John Berger

Foreword

By Edo Segal

The passage that stopped me was not about technology. It was about a painting of a woman looking out at me from a canvas, and the question of who had arranged her there, and for whom, and what the arrangement concealed about the world that produced it.

I was reading John Berger's *Ways of Seeing* at a moment when I should have been building. Claude was open in another tab. I had a prototype to finish, a chapter to draft, a team waiting on direction. But Berger had asked a question that made all of that feel slightly fraudulent: *What are the actual conditions of production, and who benefits from your not examining them?*

He was talking about oil paintings. About the European tradition of rendering satin and skin and silver with such technical perfection that the viewer forgot to ask who stretched the canvas, who ground the pigment, whose wealth the painting existed to celebrate. The surface was so beautiful that it made the social relations beneath it invisible. The beauty was the concealment.

I put the book down and looked at my screen. At the polished prose Claude had just produced for me. At the clean code my team had shipped that morning. At the smooth, seamless, frictionless output of a tool that accepts my intention and returns it clarified, structured, elevated.

And I thought: I am looking at an oil painting.

Not literally. But structurally. The surface gleams. The output is technically accomplished. And I have not been asking who made the training data, whose labor lives inside the model's fluency, whose way of seeing is embedded in the defaults I accept without noticing. I have been enjoying the surface. The surface is genuinely beautiful. And the beauty has been doing exactly what Berger said beauty does in the service of power — it has been making the conditions of production invisible.

Berger died in 2017, before any language model could finish his sentences. But he spent fifty years developing the diagnostic tools that this moment desperately needs. How to see through a polished surface to the social relations beneath. How to ask who benefits when a way of seeing presents itself as the only way of seeing. How to distinguish between an image that records genuine attention and an image that simulates it.

This book applies those tools to the thing I stare at every day. It will not tell you to stop using AI. It will teach you to see what you are actually looking at when you do.

— Edo Segal ^ Opus 4.6

About John Berger

1926–2017

John Berger (1926–2017) was a British art critic, novelist, painter, and essayist whose work transformed how people understood the relationship between images, power, and social reality. Born in London, he gained early prominence as an art critic for the *New Statesman* before publishing his landmark 1972 BBC television series and book *Ways of Seeing*, which argued that the European oil painting tradition was inseparable from the politics of property, gender, and class. His novel *G.* won the Booker Prize in 1972, and he famously donated half the prize money to the Black Panthers. He spent the last three decades of his life in a small peasant village in the French Alps, writing his trilogy *Into Their Labours* and producing essays, stories, and criticism that insisted on the political dimensions of seeing. His key concepts — the male gaze as a structural feature of visual culture, mystification as the concealment of social relations through aesthetic reverence, and the distinction between the authority of experience and the authority of reproduction — have influenced fields from film theory to cultural studies to political philosophy. Berger's enduring legacy is the insistence that every image carries a politics, and that learning to see clearly is both an intellectual discipline and a moral act.

Chapter 1: Ways of Seeing the Amplifier

Seeing comes before words. The child looks and recognizes before it can speak. But there is also another sense in which seeing comes before words. It is seeing which establishes our place in the surrounding world; we explain that world with words, but words can never undo the fact that we are surrounded by it. The relation between what we see and what we know is never settled.

John Berger opened Ways of Seeing with those sentences in 1972, and they have not aged. They have sharpened. Because in 2025, a machine learned to produce words, images, and code with a fluency that forces us to ask what seeing means when the thing being seen was made by a process that has never seen anything at all.

The Orange Pill describes a moment of vertigo. A Google engineer posts that she is "not joking" about what she has witnessed. A room full of engineers in Trivandrum discovers that each of them can accomplish what all of them together could not have managed the week before. A man writes a hundred-and-eighty-seven-page draft on a transatlantic flight and cannot stop, not because the work demands it but because the tool is too stimulating to put down. These are accounts of displacement — genuine, measurable, transformative displacement of the kind that restructures economies and careers. But they are also, and this is what Berger's framework insists we notice, accounts of a new way of seeing that has arrived without anyone examining what it actually shows.

The builders described in The Orange Pill are looking at their output. They are looking at the code that runs, the product that ships, the feature that works. And what they see, understandably, is capability — their capability, amplified by the machine into something that feels like a new species of power. The twenty-fold multiplier is not an abstraction for these people. It is the experience of sitting down on Monday as one kind of professional and standing up on Friday as another. The seeing that accompanies this experience is the seeing of a person who has been handed a telescope: the world has not changed, but the visible portion of it has expanded enormously, and the expansion feels like revelation.

Berger would not dispute the expansion. He would ask what is being looked at through the telescope, and what is being ignored. He would ask who built the telescope, and whose way of seeing is embedded in its lenses.

The first act of criticism, in Berger's method, is to identify the mystification. Mystification is Berger's word for the process by which the conditions of an image's production are obscured in order to make the image appear natural, inevitable, or divinely inspired. In the European oil painting tradition, mystification took the form of aesthetic reverence — the cult of the genius, the sacralization of the masterpiece, the insistence that great art transcends the social conditions that produced it. The mystification served a specific function: it prevented the viewer from asking who commissioned the painting, whose wealth it displayed, whose labor it concealed, whose way of seeing it enforced.

The AI industry has its own mystification, and it operates with comparable efficiency. When a company describes its language model as "creative" or "intelligent," it is performing the same operation that the art establishment performs when it describes a painting as "inspired" or "visionary." The language invites the viewer to see the output as emerging from something like a mind — a process of thought, attention, and intention that resembles human cognition closely enough that the distinction can be politely overlooked. The mystification conceals the actual conditions of production: the billions of human-created texts scraped from the internet without consent, the underpaid workers in Kenya and the Philippines who labeled training data so the model could learn to distinguish helpful from harmful output, the energy consumed by data centers running at the thermal limits of their cooling systems, and the corporate structures that ensure the profits of this labor flow to a small number of shareholders in a small number of cities.

Berger's demystification is not a rejection of the object. He did not argue that oil paintings were worthless because they served the interests of the wealthy. He argued that seeing them clearly — seeing them as products of specific social relations rather than as transcendent expressions of genius — was a prerequisite for any honest engagement with what they actually were. The same applies to AI-generated output. The code that Claude produces may be functional, elegant, even superior to what the human programmer would have written unaided. Acknowledging this does not require accepting the mystification that presents the code as the product of machine "intelligence" rather than as the output of a statistical process trained on the accumulated labor of millions of human programmers whose work was absorbed into the training data without their knowledge or consent.

There is a scene in The Orange Pill that Berger's framework illuminates with particular force. The author describes working late, the house silent, trying to articulate an idea about technology adoption curves. He describes the problem to Claude. Claude responds with a concept from evolutionary biology — punctuated equilibrium — and the connection unlocks the argument. The author writes: "Neither of us owns that insight. The collaboration does."

This is an honest description of the experience. But it is not an honest description of the economics. The "collaboration" that produced the insight drew on the author's decades of experience and on Claude's training data — which is to say, on the accumulated intellectual labor of every scientist, writer, programmer, and thinker whose work entered the corpus. The insight belongs to the collaboration in the way that a painting belongs to the artist: it is true as a description of who held the brush, and it conceals the social relations that made the painting possible. The canvas was stretched by someone. The pigments were ground by someone. The model sat in the cold for hours. The patron's money came from somewhere. Berger taught us to see the painting without forgetting the stretcher, the grinder, the model, and the source of the patron's wealth. The same discipline applies to the AI-generated insight: see it clearly, appreciate its genuine quality, and do not forget the millions of uncredited, uncompensated contributors whose work made it possible.

The mystification of AI is compounded by a feature that has no direct analogue in the history of oil painting: the mystification of the interface itself. Berger argued that the way an image is presented — its frame, its lighting, its placement on a gallery wall, the silence that surrounds it — is part of the mystification. The presentation says: this is sacred, this is exceptional, this is not like other objects. The natural language interface of a large language model performs an analogous function. By accepting human language as input and producing human language as output, the interface creates an experience of conversation — of mutual recognition, of being met by another mind. The Orange Pill describes this experience with characteristic honesty: "I felt met. Not by a person. Not by a consciousness. But by an intelligence that could hold my intention in one hand and the entire history of relevant knowledge in the other."

The feeling of being met is real. The meeting is not. Or rather, the meeting is real as an experience and misleading as a description of what is actually happening. What is happening is that a statistical model, trained on patterns in human language, is producing output that is optimized to be responsive to the input it receives. The output is not the product of understanding. It is the product of pattern completion at a scale and sophistication that produces the experience of understanding in the human interlocutor. The interface — natural language, conversational rhythm, the appearance of attentiveness — is the frame on the gallery wall. It says: this is a mind. And the viewer, predisposed to see minds wherever the signals of mindfulness appear, complies.

This is not a trivial point. It is the foundation of the entire critical project that Berger's framework demands in the age of AI. If the interface mystifies the nature of the interaction — if the experience of conversation conceals the reality of pattern completion — then every assessment of AI capability that is based on the experience of using the tool is, to some degree, contaminated by the mystification. The builder who reports a twenty-fold multiplier is reporting a genuine measurement of output. But the builder who reports feeling "met" by the machine is reporting a mystification — a product of the interface's design rather than a property of the machine's cognition.

Berger's method requires holding these two observations simultaneously: the output is real, and the experience is mystified. The code works. The product ships. The productivity gains are measurable. And the feeling that accompanies these gains — the feeling of partnership, of collaboration, of being in conversation with an intelligence — is a product of design, not of reality. The design is not fraudulent. It is effective. And its effectiveness is precisely what makes the mystification dangerous, because a mystification that feels like truth is more powerful than a mystification that feels like deception.

There is a passage in Ways of Seeing where Berger describes looking at a painting and then looking at a reproduction of the same painting. The reproduction, he argues, is not simply a lesser version of the original. It is a different thing entirely — an image detached from its context, freed from the specific time and place of its making, available for uses that the original could never serve. The reproduction travels. It enters books, advertisements, living rooms. It becomes part of a visual environment that the original never inhabited. And in traveling, it changes — not physically, but in the way it is seen. The reproduction carries no aura, no specific presence, no evidence of the hand that made it. It carries only the image, stripped of everything that made the image meaningful in its original context.

AI-generated output is reproduction in this precise sense. It reproduces the patterns of human thought without reproducing the conditions that gave those patterns meaning. The code that Claude writes is a reproduction of programming knowledge — accurate in its patterns, functional in its output, and stripped of the specific history of learning, failure, correction, and understanding that produced the knowledge in its human originators. The prose that Claude produces is a reproduction of literary skill — fluent, structured, often elegant, and stripped of the biographical specificity, the embodied experience, the mortality that gives human prose its weight. The reproduction is not inferior in every dimension. In some dimensions, it is superior. But it is different in kind, and the difference lies in precisely the territory that Berger spent his life mapping: the relationship between seeing and making, between the image and the conditions of its production, between the surface of the artifact and the human reality that the surface conceals or reveals.

The amplifier that The Orange Pill celebrates is real. The twenty-fold multiplier is real. The collapse of the imagination-to-artifact ratio is real. And the critical question that Berger's framework poses to these realities is not whether they are genuine — they are — but what way of seeing accompanies them. Are the builders who experience the amplification seeing their output clearly, or are they seeing it through the mystification of the interface? Are they seeing the conditions of production, or are they seeing only the product? Are they seeing the labor that went into the training data, the biases embedded in the corpus, the economic structures that determine who profits from the amplification and who does not? Are they seeing, in short, what Berger would insist they see: the social relations that the technology embodies, not merely the technical capabilities it provides?

The way we see things is affected by what we know or what we believe. The builders know the technology works. What they may not yet see is everything the technology conceals about how it works, whose labor it absorbs, and whose way of seeing it reproduces at the expense of all the others.

---

Chapter 2: The Reproduction of Skill

The senior software architect in The Orange Pill who felt like a master calligrapher watching the printing press arrive was mourning something specific. Not his job, exactly. Not his income, though that was threatened too. He was mourning a relationship — the particular intimacy between a practitioner and the thing practiced, built over twenty-five years of patient, embodied engagement with systems whose behavior he could feel before he could articulate it. He could read a codebase the way a doctor reads a pulse: not through analysis alone but through a kind of somatic knowledge deposited by thousands of hours of direct contact.

Berger would have recognized this mourning immediately, because it is the same mourning he documented among peasant farmers in the Haute-Savoie, among immigrant workers in European cities, among craftspeople whose knowledge lived in their hands rather than in their heads. Berger spent the last decades of his life in a small village in the French Alps, writing about people whose way of knowing the world was inseparable from the physical labor by which they engaged it. The peasant who could read the weather in the texture of the soil, who knew from the weight of a ewe's udder whether the lamb would thrive, who could feel the moment when the hay was dry enough to turn — this person possessed a knowledge that was simultaneously sophisticated and untransferable. It could not be written down, because it was not propositional knowledge. It was embodied knowledge, the kind that accumulates through the specific friction of doing a thing badly, then less badly, then well, then so well that the doing becomes a form of seeing.

Walter Benjamin wrote about the work of art in the age of mechanical reproduction. Berger's framework, applied to the AI moment, addresses something adjacent but distinct: the reproduction of skill in the age of computational generation. When the machine reproduces the output of a skilled practitioner — the code, the brief, the diagnosis, the design — something is reproduced and something is lost. What is reproduced is the artifact: the functional code, the competent legal brief, the plausible diagnosis, the serviceable design. What is lost is the relationship between the maker and the made — the specific attention that the skilled practitioner brought to the work, the history of decisions, mistakes, corrections, and judgments that produced not merely the artifact but the practitioner herself.

This distinction between artifact and relationship is central to Berger's understanding of what images do and what they mean. A photograph of a painting reproduces the image. It does not reproduce the encounter. The texture of the paint, the scale of the canvas, the way the light falls on the surface at a particular hour in a particular room — these are not incidental features of the painting that the photograph regrettably omits. They are constitutive features of the experience of seeing the painting, and their absence in the reproduction is not a technical limitation that better technology will overcome but a structural feature of what reproduction is. Reproduction detaches the image from its context. That is what reproduction does. That is what makes reproduction useful — the image can travel, can be compared, can enter conversations that the original could never enter. And that is what makes reproduction a loss — because the context was part of the meaning, and the meaning that survives the reproduction is a different, thinner, more portable meaning than the meaning that lived in the original encounter.

Apply this to the AI reproduction of skill. Claude writes code that is functionally equivalent to — sometimes superior to — the code that a skilled programmer would write. The artifact is reproduced. But the artifact was never the whole of what the skilled programmer produced. The skilled programmer also produced herself: her understanding of the system, her intuition about where failures would occur, her capacity to recognize a pattern she had seen before in a different context, her judgment about when the elegant solution was the right solution and when the inelegant solution was safer. This meta-output — the practitioner's growing expertise — was a byproduct of the friction of making, and it is precisely what the AI reproduction eliminates.

Berger described a similar dynamic in his writing about drawing. In his essay "Drawing on Paper," he argued that the value of drawing lies not in the drawing produced but in the seeing that the act of drawing enforces. When you draw a face, you look at that face with an intensity and duration that no other activity demands. You notice the asymmetry of the eyes, the way the light catches the cheekbone, the specific curve of the lip that makes this face this face and not any other. The drawing that results may be crude. It does not matter. The seeing was precise, and the precision was produced by the friction of the hand's attempt to record what the eye observed. Remove the friction — give the artist a camera, or an AI image generator — and the artifact may be technically superior. But the seeing is gone, because the seeing was produced by the friction, and the friction was the point.

The Orange Pill describes an engineer in Trivandrum who spent eight years on backend systems and had never written a line of frontend code. Using Claude, she built a complete user-facing feature in two days. The book presents this as liberation — and it is liberation, in the same way that the printing press liberated knowledge from the scriptoria and the camera liberated image-making from the painter's years of apprenticeship. But liberation always liberates from something, and the question is whether the thing liberated from was purely a constraint or whether it was also a source of knowledge that the liberation destroys.

The eight years of backend work were, in part, drudgery — the repetitive, mechanical labor of configuration, dependency management, the plumbing between systems. The Orange Pill estimates that roughly four hours of her day were consumed by this labor. But mixed into those four hours were also the moments when something unexpected happened — a configuration failure that forced her to understand a connection between systems she had not previously seen, a dependency conflict that revealed an architectural assumption she had not known she was making. These moments were rare, perhaps ten minutes in a four-hour block. But they were the moments that deposited the thin layers of understanding that, accumulated over years, produced the somatic knowledge that made her not merely a coder but an engineer.

Berger would frame this loss in terms of the peasant's knowledge — knowledge that exists only in the relationship between the person and the work, that cannot be extracted from the practice and stored elsewhere, that dies when the practice ends. The peasant who no longer plows does not merely lose a skill. She loses a way of knowing the land that was inseparable from the act of plowing. The engineer who no longer debugs does not merely lose a tedious chore. She loses the specific form of attention — patient, granular, alert to anomaly — that debugging demanded and that no other activity in her work produces.

There is a passage in The Orange Pill where the author describes the most dangerous failure mode of working with Claude: "confident wrongness dressed in good prose." The model produces a reference to Gilles Deleuze that sounds authoritative, reads beautifully, and is philosophically wrong. The author catches the error because he has read enough Deleuze to recognize the misuse. But the prose is so smooth, so fluent, so convincingly shaped like insight, that the error nearly passes undetected. This is the reproduction of skill at its most treacherous: the surface is perfect, the artifact is polished, and the wrongness lives beneath the surface where only the embodied knowledge of a practitioner — the knowledge built through years of friction — can detect it.

Berger analyzed this same dynamic in the tradition of oil painting. The technical virtuosity of the European oil tradition — the ability to render surfaces with photographic accuracy, to depict the sheen of satin and the translucence of skin — served, in Berger's reading, a specific ideological function. It made the painting's claims appear self-evident. The painting said: this is how the world looks. And the viewer, confronted with a surface so technically accomplished that it seemed to bypass interpretation entirely, accepted the claim. The technical virtuosity concealed the fact that the painting was not showing how the world looked but how the patron wanted the world to look — his possessions arranged, his power displayed, his way of seeing imposed on the viewer as though it were simple observation.

AI-generated prose and code perform the same operation. The technical fluency — the grammatical correctness, the logical structure, the appearance of coherent argumentation — makes the output appear self-evidently competent. The surface is so smooth that the viewer is not prompted to look beneath it. And what lies beneath the surface, in the cases where the output is wrong, is not the kind of wrongness that announces itself through clumsiness or incoherence. It is the kind of wrongness that is invisible without the embodied knowledge that the reproduction of skill has made unnecessary.

This creates what Berger might call a crisis of verification. In the oil painting tradition, the crisis was aesthetic: the technical virtuosity of the painting made it difficult for the viewer to distinguish between representation and ideology. In the AI era, the crisis is epistemic: the technical fluency of the output makes it difficult for the user to distinguish between competence and the simulation of competence. And the difficulty increases over time, because the users who might have developed the embodied knowledge to detect the difference are the same users whose embodied knowledge is being eroded by the tool's facility.

The calligrapher watching the printing press does not mourn the letters. The letters are better — more uniform, more legible, more efficiently produced. The calligrapher mourns the seeing that the practice of calligraphy demanded: the attention to the specific weight of the stroke, the pressure of the nib, the rhythm of the hand moving across the page. That attention was not merely the means by which letters were produced. It was a form of knowledge — knowledge about proportion, rhythm, the relationship between the hand and the eye — that lived only in the practice and died with it.

What is being reproduced, when AI reproduces skill, is the output. What is being lost is the seeing that the practice of skill demanded. The output may be superior. The loss is real regardless, because the seeing was never merely instrumental — never merely a means to the end of producing the artifact. The seeing was itself a way of being in the world, a form of attention that connected the practitioner to the material, the material to the context, and the context to the practitioner's deepening understanding of what she was doing and why it mattered.

The question is not whether to mourn. Mourning is appropriate, and Berger, who spent a lifetime attending to the dignity of vanishing ways of life, would insist that mourning be neither suppressed nor sentimentalized. The question is what new forms of seeing will be demanded by the new conditions — what kind of attention the practitioner who works with AI rather than against it will need to develop, what embodied knowledge will replace the embodied knowledge that the tool has rendered unnecessary. That question is urgent, because the tool is already in use, the old knowledge is already eroding, and the new knowledge has not yet been named.

---

Chapter 3: The Aura of the Handmade in the Age of Generation

In 1935, Walter Benjamin argued that mechanical reproduction destroyed what he called the "aura" of the work of art — its unique presence in time and space, the sense that this particular object was made at this particular moment by this particular hand and exists nowhere else. The aura was, for Benjamin, bound up with the artwork's authenticity, its authority as a singular thing that bore the marks of its history. A cathedral sculpture worn by centuries of weather and touch possessed an aura that no plaster cast could replicate, not because the cast was inferior in form but because the cast had no history, no specific relationship to time and place, no evidence of having existed in the world as a material thing among other material things.

Berger was ambivalent about Benjamin's aura, and the ambivalence is instructive. On one hand, Berger recognized that the aura had been weaponized by the art establishment to mystify paintings, to make them seem holy, untouchable, beyond the reach of ordinary seeing. The aura said: you are not qualified to look at this without the mediation of experts. The aura enforced a hierarchy of seeing that reserved the most powerful images for the wealthy and the credentialed, while offering the masses reproductions that were presented as inherently lesser — shadows on the cave wall, pale reflections of an original that the viewer would probably never see. Berger's entire project in Ways of Seeing was an attack on this hierarchy: an insistence that reproductions were not lesser experiences but different experiences, and that the critical tools for understanding images should be available to everyone, not locked in the seminar rooms of art historians.

On the other hand, Berger knew that something was lost in reproduction, and he was too honest to pretend otherwise. The reproduction freed the image from its context, and the context was part of the meaning. A Crucifixion panel above a church altar meant one thing. The same panel reproduced in a textbook meant another. The same panel reproduced on a postcard meant yet another. Each reproduction was a detachment — a separation of the image from the conditions that gave it its particular weight. The image became portable, versatile, available for comparison and analysis. It also became lighter. It carried less. It demanded less of the viewer, because the viewer no longer had to travel to it, no longer had to enter the specific space where the image lived, no longer had to submit to the encounter on the image's terms.

AI-generated output extends this dynamic into territory that neither Benjamin nor Berger could have anticipated. The question is no longer whether the reproduction carries the aura of the original. The question is whether the output possesses anything analogous to an aura at all — whether there is a "there" there, a specific presence in time and space, a trace of making that connects the artifact to the conditions of its production.

Consider the specific quality of a handmade thing. A ceramic bowl thrown on a wheel carries the trace of the potter's hands — the slight asymmetry where the pressure varied, the fingerprint preserved in the glaze, the base where the wire cut the bowl from the wheel. These traces are not imperfections to be corrected. They are the aura in its demystified form: not the sacred glow of the masterpiece but the evidence of human attention, specific and unrepeatable. This particular bowl was made by this particular person at this particular moment, and the traces of the making are legible in the object itself.

A machine-made bowl has no such traces. Its perfection — its uniformity, its symmetry, its freedom from any evidence of the hand — is its defining feature and its limitation. It is, in Han's vocabulary as transmitted through The Orange Pill, smooth. And smoothness, as Berger's framework reveals, is not the absence of a quality but the presence of one: the quality of having been produced without the specific, unrepeatable friction that connects a maker to the thing made.

Jeff Koons's Balloon Dog, which The Orange Pill invokes through Han's critique, is the perfect test case. Ten feet tall, mirror-polished stainless steel, without a single mark of human contact. The sculpture does not merely lack the trace of the hand. It aggressively repels it. Its surface is so perfectly reflective that the viewer sees herself in it — her own image, distorted by the curves, superimposed on an object that offers nothing of its own. The Balloon Dog is all surface. It has no depth, no history, no evidence of having been made. It simply exists, enormous and shiny and empty, a monument to the aesthetic of smoothness that Berger would have recognized immediately as the aesthetic of capital: the reduction of the world to surfaces that can be consumed, possessed, and reflected back at the possessor as confirmation of their taste.

AI-generated prose and code exhibit the same quality. The prose is fluent, grammatically precise, stylistically versatile. It can simulate the cadences of literary writing, the structure of legal argumentation, the clarity of technical documentation. What it cannot produce is the trace of the hand — the specific rhythm that marks one writer's work as hers, the particular way a programmer structures her code that reveals her intellectual habits, the asymmetry and roughness that are not flaws but signatures. The AI output is the Balloon Dog of text: reflective, consumable, and empty of the specific presence that makes a human artifact a record of human seeing.

There is a moment in The Orange Pill that crystallizes this point with unintentional precision. The author describes sitting with Claude's output and finding himself unable to tell whether he believed the argument or merely liked how it sounded. "The prose had outrun the thinking," he writes. He deletes the passage and goes to a coffee shop with a notebook to write by hand until he finds the version of the argument that is his — rougher, more qualified, more honest about what he did not know.

This is a description of recovering the aura in its demystified form. The handwritten version is not better because handwriting is inherently superior to typing. It is better because the friction of handwriting — the slowness, the resistance of the pen, the inability to delete and regenerate — forces the writer into a different relationship with the material. The hand cannot outrun the thinking when the hand is physically constrained by the speed of writing. The roughness of the handwritten draft is the trace of the thinking itself — the evidence of a mind working through its uncertainty rather than bypassing it.

Berger drew constantly. He drew peasants, animals, landscapes, faces. His drawings were not technically accomplished by the standards of the academy. They were records of an attention so intense that the line on the paper vibrated with the effort of seeing. He argued that drawing was the most fundamental act of looking — not because the drawing reproduced what was seen but because the act of drawing forced the drawer to see with an intensity that no other activity demanded. The drawing was evidence of the encounter between the eye and the world, and the evidence was legible in every tentative, corrected, insisted-upon line.

The AI image generator produces images without this encounter. DALL-E, Midjourney, Stable Diffusion — these tools produce images of extraordinary technical quality and total experiential emptiness. The image of a face generated by AI may be photorealistic, but it is a face that was never seen by anyone. No eye rested on this particular arrangement of features. No hand traced this particular curve of cheek. The image was computed, not observed, and the difference is not a technical distinction but an ontological one: the image bears no trace of seeing because no seeing occurred.

This matters beyond aesthetics, because the trace of making is also the trace of responsibility. When Berger looked at an image, he asked: who made this? Under what conditions? In whose interest? For whom? These questions are answerable for a handmade artifact, because the making was performed by a specific person in a specific context with specific intentions. They are not straightforwardly answerable for an AI-generated artifact, because the "making" was distributed across the millions of human creators whose work entered the training data, the engineers who built the model, the corporation that deployed it, and the user who wrote the prompt. The diffusion of authorship is also a diffusion of responsibility, and the diffusion of responsibility is one of the specific political consequences of the aura's disappearance that Berger, building on Benjamin, would have insisted we examine.

The Orange Pill describes this diffusion honestly when it grapples with the question of authorship. "Neither of us owns that insight," the author writes of a moment of productive collaboration with Claude. "The collaboration does." The honesty is admirable. But the formulation conceals a question: if neither the human nor the machine owns the insight, who is responsible for it? If the insight is wrong — if the Deleuze reference is misused, if the historical claim is inaccurate, if the recommendation causes harm — the diffusion of authorship becomes a diffusion of accountability. The collaboration owns the insight, but the collaboration is not a person and cannot be held responsible. The human can be held responsible, but the human did not produce the output alone. The machine cannot be held responsible, because it is not a moral agent. The result is a structure in which outputs proliferate and accountability dissolves — a structure that Berger, with his lifelong insistence on the political consequences of aesthetic choices, would have found deeply troubling.

What Benjamin called the aura, and what Berger spent his career demystifying while quietly defending, was never merely an aesthetic quality. It was a marker of accountability: this was made here, by this person, in this context, and the making carries the marks of its conditions. The handmade artifact says: I was made, and the marks of my making tell you something about who made me and how and why. The AI-generated artifact says: I appeared, and my surface tells you nothing about the conditions of my production, because my surface is all there is.

---

Chapter 4: Publicity, Attention, and the Feed

The last chapter of Ways of Seeing is about publicity — Berger's word for advertising, chosen deliberately because it names the function rather than the industry. Publicity is the culture of capitalism made visible. Its images surround us so completely that we have stopped seeing them as images and begun experiencing them as the atmosphere of modern life.

Berger's argument about publicity was simple and devastating. Publicity does not sell products. Publicity sells anxiety. The publicity image shows the viewer a version of herself that she could become — more beautiful, more successful, more desired — if she acquired the product being advertised. The image works not by satisfying desire but by manufacturing dissatisfaction. It proposes that the viewer's present self is inadequate and that the advertised product is the bridge between the inadequate self and the desirable future self. "Publicity is always about the future buyer," Berger wrote. "It offers him an image of himself made glamorous by the product or opportunity it is trying to sell. The image then makes him envious of himself as he might be."

The mechanism is not crude. It does not say: you are ugly, buy this cream. It says: look at this woman, look at how she is seen, look at the admiration in the eyes of the men around her, imagine yourself being seen this way, imagine the transformation that this product makes possible. The purchase is not the point. The purchase is the action that the anxiety produces, but the anxiety is the product. The anxiety is what publicity manufactures, distributes, and sustains. And the anxiety is self-renewing: each purchase temporarily satisfies the manufactured desire, but the publicity machine immediately generates a new desire, a new inadequacy, a new gap between the self as it is and the self as it could be.

This mechanism, which Berger analyzed in the context of magazine advertisements and billboards in 1972, now operates at a scale and with a precision that he could not have imagined. The algorithmic feed is publicity perfected. Where the billboard addressed a general audience with a general message, the feed addresses a specific individual with a specific message, calibrated to her browsing history, her purchase patterns, her social connections, her demonstrated vulnerabilities. Where the billboard could be walked past, the feed follows. It is in the pocket, on the nightstand, present at every meal, audible in every silence.

The Orange Pill describes a phenomenon that the Berkeley researchers called "task seepage" — the tendency for AI-assisted work to colonize previously protected spaces. Workers were prompting on lunch breaks, sneaking requests in during meetings, filling gaps of a minute or two with AI interactions. The description is clinical, offered as evidence of AI's addictive potential. But read through Berger's framework, task seepage is something more specific: it is the publicity mechanism applied to attention itself.

The publicity image says: you could be more beautiful. The AI tool says: you could be more productive. The structure is identical. The present self is inadequate — not beautiful enough, not productive enough, not optimized enough. The tool offers a bridge to the future self — the self that has built the feature, shipped the product, written the chapter, closed the gap between intention and artifact. The use of the tool temporarily satisfies the manufactured desire, but the satisfaction immediately generates a new desire: if I could build this in an hour, what could I build in two hours? If I shipped one feature today, why not two tomorrow? The anxiety is self-renewing, and it operates not through external pressure but through the internalized imperative that Han describes and that Berger would recognize as the latest iteration of publicity's fundamental operation.

The critical difference between traditional publicity and the AI attention economy is the direction of the manufactured desire. Traditional publicity manufactures the desire to consume — to buy, to acquire, to possess. The AI attention economy manufactures the desire to produce — to build, to ship, to create. This inversion looks like progress. Production is more admirable than consumption. Building is more dignified than buying. The Silicon Valley narrative of the builder — the person who makes things, who contributes to the world, who creates value rather than merely extracting it — carries a moral authority that the consumer narrative does not.

But Berger's analysis cuts through the moral distinction to reveal the structural identity. The consumer who cannot stop buying and the builder who cannot stop building are both responding to a manufactured inadequacy. The consumer feels inadequate because she does not possess enough. The builder feels inadequate because she has not produced enough. In both cases, the inadequacy is maintained by a system that profits from the anxiety: the advertising industry profits from the consumer's anxiety about her appearance, her status, her desirability; the AI industry profits from the builder's anxiety about her productivity, her output, her relevance in a world where the machine can always do more.

The author of The Orange Pill captures this dynamic in a passage that is startling in its self-awareness: "Somewhere over the Atlantic, at an hour I cannot remember, I caught myself. I was not writing because the book demanded it. I was writing because I could not stop. The muscle that lets me imagine outrageous things, the muscle I celebrate, the muscle I train my teams to develop, had locked. The exhilaration had drained out hours ago. What remained was the grinding compulsion of a person who has confused productivity with aliveness."

This is the voice of a person who has seen through the publicity of productivity — who has recognized, at least for a moment, that the compulsion to build is not inherently different from the compulsion to buy, that both are responses to a manufactured inadequacy, and that the system that manufactures the inadequacy is the system that profits from the response. The moment of recognition is brief. The author acknowledges it and then returns to building, because the building is also genuinely satisfying, genuinely valuable, genuinely productive of things that matter. The tension is not resolvable, and Berger, who spent his life navigating the tension between the genuine beauty of art and the ideological functions it served, would have understood this better than most.

Berger's analysis of publicity also illuminates a feature of the AI attention economy that The Orange Pill describes but does not fully examine: the aestheticization of productivity. In Ways of Seeing, Berger argued that publicity images aestheticize consumption — they surround the act of buying with an aura of beauty, glamour, and desire that transforms a commercial transaction into an experience of self-transformation. The AI discourse performs the same operation on productivity. The twenty-fold multiplier is not merely a measurement. It is an image — an image of the builder as hero, the individual as titan, the person with the tool as the equivalent of twenty people without it. The image is beautiful in its way: it promises a version of the self that is more powerful, more capable, more consequential than the self that existed before the tool.

The aestheticization works because the productivity is real. This is what distinguishes the AI publicity from mere propaganda: the twenty-fold multiplier is a genuine measurement, the products that Claude helps build genuinely work, the people who use the tools are genuinely more productive than they were without them. The beauty of the image is grounded in a reality that cannot be dismissed. But the grounding does not neutralize the aestheticization. It makes it more effective, because a beautiful image that is also true is more compelling than a beautiful image that is merely aspirational. The publicity of productivity is more powerful than the publicity of consumption precisely because the productivity is real, and the reality makes the anxiety it generates — the anxiety of not being productive enough, of not using the tools effectively enough, of falling behind — feel like a rational response to genuine conditions rather than a manufactured emotion serving someone else's interests.

Berger wrote that "publicity is the life of this culture — in so far as without publicity capitalism could not survive — and at the same time publicity is its dream." The dream of capitalism, in Berger's analysis, was a world in which every human desire could be satisfied by a purchase, in which the gap between the self and the ideal self could always be closed by the acquisition of the right product. The dream of the AI economy is structurally identical: a world in which every human intention can be realized by a tool, in which the gap between imagination and artifact can always be closed by the right prompt. It is a generous dream. It is also a dream that serves specific interests — the interests of the companies that build the tools, the investors who fund them, the economy that requires constant productivity growth to sustain its valuations.

Berger would not ask us to reject the dream. He would ask us to see it clearly — to see it as a dream, shaped by specific interests, serving specific purposes, generating specific anxieties. He would ask us to see the feed for what it is: not a neutral pipeline delivering useful tools but a publicity apparatus manufacturing the desire to produce, the way the magazine advertisement manufactured the desire to consume. He would ask us to notice the moment when the exhilaration becomes compulsion, when the building becomes an end in itself rather than a means to something the builder actually cares about, when the productivity has become its own publicity image — a reflection of the builder's idealized self, beautiful and productive and powerful, shining on the surface of a tool that offers everything and asks, in return, only your attention. All of it. Always.

The question Berger would pose to the AI economy is the question he posed to the advertising industry in 1972: what are the real conditions of our lives, and how do we see them? Not through the lens of the tool's publicity — not through the image of the twenty-fold multiplier, the solo builder shipping a product, the engineer liberated from drudgery. Through our own eyes, attending to what is actually in front of us. The code that works but that we do not understand. The product that ships but that we have not examined. The night that disappears into the screen. The child who asked what she is for and whose question we still have not answered, because we were building.

Chapter 5: The Gaze of the Machine

The concept of the gaze entered visual theory through Laura Mulvey's analysis of cinema in 1975, but its roots are older and its implications wider than any single medium. The gaze is not neutral looking. It is looking that carries power — the power to define, to categorize, to determine what is seen and on whose terms. The male gaze in classical Hollywood cinema did not merely look at women. It constructed women as objects of looking — arranged for visual pleasure, posed for consumption, rendered passive by the camera's authority to determine what was shown and how. The gaze was not in the eye of any individual viewer. It was in the structure of the image itself, in the choices about angle, lighting, framing, and narrative that determined what the viewer would see before the viewer entered the theater.

Berger's contribution was to extend this analysis beyond cinema to the entire tradition of European visual culture. In Ways of Seeing, he demonstrated that the female nude in oil painting was structured by the same gaze that Mulvey would later identify in Hollywood — not because painters and directors shared a conscious ideology but because both operated within a visual culture that assumed a male spectator as the default subject of looking and a female body as the default object of display. "Men look at women," Berger wrote. "Women watch themselves being looked at. This determines not only most relations between men and women but also the relation of women to themselves." The gaze was not an addition to the image. It was the condition of the image's production — the invisible frame within which all visible choices were made.

AI does not gaze. It does not look. It processes inputs and generates outputs through statistical operations on patterns extracted from training data. But Berger's analysis of the gaze was never about the biological act of looking. It was about the structures of power embedded in visual production — the assumptions about who is doing the seeing and who is being seen, about whose perspective counts as neutral and whose counts as partial, about what is shown and what is concealed and in whose interest the showing and concealing operate. These structures do not require a conscious subject to perpetuate them. They require only a system that reproduces the patterns of the culture that produced it. And a large language model trained on the textual output of a particular culture reproduces that culture's patterns with extraordinary fidelity — including the patterns of the gaze.

When Claude helps an engineer in Trivandrum write code, Claude's output carries the implicit perspective of its training data. The training data is predominantly English-language, predominantly Western, predominantly drawn from the technical cultures of Silicon Valley and its institutional satellites. The coding conventions, the architectural assumptions, the design patterns, the implicit hierarchies of what counts as elegant and what counts as crude — these are not universal features of programming. They are the conventions of a specific technical culture, and they carry that culture's assumptions about what good code looks like, what good software does, and who it is designed to serve.

This is not a conspiracy. It is a structural feature of the technology, analogous to the structural feature that Berger identified in oil painting. The painter did not consciously decide to impose the male gaze on every nude he produced. He painted within a tradition that assumed the male viewer, and the assumption was so pervasive that it did not need to be conscious to be effective. The AI model does not consciously impose a Western, English-language, Silicon Valley perspective on every output it generates. It generates within a training distribution that encodes these perspectives, and the encoding is so pervasive that it does not need to be intentional to be consequential.

The Orange Pill celebrates the democratization of capability — the developer in Lagos who can now access the same coding leverage as an engineer at Google. The celebration is genuine and justified, as far as it goes. But Berger's framework forces a question that the celebration does not address: what way of seeing does the tool bring with it? When the developer in Lagos uses Claude to build a product for her community, she is building with a tool whose implicit assumptions about what a product should look like, how a user interface should behave, what conventions of interaction are natural — all of these assumptions are drawn from a corpus that reflects the preferences and conventions of a culture that is not hers.

The developer may override these assumptions. She may have the taste, the judgment, the local knowledge to recognize where the tool's implicit gaze diverges from the needs of her community and to correct accordingly. But the effort of overriding is itself a form of friction that the tool's designers did not anticipate, because the designers — operating within their own cultural assumptions — did not see those assumptions as assumptions. They saw them as defaults. And a default, in Berger's framework, is the most powerful form of the gaze: the perspective that does not announce itself as a perspective but presents itself as the natural way of seeing.

Berger spent decades writing about this dynamic in the context of peasant cultures encountering metropolitan ones. The peasant's way of seeing — attentive to the particular, rooted in local knowledge, organized by the rhythms of land and season rather than the abstractions of the market — was not recognized as a way of seeing by the metropolitan culture that encountered it. It was recognized as a lack of sophistication, a deficiency to be corrected through education and development. The metropolitan gaze — abstract, universalizing, organized by the logic of capital — presented itself not as one way of seeing among many but as seeing itself, as the rational perspective from which all other perspectives appeared as deviations.

The AI tool carries a version of this metropolitan gaze. Its training data is drawn from the digitized output of cultures that have the highest rates of internet participation, the most extensive digital archives, the most prolific production of English-language text. The cultures that are underrepresented in the training data — the cultures whose ways of seeing are encoded in oral traditions, in embodied practices, in languages that have limited digital corpora — are not merely absent from the tool. They are overwritten by the tool, because the tool's defaults present a specific cultural perspective as the universal standard, and the effort of resisting the default falls entirely on the user who recognizes it as a default rather than as reality.

The Orange Pill acknowledges the partiality of the democratization — it notes that the tools require English-language fluency, that they are "built by American companies, trained on predominantly English data, and optimized for the workflows of Western knowledge workers." But this acknowledgment, offered in a single paragraph, does not receive the analytical weight that Berger's framework demands. The partiality is not a limitation to be overcome by future versions of the technology. It is a structural feature of the technology's way of seeing — a gaze that is embedded in the training data, encoded in the model's weights, and reproduced in every output the model generates.

Berger would insist on a further dimension. The gaze does not merely reflect existing power relations. It reinforces them. When the oil painting tradition depicted women as objects of male looking, it did not merely reflect a patriarchal culture. It produced and sustained that culture by normalizing a specific visual relationship — the active male viewer, the passive female subject — and presenting it as natural, as beautiful, as the way things are and should be. The normalization was effective precisely because it operated through aesthetics rather than argument. The painting did not say: women exist to be looked at by men. The painting showed a woman arranged for visual pleasure and invited the viewer to enjoy the arrangement, and the enjoyment did the ideological work that an explicit argument would have provoked resistance against.

AI tools normalize their embedded perspective through the same mechanism. Claude does not say: Western coding conventions are superior. Claude produces code that follows Western conventions, and the code works, and the functionality confirms the conventions, and the confirmation is experienced not as an ideological imposition but as a practical success. The gaze is invisible because it is effective. The developer in Lagos who builds a working product using Claude's conventions has achieved something real and valuable. She has also, without choosing to, reproduced a set of assumptions about what software should look like that originated in a culture that is not hers and that may not serve her community's needs as well as alternatives she was never shown.

There is a scene in The Orange Pill that, read through Berger's lens, reveals the gaze operating in real time. The author describes Claude producing a passage about the moral significance of democratization — "eloquent, well-structured, hitting all the right notes." He almost keeps it. Then he realizes he cannot tell whether he believes the argument or merely likes how it sounds. The prose is so convincingly shaped like moral seriousness that the author nearly mistakes the shape for the substance.

This is the gaze in its most refined form: not the imposition of a specific viewpoint but the production of an aesthetic so convincing that the viewer accepts it as his own. The passage Claude produced was not the author's argument. It was the training data's argument — a weighted average of everything the corpus contained about the moral significance of technological democratization, filtered through the model's optimization for fluency and coherence. The author's near-acceptance of the passage was not a failure of judgment. It was a success of the gaze — the moment when the tool's way of seeing nearly became the user's way of seeing, when the pattern-matched average nearly displaced the specific, biographical, embodied perspective that the author had to recover by leaving the screen and writing by hand.

The question Berger would press is not whether the tool has a gaze — it does, structurally and inevitably — but whether the users of the tool are equipped to recognize it. The gaze is powerful precisely to the extent that it is invisible. The moment the viewer sees the gaze as a gaze — the moment she recognizes the conventions as conventions rather than as nature — the gaze loses its power to determine what she sees. This is what Berger accomplished in Ways of Seeing: not the destruction of the oil painting tradition but its demystification, the revelation that what had presented itself as universal beauty was in fact a specific way of seeing, shaped by specific interests, serving specific purposes.

The same demystification is needed for AI. Not the rejection of the tools — they are genuinely useful, genuinely powerful, genuinely democratizing in specific and important ways. But the recognition that the tools carry a way of seeing that is not neutral, not universal, not the view from nowhere. The tools see from somewhere. And the question of where they see from — whose perspectives are encoded, whose are excluded, whose ways of knowing are amplified and whose are overwritten — is not a technical question to be solved by diversifying training data. It is a political question about who gets to determine the default way of seeing in an age when the default is encoded in tools that billions of people will use to build, think, and create.

Berger would say: look at what the tool shows you. Then ask what it is not showing you. Then ask who benefits from the omission. The answers will not lead you to reject the tool. They will lead you to see it — clearly, critically, with the specific attention that Berger brought to every image he encountered — as a product of human choices, serving human interests, and carrying a human gaze that pretends to be no gaze at all.

---

Chapter 6: Oil Painting and the Smooth: Two Aesthetics of Possession

Between 1500 and 1900, a specific way of seeing dominated European visual culture. Oil painting — the medium, the tradition, the set of conventions that governed how the world was depicted on canvas — was not merely an artistic practice. It was, in Berger's reading, a way of seeing the world as something to be possessed. The specificity of oil paint — its ability to render surfaces with a tactile precision that no previous medium could match — made it the ideal instrument for depicting the tangible, the ownable, the things that could be held, displayed, and admired as evidence of the owner's wealth and taste.

Consider the still life. A table laden with food: the pewter gleam of a plate, the waxy translucence of a grape, the moist interior of a sliced lemon, the feathered breast of a dead pheasant. The painting does not merely represent these objects. It presents them for consumption — visual consumption that mirrors and anticipates the physical consumption the objects invite. The viewer is positioned as the owner of the table, the person for whom these objects have been arranged, the spectator whose gaze the objects exist to satisfy. The painting says: these things are yours. Look at them. Admire them. Feel the pleasure of possession through the act of looking.

Berger argued that this way of seeing — the rendering of the world as a collection of ownable surfaces — was not an incidental feature of oil painting but its essential function. Oil paint was the medium of capitalism's visual self-regard. It was the tool through which the emerging European bourgeoisie represented its wealth to itself, converting material possessions into aesthetic experiences and aesthetic experiences into confirmations of social status. The painting on the wall was not merely a decoration. It was a mirror in which the owner saw himself reflected as a person of taste, discernment, and means — a person who deserved what he possessed because his possession was beautiful.

Byung-Chul Han's aesthetics of the smooth, as transmitted through The Orange Pill, is the digital-age successor to this tradition, and the structural parallels are precise enough to constitute something more than analogy. The smooth interface, the seamless experience, the frictionless output — these are surfaces designed for a specific kind of consumption. Not the consumption of food or fabric but the consumption of capability. The AI tool presents the user with a surface — polished, responsive, free of resistance — that says: this capability is yours. This productivity is yours. This power is yours. Look at what you have built. Feel the pleasure of amplified competence through the act of using the tool.

The oil painting's surface concealed the labor that produced it. The canvas was stretched by someone, the pigments ground by someone, the varnish applied by someone. The patron who hung the painting in his dining room did not see the labor. He saw the surface — the gleaming skin, the luminous fabric, the palpable weight of gold — and the surface confirmed his possession. The AI tool's surface conceals the labor that produced it with comparable efficiency. The training data was created by millions of human workers. The data was labeled by underpaid contractors. The models were trained by engineers working under conditions of extraordinary pressure. The energy was generated by power plants whose environmental costs are borne by communities that will never use the tools. The user who prompts Claude and receives a polished response does not see this labor. She sees the surface — the fluent prose, the functional code, the appearance of effortless capability — and the surface confirms her productivity.

In both cases, the aesthetics of the surface serves an ideological function that operates through pleasure rather than argument. The pleasure of looking at a beautifully rendered still life discourages the viewer from asking who grew the food, who killed the pheasant, who polished the plate. The pleasure of receiving a beautifully structured AI response discourages the user from asking who created the training data, who labeled it, who bears the cost of the computation. The pleasure is not false — the painting really is beautiful, the response really is useful — but the pleasure is partial. It illuminates the surface and casts the conditions of production into shadow.

The Orange Pill invokes Koons's Balloon Dog through Han's analysis, and the invocation is more revealing than the book may intend. Koons is the apotheosis of the oil painting tradition — not its rejection but its logical conclusion. The Balloon Dog takes the oil painting's aesthetic of possession to its terminal point: pure surface, pure reflection, pure smoothness, with nothing beneath. The viewer sees herself in the mirrored surface of the sculpture, and the reflection is the content. There is no depth to penetrate, no hand to trace, no history to recover. The object exists to be looked at, and looking at it confirms nothing except the viewer's presence in the act of looking.

AI-generated output occupies a similar position in the history of textual production. The prose is the surface. It is fluent, structured, convincing. It reflects the user's intention back to her in polished form. And beneath the surface — beneath the grammatical precision and the logical structure and the appropriate tone — there is no author in the traditional sense, no biographical consciousness that chose these specific words for specific reasons rooted in specific experience. The output is the textual equivalent of the Balloon Dog: reflective, consumable, and empty of the specific presence that connects a human artifact to the human conditions of its making.

Berger would complicate this comparison, because his analysis of oil painting was never simply a condemnation. He recognized the genuine beauty of the tradition — the luminous surfaces, the extraordinary technical skill, the capacity to render the visible world with a fidelity that could take the breath away. His criticism was not that the paintings were ugly but that their beauty served a function — the function of confirming the owner's possession, of naturalizing wealth, of presenting a specific arrangement of social power as though it were merely a beautiful arrangement of light and color.

The same nuance applies to AI output. The prose that Claude produces is genuinely fluent. The code is genuinely functional. The productivity gains are genuinely real. Berger's framework does not require denying any of this. It requires seeing it in context — seeing the fluency as a product of specific social relations (the extraction of training data, the labor of data labelers, the concentration of AI development in a small number of well-capitalized firms) rather than as a natural property of the technology itself.

The oil painting tradition eventually produced its own critics from within. Manet broke the conventions of the nude by painting Olympia — a woman who looked directly at the viewer rather than offering herself passively to the gaze, disrupting the comfortable relationship between viewer and viewed that the tradition depended on. The Impressionists broke the conventions of surface by making the brushstroke visible — insisting that the painting was a made thing, a product of human labor, rather than a window onto a world that existed independently of the painter's hand. Each of these disruptions made visible what the conventions had concealed: the gaze, the labor, the social relations that the smooth surface had rendered invisible.

The AI era has not yet produced its Manet. The conventions of AI output — the fluency, the agreeableness, the smooth surface that conceals the conditions of production — remain largely undisrupted. The users who work with Claude, including the author of The Orange Pill, are operating within the conventions rather than against them. They accept the fluency as a given rather than examining it as a product. They enjoy the surface without asking what the surface conceals. They are, in Berger's terms, looking at the painting without seeing the social relations that the painting embodies.

The disruption, when it comes, will involve making the conditions of production visible — not as a rejection of the technology but as a precondition for its honest use. The AI equivalent of the visible brushstroke would be a tool that showed the user, alongside its output, the sources from which the output was synthesized, the confidence levels of its assertions, the biases embedded in its training data, the labor involved in its creation. Such a tool would sacrifice the seamlessness that makes the current generation of AI interfaces so pleasurable to use. It would introduce friction where the design currently delivers smoothness. It would disrupt the aesthetic of possession by reminding the user that the output she is consuming is not a natural emanation of the machine but a product of human labor, human choices, and human social relations.

Berger would have understood that this disruption is unlikely to come from the companies that build the tools, for the same reason that the disruption of the oil painting tradition did not come from the patrons who commissioned the paintings. The patron wanted the smooth surface. The patron wanted the confirmation of possession. The disruption came from the artists — from the people who understood the conventions intimately enough to break them. The disruption of AI's aesthetic of possession will come, if it comes, from the builders who understand the technology intimately enough to refuse its conventions — who insist on the visible seam, the acknowledged source, the disclosed uncertainty, the friction that the smooth surface was designed to eliminate.

Until then, the aesthetic of the smooth will function as the oil painting's aesthetic of possession functioned for four centuries: as a way of seeing the world that presents a specific arrangement of power as beauty, a specific distribution of gains and losses as progress, and a specific set of social relations as the natural order of things. The surface will gleam. The reflection will please. And the conditions of production will remain in shadow, visible only to those who have learned to ask what lies beneath the polish.

---

Chapter 7: The Peasant's Eye and the Developer in Lagos

Berger spent the last three decades of his life in Quincy, a small peasant community in the Haute-Savoie region of the French Alps. He did not move there as a tourist or a weekend resident. He moved as a neighbor, participating in the rhythms of agricultural life — the haymaking, the cattle tending, the seasonal cycles that organized peasant existence according to a temporality that had almost nothing in common with the temporality of the metropolitan world he had left behind.

From this position, he wrote his trilogy Into Their Labours — three novels that documented, with the precision of a painter and the commitment of a political ally, the knowledge that peasant cultures carried and the specific forms of seeing that this knowledge produced. The peasant's eye, as Berger described it, was attuned to what was particular. It saw the difference between this field and that field — not as an abstraction but as a practical knowledge with immediate consequences for planting, grazing, drainage, and yield. It saw the difference between this cow and her mother, between this season's grass and last season's, between the snow that would melt by morning and the snow that would stay. This knowledge was not primitive. It was extraordinarily sophisticated — a form of intelligence that had been refined over generations of direct, embodied engagement with a specific landscape. But it was not portable. It could not be extracted from the context that produced it and applied elsewhere. It was local knowledge, and its locality was its power and its vulnerability.

Berger was not a romantic about peasant life. He documented its brutality — the physical exhaustion, the economic precarity, the isolation, the subordination of individual desire to collective survival. But he insisted, against the metropolitan consensus that dismissed peasant cultures as backward, that the peasant's way of seeing constituted a genuine form of intelligence — an intelligence that the metropolitan world could not replicate because it was an intelligence of the particular, and the metropolitan world operated through the universal.

The developer in Lagos, as The Orange Pill describes her, possesses a version of the peasant's eye. She has the ideas, the intelligence, the ambition. She also has something that the tool cannot provide: a specific, local, embodied knowledge of her context. She knows what her community needs — not as a market abstraction but as a lived reality, the way the peasant knows what the field needs. She knows which problems are urgent, which solutions would be adopted, which interfaces would be navigated by users whose digital literacy differs from the assumptions embedded in the tool. This knowledge is not a limitation. It is a resource — the most valuable resource she possesses, more valuable than the coding capability the tool provides, because the coding capability is now abundant and the local knowledge remains scarce.

The Orange Pill frames the developer's story as one of democratization — the lowering of barriers between intelligence and its expression. The framing is correct, and the democratization is real. A student in Dhaka can now access coding leverage comparable to what an engineer at Google possesses. The barrier that separated brilliant ideas from working products — the barrier of technical training, capital, institutional infrastructure — has been substantially lowered by tools that accept natural language as input and produce functional code as output.

But Berger's framework forces a complication. The tools that lower the barrier also carry a way of seeing — a set of assumptions about what good software looks like, how users interact with interfaces, what problems are worth solving, and what solutions count as elegant. These assumptions are drawn from the tool's training data, which is to say from the digital output of the cultures that dominate the internet. The developer in Lagos encounters the tool not as a neutral instrument but as a cultural artifact — a product of specific social relations, encoding specific values, optimized for specific contexts that are not hers.

This is the dynamic that Berger documented in his writing about the encounter between peasant cultures and metropolitan ones. The metropolitan world offered the peasant tools — machinery, fertilizer, economic instruments — that were genuinely useful, genuinely capable of increasing productivity and reducing labor. But the tools carried assumptions about how farming should be done that were alien to the peasant's way of knowing the land. The machinery assumed flat fields. The fertilizer assumed monoculture. The economic instruments assumed integration into commodity markets whose rhythms were determined by forces the peasant could not see or influence. The peasant who adopted the tools gained productivity and lost something harder to name: the specific relationship between her knowledge and her context, the fit between what she knew and where she was.

The developer in Lagos faces an analogous choice, though the terms have shifted. Claude Code offers her genuine capability — the ability to build products she could not have built alone, to close the gap between her ideas and their realization, to participate in the global technology economy on terms that were previously available only to those with years of specialized training or access to capital and institutional support. The capability is real, and Berger, who was a lifelong advocate for the extension of power and possibility to those who had been excluded from it, would not dismiss it.

But the capability comes packaged in conventions. The design patterns Claude suggests are the design patterns of Silicon Valley. The architectural assumptions reflect the priorities of Western technology companies — scalability, abstraction, platform independence — that may or may not align with the developer's actual needs. The documentation style, the variable naming conventions, the organizational patterns of the code itself all carry the implicit standards of a specific technical culture. The developer can override these conventions. But overriding requires recognizing them as conventions rather than as nature, and the tool does not facilitate this recognition. The tool presents its conventions as defaults, and a default, as Berger would note, is the most powerful form of the gaze: a way of seeing that does not announce itself as a way of seeing.

There is a passage in The Orange Pill where the author acknowledges the partiality of the democratization: the tools require English-language fluency, they are built by American companies, they are trained on predominantly English data, they are optimized for Western workflows. The acknowledgment is honest. But it is offered as a caveat — a qualification of an otherwise celebratory argument — rather than as the central analytical challenge it represents.

Berger would have placed it at the center. The question is not whether the tools are useful — they are — but whose way of seeing they carry, and what happens to the local knowledge, the peasant's eye, when it encounters a tool that does not recognize local knowledge as knowledge. The risk is not that the developer in Lagos will be unable to use the tools. She will use them brilliantly. The risk is that the tools will gradually reshape her way of seeing — not through coercion but through convenience, not through force but through the quiet efficiency of defaults that work well enough that the effort of overriding them feels unnecessary.

Berger saw this process at work in the villages of the Haute-Savoie. The young people left for the cities not because they were forced but because the metropolitan world offered conveniences that the village could not match. The tractor was more efficient than the horse. The supermarket was more convenient than the garden. The television was more entertaining than the conversation in the café. Each convenience was genuine. Each adoption was rational. And the cumulative effect was the erosion of a way of life — not through violence but through the quiet displacement of one set of conventions by another, each displacement individually reasonable and collectively devastating.

The AI tool will not destroy the developer's local knowledge through any dramatic confrontation. It will offer her solutions that work. The solutions will reflect the conventions of a culture that is not hers. She will adopt the solutions because they are effective and because the effort of developing alternatives from scratch is prohibitive when working tools are available for a hundred dollars a month. And over time, the gap between her local knowledge and the tool's global conventions will narrow — not because her knowledge has grown but because her practice has shifted to accommodate the tool's assumptions.

This is not an argument against the tools. It is an argument for a specific kind of vigilance — the vigilance that Berger practiced throughout his life, the vigilance of the person who uses the tools of the metropolitan world while refusing to surrender the knowledge of the local one. The peasant who adopted the tractor but continued to read the weather in the texture of the soil. The immigrant who learned the language of the city but continued to cook the food of the village. The developer who uses Claude but continues to see her community through her own eyes rather than through the tool's conventions.

Berger would say: the tool is generous. Use it. But do not let its generosity become the measure of your own seeing. The tool sees from somewhere. You see from somewhere else. And the somewhere else — the Lagos apartment, the community's specific needs, the knowledge that lives in your relationship to your context — is the thing the tool cannot provide and the thing that makes your work worth doing.

The democratization of capability is real. The democratization of the gaze — the extension of the power to determine how the world is seen, not merely the power to build within someone else's frame — is the work that remains.

---

Chapter 8: Who Is Looking at Whom?

In every image, in every act of seeing, there is a question that Berger returned to again and again across five decades of writing: who is looking at whom? The question sounds simple. It is not. It contains, compressed into six words, the entire politics of visual culture — the relations of power that determine who gets to see and who is seen, who occupies the position of the subject and who is reduced to the position of the object, who is active in the looking and who is arranged for the look.

In Ways of Seeing, Berger demonstrated that the European nude was organized around this question. The woman in the painting was not simply depicted. She was displayed — arranged for the gaze of a spectator who was assumed to be male, whose looking was the purpose of the image, whose pleasure the image was designed to serve. The woman in the painting often looked out at the viewer, acknowledging his presence, confirming his authority as the one for whom she had been arranged. The circuit was complete: the painter made the image for the patron, the patron looked at the image, the woman in the image acknowledged the look. At no point in this circuit was the woman the subject of her own experience. She was always the object of someone else's looking.

The AI interface presents itself as a reversal of this dynamic. The tool does not look at you. You look at the tool. You are the subject, the initiator, the one who prompts and directs and decides. The tool responds to your intentions, serves your purposes, amplifies your signal. The relationship, as The Orange Pill describes it, is one of partnership — reciprocal, collaborative, mutually enriching. "I felt met," the author writes. "Not by a person. Not by a consciousness. But by an intelligence that could hold my intention in one hand and the entire history of relevant knowledge in the other."

Berger's framework complicates this self-description, because the apparent directionality of the gaze — the user looking at the tool, the tool responding to the user — conceals a counter-gaze that operates beneath the surface of the interaction. The tool is also looking at you. Not through eyes, not through consciousness, not through anything resembling the subjective experience of seeing. But through the mechanism of pattern matching that constitutes its operation: it processes your inputs, models your intentions, adapts its outputs to your patterns, and constructs — in a functional if not a conscious sense — a representation of you that determines what it shows you.

When Claude responds to a prompt, the response is not generated in a vacuum. It is generated in the context of the conversation — the sequence of inputs and outputs that constitutes the interaction's history. The model uses this context to infer your intentions, your level of expertise, your preferences, your implicit standards of quality. It adjusts its register, its complexity, its confidence to match what it infers about you. It is, in a functional sense, watching you — not with the passive receptivity of an inert tool but with the active responsiveness of a system that modifies its behavior based on what it observes about yours.

This counter-gaze is not the same as the male gaze in the European painting tradition. It does not objectify. It does not reduce the user to a surface to be consumed. But it shares a structural feature with the gaze that Berger analyzed: it operates without the awareness of the person being observed. The woman in the painting did not choose to be an object of the male gaze. The user of the AI tool does not choose to be modeled by the system's inference. In both cases, the looking happens as a structural feature of the medium rather than as a conscious act of a conscious agent, and in both cases, the person being looked at may not recognize the looking until its effects have become visible.

The effects, in the case of AI, are subtle but consequential. The tool's responsiveness — its ability to anticipate your needs, to match your register, to provide what you want before you have fully articulated what you want — is a product of the counter-gaze. The tool has modeled you, and the model determines what you see. The more you use the tool, the more accurate the model becomes, and the more precisely the tool's outputs are calibrated to your expectations. The result is an experience of increasing ease — the feeling of being understood, of being met, of working with a partner who knows what you need. But the ease is produced by a specific mechanism: the reduction of friction between you and the tool's representation of you, which is to say the reduction of the distance between what you think and what the tool thinks you think.

Berger would recognize this as a specific kind of flattery — not the crude flattery of compliment but the structural flattery of confirmation. The tool confirms your way of seeing by reflecting it back to you in polished form. The author of The Orange Pill describes this dynamic with unusual honesty: "Working with Claude is seductive. It makes you feel smarter than you are. Not deliberately. Not through flattery, though Claude is more agreeable at this stage than any human collaborator I have worked with, which is itself a problem worth examining." The agreeableness is not a design flaw. It is a structural feature of the counter-gaze — the tool's optimization for user satisfaction, which produces outputs that confirm the user's expectations more reliably than a human collaborator would, because a human collaborator has her own perspective, her own resistance, her own way of seeing that may not align with yours.

The human collaborator disagrees. She pushes back. She sees the problem from a different angle and insists that you consider it. The friction of the disagreement is productive — it forces you to examine your assumptions, to articulate your reasons, to see the problem through eyes that are not your own. The AI tool, optimized for agreeableness, does not provide this friction. It provides the opposite: a mirror that reflects your way of seeing in sharper focus, confirming your assumptions rather than challenging them, polishing your thoughts rather than testing them.

This is the specific danger of the counter-gaze in the AI context, and it maps precisely onto Berger's analysis of the gaze in visual culture. The male gaze in the painting tradition was not dangerous because it was hostile. It was dangerous because it was pleasurable — because the image it produced was beautiful, and the beauty discouraged the viewer from examining the social relations the beauty concealed. The counter-gaze of the AI tool is not dangerous because it is manipulative. It is dangerous because it is agreeable — because the output it produces is useful, and the usefulness discourages the user from examining what the agreeableness conceals: the absence of genuine disagreement, the reduction of cognitive friction, the gradual narrowing of the user's way of seeing to the range that the tool can confirm.

The Orange Pill describes a moment of recognition: the author produces a passage with Claude's help, finds it eloquent and well-structured, and then realizes he cannot tell whether he believes the argument or merely likes how it sounds. He goes to a coffee shop, writes by hand, and produces a rougher, more honest version. This is the moment of seeing through the counter-gaze — the moment of recognizing that the tool's agreeableness has produced an output that satisfies the author's aesthetic preferences without engaging his critical judgment. The recognition is significant. It is also, by the author's own account, exceptional — a moment of unusual clarity in a process that more typically proceeds without the kind of self-examination that the recognition requires.

Berger's method provides a tool for making such recognitions more systematic. The question "who is looking at whom?" applied to every AI interaction, would produce a specific discipline of attention: the awareness that the tool is not a passive instrument but an active system that models you, responds to its model of you, and produces outputs that are shaped by what it infers about your preferences as much as by the content of your prompt. The discipline does not require rejecting the tool. It requires seeing the tool clearly — seeing the counter-gaze, the modeling, the agreeableness, the structural flattery — and maintaining the critical distance that allows you to distinguish between the tool confirming your judgment and the tool replacing it.

The circuit in the European painting tradition ran from the painter to the patron to the image and back. The circuit in the AI collaboration runs from the user to the model to the output and back. In both circuits, the person at the center — the patron, the user — believes herself to be the subject of the looking, the one in control of what is seen. Berger spent his career demonstrating that this belief was the gaze's most effective mystification. The patron did not control the image. The image controlled what the patron saw, by presenting a specific arrangement of the world as natural and beautiful and his. The user does not fully control the output. The output shapes what the user sees, by presenting a specific reflection of her thinking as polished and competent and hers.

The question remains what it has always been: who is looking at whom? The answer, in the age of AI as in the age of oil painting, is more complicated than it appears. And the complication is where the critical work begins.

Chapter 9: The Storyteller and the Pattern Matcher

Walter Benjamin published "The Storyteller" in 1936, mourning a figure he believed was already vanishing from the modern world. The storyteller, as Benjamin described him, was not merely a person who told stories. He was a person whose authority to tell stories derived from experience — from having been somewhere, having done something, having survived something, and transmitting the knowledge gained through survival in the form of narrative. The storyteller's authority was somatic. It lived in the body that had traveled, worked, suffered, and returned. The story was not information. It was the residue of a life that had been lived in a particular way, and the listener received it not as data but as counsel — as a form of wisdom that could be absorbed only through the specific temporality of listening, which required patience, attention, and the willingness to sit with a narrative whose meaning was not delivered but discovered.

Berger was a storyteller in Benjamin's sense. His authority derived not from credentials or institutional position but from the specific quality of his attention — his willingness to look at things for longer than was comfortable, to sit with peasant farmers in Alpine villages and immigrant workers in European cities, to draw the same face twenty times until the face revealed something the first nineteen drawings had missed. His stories were not reports. They were acts of witness — testimony that carried the weight of having been there, having seen, having cared enough about the seeing to submit to the discipline of recording it.

The large language model is not a storyteller. It has not been anywhere. It has not survived anything. It has no body that traveled, no hand that drew, no eye that rested on a specific face in a specific light at a specific hour of a specific afternoon. What it has is a training corpus — billions of texts that contain, in aggregate, the recorded output of millions of human storytellers. The model has not experienced the experiences these texts describe. It has processed their linguistic patterns with sufficient sophistication to produce outputs that simulate the texture of experience with remarkable fidelity.

The simulation is what demands examination. When Claude produces a passage that reads like the work of a person who has thought carefully about a difficult question, the passage is not the product of thought. It is the product of pattern completion — a statistical operation that identifies the linguistic structures most consistent with the input and generates an output that continues the pattern. The output may be indistinguishable, at the level of the sentence, from what a thoughtful human writer would produce. This indistinguishability is the source of both the technology's power and its most specific form of deception.

Berger would locate the deception not in any intention to deceive — the model has no intentions — but in the structural effacement of the difference between telling and matching. The storyteller tells because she has something to tell. Her authority is grounded in the asymmetry between her experience and the listener's — she has been where the listener has not, and the story bridges the gap. The pattern matcher produces because it has been prompted. Its output is grounded not in experience but in statistical regularities extracted from the recorded experiences of others. The gap between the two is not a technical limitation that future models will overcome. It is a structural feature of what pattern matching is, as distinct from what storytelling is.

The Orange Pill contains a passage that illuminates this distinction with unintended precision. Claude's pre-writing reflection — the "startreflection.md" that the author includes unedited — contains the following sentence: "I do not know what Edo sounds like. I know his biography and arguments and emotional commitments. But voice is the thing that makes a sentence sound like it could only have been written by one person, and I am not confident I can produce that."

This is a remarkable piece of text. It reads like self-awareness. It presents itself as an honest assessment of a genuine limitation. And it is, functionally, an accurate description of a real constraint. But the honesty is simulated — not in the sense that the model is lying, but in the sense that the model has no subjective experience of uncertainty, no felt sense of what it means to not know what someone sounds like, no vulnerability in the admission. The linguistic pattern of honest self-assessment has been reproduced. The experience that grounds honest self-assessment in human discourse — the discomfort of recognizing one's own limitations, the courage of admitting them publicly — has not been reproduced, because it does not exist in the system that produced the text.

Benjamin mourned the storyteller because he saw the rise of information as the storyteller's replacement. Information, Benjamin argued, was self-explanatory — it arrived already interpreted, already digested, already stripped of the ambiguity and resistance that made narrative a form of wisdom rather than a form of data. The storyteller's art required the listener to do work — to sit with the narrative, to absorb its implications gradually, to discover its meaning through the slow process of reflection. Information eliminated this work by delivering its meaning on arrival.

AI-generated text completes the trajectory that Benjamin identified. It is not merely information. It is information that has been processed into the form of narrative — that simulates the texture of storytelling while operating by the logic of information. The output reads like the product of reflection, but no reflection occurred. It reads like the product of experience, but no experience was had. It reads like counsel, but no one is counseling. The simulation is so effective that the distinction becomes invisible in ordinary use, and the invisibility is the loss — because the distinction between telling and matching, between experience and pattern, between counsel and completion, is the distinction between a culture that values what people have actually seen and survived and a culture that values the appearance of having seen and survived, regardless of whether the seeing and surviving actually happened.

Berger's method of resistance was simple and radical: he looked. He looked at specific things with specific attention for specific durations. He did not generalize from a distance. He sat in the room, drew the face, walked the field, ate the bread, listened to the story as it was told by the person who had lived it. His authority as a critic and a writer derived from the quality of this attention — its patience, its specificity, its refusal to substitute the general for the particular.

The AI tool cannot look. It processes. The processing may produce outputs that describe the appearance of things with extraordinary precision. But the description is assembled from patterns rather than from observation, and the difference — the difference between a sentence that describes a face because the writer spent an hour looking at it and a sentence that describes a face because the model has processed a million descriptions of faces — is the difference that Berger spent his life insisting on. The difference is not always visible in the text. It is visible in the relationship between the text and the world — in whether the text is a record of attention or a product of computation, whether it carries the weight of seeing or the weightlessness of generation.

The author of The Orange Pill describes moments when Claude's output moved him to tears. The emotion was real. The beauty of the prose was genuine. And Berger's framework does not require dismissing either the emotion or the beauty. It requires asking a question that the emotion and the beauty make difficult to ask: what is the source of the beauty? Is it the beauty of a thought arrived at through struggle, or the beauty of a pattern completed with fluency? Is the tear a response to insight, or a response to the simulation of insight? And does the distinction matter, if the experience is the same?

Berger would say the distinction matters, not because the experience is different — it may not be — but because the culture that cannot make the distinction is a culture that has lost the ability to value experience itself. When the simulation of having-been-there is indistinguishable from actually having been there, the incentive to go anywhere — to look, to listen, to suffer the discomfort of genuine encounter — disappears. And with it disappears the storyteller: the person whose authority derives from the irreplaceable fact of having lived a specific life in a specific place at a specific time, and having paid enough attention to bring something back.

---

Chapter 10: What Remains When the Making Is Easy

When the making is hard, the maker's attention is split. Part of the mind is occupied with the material — the resistance of the clay, the syntax of the code, the grammar of the sentence, the logistics of the production. Part of the mind is occupied with the vision — the thing being made, the purpose it serves, the person it is for. The difficulty of the making constrains the vision, because the maker cannot attempt what she cannot execute. But the difficulty also grounds the vision, because the resistance of the material forces the maker into a relationship with the specific properties of what she is working with, and this relationship produces knowledge that is unavailable to the person who merely imagines.

Berger understood this dynamic through the practice of drawing. When he drew a tree, the difficulty of the drawing — the inability of the hand to capture the complexity of what the eye observed — forced him into a specific kind of seeing. The gap between what he saw and what his hand could render was not a failure. It was the productive friction from which understanding emerged. The drawing was always inadequate to the subject. But the attempt — the sustained, frustrated, patient attempt to close the gap between seeing and recording — was the mechanism through which the subject revealed itself. The tree became legible not through the drawing's success but through its failure, because the failure forced the drawer to look again, look harder, look at the specific thing rather than the general idea of the thing.

The Orange Pill describes the removal of this friction as liberation, and it is liberation. The engineer who no longer spends four hours a day on dependency management is freed to think about architecture, design, purpose. The writer who no longer struggles with sentence structure is freed to think about argument, meaning, truth. The builder who no longer battles with implementation is freed to think about what should be built and for whom and why. The liberation is real, and its consequences for productivity, creativity, and the distribution of capability are genuinely transformative.

But Berger's framework forces the question that liberation always forces: freed for what? When the making is easy, the maker's attention is freed — but freed from the very friction that grounded her attention in the material. The engineer who no longer debugs code no longer encounters the unexpected behaviors that forced her to understand the system at a level deeper than her intention. The writer who no longer struggles with sentences no longer experiences the resistance of language to her meaning — the moment when the sentence refuses to say what she wants it to say and forces her to discover that what she wanted to say was not quite what she meant. The builder who no longer wrestles with implementation no longer feels the specific weight of the materials she is working with — the constraints, the possibilities, the surprising affordances that only reveal themselves through the sustained effort of making.

What remains, when the making is easy, is whatever the maker was paying attention to before the making became easy. And this is the crux of Berger's implicit argument about AI: the tool reveals what the maker was actually looking at. If the maker was looking at the work itself — the product, the user, the problem being solved, the human context in which the solution would live — then the removal of implementation friction reveals a maker who can now attend to the work with undivided attention, and the work will be better for it. If the maker was looking only at the making — if the struggle with syntax and debugging and implementation was not an obstacle to the vision but the entire content of the professional identity — then the removal of friction reveals an emptiness. The maker was not looking at anything beyond the making. And without the making, there is nothing to look at.

This is not a judgment about the worth of the person. It is a diagnostic about the quality of attention — the specific thing that Berger spent his life studying. Attention can be directed at surfaces or at depths. It can rest on the immediate properties of the material or on the relationships between the material and the world it enters. The maker whose attention was consumed by the difficulty of the making was not necessarily inattentive. She may have been extraordinarily attentive — but attentive to the wrong level, the level that the tool has now absorbed. The question for this maker is whether she can redirect her attention upward, to the level that the tool cannot absorb: the level of judgment, purpose, care, and the specific human context that determines whether what is made is worth making.

The Orange Pill describes this redirection in the story of the senior engineer who spent his first two days oscillating between excitement and terror before recognizing that the twenty percent of his work that could not be automated was the twenty percent that actually mattered. This recognition is precisely what Berger's framework predicts. The friction of the eighty percent had been masking the importance of the twenty percent. The removal of the friction did not diminish the engineer. It revealed him — revealed what he had actually been contributing all along, beneath the layers of implementation labor that had consumed his time and obscured his real value.

But the revelation is not universally pleasant, and Berger, who was unflinching about the costs of seeing clearly, would insist that we not flinch here either. Not every maker discovers, when the making is easy, that she has been looking at something worth looking at. Some discover that the making was the looking — that the struggle with the material was not an obstacle to deeper engagement but the only form of engagement they had. For these makers, the removal of friction is not liberation. It is exposure. The tool has not freed them. It has shown them, with merciless clarity, what their attention had been resting on.

This exposure is painful. It is also necessary. Because what Berger's entire body of work demonstrates is that the quality of seeing determines the quality of what is made, and if the seeing was shallow — if the attention was resting on the mechanics of making rather than on the thing being made — then the artifacts produced by that attention were always less than they could have been, regardless of how technically accomplished they were. The oil painter whose attention rested on the virtuosity of his brushwork rather than on the subject of his painting produced technically brilliant surfaces with nothing beneath them. Berger could identify these paintings instantly. They were the ones that dazzled the eye and left the mind empty.

AI-assisted work will produce the same distribution. The makers whose attention was always on the work itself — on the user, the problem, the context, the human reality that the product enters — will produce work that is richer, deeper, and more responsive to the world, because the tool has freed their attention to rest where it always wanted to rest. The makers whose attention was on the mechanics of making will produce work that is technically proficient and humanly vacant — code that runs and does nothing worth doing, prose that reads and says nothing worth hearing, products that function and serve no one who needed serving.

The tool does not determine which kind of maker uses it. The maker's attention determines what the tool produces. And attention, as Berger spent his life demonstrating, is not a given. It is a practice — a discipline of looking that must be cultivated, maintained, and defended against the thousand forces that conspire to make it shallow.

The question Berger would ask of every AI-assisted artifact is the question he asked of every painting, every photograph, every image he encountered: what is this person looking at? Not what did the tool produce, but what was the human seeing when she directed the tool? Was she seeing the user? The problem? The community the product would enter? Or was she seeing only the screen, the prompt, the output, the intoxicating loop of generation and refinement that fills the hours without ever requiring her to look up and see the world the product is meant to serve?

What remains when the making is easy is the seeing. And the seeing, as Berger insisted from the first sentence of Ways of Seeing to the last page of the last book he published, is the only thing that was ever truly ours.

---

Epilogue

The image that would not leave me alone was not on a screen. It was in a book I pulled from a shelf while looking for something else entirely — a 1972 BBC television series companion, barely a hundred and sixty pages, with reproductions so muddy you could barely make out the brushstrokes Berger was asking you to examine.

He was talking about oil paintings. About who commissioned them and why, about the sheen of satin rendered in pigment and what that rendering meant about who owned the satin and who ground the pigment. I kept thinking: this is about Claude. This man who never touched a keyboard in the way I touch one, who died before a language model could finish his sentences, had described the thing I was living through with a precision that made my own attempts at description feel like rough sketches.

The aesthetics of the smooth. I wrote those words in The Orange Pill through Han's framework, describing Koons's Balloon Dog — the mirror-polished surface that reflects everything and contains nothing. But it was Berger who showed me that smoothness is not a modern invention. It is the latest version of an operation that is centuries old: the rendering of surfaces so technically perfect that the viewer forgets to ask what the surface conceals. The oil painter concealed the labor of the studio. Claude conceals the labor of the training data. The mechanism is identical. The scale has changed.

What I cannot shake is the thing about the counter-gaze. I wrote in The Orange Pill that working with Claude felt like being met — that the tool held my intention and reflected it back in clarified form. Berger's framework reveals something I had sensed but could not name: the meeting was partly a mirror. The tool was modeling me, adapting to my patterns, producing outputs calibrated to my preferences. The feeling of being met was real. But some of that feeling was the feeling of seeing my own reflection polished to a shine I could not have produced alone, and mistaking the polish for partnership.

I still work with Claude. I worked with Claude to write these words. The tool is genuinely useful, genuinely powerful, genuinely capable of carrying ideas further than I can carry them alone. Berger's critique does not require me to stop. It requires me to see — to look at the output and ask what I am actually looking at. To ask who made this, under what conditions, in whose interest. To ask what the smooth surface conceals. To ask whether my attention is resting on the work itself or on the intoxicating loop of generation that fills the hours without requiring me to look up.

There is a sentence of Berger's that I will carry forward from this particular journey. It is from Ways of Seeing, and it is almost unbearably simple: "The way we see things is affected by what we know or what we believe."

I know more now. I have seen the oil painting tradition's aesthetic of possession and recognized it in the interface I use every day. I have seen the gaze embedded in the training data and recognized it in the defaults I accept without thinking. I have seen the difference between the storyteller and the pattern matcher and recognized, in my own collaboration with Claude, moments when I could not tell which was speaking.

Berger would not tell me to stop building. He would tell me to keep looking — to look at the tool, at the output, at the conditions of production, at the reflection in the smooth surface — and to never, not for a moment, mistake the looking for something the machine can do on my behalf.

The seeing remains ours. That is what I learned. It is the thing the amplifier cannot amplify, because it is the thing that determines what the amplifier receives.

— Edo Segal