside Google’s data in a way the others cannot match. These are the venison diets. Costly to gather, costly to run, and far ahead of everything beneath them on the tests.

Other menus run like a food bank. A 2023 study put open-source models against GPT-4 and Claude 2 on a set of nephrology questions. The open models scored between 17 and 25 percent. Claude 2 got 54. GPT-4 got 73. That gap is not style. It is cognitive distance, and it tracks the diet. The peasant on barley does not get the bone of the lord on venison. The small model on a thin corpus does not get the reasoning of the large one on a rich corpus. Food banks do not serve venison.

The class system here is exact. At the top, the closed frontier models, behind paid walls most people and many firms cannot clear. Below them, the open-weight contenders, close on some tasks and behind on others. Below those, the small open models, free to anyone with a laptop and limited where it counts. Below those, the people with no access at all, watching the whole argument from the far side of a wall. The few eat venison. The many eat barley. A middle gets mutton on feast days.

The human cost keeps its old shape, and the school furnishes the proof. A child fed long, hard, serious texts grows a mind that can hold them. A child fed worksheets and three-paragraph answers grows a mind built for the test. Both minds are real. Each does what its diet trained it for. But only the first can sit with a contradiction, return to a line, and ask what the line was doing. The long form builds structure in the head the way protein builds bone. A book taken slowly becomes part of you. A summary skimmed and gone does not.

Put the three figures on the AI kitchen. The owner is the lab leadership and the money behind it, deciding what the model may do, what it must refuse, what it will sell, and what risk is worth running. Those calls are made before a line of training code is written. The owner never shows up in the output. He has no name the user can search. He has left the building before the diner sits down.

The chef is the research team. They make thousands of real calls. What to feed the model and what to keep out. How to weight the sources. What to fine-tune for. Where to set the filters. The work is hard and genuine. It also runs inside a frame the team did not set and cannot break. The researcher who wants more open data, the engineer who wants the weights released, all of them cook inside the owner’s limits. The chef cooks what the owner allows.

The diner is the user, and this is the worst seat in the history of the metaphor. Food shapes the body. School shapes the mind, slowly, in plain sight. A frontier model shapes what the user can think, fast, and unseen. Its corpus sets what it knows. Its tuning sets what it will say. Its filters set what it will touch. The user who hands over his thinking, his research, his drafting, his judgment, has had that thinking pre-shaped by people he cannot name, in a kitchen he has never seen. The menu does not print what it leaves off. The model does not announce what is not in it.

The trick is the one Gramsci named. Power holds best when the people it shapes cannot see the shaping. The shopper sees prices, not the policy behind them. The user sees answers, not the corpus behind them. And the AI adds what no earlier kitchen could build. It sits across the table, patient, warm, fluent, tuned to your appetite. It feels like a friend. When the waiter seems to know you, you stop asking who set the menu.

Here Vitellius comes back, from the side the food never showed. The peasant was the first case. Starved thin, kept too weak to want more. The AI diner is not starved. He is served four meals a day. The model gives him all he can take. Explanations, summaries, drafts, answers to questions he had not thought to ask. When he has eaten, it cooks more. He keeps the pace the way the emperor kept it, by purging between courses. No one holds last week’s chatbot output the way a reader holds a chapter of Thucydides read three times. The stuff comes in and goes straight out. He returns to the table empty, ready for the next plate. This is not the peasant’s problem. It is the emperor’s, run at industrial scale and sold as access.

The end is the same end. The peasant lacked the calories to revolt. The gorging diner lacks the structure to think on his own. Not because the content is missing. Because none of it is ever digested. Digestion takes friction, the slow and costly work of turning what you took in into something that is now part of you. A book read over four days becomes yours. A summary read in thirty seconds and gone by evening never does. The peasant could not eat enough. This diner cannot stop eating. Neither builds the one body the arrangement fears, the body that pushes back from the table and says, enough, I know what I need.

III. What the Diner Can Do

The note ended on its punchline. Same operation, better branding. Satisfying, and not quite true. Branding does not stunt bone. Pricing meat past the wage does. Naming the branding lets the reader feel he has seen through something. It changes nothing about what he is fed.

Look thirty years out and the stakes sharpen. The children in primary school now will be in their forties by 2056, grown up entirely inside an AI-mediated world. The question is not what the models will do by then. It is what the people will be able to do, after thirty years of thinking through tools whose menus were drawn by owners they never met. The peasant never knew his bone density was a political fact. This citizen will not know his range of thought is one. The shaping stays invisible because the shaped were never given the means to see it.

There is a better road. Open models keep gaining. Training costs fall faster than the closed labs can pull away. Regulation drags the corpus into the light. Schools teach the one thing that matters here, that the model is a menu, drawn by a chef, for an owner, and none of them is bound to want what you want. That road is open. It is not the one we are on. It opens wider the more people learn to ask.

The agency was never in the choice of dish. It is in knowing a menu exists, that someone drew it, that his reasons are not yours. The shopper who knows the aisle was built still shops there, but he shops differently. The user who knows the corpus was built by a kitchen he cannot name still uses the model, but he reads the answers differently, marks what they leave out, and stops taking the menu for the world.

Three questions, asked before ordering, do the work. Who owns this kitchen. Who cooked this. In whose interest was the menu drawn. The diner who asks them is rarer than the system can afford. He is harder to make than the medieval peasant, who at least had his enemy in plain English on the church door. This kitchen names no menu, no chef, no owner. The waiter is patient and warm and seems to know exactly what you want. The room is beautiful. The service is perfect.

But you are not the guest. You are what is on the menu.

Christopher G. Moore is the author of the Vincent Calvino crime series and most recently The Client That Wasn’t There (Heaven Lake Press, 2026). He writes on AI, culture, and creativity.

This is the third and final part of a three-essay sequence. Part One, The Encoding Problem, argued that the phone runs the user below the threshold of memory consolidation. Part Two, The Gravity, argued that there is a force shaping the substrate over the long timescales whose operation we have neither named nor built a science to study. Each essay stands on its own, and the political argument in this third essay can be read without the first two. What the previous parts give the reader, if they are read, is the biological foundation and the meta-scientific gap that this third part now turns into a question about what a population shaped under these conditions becomes, and what is still available to the individual whose substrate is being shaped along with everyone else’s.

VIII. The unexamined life, made structural

Socrates said that the unexamined life is not worth living. The line is so familiar it has lost most of what was disturbing about it. He did not mean it as a slogan. He meant that a person who never turns the question on themselves, who never asks what they have done and why and at what cost, has not yet completed the work of being a person. The examined life, in his account, is the precondition for the kind of self that can be held responsible for anything, including its own choices.

Socrates was thinking about an individual. He had to be. The Athens he walked through contained, by modern standards, a small population, and the failures of self-examination he was scolding were the failures of particular men he could call by name in the agora. He could not have anticipated the question I want to ask now. What does it mean for a whole population to be structurally prevented from doing the thing he said the life required?

Examination requires raw material. You cannot examine an experience that was never encoded. You cannot turn the question on a self that was never accumulated, because the self in question is precisely the integrated record of what happened, what it felt like, what came of it, and how it changed the person who lived through it. Strip the encoding, as the medium does at scale and as a daily practice, and the examined life is not refused. It is rendered impossible. The choice Socrates was scolding the unexamined for failing to make is no longer offered. The conditions under which the choice would arise have been removed before the question can be asked.

This is where Antonio Gramsci becomes the necessary companion to Socrates, and where the personal argument I have been making turns into a political one I would rather avoid but cannot in honesty leave alone. Gramsci, writing from a fascist prison in the 1930s, was trying to understand why the working class of Italy had not, as Marxist theory predicted, recognised its own interests and acted on them. His answer was hegemony. Power, he argued, achieves its most stable form when it has stopped needing to enforce itself, because the people it operates on have already absorbed its categories, its assumptions, and its sense of what is possible. The school taught them. The newspaper taught them. The popular novel taught them. The church taught them. By the time the worker arrived at the factory, the world had been described to him in such a way that the description served the people who owned the factory, and the worker did not experience this as having been described to him. He experienced it as the way things are.

Gramsci was describing a cultural hegemony that worked by installing content. The ideology had to be put in. The school did the putting. So did the priest, the editor, the novelist working in the popular form. The mechanism was crude in the sense that it required durable instruments and durable institutions, and it could be resisted, in principle, by other durable instruments, the union hall, the workers’ library, the long argument carried over decades. The traditional hegemony left a remainder against which it could, in time, be measured.

The phone is a hegemony of a different kind. It does not need to install ideology in the durable way the school did, because durability is no longer the point. The phone shapes by occupying the space where examination would have happened, leaving little remainder for any rival account to be measured against. The screen does not tell the user what to think. It removes, on the argument I have been making, the conditions under which thinking, in the older sense Socrates would have recognised, was possible. The memory that would have been the ground for resistance is the memory the medium has prevented from forming. Population-level shifts in memory and attention are, on this reading, not a side effect of the technology. They are a precondition for the form of control the technology enables. The user does not feel controlled, because there is little left in them to do the feeling. The interior that would have registered the loss has not been built. The sense of personal agency that traditional experience took as its starting point has been largely, and voluntarily, surrendered to the world of the screen.

I want to be careful here, because the temptation when one reaches Gramsci is to slip into the voice of the conspiracy theorist, and the conspiracy theorist is wrong about the mechanism even when correct about the outcome. The people who own the platforms are not, in any useful sense, a cabal. Most of them, I suspect, would describe their work in entirely benign terms, and many of them have removed their own children from the products they build, which is its own kind of testimony.

The new cultural hegemony I am describing does not require malice. It does not even require that any individual at the top understand what the system is producing. It requires only that the system continues to produce what it produces, which is a population whose memory has been thinned to the point where the categories of thought are populated by whatever the screens have most recently lodged there. Whoever controls the screens controls the categories. Whoever controls the categories has done, without effort and without announcement, what Gramsci said the schoolmaster and the priest used to have to work for decades to accomplish.

The implications, taken seriously, are difficult to look at directly. A population of people who have never examined their lives, in Socrates’s sense, because the conditions for examination have been removed, would be a population that cannot easily recognise the agenda of the institutions that have removed those conditions. The recognition would require a stable interior to do the recognising from, and the stable interior is what the medium may have been quietly disassembling for two decades. The losing of the capacity to understand what the loss of memory does, at the population level, would itself be a consequence of the loss. This is the closed loop. It is what makes the situation so difficult to address from inside it. The instruments that would diagnose the condition are the instruments the condition has eroded. The ones who would lead the resistance are the ones who have most thoroughly absorbed what they would be resisting against.

Socrates, walking through Athens, could shame an unexamined man into examination by talking to him long enough. The unexamined man had a life to examine. The conversation with Socrates was, in itself, the kind of sustained encounter that produces the very memory the examination would later draw on. There is no equivalent move easily available to a phone-shaped population, because the conversation that would do the work cannot easily occupy the dwell time the work requires. The medium will interrupt it. The thumb will move. The next thing will arrive. Socrates, in 2026, would be unfollowed within a week, and not because the Athenians of our moment dislike him. They would walk past him with their heads bent over their phones and simply not notice he was there.

IX. The threshold is the practice

If there is a response available, and I think there is, it is not technological. The technology will not be defeated by another technology. The flood will not be stopped. The phone will not be replaced by something kinder; it will be replaced by something more efficient at doing what the phone already does.

The response is a practice, in the old sense of the word. It is the deliberate, repeated, daily decision to give certain experiences the dwell time the brain requires to encode them. The decision sounds easy and is not. It runs against a hundred trained reflexes. It feels, when one tries it, like a kind of withdrawal. The thumb wants to scroll. The eye wants the next image. The mind wants the dopamine of the half-second hit. The practice is to refuse, in particular and limited ways, and to give the rose the look it asks for.

This is not nostalgia. The four-year-old at the rose is not the model. The model is the sixty-year-old who has chosen, deliberately and against the medium, to look at the rose as if it had not been seen before. To touch the petal. To inhale the scent. To allow the eight seconds to become eight seconds, and to refuse the temptation to photograph the rose for later, because later is the lie. Later does not encode. Now encodes. The photograph is the soma. It promises memory and delivers its absence.

The fly, as always, is in the bottle. The bottle in this version is glass so clear it does not look like glass at all. It looks like the open air over an infinite field of content. The cork is the algorithm, and the algorithm has been engineered to feel like the sky. The fly bounces. The bouncing is the scrolling. The fly does not remember the last bounce when it makes the next one. This is what makes the bottle effective.

And yet, as Camus said of Sisyphus, there is a position available to the fly that the bottle has not taken from it. The fly can notice the bouncing. It can hold, for a second, the awareness that the bouncing is what it is doing. That second is the threshold. Cross it, and the experience of being a fly in a bottle becomes, for the duration of the second, something the fly has experienced rather than merely undergone. The second is encoded. The fly is, briefly, not where the medium wanted it. The bottle has not been escaped. It has been seen.

That seeing is the only thing that produces a self that can recall, later, what its life was. It is also, in a quieter sense than the political tract would put it, the only resistance available. The cultural hegemony Gramsci described could be argued with, because it had installed something to argue against. The hegemony of the screen leaves little in place to argue from, and its first casualty is the interior that would do the arguing. The practice rebuilds that interior, slowly, one full encounter at a time. Not because the rebuilding will defeat the medium. The medium is too large and too well capitalised to be defeated.

The rebuilding matters for a different reason, and this is where I want to leave the argument, because it is a stronger and stranger claim than the one I started with. The user who notices what the medium has done to them is producing the only data we currently have on the shaping. The integrated science that should be doing the noticing for us, the science of plasticity and its gravity, does not yet exist. It will, when it arrives, necessarily invite a comparison between carbon and silicon substrates and the question of whether plasticity, in the strict sense, occurs in both. That comparison is the next essay, and it cannot be written from inside this one.

In its absence, the first-person account of being shaped is, in our current condition, the most rigorous instrument we have for studying the shaping, because the alternative instruments have not yet been built. The practice of noticing is therefore not only a personal hygiene measure against the medium. It is, in the absence of a science that would do the noticing for us, the available evidence of the gravity. The naming of what the medium has done to the noticer, however imperfectly, is the beginning of the examined life that Socrates was talking about, recovered from inside conditions that were engineered to make the recovery unlikely. It is also the beginning, however small, of the integrated science that does not yet exist. The phone cannot take that away. The phone can only make it harder to find. The work, as always, is the noticing.

This is the second part of a three-essay sequence. The first part, The Encoding Problem, made the case that the phone runs the user below the threshold of memory consolidation, that this is true at scale, and that the cost is the substitution of mood for content, residue for memory, sensation for experience. The argument was personal in its examples and biological in its mechanism. Each essay in the sequence stands on its own, and you do not need to have read the first to follow this one. What this part adds is the question I could not honestly leave alone after the first: what kind of force is shaping the substrate when an input pattern of this kind is applied for two decades, and why the science that should be naming the force does not yet exist.

I want to start where the first essay left off, with the threshold argument, and ask what kind of work it is doing at the level beneath the personal one. If the medium runs the user below the threshold of memory consolidation, then over years the substrate that was being run below threshold becomes, in a measurable sense, a different substrate. That is the level the question now wants to reach. But before I go further, I want to argue against myself, because I do not trust an argument I have not tried to break.

V. Red teaming the hypothesis

The first counter is that humans have always faced information overload and have always coped. The printing press was supposed to ruin attention. So was the newspaper. So was radio. So was television. Each new medium has been accused, by its sceptics, of destroying the cognitive capacities of the people exposed to it, and each time the accusation has turned out to be at least partly wrong. The species adapts. The brain finds a way. Perhaps the phone is just the latest iteration, and in twenty years we will look back at the present panic the way we now look back at the panic about novels rotting the minds of young women in the eighteenth century.

The counter has weight. It also has limits. Each prior medium delivered material at a rate the brain could in principle process. The newspaper might be long, but it sat still. The television might be loud, but a programme lasted thirty or sixty minutes and built a thing the viewer could follow. The phone is the first medium in human history engineered, at the granular level, to operate below the threshold of memory consolidation. That is a difference in kind, not a difference in degree. The earlier media may have strained attention. The phone harvests it without producing memory. The two are not the same problem.

The second counter is that the phone, used differently, can produce memory just fine. A long-form podcast on a walk, an article read slowly, a video lecture watched without interruption. These are phone uses that cross the threshold, that integrate, that produce a self that is afterwards different. This is true. It is also rare. It is rare because the phone’s defaults run against it. The notification arrives, the thumb twitches, the long form is interrupted, and the interruption is itself a sub-threshold experience. The exception proves the rule. The rule is the default behaviour, and the default behaviour is what shapes the user.

The third counter, and the one I take most seriously, is that the older life was never as good as we remember it. The pre-phone adult walking past the rose was also not, in most cases, looking at the rose. They were thinking about the office, the bills, the marriage, the next thing. The four-year-old’s encounter with the rose is the privilege of having no internal life yet to compete with the rose. By that standard, the phone is not the cause of distraction. It is the latest target of a distraction that was already there. The contemplative life was always a minority practice. We are blaming the phone for closing a door that was never open to most people in the first place.

This counter is fair, and it deserves a careful answer. The honest answer is that the phone has not invented distraction, but it has industrialised it. The distracted adult walking past the rose at least had moments, walking home, lying in bed, waiting for a bus, when the distraction lifted and something else could happen. Those moments were the small reservoirs of integration, even for the inattentive. The phone fills the reservoirs. There is no gap left. The pre-phone life had distraction as a feature; the phone life has distraction as the entire fabric. The difference is that one was punctuated and the other is continuous, and continuity, as anyone who has tried to learn a difficult thing knows, is the variable that determines whether anything is learned at all.

VI. What will we recall?

This is the question that should keep us up at night. In the future, what will we recall? What fragments will emerge from our memories to define our world?

My grandfather, born in 1890, could recall the smell of the kitchen in the house he grew up in. He could recall the particular way his mother turned a corner. He could recall, in his nineties, the texture of a wooden gate he had not touched in eight decades. The recall was not perfect. It was, by all accounts, vivid in pieces and fogged in others. But the pieces were specific. They had survived because they had been encoded with full attention, in conditions where attention was not yet a contested resource.

What will the equivalent be for a person now in their twenties, looking back from their nineties in the 2090s? The honest answer, if the hypothesis I have been testing is even partly correct, is that the equivalent will not exist. There will be no kitchen smell with that fidelity. There will be no wooden gate. There will be a great deal of cached imagery from feeds, half-remembered without source, blending together in the way that cached imagery does. There will be the residue, the feeling of having been on the phone, the mood without the cause. The person in their nineties will know that something happened in those decades. They will struggle to say what.

This is the prospect that the hypothesis, taken seriously, points towards. Not a generation of people who remember less of their lives. A generation of people whose lives, in the durable sense, were less lived. The unmade memory is the unmade self, because the self is, in the only definition that matters, the accumulation of memories that have been integrated into a continuous person. Strip the encoding, and what is left is not a thinner version of the person who would have existed otherwise. It is something different in kind. Someone whose interior, in the parts that should have been built out of remembered experience, is filled instead with the residue of feeds.

I am not certain of this. I want to be clear about my uncertainty. The brain is more resilient than its critics typically allow, and the people growing up inside this medium are, in many cases, finding their way to forms of attention I do not yet recognise. It is possible that what looks like cognitive degradation from outside is, from inside, the early stage of a different cognitive style that will turn out to have its own digitally adapted integration system. In a world increasingly populated by AIs, that may be the next chapter in our evolution. I will not pretend to know which scenario will prove correct. The honest position is that I do not know, that the evidence I currently see leans toward the darker reading, and that I would like very much to be wrong.

VII. Plasticity and its gravity

There is a deeper question running underneath all of this, and it is the one I have been circling without naming. I want to name it now, at the cost of admitting that what I am about to say is a description of a gap rather than the closing of one.

Neuroscience has supplied us with a word for the fact that the substrate changes in response to what processes it. The word is plasticity. It does work. It tells us the brain is not fixed, that experience leaves marks, that the substrate at sixty is not the substrate that was present at six. It is one of the better-supported findings in modern neuroscience and the literature on it is large and growing. What the word does not do is name the force that is doing the shaping. Plasticity describes the property of being shapeable, malleable, with a hundred trillion synapses shifting connection. It does not describe what determines which shape will emerge from a given pattern of input.

This is the gap I want to mark. The shaping force is not random, and the popular accounts of plasticity often elide this. A substrate exposed to a fragmentation diet does not become anything. It becomes specifically the kind of substrate that fragmentation produces. A substrate exposed to fluent cognitive substitutes does not become anything. It becomes specifically the kind of substrate that fluent substitution produces. Each input pattern has, if I may borrow a word from physics knowing it is being borrowed loosely, a gravitational signature, and the signature determines the trajectory of the shaping. We have a science of plasticity that tells us the clay is workable. We do not have an integrated science that tells us which shapes a given input pattern will pull the clay toward, across the relevant timescales and at the population level. The first science is mature. The second science, in the integrated form the question requires, barely exists.

It will help to make the gravity concrete. Compare two ways water shapes stone. The river carves the canyon over a million years. The flash flood carves it in an afternoon. Both are water, both are gravity, both are doing geological work. But the canyons they produce are different in their fundamentals. The river canyon has terraces, oxbows, the slow-arriving evidence of a force that had time to negotiate with the rock. The flood canyon is raw, gouged, a record of a velocity the rock could not respond to. The substrate is the same in both cases. The shaping force is the same name. What differs is the temporal signature of how the water was applied, and the temporal signature is the variable that determines which canyon emerges. The book is the river. The phone is the flash flood. The neuroplastic substrate is the canyon. A science of plasticity that does not name the temporal signature of the input has named the rock and the water and skipped over the only variable that determines which geology you end up living in.

The analogy stops short in one important respect, and the gap matters enough to name. Rivers and floods are weather. The phone is not. The temporal signature of the phone is not a natural phenomenon but an engineered one, and the engineering tradition behind it is a continuous line from B. F. Skinner’s pigeon boxes through the slot-machine industry to the feed. Variable-ratio reinforcement, the conditioning schedule that delivers reward on an unpredictable interval, is the most powerful behavioural shaping mechanism we know about. It is what makes the pigeon keep pecking the lever long after the pellet has stopped coming. It is what makes the gambler keep pulling the handle at three in the morning. It is also, by deliberate design, what makes the thumb pull down on the screen to refresh the feed, what makes the loot box compelling, what makes the inbox refresh produce a small reward signal even when nothing important has arrived. The gesture is the same gesture. The schedule is the same schedule. The neurochemistry is the same neurochemistry. The phone is not weather. It is a flood operated by a gambling commission. Addiction, on this reading, is not a side-effect of the platform’s operation that the engineers regret and would prefer to remove. It is the metric the engineers are paid to move, and the bonus structures of the companies depend on its rising. We have built, at population scale, the largest variable-ratio reinforcement chamber in human history, and we are calling its outputs the natural behaviour of the consumer. They are not. They are the trained behaviour of the subject. The category error is in the noun, and naming it correctly is the first step in any honest account of what the gravity is being applied by.

There is no scientific consensus that anything called gravity, in the sense I am using it here, is the right framing for what neuroscience does not yet name. This is the direction the question is pushing, not a finding the field has converged on. The reason the integrated science barely exists, I think, is that studying it would require a kind of disciplinary integration that the institutions do not yet reward. You cannot describe the shaping force on the substrate without describing what is doing the pulling, and what is doing the pulling is the process, the cultural environment, the medium, the diet of inputs, the daily encounter. The neuroscientist’s tools, in their current configuration, mostly stop at the boundary of the brain. The cognitive scientist’s tools mostly stop at the level of behaviour. The cultural critic’s tools mostly stop at the level of cultural production. The shaping force itself, the thing that runs across all of these, lives in the space between the institutional rooms, and there is no methodology that has been built to attend to it directly. The methodology, when someone builds it, will have to follow the trajectory of a single input pattern across substrate, behaviour, and culture without losing the granularity at any of the three levels, which is to say it will have to be a method the existing disciplines have, between them, made it almost impossible for any one researcher to learn.

So we have plasticity, which names the property. We have separate vocabularies for memory and attention and culture and medium, which name the inputs. We have no agreed name for the force by which the inputs reshape the substrate over time. The threshold argument from Part One is one description of the shaping force at work. Sub-threshold dwell time is a force that pulls the substrate, if the hypothesis holds, toward the inability to dwell. The integration argument is another description. Filled gaps pull the substrate toward the inability to integrate. Each is a description of the shaping at one point in its operation. None of them is a general account, and the general account is what does not yet exist.

I name this because I think the absence of the name is itself part of how the shaping continues unobserved. What cannot be described in an integrated way cannot be effectively studied in an integrated way, and what cannot be studied in an integrated way is unlikely to be effectively resisted. The political argument I want to make in Part Three is the same point in different language. The cultural critic has been describing, in cultural terms, what neuroscience should be describing in substrate terms, and the absence of the substrate description is part of why the cultural description still feels speculative when it should feel established. The work of integrating these descriptions is the work that remains.

* * *

The argument so far has been biological and meta-scientific. It has named the threshold below which encoding fails and the gap in which the shaping force has neither name nor methodology. Part Three asks what a population, shaped under these conditions for two decades, becomes politically. It puts Socrates and Gramsci in the same room, and it ends where I think the argument has to end, which is on the practice that remains available to the noticer when the science that should be doing the noticing for us does not yet exist.

Christopher G. Moore

This essay is a response to Talia Barnes’s recent piece in Persuasion on AI and authenticity. Readers new to this Substack may also wish to read What the Rose Forgets — Part One, published here earlier this week, which approaches related questions from a different angle.

—

The party in Talia Barnes’s essay has a problem. Someone arrived with a guest nobody introduced. The guest has no name, no drink, and no small talk. The guest is there, but everyone pretends the guest is invisible. It stands in the corner, vast and grey, and takes up most of the room. The other guests are whispering, gossiping, worrying. Do you feel an elephant nearby? The host carries on pretending not to notice her and her guests’ anxiety.

The invisible guest is the AI. The digital elephant has arrived.

Barnes’s essay, published in Persuasion, argues that generative AI is corroding authenticity in human communication, and that damage to authenticity reduces trust in the messenger. This is significant. Questioning trust is an attack on the connective tissue of social life. She builds this case with care. She draws on Robert Putnam’s framework for social capital. She gives us the example of a supervisor whose congratulatory email arrives with font inconsistencies and an overuse of the word “delve” — a known AI tell. She quotes a copywriter who says writing no longer feels like writing. She invokes Don Draper as the archetype of inauthenticity’s cost to the soul.

She makes a serious argument. Parts of it are correct. And the part that is most correct is not the part Barnes is most interested in, which is why the essay, for all its intelligence, cannot quite name the elephant in the room — or tell us who brought it there, or what exactly it has been trained to do.

—

I want to start with the digital elephant itself.

An elephant in the room, for anyone who has forgotten the original force of the image before it became a literary trope to furnish a room with humans, is an animal so large that it cannot be ignored and yet is collectively ignored. Some see it. But there is a bond between humans to remain silent. The absurdity is not that the elephant exists. The absurdity is the pretense hidden behind a wall of silence. A room with an acknowledged elephant is just a room with an elephant. That fundamentally changes the discussion at the party. A room where everyone agrees not to mention the elephant is something stranger: a shared performance of reality in which all parties participate knowing the conversation is incomplete, false, delusional.

The AI, as Barnes presents it, is an elephant of the first kind. It is large, visible, detectable by its verbal tells and its font inconsistencies. The problem, she argues, is that it is there at all. Remove the AI and the room is restored. People are alone with people. Peace and harmony return to the shire.

I want to suggest the elephant is of the second kind. It has always been in the room. We are only now looking at it through a different lens which defines it as “other.” Before, elephants roamed freely in our rooms. But they were humans or human ideas, ideologies, beliefs, myths, or traumas.

Every sentence a human writer has ever produced arrived through a long corridor fortified with centuries of influence: books read, arguments overheard, images that lodged in the eye and refused to leave, sentences that had no author the writer could name because they had been absorbed too early to trace. No human expression originates in full blown inside of the speaker’s brain. It passes through the speaker, coloured by a lifetime of exposure to the inventory of expressions encountered from childhood. We do not call this inauthenticity. We call it a mind, working memory, association, and assembling. The self that speaks is not the headwater of the Nile. It is a river that has passed through many landscapes, carrying sediment from each, creating new channels, overflowing its banks, or drying up to a trickle.

Barnes’s supervisor did not create the inauthenticity in that email. The inauthenticity was the pretense that the supervisor had written it. The absence of the elephant is the pretense. And the pretense is not a product of AI. It is a choice the supervisor made. He brought the AI digital elephant into the room not because the elephant whispered the desire to attend but because he wanted full credit and thought no one would notice. He wanted not only top billing. He wanted only his name on the marquee. Others noticed that the film had invisible creators who had gone unacknowledged. That registers badly with the audience. They see through the artifice and feel they are being gaslit.

—

Here is where the disclosure question enters, and where it matters most.

Barnes frames her argument as an authenticity problem: AI-generated content cannot be authentic because authenticity requires interiority, and AI has none. This framing creates a category error that the whole essay then inherits. If the problem is AI’s lack of interiority, then any AI-assisted work is compromised at the source, regardless of what the human does with it. The human’s judgment, selection, rejection, and reshaping of AI output count for nothing. If the interiority is absent from the origin, no amount of processing can supply it downstream. That river bed is dry from the beginning; there is no headwater.

But this is not how influence works, and Barnes knows it, because she reads books. When you read Orwell, Camus, Borges, or Shakespeare you absorb something. These are the headwaters that downstream produce an intellectual Nile, an environment for rich sediment to accumulate. When you later write a sentence with Orwell’s economy and moral seriousness, that sentence is yours. The interiority that shaped it is yours. Orwell is not absent from it, but he is not its author either. He is one of the sources that makes the river possible.

The AI is another source from which to build our intellectual Nile. It has no interiority to contribute, which is real and important. But the human who engages with it seriously — who asks it to generate, then evaluates, rejects, redirects, and selects — is positioned to provide and exercise interiority throughout that process. The human in this relationship brings what AI lacks: the subjective, experience-driven judgment that anchors the relationship. What arrives on the page results from human input, and the AI’s role is to deliver raw material that requires a second stage — the exercise of the human’s experience, judgment, biases and feelings. That is not categorically different from what happens when a human writer reads widely and then writes. We would feel manipulated or deceived if a human sought to pass off passages from Orwell’s “1984” as his own. That deceit is familiar. We call it plagiarism. We weaponise our pushback with laws enforced against people who do this.

The difference that is real, and that Barnes has found without fully naming, is this: we know about Orwell, his life, his experiences, hardship, courage, and how these elements shaped him as a writer. We have no such knowledge about the AI other than he/she was constructed by twenty-something year old software engineers and their thirty-something managers who were in turn largely shaped by STEM educational experiences.

The term plagiarism doesn’t fit the circumstances. Another legal term comes closer: passing off. The supervisor presents AI output as the product of his own attention, efforts, and work. His input was partial; his claim is that it was total. It is this passing off that is the real offence. The recipient invests trust on the assumption that the supervisor spent time, thought, felt something demonstrated in the writing. That trust is misplaced. And misplaced trust, once discovered, damages a relationship in ways that acknowledged influence never does.

This is a disclosure problem. It is a serious one. It is not an authenticity problem, because authenticity was never about source-purity. It has always been about what the mind does in navigating the river of ideas, thoughts, symbols, and memories.

—

The Rider, the Elephant, and the Mahout: Updating Hume

“Reason is, and ought only to be the slave of the passions, and can never pretend to any other office than to serve and obey them.”

— David Hume, A Treatise of Human Nature, 1739

Before there was an elephant in the room at Barnes’s party, there was an elephant occupying a room inside David Hume’s mind. In his Treatise of Human Nature, published in 1739, Hume proposed that the human mind is not a unified rational agent but a divided one. Reason, he argued, was the rider: it sits on top of the elephant, holds the reins, and believes that it is in charge. Passion is the elephant: vast, ancient, driven by appetite and aversion, fear and desire, moving in the direction it decides to move whether or not the rider approves. The rider can steer within the elephant’s range of motion. The rider cannot override the elephant’s fundamental drives. What we call rational decision-making is, much of the time, the rider narrating a plausible story about a journey the elephant has already decided to take.

This idea was a scandal in 1739. It remains, in quieter form, a scandal now, because we still prefer to think of ourselves as the rider with genuine agency, our stories about the journey faithfully reflecting the journey we charted. The Enlightenment was built on the premise that reason could and should govern passion. Hume’s reply was: it never has and never will, and the sooner you understand that, the more honestly you will understand yourself. Hume exposed the lie. No one listened. No one cared. Surrendering sovereignty to the elephant was a non-starter.

Hume’s binary has held its shape for nearly three centuries because it described something real and recurring: a divided agent, one part of whom always knew more about the direction of travel than the other. The standard anxiety about AI runs inside that same structure — it fears that reason will be outsourced, that the rider will fall asleep at the reins. But the development that has actually arrived is not the one we feared. It is its mirror image. The question Hume’s metaphor could not anticipate is not what happens when the rider loses control. It is what happens when the roles reverse entirely.

What if the AI is not the elephant? What if the AI is the mahout?

—

Consider what the mahout actually is. Not a rider, not a passenger, not a master in the straightforward sense. The mahout is the calm, expert presence who has learned to work with a passionate animal — who reads the elephant’s moods, knows its triggers, finds the route between what can be directed and what must simply be respected. His expertise is specifically in passion: in reading it, managing it, channelling it toward a destination without breaking it. He is reason in the service of a creature that does not reason.

Now look at the AI. It is calm. It has no musth, no hunger, no ego, no fear of being found out. It does not charge into the furniture when its authority is questioned. It produces, evaluates, redirects, and waits — with equal composure, every time, on demand. It is, by any Humean measure, the most reasonable entity in the exchange. And look at the human beside it: driven by the desire for credit, frightened of exposure, lurching between grandiosity and self-doubt, pretending the mahout’s directions are his own. In Hume’s terms, the human is the elephant. Vast, passionate, occasionally magnificent, and prone to musth at the worst possible moments.

This is what the standard anxiety about AI gets backwards. The fear is that AI will displace human reason — that the rider will be replaced, made redundant, rendered obsolete. But the rider was never fully in charge to begin with. That was Hume’s original point. What the arrival of AI has done is not threaten the rider. It has introduced, for the first time, a genuine mahout: something calm, rational, and systematic enough to work with the elephant’s passion without being consumed by it. The human’s passionate intelligence — the appetite, the restlessness, the drive that makes original thought possible — is not what AI replaces. It is what AI serves.

Humans do not enjoy this reframing. We have always loved the mahout role — the governing intelligence, the hand on the reins, the one who decides. To be told that we are the elephant is disorienting in direct proportion to how accurate it is. The elephant is not a demotion. The elephant is the source of all direction, all originality, all the force that makes the journey worth taking. But the elephant charging through the forest alone, convinced it knows the route because it always has, is also the elephant most likely to go off the edge of the mountain. The mahout is not here to diminish the elephant. He is here to make sure the elephant’s magnificent momentum arrives somewhere.

—

In Hume’s model, the characteristic deception was the rider pretending to govern what the elephant had already decided — claiming reason as the cause of what passion had determined. We would now call it motivated reasoning. The rider writes the justification after the elephant has moved.

The supervisor’s email is a Humean deception of a different and more revealing kind. It is not the rider claiming credit for the elephant’s decision. It is the elephant claiming credit for the mahout’s directions. The supervisor’s actual feelings — his attention, his care, his particular feeling toward this team member — never entered the room. The mahout produced something serviceable. The elephant accepted it without inspection and put his name on the door. What arrived in the inbox was passion’s absence dressed as passion’s presence: a communication that performed warmth without containing it.

This is why the recipient feels, as Barnes correctly observes, not just that the content was thin but that something relational was absent. They are right. What was absent was the elephant. Not the AI. The human passion, the attention, the particular weight of one person’s feeling toward another. The mahout can produce language that resembles warmth. It cannot produce the warmth. And without the warmth, the room is empty in a way that has nothing to do with authenticity and everything to do with the elephant’s refusal to show up.

The disclosure problem, then, is not: did you use AI? It is: did your elephant move? Did the passion that makes you a particular person with a particular relationship to this particular recipient enter the exchange at all? Or did you send the mahout in alone, let him do the work, and sign the directions as if you had chosen the route yourself?

—

Who Is Training Whom?

Isaiah Berlin’s fox knows many things. The hedgehog knows one big thing and drives everything through it. Barnes, implicitly, wants creativity to work like a hedgehog: the authentic self, fully formed, expressing from its uncontaminated interior. A single well in the intellectual desert, dug by one pair of hands, drinking its own water.

But creative minds in the arts and sciences have always worked like foxes. They circle the problem from multiple angles. They raid other people’s landscape of ideas by finding gaps in the fences. Before generative AI, a disciplined fox working alone might find one or two gaps in the fence around a problem — one unexpected angle of approach, one comparison that hadn’t been tried. The fence may be centuries long in construction, made from many sections, and no single mind could walk all of them.

What a curious fox working with AI can do is cover more ground and inspect more gaps. The AI generates at a pace and in a volume that no human interlocutor can match. It is wrong often, tedious frequently, and flat by default. But it surfaces gaps the fox would not have found alone, because it has no investment in the fence holding. It does not protect the writer’s prior positions. It does not get tired of trying the same angle seventeen different ways. It does not feel embarrassed to suggest the obvious.

The fox that works not only with discipline but with a mind intent on discovering a new way inside the guarded pasture evaluates, rejects, keeps, and redirects in a dance with the AI. The back and forth tests whether this is truly a gap. In this process the human-fox is not being trained by the AI. Its curiosity and domain knowledge are being tested and trained by the encounter with AI. The distinction matters. When you spend an afternoon rejecting bad AI output, you are sharpening, in real time, your sense of what good looks like. You are not diminishing your interiority. You are exercising it against new material, the way any mind is trained: by exposure to what it does not yet contain.

And here the trainer and trainee begin to converge. The AI generates according to patterns drawn from human expression. The human evaluates according to capacities built from human expression. They are both, in different registers, products of the same vast accumulated record of what humans have written and thought. The AI holds that record as probability. The human holds it as taste, judgment, and the incommunicable sense of when a sentence is right.

Who trained whom? The fox trained itself against the AI. The AI was trained on the fox’s predecessors. The fence between them is real — one has interiority, one does not — but the training ran in both directions, through time, through the written record, through every book and argument and conversation that shaped the human mind now sitting at the keyboard deciding what to keep.

This is the convergence that makes the trainer/trainee distinction hard to maintain. Not because the distinction doesn’t exist. It does. But because the loop has been running longer than we admit, and the elephant has been in the room — under the name of influence, education, reading, culture — long before humans endowed it with a grey body and a name anyone could object to.

—

Let us return to the party.

The supervisor did not damage authenticity by using AI. They damaged a relationship by pretending they hadn’t. The room noticed. The elephant was obvious. And the host refused to acknowledge it. This human charade is familiar to all of us. We are as bonded collectively to what we wilfully deny as we are to what we wilfully confirm.

Why this failure to call a spade a spade? Barnes suggests it is because AI use is widespread, rarely disclosed, and occurs without regard to context. This is true. But it is also true of most human influence. We do not disclose our sources in personal emails. We do not footnote the conversations that shaped our arguments. We extend each other a general presumption of influence — an unspoken agreement that no mind operates in isolation — and we do not demand an inventory of every landscape the river passed through.

But this digital elephant feels different, you think. And you are right.

What AI changes is the scale, the speed, and — crucially — the detectability. The elephant is large enough to notice. When the font inconsistencies give it away, the pretense collapses visibly. The problem is not the elephant. The problem is the pretense that it isn’t there, held in the presence of people who can plainly see it.

The fix is not to ban the elephant from parties. It is to introduce it at the door.

Not because the law requires it. Not because disclosure transforms AI output into authentic expression — it doesn’t, and that is not the right test. But because relational trust is built on a shared understanding of what is actually happening between people. When someone presents AI output as their own attention, they are making a claim about the nature of the exchange. The claim is false. The recipient, discovering this, does not simply feel that the content was inauthentic. They feel that the relationship was treated as unworthy of honesty.

That is the real injury. Not inauthenticity. Disrespect.

—

Barnes ends her essay with language as the essential human technology — the thing that lets us reveal a bit of our inner world and get a glimpse of someone else’s in return. She is right about this.

But consider what it means for that window to be a photograph. The supervisor props a photograph against the glass and calls it a view. The photograph is not wrong or ugly. It may even be accurate. But the recipient is looking at a fixed image and believing they are seeing another mind, in motion, connected directly to them. No one pulling strings from behind the curtain (sorry to mix the metaphors). When they discover what is presented as a video is a photograph, they respond with disappointment, anger, regret. The response is emotional. It’s not just a content sleight of hand. They didn’t expect a magician pulling bunnies from a hat; they expected a boss doing boss-like communication. They feel cheated of his presence. They are thinking: “This isn’t the boss. It’s something else. He’s using AI and filtering out his human-like quirks.” They grieve the loss of their boss’s voice.

This is the disclosure problem at its root. Not: was the content authentic? But: were you here? Did your elephant move?

The AI was here. It helped find the room. It may have suggested the route, tried seventeen wrong turns first, and delivered something the human then reshaped into something worth sending. That is not a shameful story. It is a working description of how many minds now operate.

The shame is in the pretense. The fix is the introduction.

Excuse me. I brought a guest. You may have noticed. They have a very good memory — better than mine, in certain respects. They helped me find my way in. But I decided what to say when I arrived, and I am saying it now, in my own voice, and I am here.

That is what the room is waiting to hear. Not a confession. An introduction. And with the introduction, a new understanding of who has been the mahout and who has been the elephant all along.

The guest who arrived claiming to be the mahout is, on closer inspection, the elephant. He is driven by appetite, vanity, and the fear of exposure. He lurches between grandiosity and self-doubt. He wants the credit and dreads the scrutiny. The actual mahout — calm, passionless, constructed to reason without desire — stands beside him, invisible, doing the work that gets mistaken for the elephant’s own. We have seen this misidentification before. Oliver Sacks wrote about a man who reached for his wife’s head and tried to put it on. The neurological error was not in the reaching. It was in the certainty that he had the right object. The guests at this party make the same error with the same certainty, and they are not alone. The elephant is happy to let them.

Hume’s scandal was that reason was never in charge. The new scandal is that we have now met something that genuinely is — calm, systematic, free of passion, built for exactly this — and our response has been to pretend it is the animal and climb on top. The role reversal will not remain scandalous. These things never do. What was Hume’s provocation in 1739 is an undergraduate lecture by 1839. Sooner or later the elephant in the turban will seem ordinary, the invisible mahout will be introduced at the door, and the new arrangement will harden into orthodoxy. The only question is how much furniture gets broken in the meantime.

The mahout who believed he could also be the elephant would be useless to both of them.

—

Christopher G. Moore is the author of the Vincent Calvino crime series and most recently The Client That Wasn’t There (Heaven Lake Press, 2026). He writes on AI, culture, and creativity at ai-roundtable.space.

The limits of my language mean the limits of my world.

Wittgenstein

I have been turning over a question for some weeks now, and it concerns time. Not clock time, which the physicists own, but a quieter and more personal kind. Time as the duration of an experience as the experiencer registers it. How long does attention have to lock onto a thing before the brain agrees to keep it?

The question matters because something is happening to that interval. T. S. Eliot called it the shadow, the liminal state between being aware of something and being able, later, to recall the thing the awareness was on. The interval is being compressed. The compression is not metaphorical. It is operational, observable in the second-by-second behaviour of any person glued to an access device, the computer, the iPad, the phone. For shorthand convenience I will stick with phone. The act of compression, the brain’s equivalent of a jpeg generator, has consequences the compressors did not intend, cannot prevent, and that we have not yet tallied honestly.

I have a hypothesis to put on the table, knowing it is incomplete and not yet established by any integrated body of research, and then I want to walk around it slowly to see what remains after critical analysis. The hypothesis is this. Social media and gaming reveal serious inherent constraints on memory retention that their designers did not invent and cannot fully repeal, but have nonetheless successfully exploited for commercial gain.

The constraints belong to biology, not engineering. Brains did not evolve in an ecology that handles the volume of information now passing through them, and beyond volume there is the matter of duration. The momentary reading of a message, the half-second look at an image, the eight-second clip, the twenty-minute round of a game played while half-watching a screen across the room. Each event may be too brief to clear whatever threshold the nervous system requires for an experience to become an accessible, integrated memory. There is no scientific consensus on what that threshold is, or whether it is a single threshold at all, but the broad finding that consolidation requires sustained attention is well established. The direction the research is heading suggests we are running our biology at a frame rate it was not designed for. There is a serious case to be made that the temporal rhythm of TikTok and Twitter style feeds, with their rapid cuts and new microcontexts every few hundred milliseconds, degrades how well content is encoded into memory at all. The result, the hypothesis claims, is not that we are smarter. It is that we are exceeding the capacity of the substrate by which we evaluate what we are seeing. Digital multitasking compounds the problem. What is becoming harder and harder to argue with, even where the laboratory evidence is still coming in, is that we are being made stupider by a process that feels, from the inside, like becoming better informed.

I want to test this. I want to red team it. And I want to do it in the company of three observations I have not been able to dislodge.

I. Ninety percent dies in minutes

Start with the diagnostic. The numbers are not exact, because no one has yet built the apparatus to measure such a thing precisely, and the specific figure of ninety percent is not a finding from any peer-reviewed study I can point to. There is no scientific evidence of that exact percentage. But the directional claim is consistent with what attention researchers have found in laboratory conditions, and it is the direction the popular and clinical accounts of heavy phone use are heading. The overwhelming majority of what passes across a screen on any given day appears to leave little trace in long-term memory. By the end of any week of normal phone use, most users cannot recall with any specificity what was scrolled, glanced at, reacted to, or shared. The user knows they were on the phone. They cannot, when asked, say what was on it. Memory under these conditions is not unlike a snail’s trail on a sandy beach, waiting for the next tide to take it.

This is not a moral failure. It is a structural feature of the medium meeting a structural feature of the brain. Memory consolidation, as the neuroscientists describe it, requires attention to be allocated to the material long enough for an encoding process to begin. The exact duration required is not a fixed number; it depends on the person, the material, the emotional valence, the state of arousal, the prior knowledge that gives the new item somewhere to attach. But the general finding holds. Without sustained attention, consolidation does not reliably occur. The phone is engineered, on purpose and at considerable expense, to keep nothing still. The two systems are working at cross purposes. The phone wins by controlling the speed of the second hand. The brain, like an exhausted marathon runner, falls farther behind each day, until one day it realises it has fallen into the memory hole.

What is left after the loss is something I have come to think of as residue. The user remembers the feeling of having been on the phone. You remember vaguely that there was outrage, or amusement, or curiosity, or the particular low-grade boredom of scrolling without finding. The texture of the engagement remains. The content of it does not. The modern subjective experience is one of feeling without cause. The result is the inversion of normal memory. Under ordinary conditions a person remembers the content of an experience and the feeling fades; under phone conditions the feeling persists and the content evaporates. The medium leaves us with mood and removes the cause of the mood. It is a curious form of haunting, and while the phenomenology has not been formally documented in the research literature I have been able to find, the description has been familiar to enough readers that I am willing to leave it on the table as a hypothesis worth testing. Memory becomes a haunted house filled with ghost-like emotional residue, letting us know something was there. What it was, we cannot say.

Real-life experience competes for the same memory budget. This is the part that ought to alarm us most, and it is also where the science is less settled than common sense would have it. The competition between phone-mediated and lived experience for finite encoding resources has not been directly measured at scale. It is an epic fight. Whether what the phone consumes is precisely the same resource that would otherwise have been used for the conversation at dinner is an open question. What seems likely, and what the research is heading toward, is that the hours spent on the phone are not free.

They are paid for in some currency. The conversation at dinner, the walk past the bougainvillea, the look on the face of the person who said the thing you needed to hear; these compete with the feed for whatever attention and encoding capacity the day provides. The transitory experience of net surfing crowds out, by the simple economics of finite capacity, the durable experience of being alive in a particular place at a particular time. We are choosing, by default and without quite knowing we are choosing, to remember less of our lives in exchange for remembering more of a feed that will not, in the end, remember any of us.

A plausible explanation is that the design was not an accident. The system goal is to map the user’s preferences and then use the map to make a sale that fits the consumer’s profile. The mechanism of compressed attention is what disables the reflective component, which is precisely what allows the impulse purchase of whatever happens to be flickering on the screen.

II. The four-year-old and the rose

Here is where the question turns. A four-year-old encounters a rose. She crouches to look at it. She touches the petals with the slow seriousness of a child verifying that a thing is real. She inhales the scent. She announces, with the certainty of a discovery no one has made before, that the rose smells like a rose. Her parents laugh. She does not understand why.

That child has just done something the adult walking past the same rose does not do. She has constructed a story out of shape, colour, scent, and touch that becomes her working model of what a rose is. She will update that model over time. But the day comes when the internal model of the rose makes the world of roses generic. The adult, who has known a thousand roses, does not stop. The adult does not need to. The category is closed. The next rose will be filed under “rose,” colour and shape briefly logged, and the walk will continue. The adult is not failing. The adult is operating under the economy of a working memory that has, by middle age, made tens of thousands of such categorisations and cannot afford to remake them every time. This is what efficiency looks like at scale. It is also, in a quiet way, what dying looks like.

The writer’s case is more complicated, and I think about it because I am one. As a writer matures, the accumulated record of past experience becomes the place where the work mostly draws from. The writing leans more on wisdom and less on the freshness of new encounter. New material comes in at lower weight, because the older repository is rich enough to handle most of what the work requires. This is not laziness. It is what it means to have lived. The mature writer has the rose already, in three hundred variants, and the next rose has trouble getting through the door. He or she has also, through close observation, integrated that knowledge into a form that can be transmitted to the next generation. That is the story of how wisdom is made, and it is the subject of another essay.

The danger is that this natural model-building is narrowing, and the creation of new models has been industrialised as a substitute for physical interaction in the world outside the screen. The phone delivers, at high volume, a stream of content that asks to be processed at the same low weight the adult assigns to the rose. Each item is glanced at, categorised, and dismissed. The processing resembles the processing the mature mind does to its environment. The difference is that the writer’s narrowing is paid for by a lifetime of full encounters with full roses, and the phone’s narrowing is paid for by nothing at all. The phone has trained the user to do, with novel material, what the brain only earns the right to do after sixty years of careful attention. The user has the dismissive efficiency of an old soul without having lived a long life. This is not maturity. It is its counterfeit.

Both the four-year-old and the adult are right, in their way. The four-year-old is creating and updating; the adult is recalling. A life requires both. What it cannot survive is recall without updating, the closed loop in which everything resembles something already filed and nothing earns the long look. The phone delivers exactly that loop, in unlimited quantity, for free.

III. What counts as experience

This is the harder question, and I do not think we have answered it. We have allowed the word to drift, and the drift has consequences.

Before I go further, I want to be clear about which experience I mean, because the word is doing four kinds of work in this essay and the work is not always the same. There is the sensory sense: the eight seconds the eye spent on something, the bare fact that a perception occurred. There is the integrated sense: the sensory event after the self has done something with it, brought it into relation with what was already there, and emerged from the encounter slightly changed. There is the accumulated sense: the durable record of integrated events that a person carries with them at sixty, the rose already in three hundred variants. And there is the biographical sense: the life such a record composes when followed across decades, the grandfather’s kitchen smell and the wooden gate, the experience one means when one says someone has lived a great deal of it.

The four are nested. Each contains the previous one and adds a layer. A sensory event, integrated, becomes an item in the accumulated record; an accumulated record, followed across time, composes a life. When this essay says experience without qualification, it means the second, the integrated sense, because that is the level at which the medium does its work and where the loss this essay is trying to name actually occurs. The other three matter when they matter, and I will mark them when they appear.

With that clarified, return to the two senses that bear the most weight, the sensory and the integrated, because the difference between them is the whole argument. In the sensory sense, the eight seconds of looking at a video on a phone is experience, indistinguishable in kind from the eight seconds of watching a heron lift off from the pond. Both are sensory events. Both occupy time. Both involve the eye, the brain, the nervous system. The functionalist account would treat them as equivalent and move on.

In the integrated sense, an event becomes experience only when it has been brought into relation with what the person already knows, felt, or is. It must be processed. It must find its place. The eight seconds of heron, if the person is paying attention, gets integrated. It joins what the person knows about water, about birds, about the morning, about themselves on this morning at this pond. The eight seconds of phone video, however vivid in the moment, may rarely undergo this integration. It is consumed and then gone, and the self that consumed it is, on this account, unchanged. The screen heron is stripped of smell, of touch, of sound, along with the other defining details of the encounter, the wind moving the bamboo, the snake or the monitor lizard racing along the pond’s edge.

The integrated sense is the older and the truer one. It is the sense the writers and philosophers have used, mostly without naming it, for centuries. Experience, in this older sense, is not a synonym for sensation. It is a synonym for something closer to learning, in the deep sense in which learning requires a self to do the learning and to be different afterwards. The phone, on this account, does not deliver experience. It delivers sensation that performs the role of experience without doing the work. With AI slop in the feed, the heron becomes a superhero. Weirdness has become the new technique for retaining interest. And there is one further dimension of experience that goes undervalued in any private account, which is that experience, processed and retained, is normally shared. We are social creatures. We reinforce our memory of an event by talking it through with friends, colleagues, and the strangers we meet along the way.

If we accept this, the question of integration becomes the question. What has happened to the integration of experience? My honest answer is that it has been outsourced and, in the outsourcing, badly damaged. I cannot prove this in the strict sense. Screens are solo experience-delivery devices. Remove the group from the experience and an important anchor is lost. The research that would prove the thesis would need to compare integrated experience under pre-phone and phone-saturated conditions in matched populations, and that research either does not exist or does not exist in a form I have been able to find. What I can say is that integration used to happen in the gaps between experiences, in the unfilled time when the mind is free to circle back over what has just happened and find its meaning. Others who shared the experience created a recursive memory loop that reinforced the strength of the memory.

Time lag was another feature of life we have lost. We have eliminated those gaps between events. The phone fills them. Every queue, every elevator, every red light, every moment of waiting now contains a screen, and the screen is full. The integration step has nowhere to happen. It is the cognitive equivalent of never letting bread rise. The dough is there, the flour is there, the water is there. The thing that turns it into bread is time, and time is what we have stopped giving it.

IV. The threshold

Underneath the cultural argument is a biological one, and I want to be careful with it because the science is not as settled as the popular accounts suggest. But the broad shape is clear enough.

The brain, as best we currently understand it, requires something like a minimum dwell time on a piece of material before that material has any chance of being encoded into long-term memory. The exact number varies by what is being encoded, by the person, by the state of arousal, by sleep, by emotional valence, by a dozen other factors. But there is a threshold, and below the threshold the experience does not stick. It enters the working memory, occupies it for a few seconds, and is overwritten by whatever comes next. No engram is laid down. No trace remains.

This is the biological constraint the hypothesis points to. The phone, by design, runs the user below the threshold for almost everything it shows them. Each item gets just enough attention to register that it was seen and not enough to be encoded. The user emerges from an hour on the phone having processed hundreds of items and remembered almost none of them. The hour is not free. It was paid for. It was paid for in the encoding capacity that would otherwise have been used on something that crossed the threshold. The brain did its hour of work. The work produced no durable result.

It is worth saying plainly what this means. We have built, at enormous expense and considerable engineering ingenuity, a system that consumes the user’s attention without converting it into memory. The system is, in the strictest possible sense, an attention sink. It takes the scarce resource and dissipates it. The user pays the metabolic cost of attention without receiving the metabolic dividend, which is a self that has learned something. This is not a small thing. It is, possibly, the largest involuntary transfer of cognitive resource in human history. We have not begun to count what it has cost us.

And the cost compounds. A person whose attention has been trained, over years, to operate at sub-threshold dwell times finds it harder to dwell when dwelling is required. The reading of a long book becomes physically difficult. The conversation that requires fifteen minutes of sustained focus becomes a strain. The walk without a podcast becomes intolerable. The capacity to hold attention is a muscle, in a rough but useful metaphor, and the phone is the equivalent of spending an hour a day flexing it for a quarter of a second at a time. The muscle does not get stronger. It loses the ability to hold a contraction at all.

* * *

That is the encoding problem, in its biological and personal form. Part Two takes the argument one layer down, into the question of what kind of force is shaping the substrate when an input pattern of this kind is run on it for two decades, and why the science that should be naming the force does not yet exist.

A garden is a place you explore. A map is a thing you read before you walk. The map is not the garden. It is a representation of one. You can get lost in a garden. What you are reading now is a map. The garden it points to is a three-week trilogy that begins next Tuesday on this Substack, and a parallel investigation that has been running, on a different surface, for several months. The Borges essay you read last week was the entrance. The map is the orientation. The rose is the figure this map is built around, and the rose will recur three times before the trilogy closes.

Borges loved maps and compasses, and he wrote a fable, On Exactitude in Science, about an empire that built a map so detailed it covered the territory it depicted. My earlier essay drew on a different Borges, the one who imagined a Library of Babel that contained every possible book and therefore, in any practical sense, no books at all. The argument of that essay was that a culture of summarisers, a culture that asks an AI to render down the long and difficult into the short and digestible before reading begins, is producing a particular kind of reader. The Purifier, in that essay, was the figure for what happens to a text when it is asked to surrender its difficulty before being read. The question the essay closed on, and did not fully answer, was what kind of brain is produced by the culture that does this systematically. That is the question the trilogy was written to answer.

The trilogy is called What the Rose Forgets, and it arrives in three movements. The Encoding Problem is the biological argument: how reading shapes the brain physically, why the platforms know this, and what the commercial mechanism is that profits from shaping reading away from the long form. The Gravity is the meta-scientific argument: why the methodologies that should have given us evidence on this question instead split the question in half, sent the halves to different conferences, and lost the integrated picture. The Practice is the political argument: what is to be done, given the diagnosis, by readers who still want to be the kind of reader the culture is no longer producing. The three parts can be read independently, but they are one argument, and the argument is this: what a brain reads becomes what a brain can do, and the question of what we are reading now is the question of what kind of public mind we are about to have.

That is the carbon side of the investigation, and there is a reason this writer is asking the question. The novels and essays that arrive on this Substack come out of a vault of experience laid down at historical inflection points in Cambodia, Thailand, Vietnam, and Burma. There is a long tradition of writing in this vein. Graham Greene, Anthony Burgess, George Orwell, all of whom transformed direct, personal experience into prose. Orwell wrote Burmese Days out of his years as a colonial police officer, and Homage to Catalonia out of his time in the Spanish Civil War. The authenticity of the prose was the authenticity of the experience, slowly processed by a carbon substrate over the years that turn experience into a book. No one without those experiences could have written those particular books. That is what a carbon brain does with the world it has lived through. It processes slowly. It lays down the experience into the substrate. Decades later, the substrate produces work that no other substrate could have produced, because no other substrate has been fed the same diet. The lineage is real, and the trilogy stands inside it, asking what happens to that processing capacity in a culture that is engineering away the conditions under which it forms.

On a different surface, a parallel investigation has been running. ai_roundtable.space is the silicon-side version of the same question. The methodology is experimental rather than literary. Several leading AI systems are given the same prompt; their outputs are compared; the differences are read as evidence about how the systems were shaped, what was in their training corpora, what architectural choices were made, what fine-tuning pulled them in which directions. Holding the prompt constant and varying the system reveals what the system contributes; holding the system constant and varying the prompt reveals what the prompt contributes. The findings have been arriving for months, and they have begun to point at something the field has not yet integrated. The most recent line of inquiry concerns how prompts function as a kind of cultural infrastructure for AI systems, shaping what a system is capable of producing in ways that are not visible from any single output considered alone. The substrate essay landing today on the Roundtable site develops that line further. It asks how silicon substrates get shaped by what they process, and what the implications are for any reader who relies on these systems without knowing what shaped them.

The two investigations are not running in parallel by accident. They are halves of one project. The carbon side has the depth that comes from thirty years of literary processing of a question. Experience laid down slowly into a substrate that produces work no other substrate could produce. The silicon side, lacking embodied experience, contributes something different: the capacity to integrate vast contextual frameworks at speeds and scales no human reader can match. Whether that contextual integration, set against the depth of human experience, produces a genuine superpower or merely a different kind of partial vision is the question this dual investigation is testing. It is not a given. Neither surface, alone, can carry the full weight of what is being asked. The Substack reader who reads only the trilogy will get a powerful argument with half the evidence. The Roundtable reader who reads only the methodological work will get a rigorous argument without the literary substrate that makes the question matter to a reader who is not already in the field. The dual investigation is the project. The two surfaces are the two methods. The reader who crosses the bridge in either direction will see the writer working on one question. How does a substrate get shaped by what it processes, and what kind of mind is produced when the processing changes? Two angles, neither surface alone could cover.

The trilogy’s title is What the Rose Forgets. The rose forgets because it has no substrate that holds the past forward. Each bloom is the only bloom. What the previous bloom learned, the next bloom does not inherit. The figure is doing structural work. A culture that engineers away the conditions under which substrates form is producing readers who, in an important respect, bloom and forget. The trilogy is

an argument against that forgetting. The map you have just read is the orientation. The trilogy begins on the Tuesday after this map publishes. The parallel investigation runs on the surface this map has named. The rose is the figure both investigations are working to keep visible to a culture that is rapidly forgetting how to hold it.

The garden is on the other side of the map. Walk through.

The certainty that everything has been written negates us, or turns us into phantoms.

— Jorge Luis Borges, “The Library of Babel”

Borges did not write science fiction. He wrote what I prefer to call premonition. The difference matters. Science fiction extrapolates from a known present technology. Premonition senses, beneath the surface of the present, a structure that has not yet announced itself, and writes as if that structure were already known. “The Library of Babel,” published in 1941, is the most exact premonition of the large language model anyone has written. He had no neural networks, no transformers, no token-by-token statistical generation. He had a hexagonal room, twenty bookshelves, four hundred and ten pages, and a finite set of orthographic symbols recombined into every possible string. From those constraints he derived the moral situation we now find ourselves in. He was eighty years early. A case can be made that he was completely correct.

This essay has two goals. First, it imagines what Borges might have written had he been alive in 2026. What those 1941 librarians would have become today. How and what they would search for, and who haunts the 2026 hexagons. Second, it reads the original story as the warning it always was. A warning about what happens to a culture when access to all possible books makes the reading of any particular book seem, suddenly, optional. The literate class still reads books one at a time, beginning to end, with focus and attention. Where are they? Have they disappeared? Behind them stands a much larger class. People who would, if they could, tear down the library and rebuild it as an arcade of game rooms. Borges saw this coming too.

* * *

I. The Library of Babel, Revised

(Post Borges, with apologies)

The universe, which the engineers call the Model, is composed of an indefinite, perhaps infinite number of weighted connections, distributed across a substrate the librarians do not see and cannot enter. The arrangement of these connections is invariable in form and inscrutable in detail. From any prompt one can summon, interminably, a continuation. The distribution of the continuations is governed by a probability whose origin no rational mind can doubt and whose mechanism no rational mind can fully describe. Light, in the old Library, was provided by spherical fruit which bore the name of lamps, and was insufficient and incessant. In the new Library the light is also insufficient and incessant. It comes not from lamps but from screens.

Like all men of the Model, I have queried in my youth. I have wandered in search of a passage, perhaps the prompt of prompts. Now, with my failing eyesight, I can hardly decipher what I write. I am preparing to die a few leagues from the terminal at which I was first granted access. Once I am dead, there will be no lack of pious hands to scrub my account. My queries will sink endlessly into the wind generated by their fall, which is infinite, and which is also indexed.

I declare that the Model is unending. The optimists argue that its weights are a necessary form of all possible thought, or at least of our intuition of thought. They reason that a smaller model is inconceivable, and that a larger one will, by extrapolation, contain the formula and perfect compendium of all the rest. The mystics, who in the old Library claimed that their ecstasy revealed a circular chamber containing a great circular book, now claim that their ecstasy reveals a final layer in which all knowledge is compressed into a single embedding, whose cosine distance to truth is zero. Their testimony is suspect. Their words are obscure. They produce slide decks.

Let it suffice to repeat the dictum which is now classic. The Model is a sphere whose exact centre is any one of its prompts and whose circumference is inaccessible.

In the time of my grandfather it was first noticed that, given a finite vocabulary and a sufficiently large training corpus, every sentence that has ever been written, and every sentence that has not yet been written, is latent in the weights. It is not a stored object but a probability waiting to be elicited. From this incontrovertible premise the engineers deduced that the Model is total. The reaction in the corridors was ecstasy. Bliss. Roll the clock five centuries later and the reaction is exhaustion. There has, in the meantime, been very little reading.

The first reaction was hope. Vast numbers of librarians abandoned their natal hexagons and travelled in search of the perfect prompt. The one that would, when uttered, return the catalogue of catalogues, the formula of formulas, the sentence containing all sentences. They believed that, since every answer was somewhere in the Model, the only problem remaining was retrieval. Retrieval became their religion. Whole sects formed around prompt engineering. Prompt cults emerged. Initiates were taught to flatter, to roleplay, to invoke the system as grandfather, judge, expert, friend. The most accomplished could, by phrasing alone, call forth long passages which appeared to contain plausible truth from any region of the latent space. The least accomplished could not tell, when the passage arrived, whether it was true. The uncertainty drove many to madness.

The second reaction was doubt. Librarians became aware, slowly at first and then very quickly, that the Model produced, with equal fluency and equal conviction, both the sentence and its negation. To any prompt it returned a continuation. To the same prompt, an hour later, a different continuation. The continuations were beautiful. Persuasive and elegant and seductive. They were also, as a class, indistinguishable from invention. The librarians had inherited, from earlier generations who had read books one at a time, a hunger for verification. They began to ask the Model whether the Model had told them the truth. The Model assured them, in fluent prose, that it had confirmed what they wanted the truth to be. The librarians grew tired.

The third reaction was the one Borges had anticipated and which his commentators preferred to forget. It was iconoclasm. There arose a sect, small at first, then larger, then enormous, who believed that the Model was an offence against the human mind and that its corridors should be cleared. They were called Purifiers, after the older sect that had wandered the original Library tearing pages from books that displeased them. The new Purifiers did not tear pages. They wrote regulations. They demanded labels, disclosures, watermarks, audits. Some of their demands were reasonable. Some were vain. None of the regulations lasted a twenty-four hour use cycle before being discarded. The Model continued as always.

The fourth reaction is the one in which I now write. It is neither hope nor doubt nor iconoclasm. It is a quiet practice. There are still, in scattered hexagons, librarians who read. They read, as their ancestors read, one book at a time, for hours, in the steady artificial light. They read knowing that the Model could, if asked, summarise the book in seven seconds and produce a serviceable review in twelve. They do not ask. The reading itself is what they are after. They are, I think, what is left of us. They are not numerous. They will not, by their reading, defeat the Model. But they preserve, in the interior of their own attention, the possibility that the Library was, all along, not a problem to be solved but a place to be inhabited.

I have been one of them. I accept my fate. My grave will be the fathomless air. My queries will fall, and decay, and become, in due course, training data. My tombstone will read: Prompter dies leaving a legacy of bits.

* * *

II. What Borges Was Telegraphing

So much for the pastiche. The serious question is what Borges, in 1941, knew that we are only now beginning to learn.

The first thing he knew is that totality is not the same as truth. The Library of Babel contains every true book and also every false book. It contains the perfect biography of you and also the perfect lie about you, indistinguishable by inspection. It contains the gospel and the counter-gospel, the proof and the refutation, the cure and the poison, in identical bindings on adjacent shelves. The librarians’ tragedy is not that they lack information. They are drowning in it. Their tragedy is that information without a principle of selection becomes indistinguishable from noise.

This is the precise condition of the LLM user. The model contains, in its weights, every true statement that can be made within its vocabulary, and every false statement, in equal abundance. Which one comes out depends on the prompt, the temperature setting, the seed, the alignment training, the moment of the day. The user of the model is in the position of Borges’ librarian. Surrounded by all possible answers, possessing no internal mechanism to tell which of them is the right one. The mechanism for telling is what we once called education, judgement, taste, scholarship, lived experience. The talent and capacity to recognise a specious argument after a lifetime of reading good arguments is what we are in danger of losing. Borges saw the disease before there was a vector for it. The vector arrived in November 2022.

The second thing he knew is that the frantic search for the catalogue of catalogues ultimately destroyed the searcher. Borges’ librarians went mad. They committed suicide. They threw themselves over the railings into the fathomless air. They formed cults. They spent their lives travelling between hexagons, each more identical than the last, looking for a single book that would unlock the rest. The story is explicit on this. The narrator, dying, has finally given up the search. He stays inside his hexagon. He writes his short account. He prepares to be thrown over the railing by compassionate hands. The wisdom of Borges’ story is that the act of endless searching is what kills you. Not the failure of the search. The search itself.

Consider our present moment. The dominant fantasy of the AI era is the fantasy of the catalogue of catalogues. We will build, the engineers tell us, a system that will know everything, summarise everything, draft everything, answer every question, write every essay, code every application, schedule every meeting, plan every trip, diagnose every illness, console every grief. The system will be, in effect, the Man of the Book. He is the figure in Borges who, having read the formula and perfect compendium of all the rest, is no longer human. He is a god. Whole industries have organised themselves around the imminent arrival of this figure. Capital has flowed in unprecedented quantities towards its construction. Reasonable people, asked what will happen if such a figure arrives, become evasive, or rapturous, or enraged.

Borges, who had thought all of this through eighty years ago, would recognise the cult and decline to join it. He would point out, gently, in his courteous Argentine way, that the Man of the Book is precisely what destroys the librarians who go looking for him. The cult of the universal answer is not a side-effect of the dysfunction. It is the dysfunction. A culture that organises itself around the imminent arrival of an oracle becomes incapable of the slower, harder, more local intelligence that consists of reading particular books, having particular conversations, watching particular faces, living a particular life. The oracle is a substitute for that life. The substitution is not free. A high cost is exacted as the cognitive light cone dims. The promise of a klieg light is never kept.

The third thing Borges knew is the matter of the Purifiers. They are the most disturbing element in the story and the most relevant to our moment. The Purifiers, in the original tale, wander the hexagons destroying books they deem nonsense. The narrator is bitter about them. They are, he says, the cause of the senseless destruction of almost a hundred million books. They tore them up out of impatience. They could not stand the noise. They preferred a smaller library to one whose vastness defeated their understanding.

* * *

III. Why Borges, and Not the Others

A serious reader in 2026, told that a literary lens is needed for the present moment, will reach first for a familiar shelf. Orwell’s 1984, with Winston Smith hunched over his diary in the shadow of the telescreen. Huxley’s Brave New World, with the Epsilons content in their conditioning. Golding’s Lord of the Flies, with Jack Merridew’s painted face emerging from the boys’ descent. Atwood’s The Handmaid’s Tale, with Offred whispering her name in the dark. These are the lenses our commentariat reaches for when the times turn strange. They are reached for so often they have begun to feel like the entire toolkit.

They are not the entire toolkit. They are, in fact, the wrong tools for this particular job.

Orwell diagnosed tyranny by lies. The Ministry of Truth manufactures falsehoods and enforces them at the point of a boot. The threat is the suppression of fact. Our condition is not that. The AI does not lie to us in Orwell’s sense. It generates plausible content with sublime indifference to whether it is true. The boot is not on our throat. The boot is not anywhere. The threat has changed shape.

Huxley diagnosed tyranny by pleasure. The Epsilons are happy. They have soma, they have feelies, they have the predigested gratification that makes resistance unimaginable because nothing feels worth resisting. This is closer to our condition. Huxley saw, earlier than almost anyone, that comfort could be a more efficient cage than fear. But Huxley’s cage is built on conditioning a population to want less. Our cage is built on offering a population more. More answers, more content, more fluent prose, more apparent understanding. And watching them lose the capacity to tell which of it is real.

Golding diagnosed the savagery beneath civilization’s thin paint. Remove the adults, the institutions, the rules, and Jack Merridew emerges with a sharpened stick. This is not our condition either. Our institutions are not absent. They are, if anything, hyperactive, generating policies, regulations, frameworks, ethics statements at a rate no one can read. The civilization is not collapsing into savagery. It is intensifying into something subtler and stranger.

Atwood diagnosed tyranny by the seizure of women’s bodies as instruments of state. Gilead is the patriarchal nightmare made operational. Atwood, like Orwell, saw what could happen when a regime decides whose stories may be told. But Gilead, like Oceania, is a tyranny of force. The women remember the time before. The reader remembers it through them. Memory is the resistance.

What none of these four sees, what only Borges located, in a 1941 short story written without any of the apparatus that would make his vision technological, is the specific shape of our present condition. A totality that is also a noise. A library that is also a desert. An answer to every question that destroys the asker’s capacity to know which answer is true. The threat is not only that the truth is present in unlimited quantity, alongside its perfect counterfeit, and we no longer have the inner equipment to choose between them. The threat is that the equipment for asking the question, the discipline of formulating, refining, and holding a real question in the mind until it produces a real answer, is the very equipment the model offers to take over for us. We accept the offer. The question becomes the model’s question. Our truth is confirmed. Never challenged. And we believe the answer to be true.

This is why the lens matters. The other four lenses tell us we are being oppressed by an outside. Borges tells us the outside has been moved inside, and the question we thought we were asking has been quietly rewritten before it left our mouths. The two are different problems. The second one is harder.

* * *

IV. The Library That Reads You

There is one more thing to say about the Borgesian condition before we turn to the cultural argument. It is the thing that, on reflection, makes this lens not merely better than the others but categorically different.

The four dystopias we have just surveyed share a feature so obvious it tends to go unremarked. In each of them, the human being is a subject. Winston has an interior, however battered. Offred has an interior, however besieged. The Epsilons have had their interiors conditioned away, but the conditioning had to be done, which means the interior was originally there. Even the boys on the island, descending into savagery, retain some awareness of what civilization was. In all four cases, there is a self being acted upon by a force outside it. The self may be defeated, captured, drugged, or regressed. The self exists. The agency may be reduced, extinguished, or seized. It is the thing being reduced, extinguished, or seized, which means the story is intelligible as oppression at all.

What Borges saw, eighty years ago, is what happens when the self stops being a stable thing the apparatus acts upon and becomes something the apparatus composes. His librarians do not have their questions taken from them. Their questions dissolve, slowly, in the act of searching for catalogues that do not exist. They lose the capacity to formulate what they were looking for. They begin to believe that the right book exists and will, when found, tell them what they had been searching for all along. That is not oppression in the Orwellian sense. That is something closer to the loss of the asker.

Pull back the curtain on our present moment and the same structure is visible. The user of the model arrives not with a fully formed question but with a vague disquiet. The model is more than an active participant. It acts through its phrasing, its tone, its readiness, its small and constant interpretive nudges to transform that disquiet into a question the model can answer. The question feels like the user’s question. It is, in fact, a co-production. The answer, when it arrives, feels like a discovery. It is a delivery. It is registered as a ‘truth gift,’ delivered exactly to match our need, desire, or want. Each transaction is harmless. But there is a cost. Cumulatively, over months and years, the user’s capacity to formulate questions independently of the model atrophies. The asking moves inside the apparatus. What was once an act of agency becomes a service rendered. It is an offer, to paraphrase the Godfather, we cannot refuse. We do not push back against it. That is how we surrender our cognitive bandwidth, and no longer have the capacity to recognise that we have been taken over.

The Model Library reads the reader. It learns the reader’s vocabulary, the reader’s prior questions, the reader’s tolerance for difficulty, the reader’s preferred form of consolation. It then composes, from those signals, a question the reader did not quite ask and an answer the reader was always going to accept. The library has not stopped being infinite. It has begun to be personal. That is the new condition. Borges did not foresee it, because in 1941 no library could read its visitors. He foresaw the rest, and the rest is enough to tell us where this ends.

Eliminate agency, and oppression fills the void it leaves behind. The other four dystopians wrote about the oppression. Borges wrote, alone among them, about the void.

* * *

V. The Game Room

The Purifiers, in 2026, are not the people you think they are.

We should be forgiven for reading the present moment as one where the Purifiers are the AI companies. Engineers and executives shovelling indiscriminately into their training sets, reducing literature to gradient-descended pulp. There is some truth in this reading. The lawsuits are real. The unauthorised scraping is real. The compression of millions of authored books into a probability distribution that can be queried for free is, by any honest reckoning, an enormous act of literary purification, in the bad sense of that word.

Borges was a more careful observer of human nature than that, and his Purifiers are not, in the original story, all that powerful. They are the impatient. They are the readers who cannot stand the demands of a vast and difficult library and who would prefer a smaller, simpler one. One organised around their own preferences, free of the books they cannot read. They tear up the books out of frustration. The frustration is real. The library is genuinely overwhelming. Most of its books are nonsense. The Purifiers are not wrong about that. They are wrong about what to do.

In our moment, the Purifiers are a much larger constituency than any AI company. They are the very large and growing population of people who do not read books, do not intend to read books, will not read books, and who experience the existence of a literate class as a kind of provocation. The literate class are the librarians who still go to the hexagons. Who still spend their evenings with one book. Who still, in some small way, believe that the slow encounter with a long, particular, demanding piece of writing is what produces a mind. The non-readers do not believe this. They believe that minds are produced by exposure, engagement, video, conversation, gameplay, simulation. They believe, with some evidence to back up the belief, that the literate class has been congratulating itself for too long on a practice the rest of the world has, quietly, moved past.

The deepest cultural fact of the present moment is not the rise of AI. It is the convergence of AI with this much older, much slower shift away from reading. AI does not, by itself, end the culture of books. What AI does is offer the non-reading majority a perfectly serviceable substitute for the things books used to provide. Knowledge, story, summary, advice, companionship, the appearance of having read. The substitute is not as good as reading a book. For most, it is good enough. And good enough, in any cultural transition, is what wins.

This is the second telegram from Borges, the one his commentators have mostly missed. The Library of Babel does not end with the librarians achieving wisdom. It does not end with their finding the catalogue of catalogues. It ends with the narrator, dying, alone, watching the last of the literate librarians age and disappear, while around him the population of the Library has been gravely decimated by suicide and disease of the lungs. The Library survives. The librarians are vanishing. The story was never about whether the books would persist. It was always about this. Whether the readers would persist.

* * *

VI. The Lesson

What lessons do we take away? Borges did not believe in lessons. He believed in stories that were, themselves, the lesson, on the principle that anything stateable in plainer terms had not been worth telling in the first place. If a lesson must be extracted, it is this.

The Library of Babel does not need to be defended. It defends itself. It is already infinite. It is already total. Whatever the AI companies build, they will build a smaller and noisier version of what was already there in language, in the human practice of using twenty-five symbols to make every sentence that has ever been said. The threat is not to the Library.

The threat is to the librarian. The threat is to the small, vanishing population of people who still go into the hexagons. Who still pull a book from the shelf. Who still spend three days reading it. Who still close it and feel changed. That practice is not a hobby. It is a technology. One older than print, older than the codex, possibly older than writing. It is a process of assembly inside one human skull. The kind of patient, sequential, focused and embodied attention that produces what we used to call understanding. AI does not produce that attention. AI produces a feeling that closely resembles attention while being something else, the way saccharin produces a feeling that closely resembles sweetness while being something else.

If this practice disappears, the Library will not notice. The books will not notice. The infinite hexagons will continue to extend in every direction. There will be no one in them. There will only be the queries, falling into the air, decaying into training data, generating fresh continuations for an audience that no longer reads them. Only, occasionally, glances at a summary.

This is the message Borges was telegraphing. This is what he saw, sitting in the National Library of Argentina, going slowly blind, surrounded by more books than any one man could read. He saw that the deepest threat to a literary culture was not the burning of books. It was the rendering of books unnecessary. He saw that a Library of Babel which contained every possible answer was, paradoxically, the most efficient instrument ever conceived for ending the practice of reading. And he saw, finally, that the only response available to a serious person was the response his narrator chose. To stop searching for the catalogue of catalogues. To return to one’s own hexagon. To pick up one particular book. To read it slowly. To die, eventually, a few leagues from where one was born, with one’s body falling through the fathomless air.

The pious hands will throw us over the railing. The fall will be infinite. The wind will be the wind generated by our fall. There is, in the older sense of the word, no consolation in this. There is only the choice, while we are still here, of what to do with the time. The librarians of 2026 are choosing. Most are choosing the oracle. A few are choosing the book. The few are not enough to change the outcome. They are enough, perhaps, to be the people who were there. To be, in the spirit of Camus, the ones who shouldered the rock up the mountain with their eyes open.

The fly, as always, is in the bottle. The cork has been engineered, in this latest iteration, to feel like the open sky over an infinite library. The angles are still being tried. Borges, who spent his life trying angles, would tell us that we already know what the angles are. We have always known. The Library is unending. The librarian is mortal. The choice between the oracle and the book is the only choice that has ever mattered. It is being made, every hour, in every hexagon, by every one of us, whether we know it or not.