AI, Authorship, and Art

The reactive pattern is familiar. The outcome usually looks different in hindsight.

In 1476, Parisian scribes destroyed a printing press. Not metaphorically — they physically dismantled it. The monk Filippo de Strata had already written to the Doge of Venice warning that the press would corrupt scholarship, flood the world with error, and eliminate the sacred labor of copying. The abbot Johannes Trithemius published a treatise arguing that scribes, by working slowly and deliberately, produced something spiritually superior to the machine-pressed page. He had it printed on a press.

We tell this story now with a kind of fond condescension. Of course the scribes were wrong. Of course the press didn’t kill writing — it created it, in the modern sense. We know how it ends.

We are not as good at recognizing when we are inside the same story.

The Pattern

Here is what keeps happening, across centuries of new tools:

1839. Louis Daguerre presents the daguerreotype in Paris. The French painter Paul Delaroche reportedly says, “From today, painting is dead.” Within a generation, painting had produced Impressionism, Post-Impressionism, and Cubism — movements that only became possible because photography had released painting from the obligation to document reality accurately. Photography didn’t kill painting. It clarified what painting was actually for.

1964. Robert Moog demonstrates his synthesizer at the Audio Engineering Society convention. The American Federation of Musicians responds by banning the Moog from commercial recording sessions. Session musicians feared replacement; the AFM imagined the device could simply replicate any instrument on demand. Four years later, Wendy Carlos released Switched-On Bach and demonstrated that playing a synthesizer required genuine musicianship — different skills, not lesser ones. Today the synthesizer is as uncontroversially an instrument as the piano.

The 1980s. Writers discover word processors. Gore Vidal warns that the word processor is “erasing literature.” Critics argue that the ease of revision creates a false illusion of polish, that real writing requires the physical friction of the typewriter or the pen. The novelist Stanley Elkin, who had multiple sclerosis, called his Lexitron word processor “the most important day of my literary life.” Anne Rice said it changed what she was capable of doing as a writer. The reactive moment faded; the word processor became invisible, which is what good tools eventually do.

The 1990s. Photoshop arrives and traditional painters dismiss digital art as work “for hacks who can’t draw.” Today, digital painting and concept art are established professional disciplines, and no serious person contests their legitimacy.

The arc is always the same: a new tool appears, the existing practitioners of a craft declare it incompatible with real art, the tool gets absorbed, and the definition of craft expands to include the new skills the tool requires. The art doesn’t change. The instruments for making it do.

What the Thoughtful Voices Actually Say

The AI-and-art discourse gets flattened into two camps: enthusiastic technologists who say AI will democratize creativity, and alarmed artists who say it will destroy it. The most interesting voices refuse both positions.

Stephen Marche, the novelist who wrote Death of an Author using AI as a core collaborator, describes the work as “curatorial.” He compares it directly to Photoshop and CGI — tools that required new skills to use well, that changed what was technically possible, but that didn’t change what filmmaking or design fundamentally is. On the question of replacement: “Any studio executive who thinks they’re gonna hire AI rather than a writer to create a script is out of his mind.” The work, he argues, still requires a person whose sensibility shapes every decision.

Robin Sloan, the novelist behind Mr. Penumbra’s 24-Hour Bookstore, has been writing with custom-trained neural networks since 2016. He describes the practice as “snapping together conveniently-packaged blocks of human intellect and effort” — a description that sounds less like creative abdication and more like every writer who has ever reached for a thesaurus, a rhyming dictionary, or a well-worn copy of Strunk and White. His view on AI replacing novelists: the question is “basically nonsensical.” People read novels for the perspective of a person, with the inner life of a person, about existence as a person.

Holly Herndon, the musician and AI researcher who has built some of the most sophisticated AI-assisted work in contemporary music, pushes back on the idea that AI tools are passive or automatic. “Every project that I’ve done with AI has been extremely manual. It’s not just some automated process where it’s type in a few words and art is done. It’s usually very laborious, many decisions made.” She and her collaborator Mat Dryhurst describe AI as a “coordination technology” — closer to choral singing, with all the human arrangement that requires, than to a vending machine for music.

Margaret Atwood, at 84 and still writing prolifically, calls AI “a crap poet — really bad.” She is neither in denial about AI’s capabilities nor catastrophizing about them. Her posture is the right one: clear-eyed about the tool’s current limits, undisturbed about its implications for real writing.

Kazuo Ishiguro finds AI “alarming and exciting at the same time.” He thinks it may produce “a new kind of literature, like the way modernism transformed the novel” — a framing that echoes the photography-and-painting story exactly. New tools don’t kill existing forms; they create pressure that generates new ones.

Brian Eno has been working with generative systems since the 1990s, coining the term “generative music” and arguing for AI-assisted composition as a legitimate creative practice decades before ChatGPT made it a culture war. His work is a reminder that the argument isn’t new — only the scale and accessibility of the tools.

The Case Against, Taken Seriously

The strongest skeptical voice belongs to Ted Chiang, the science fiction writer, whose August 2024 New Yorker essay “Why A.I. Isn’t Going to Make Art” is the most careful version of the dissenting argument.

Chiang’s claim is precise: art results from making many choices, and text prompts don’t give you enough choices to constitute authorship. “An artist needs to have control of every aspect of a painting. A writer needs to have control over every sentence in a novel.” His concern isn’t tool-use in general — it’s specifically the fantasy of delegating the choice-making to the AI and calling the output yours.

This is a serious argument. It’s also, looked at carefully, a strong endorsement of the “workbench, not generator” philosophy. The writer who uses AI to brainstorm rhymes, surface synonyms, test a line from a different angle, or identify a structural problem — and who then makes thousands of choices about what to keep, what to cut, and what to rewrite entirely — is doing exactly what Chiang says real authorship requires. Chiang’s critique is of the fantasy of creative abdication. It doesn’t touch the artist who is genuinely in the driver’s seat.

Nick Cave goes further and deeper. In his Red Hand Files response to a fan who had used ChatGPT to write lyrics in his style, Cave wrote: “Songs arise out of suffering… algorithms don’t feel. Data doesn’t suffer.” His objection isn’t to AI tools; it’s to the idea that the output of a system with no inner life could constitute art in the full sense. This is a philosophical claim about what art is for, not a claim about what tools artists may legitimately use.

Both arguments, taken seriously, point in the same direction: the presence of the human — with their choices, their experience, their suffering, their taste — is what makes the art. The tool is only as meaningful as the person holding it.

The Real Concerns

Three legitimate concerns deserve honest acknowledgment, not dismissal.

Training data and consent. The lawsuit landscape — Authors Guild v. OpenAI, the Anthropic settlement — reflects a real ethical question about whether AI companies should be able to use artists’ work to train systems without consent or compensation. This is a question about how tools are built, not whether they are legitimate to use. The same distinction applies to many technologies: the ethics of manufacturing a thing and the ethics of using it are separable. But artists raising this concern are not being irrational, and the concern doesn’t simply evaporate when you find the tool useful.

Homogenization. There is genuine empirical evidence that AI-assisted creative work, in aggregate, tends to converge on the popular and the familiar. This is not a reason to avoid AI tools — it is a reason to use them with strong aesthetic intent. The antidote to homogenization isn’t abstention; it’s taste.

Economic displacement. The WGA strike, the Spotify platform’s “perfect fit content” program replacing human artists with cheap AI-generated tracks, the displacement of working illustrators — these are real. Every previous tool transition involved displacement alongside new opportunity. The honest position is to acknowledge both, not to flatten either.

A Workbench, Not a Generator

Here is the principle that resolves the question, for me:

Artists use tools. They always have. The troubadour used a system of modes and chord progressions that constrained and enabled their composition. The poet used received forms — the sonnet, the villanelle, the ballad — as scaffolding. The novelist used genre conventions as a kind of grammar. The songwriter used rhyme schemes and prosody systems that nobody invented fresh for their particular song. None of this made the art less theirs. It made it possible.

A thesaurus doesn’t write your sentences. A rhyming dictionary doesn’t find your metaphors. A chord chart doesn’t make your decisions. These are tools that expand the space of what you can see and consider, while the choices remain entirely yours. Used well, AI can occupy the same category — and do something even a thesaurus can’t. A thesaurus shows you words you already know exist. A good tool takes you somewhere you didn’t know to look. Economist Thomas Schelling put it plainly: “One thing a person cannot do, no matter how rigorous his analysis or heroic his imagination, is to draw up a list of things that would never occur to him.” That’s a key problem which contributes to writer’s block — not always a failure of craft, but a failure of discovery. The next line isn’t hiding behind a better word choice. It’s on the other side of an unexplored rock. Tools, at their best, help you turn rocks over.

This is the part of the AI conversation that gets least attention. We debate whether AI replaces artists. We argue about whether prompting counts as authorship. But the more interesting question — the one working artists are actually asking — is whether AI can help them find something they didn’t already have. Not choose between options, but discover a possibility that wasn’t yet visible. Brian Eno spent decades on exactly this problem, building his Oblique Strategies card deck for one purpose: to break the grip of studio pressure and reveal the tangential move the conscious mind couldn’t see. Igor Stravinsky said constraints don’t limit the artist — they free them, by narrowing the field until the unexpected becomes inevitable. The Surrealists called it automatism. The Oulipo called it productive constraint. The name changes; the mechanism doesn’t. A tool that widens your adjacent possible — that shows you the terrain just past the edge of your current map — is doing something closer to inspiration than to autocomplete. That’s a longer conversation, and one worth having on its own: [AI as Discovery Engine: What Harrington Is Actually For →]

The critical word is used. The tool doesn’t determine the outcome. The person holding the tool does.

This is the philosophy behind Harrington. Not “AI will write your songs,” but “AI will help you write better ones.” Not answers, but Harrington’s tools at the right scale — Song Coach and Theme Check for the whole song, Structural Guidance and Chord Progression for sections, Spark, Suggest, Refine, and Rhyme for lines. Not a generator, but a workbench: structured space to work, Harrington’s tools to analyze what you’ve made, assistance when you’re stuck, and then the thing gets out of the way.

The authorship isn’t shared. The capability is extended.

Harrington does not use your songs to train models. Your songs are yours — you can export them and delete your data. The tool exists to serve the songwriter — not the other way around.

Authorship is personal. The license is for one person. The song is yours.

The scribes were wrong about the printing press. The painters were wrong about photography. The union was wrong about the Moog. They were wrong not because the tools posed no risks — some of those risks materialized — but because they located the threat in the wrong place. The threat was never the tool. It was always the question of who was in control of it, and whether the human at the center of the work remained genuinely present.

If you’re present — if your choices, your voice, your experience, your suffering, your taste are in every decision — the tool is just a tool.

And you’re the author.