As generative AI’s impact on the creative industries grows, I’ve been thinking about what produces art, and why graphics produced by diffusion models and text produced by LLMs are (by and large) not. And I think that, ultimately, art is the result of creative decisions, whereas generative AI removes decision-making. **Prompting is ideating; it is not deciding. **Craft — the execution of decisions — is what turns ideas into art.
Art emerges through the labor of articulation
Decisi…
As generative AI’s impact on the creative industries grows, I’ve been thinking about what produces art, and why graphics produced by diffusion models and text produced by LLMs are (by and large) not. And I think that, ultimately, art is the result of creative decisions, whereas generative AI removes decision-making. **Prompting is ideating; it is not deciding. **Craft — the execution of decisions — is what turns ideas into art.
Art emerges through the labor of articulation
Decisions are rooted in taste
“Art only happens where there is room for options, where things can be fundamentally otherwise,” writes Brian Eno. That is, where there is artistic choice. Aesthetics are subjective; there are no “correct” artistic decisions, simply individual choices guided by taste, intention, genre, medium, and ability.
Creative decisions rely on the artist’s taste. Taste is not quantifiable — it wouldn’t make sense to try to rank artworks in order of their tastefulness — nor is tastefulness as simple as a binary yes/no. We develop our unique taste through discernment, making aesthetic judgments — decisions, conscious or not — about what we believe is important in art. Artists build our craft by making decisions and evaluating them. Through practicing our craft, developing taste, and learning our genre’s conventions, each artist develops our own artistic “patterns of speech”: a suite of phrasing and approach to developing a storyline or melody. Although genres and mediums share conventions, these are constantly evolving, allowing artists to exercise individuality even within established spheres.
The decisive moment is a theory of photography from Henri Cartier-Bresson. Framing a photo is meaning-making by decision, which literally allows viewers to see “through the artist’s eyes.” The photographer must anticipate, frame, choose, commit: once the photographer depresses the shutter of a single-lens reflux camera, the device is physically occupied in exposing the film. The photographer chooses which moment will make the image they want to create; each artist’s decisive moment is distinct to their sensibilities.
Our artistic vision is bound to our creative labor
Taste is embedded in craft: the artist exercises aesthetic judgment continuously as they work. Creative decisions arise from a meeting of taste and skill — what can we physically execute that most closely matches what we want to make? How much are we willing to stretch ourselves, if what we wish to make exceeds our current capabilities?
Art is shaped by the constraints and affordances of its medium*. A medium’s limitations and potentialities directly inform our artistic labor. Artists choose a medium to work in both based on our desired outcome and how we want to spend our time.
Before the pandemic, I wrote an outline for a graphic novel retelling of The Little Mermaid. Although I’m a prose writer, I chose the comic format for the project because I wanted to contrast two illustration styles as a storytelling feature (a la Pleasantville‘s use of color)… but a mute main character meant I’d need to figure out how to indicate signed speech. In the end, I decided I didn’t want to spend the next decade learning how to draw a graphic novel — the challenges a comic treatment would need aren’t where I want to spend my limited creative energy.
Robin Sloan observes that for many artists and projects, the initial creative spark is less about the content, and more about the shape of the art:
“[E]xperienced creative people do not begin projects with emotive ideas or “stories” (🙄) untethered to a specific medium. Rather, they begin with a sense of what they want to do in a medium they love. […] Creative work begins with an impulse for WORK, the kind of thing you want to spend your time doing, to which narrative and emotional material is quickly added.”
Which constraints sound exciting rather than limiting? As a writer, which affordances and tropes do we want to play with, and what story will suit those? What tools do we like to use? (Personally, I’m itching to write something super voice-y, which is not my usual style.)
Ideas are scaffolds for artistic decisions
Most creative concepts are less complete than they feel in the mind; articulation forces us to confront an idea’s hollowness. It quickly becomes apparent how hazy one’s concept is when forced to spell it out for someone else’s understanding. “You had to be there” doesn’t fly for a novel. The writer must decide whether a character goes home to “a sagging Craftsman, moss growing heavy on the cedar shingles” or “a modernist monster, all angles and glass.” (In writing we call this White Room Syndrome, where conversations seemingly happen in a void.)
**Attention — in the form of time working — helps the writer realize they have not said what they thought, or that their thought is muddy and needs more thinking to become coherent. **“There are innumerable ways to write badly,” notes Verlyn Klinkenborg in Several Short Sentences About Writing. “The usual way is making sentences that don’t say what you think they do.” Articulation does not come cleanly or smoothly, most of the time. It *is *labor. But it is necessary labor, if we want to write clearly and meaningfully. Reality has a surprising amount of detail; so does art.
Novelist Cory Doctorow writes that “art is an act of communication,” which means that we must know what we are trying to communicate:
As a writer, when I write a novel, I make tens – if not hundreds – of thousands of tiny decisions that are in service to this business of causing my big, irreducible, numinous feeling to materialize in your mind. Most of those decisions aren’t even conscious, but they are definitely decisions, and I don’t make them solely on the basis of probabilistic autocomplete. One of my novels may be good and it may be bad, but one thing it definitely is is rich in communicative intent. Every one of those microdecisions is an expression of artistic intent.
Readers and viewers experience a completed piece of work as basically inevitable, as if it could never have been anything else… but the author thought about every word on the page, the painter thought about every brush stroke, the videographer thought about every cut. **While a piece of art is being created, it exists as a multiverse of possible final works that it could turn into, and every decision discards one potentiality after the next, until only the final work remains. **Creative decisions both expand nothing into something, and distill “everything” into one specific thing.
Articulation is what makes the art — not the concept.
Working on producing our artistic vision refines the concept. The artwork solidifies with our labor, through our labor. No one can conceptualize big works like a novel as a whole upfront — the book always evolves as it is translated from idea to writing. Word by word, page by page, a fog we didn’t notice at first lifts from our artistic vision. Characters are merged, subplots are removed or added. Writers talk about “butt in chair” as a shorthand for putting in the time; no matter how much you daydream about your idea, if you never try putting the vision into language, it will never become a book. **Ideas are easy; execution is hard. **
Details drive meaning
The details *are* the artwork
Sci-fi writer Samuel R. Delany shares an anecdote about a printing error in one of his books: in one version, the last line — a single word, the answer to a question — was omitted. People who read the version without the last line *hated *the book. Delany’s point is that every word of a novel matters, because the reader’s experience is shaped and reshaped continually throughout the reading, every word modifying their overall experience.
The construction of a sentence may not seem of much importance, so long as it does the job of conveying information… but facts and actions are the least of what a sentence tells a reader. Words convey more than their direct meaning; they are shaded in tone and color and class*.
Readers co-create the story with the author as we read, drawing upon our own experiences to interpret the story world as constructed by the author’s words. Book cover designer Peter Mendelsund writes:
“Words are effective, not because of what they carry in them, but for their latent potential to unlock the accumulated experience of the reader. Words “contain” meanings, but, more important, words potentiate meaning… This is a word’s dormant power, brimming with pertinence.”
Translator Jim Rion meditates on what “accuracy” in translation means when the author’s original word carries cultural resonance in its native tongue whose emotional tenor would be lost on a reader from another culture. In a novel, using a word that clashes can ruin a scene’s mood or undermine characterization; words with conflicting undertones can confuse the reader. Earlier this year, I read a novel with a “surprise” villain plotline revealed at the climax, and it worked in large part because the author had employed gothic descriptions and a haunted tone throughout the book, even though it was a Romance.
In fantasy and sci-fi, metaphors carry the danger of drawing too directly upon current cultural references and modern concepts. Would a sci-fi character who’s never set foot on a planet compare anything to a forest? Would a high fantasy elf think about the speed of light?
It’s easy to lean on tired language, especially in first drafts where the writer is just trying to get basic action on the page, focused on plotting rather than phrasing. “A cliché isn’t just a familiar, overused saying,” Klinkenborg writes. “It’s the debris of someone else’s thinking.”** Stereotypes and cliches are comfortable placeholders, essentially invisible, often used when the artist has not thought through precisely what they mean.** Generated text is rife with stock phrases and biases because generalizations litter the online communication that comprised a significant portion of AI model training data. Artist Hito Steyerl puts it: “They converge around the average, the median; hallucinated mediocrity. They represent the norm by signaling the mean.”
Articulation underpins meaning
Writer Baldur Bjarnason observes:
“The side that creates sets in place structures, pacing, rhythms, symbols, and signals that play with conventions and precepts to convey a message to the other side. The “reader” decodes, interprets, deconstructs, and reassembles a meaning that’s built out of their experiences and expectations using their skills at understanding media. […] This is why generated media is often so vacuous, there is less intent behind it – mostly references and formulaic adherence to accepted conventions – leaving the burden of constructing meaning entirely on the consumer, and there’s a hard limit to how much meaning can be made out of what’s effectively an empty balloon: all surface filled with air.”
Humans are excellent at pattern recognition, so we may discern meaning in generated text or graphics the same way pareidolia makes us see Jesus in toast. Echoes of humanity can break through in statistical convergence or data gaps, but these haunts just make generated material more uncanny. When a book inspires strong feelings in the reader, that is the result of the artist grappling with how to evoke a reaction in the reader using mere words.
Meaningful art comes from the artist reckoning with emotions. The decisions needed to create an artwork require emotional labor — the writer must decide not only what emotion to convey, but how to translate embodied sensations. Our feelings — both readers’ and writers’ — derive from experiences, and the meaning they hold for us. **Writers adapt remembered embodied truths to the situations of the story, supplying meaning to the imaginary by providing a real emotional core. **Ursula K. Le Guin puts it:
“Experience is where the ideas come from. But a story isn’t a mirror of what happened. Fiction is experience translated by, transformed by, transfigured by the imagination. Truth includes but is not coextensive with fact. Truth in art is not imitation, but reincarnation.”
Generative AI removes articulation from production; generating a text no longer requires the “writer” to draw upon their own emotional experiences. Ideas can now be input to the machine to produce writing-like text and art-like graphics. Without doing the emotional work of articulating why a story element is meaningful to the protagonist and the reader, authors abandon the reader to the banalities of statistical likelihood. To not choose the words of a story is to surrender crafting the reader experience and relinquish valuable tools for developing character, mood, and theme. Skip this labor, and the end result may be muddled, facile, or underdeveloped. “If you are not thinking about the things you produce,” writes historian Dr. Eleanor Janega, “then they inherently have no meaning.” Sharing our feelings, even with their origins obscured, requires the artist to be vulnerable, to expose something real of ourselves — but it is this sense of connection to another human that readers and viewers crave from a work of art.
“The labor of articulation binds us to our experience and in relationship with others.” — L.M. Sacasas
“The emotional wallop of a story is created by its totality,” posits author Donald Maass. “Readers experience that wallop when they must not just form an opinion about a story, but when they must study, question, and form an opinion about themselves.” The writing process asks that of writers, too. Articulation is transformative; creative decisions make us as much as they make the art. A work of art is the artefact of the artist engaging with whatever questions the artwork poses, being changed in the process.
Writer Mandy Brown articulates the conundrum of genAI’s supposed efficiency (emphasis mine):
This is one of the many reasons why I find the current conversation about so-called generative AI so immensely frustrating: there’s all this hype about making everything easier and faster, about how we can eliminate all the work involved in the making of words and images. But no one arguing for this seems to have asked what’s left when the work is gone. What is the experience of asking for something to appear and then instantly receiving it? What changes between the thought and the manifestation? I fear that nothing changes, that nothing is changed in such a making, least of all ourselves.
Further reading:
Why Write? Even if no one will read it and no one is sure what’s AI and what’s human by Patrick Grafton-Cardwell
Duck duck duck dichotomy by Annie Mueller
Midst Press — see a timelapse of poems being composed
See also: