There is a particular silence that settles over a gallery opening when someone, usually by accident, reads the wall text aloud. It is not reverent. It is the silence of twenty people realizing, at roughly the same moment, that the paragraph they just skimmed—the one promising the work “interrogates liminal spaces of post-digital embodiment”—could have been written by a moderately depressed intern with a thesaurus and a looming deadline. Nobody laughs. Laughing would mean admitting the whole room has been running on polite fiction. But the thought hangs there, above the canapés: this sentence was not written by a human being in any sense that matters. It was assembled. It was output.
I have been thinking about that silence lately, because the art world has found a new tool that makes the old ones look almost charming. AI script generators—the kind you stumble across when you search for a script writing app that promises to turn a premise into a formatted screenplay—are the latest entry in a long, distinguished parade of devices that promise to swap labor for language. The pitch is seductive: type a logline, pick a genre, and watch the machine spit out scene headings, character names, dialogue, even that sacred Courier 12-point formatting that signals you are a professional. It is, in its way, a miracle. But the miracle is not that a computer can write a screenplay. The miracle is that we are surprised.
Because if you have ever written an artist statement, a grant proposal, or a gallery press release, you already know the feeling of feeding a set of inputs into a mental template and watching the sentences emerge, almost unbidden, like water from a spigot. You know the rhythm: “My practice investigates…” / “Through a research-based approach…” / “The work gestures toward…” / “In conversation with…” The words are not chosen; they are selected from a finite menu of acceptable phrases, arranged according to conventions so stable they might as well be code. The AI script generator is not doing something new. It is doing something old, faster, and without the pretense that a soul was involved.
This is the real anxiety, the one nobody wants to name at panel discussions. It is not that AI will replace writers. It is that AI reveals how much of what we call “writing” was already algorithmic. The press release, the grant narrative, the curatorial statement, the screenplay—these are genres with grammars. And grammars can be learned by anything with enough training data and a loss function.
The Template Before the Machine
Consider the screenplay. Before AI could generate one, screenwriting was already a discipline of rules. The font must be Courier 12-point. The left margin must be 1.5 inches. Dialogue blocks must be centered under character names in all caps. Scene headings must declare INT. or EXT. followed by location and time of day. As StudioBinder’s guide to screenplay format explains with the serene authority of a manual, “one page of script format equals roughly one minute of screen time.” This is not a suggestion. It is a law, and laws are algorithms. The screenwriter who internalizes these rules is already operating a kind of biological script generator, converting story ideas into formatted pages through a repeatable procedure. The AI just removes the middleman.
The art world has its own formatting laws, though they are enforced by social pressure rather than production schedules. An artist statement must be written in the first person but sound like it was translated from a language that does not exist. A grant proposal must demonstrate “community engagement” by describing a workshop where participants will be photographed looking thoughtful. A press release must include the phrase “is pleased to announce” and a quote from the artist that the artist did not say. These are not creative choices. They are compliance. And compliance is what machines do best.
I once received a press release for a group show titled “Dialogic Topographies of the Ephemeral.” The release explained that the exhibition “creates a platform for multivalent voices to negotiate the contested terrain of memory and materiality.” I read it three times. I still do not know what it meant. But I know exactly how it was made: someone opened a document, typed a list of approved nouns (terrain, memory, materiality, platform, voice), added a few verbs (negotiate, contest, create, interrogate), and stirred. The result was not writing. It was generation, in the sense we now use for machines. The only difference is that the human generator took longer and probably charged more.
The Grant Proposal as Procedural Generation
If you want to see the algorithmic soul of art writing in its purest form, read a stack of grant proposals. Not the successful ones—those are just the same algorithm with better luck. Read the ones that failed. They are identical in structure, vocabulary, and emotional temperature to the ones that succeeded. They all describe a “research-based practice” that “engages with urgent questions.” They all promise “outputs” that include a publication, a public program, and something described as “a new body of work.” They all include a budget line for “artist fee” that is exactly the amount the funder expects to see, calculated backward from the total grant cap like a Sudoku puzzle.
This is not writing. This is filling out a form in complete sentences. The form has fields—Project Description, Artistic Merit, Community Impact, Budget Narrative—and each field expects a specific kind of content, delivered in a specific tone, with specific keywords. An AI trained on a corpus of funded proposals could produce a competitive application in seconds. The only thing stopping it is that funders still pretend to read them. But do they? Or do they scan for the keywords, check the budget math, and move on? The algorithm may already be on both sides of the transaction.
The Authors Guild, in its AI Best Practices for Authors, warns that “AI outputs, by contrast, are generic mashups of pre-existing works ingested during training” and insists that “it is your original voice, thinking, and creativity that make you the writer that you are.” This is a noble defense of human authorship. But it also inadvertently describes the problem: if AI outputs are generic mashups, what do we call a press release that is a generic mashup of every press release ever written? What do we call an artist statement that is a generic mashup of every artist statement the writer has ever read? The Authors Guild is worried about machines imitating humans. I am worried about humans who have been imitating machines for so long they forgot the difference.
The Screenplay as Structural Confession
The screenplay is the most honest genre in this conversation because it never pretended to be anything other than a template. Screenwriting manuals do not talk about “finding your voice.” They talk about page counts, plot points, and the exact moment the inciting incident should occur (page 10, if you are following the gospel of Syd Field). The three-act structure is an algorithm. The hero’s journey is an algorithm. Save the Cat is an algorithm with a catchy name. The entire discipline is built on the premise that stories can be reverse-engineered into component parts and reassembled according to proven patterns. An AI script generator is not violating the sanctity of screenwriting. It is fulfilling screenwriting’s deepest assumptions about itself.
This is why the panic about AI in creative fields is so selectively applied. Nobody panics when a spreadsheet automates accounting. Nobody panics when a template automates invoice generation. But when a machine automates the production of a document that looks creative—a screenplay, a poem, a press release—we suddenly discover our attachment to the idea that a human soul was present in the making. The attachment is touching. It is also, in many cases, a fantasy. The soul was not present in the press release. The soul was not present in the grant proposal. The soul was barely present in the artist statement, which was written at 2 a.m. the night before the application deadline, with the artist’s real thoughts carefully removed to make room for the word “dialogic.”
What the Machine Exposes
The AI script generator is a diagnostic tool. It reveals, by succeeding, which writing tasks were never creative to begin with. If a machine can write a competent screenplay from a prompt, the screenplay was always a kind of engineering problem—a problem of structure, pacing, and format compliance, with creativity confined to the premise and the occasional surprising line of dialogue. If a machine can write a competent grant proposal, the grant proposal was always a kind of bureaucratic performance, with creativity confined to the budget justification. If a machine can write a competent press release, the press release was always a kind of advertising copy, with creativity confined to the choice of which synonym for “exciting” to use.
This is not a condemnation of these genres. It is a clarification. The problem is not that machines can do them. The problem is that we have been calling them “writing” and paying humans to perform them as if they required a self. The humans, in turn, have learned to perform the selflessness required, suppressing their actual observations, their actual doubts, their actual sense that the work they are describing is, in some cases, not very good. The result is a literature of compliance: documents that say exactly what they are supposed to say, in exactly the way they are supposed to say it, with exactly the expected emotional temperature (earnest, slightly urgent, never angry). This is the literature AI was born to write.
I am not arguing that all art writing is algorithmic. I am arguing that the algorithmic parts have expanded to fill the available space, crowding out the parts that might have been alive. The grant proposal that actually describes a project in plain language, with specific details and a genuine argument for why it matters, is so rare it feels like a literary event. The press release that admits a show is “uneven but interesting” would be a scandal. The artist statement that says “I make these because I like making them, and I hope you like looking at them” would be treated as a joke, even though it is more honest than ninety percent of the statements currently on gallery walls.
The Polite Fiction and Its Discontents
The art world maintains its algorithmic writing because the alternative is terrifying. The alternative is admitting that we do not always know what we mean. The alternative is admitting that some work is bad, some ideas are half-formed, and some exhibitions exist because a dealer needed to fill a slot in the calendar. The algorithmic language—the “interrogates,” the “liminal,” the “gestures toward”—is a protective layer, a way of saying something while saying nothing, a way of filling the silence that might otherwise be occupied by an honest question. “What do you actually think?” is the question the press release is designed to prevent.
AI script generators do not have this problem. They do not have opinions. They do not have doubts. They have training data and probability distributions. They produce the most probable next word, given the prompt and the corpus. This is, in a technical sense, exactly what the grant writer does when they reach for “community-engaged practice” instead of describing the actual community they actually engaged. The difference is that the grant writer could have done otherwise. The grant writer could have written a sentence that sounded like a person. The machine cannot. The machine can only produce the most probable output. And the most probable output, in art writing, is the one that sounds like every other art writing ever produced.
This is the real indictment. Not that AI can imitate us. That we have made ourselves so easy to imitate.
The Counter-Tradition
There is, of course, another kind of art writing. It exists in the margins: in the auction catalog footnote that admits a work’s provenance is “uncertain,” in the restoration report that describes exactly what was damaged and how, in the art handler’s muttered assessment of a sculpture’s structural integrity, in the critic’s review that begins with a physical description of the gallery’s lighting before it says anything about the art. This writing is specific. It is grounded in observation. It uses words that refer to things in the world, not just to other words in the genre. It is, in short, the kind of writing a machine cannot do—not because machines are bad at language, but because this kind of writing requires having been in a specific room, looking at a specific object, with a specific body, on a specific day. The machine has no body. The machine has never been in a room.
The art handler who tells you, while uncrating a painting, that “the stretcher bars are pine, which means it was made in Europe, probably Germany, probably before 1990” is doing something no AI can do. They are applying embodied knowledge—knowledge acquired through hands and eyes and years of lifting things—to a particular object in a particular moment. The critic who notices that the wall text font size at a major museum correlates with the artist’s market is doing something no AI can do. They are making a connection that requires pattern recognition across domains (typography, economics, institutional behavior) and a willingness to state the uncomfortable obvious. The artist who writes a statement that says “I started this because I was angry about something I saw on the train, and I kept going because the anger turned into a shape I liked” is doing something no AI can do. They are telling the truth about a process that actually happened.
This is the counter-tradition, and it is the only tradition worth defending. It is not threatened by AI. It is threatened by the same forces that have always threatened it: the pressure to conform, the fear of sounding unprofessional, the institutional demand for language that can be quoted in a fundraising brochure without alarming the trustees. AI is just the latest excuse to avoid doing the harder thing.
The Prompt as Mirror
So what does it mean when “generating a screenplay” becomes structurally identical to writing an artist statement? It means we should stop being surprised. It means we should look at the prompt—the little box where you type your logline, your genre, your desired tone—and recognize it for what it is: a mirror. The prompt is the grant application form. The prompt is the gallery’s submission guidelines. The prompt is the curator’s email asking for “a short text on your practice, 300 words max, by Friday.” The prompt is everywhere, and we have been filling it out for years, producing the expected output, collecting the expected response, moving on to the next prompt.
The AI script generator did not create this condition. It inherited it. The condition was created by an art world that decided, somewhere in the late twentieth century, that professionalization required a professional language—a language that could be taught in MFA programs, assessed by grant panels, and reproduced by graduates with minimal variation. The language succeeded. It is now so stable that a machine can learn it in an afternoon. The machine is not the villain of this story. The machine is just the character who arrives in the final act and says aloud what everyone has been thinking.
The question is not whether AI will replace art writers. The question is whether art writers will continue to do work that is replaceable, or whether they will reclaim the territory that machines cannot enter: the specific, the embodied, the honest, the funny, the wrong-in-an-interesting-way. The territory where a sentence sounds like it was written by a person who has actually looked at the thing they are describing, and who has something to say about it that is not already contained in the prompt.
That territory is small. It is shrinking. But it is still there, in the art handler’s comment, in the restoration report, in the critic’s aside, in the artist’s late-night notebook entry that will never become a statement because it is too true. The machines cannot follow us there. The question is whether we are willing to go.