Category: Business of Writing

I happen to know a painter. Imagine for a moment, that I just moved into a new house, and I decide that I want new art to decorate the place. I tell my artist friend that I’ll pay her to create twenty paintings to my specifications.

Now I’m edgy, and I’ve already painted the walls black, so I decide what I really want is a bunch of paintings that look like H. R. Geiger’s work, specifically the movie Alien. I tell my artist friend that I want a shadowy techno-biological spaceship bridge for the entryway, and a carapaced, eyeless monstrosity for the living room, and for the bedroom, three damp corridors that look like the inside of an esophagus.

She rolls her eyes, but agrees to the terms. She paints the paintings and presents them to me. I like them, mostly, but I want a few changes. She makes a few modifications, I pay her the agreed-upon sum, and the deal is complete.

Here’s the question: am I an artist? I clearly have an artistic vision for my house—a vision so singular and intense that nobody seems to want to come over anymore. But should I be credited in the act of creation?

The Inversion

We already have a term for arrangements like this. It’s been around for quite some time. This kind of arrangement is artistic patronage. I pay you, and you create art to my specifications.

The general concept has evolved, from nobility and the wealthy supporting individual artists, to platforms like Kickstarter and Patreon allowing fans to support their favorite artists, to the many and varied forms of modern contract-worker abuse, where artists are treated more and more like machines that turn money into art, and preferably at the highest possible rate of exchange.

The twist in our current moment is that AI has provided a new type of patronage—cheaper and faster and even more heavily mediated by Silicon Valley technology. Instead of the wealthy paying human artists, the AI companies are attempting to build an ecosystem where mostly less wealthy patrons pay LLMs, and indirectly, their wealthy investors.

Since the process of training LLMs divorces the training inputs (the work of human artists) from the outputs, it has so far been treated (from a legal perspective) as transformative of the original work, and therefore fair use. And since machines cannot be copyright holders yet, nobody owns the rights to the resulting output.

Expression and Skill

Proponents of LLM-generated art argue that gen AI is a tool like Photoshop, and prompting is simply a new artistic skill. They argue it can be equally expressive.

As an extreme comparison, let’s look at scanning or photocopying. Photocopying generates an artistic output, a nearly exact copy of an artwork, but has little expression or skill. The only skill is identifying art that you want to copy.

AI images are more than this, certainly, but how much more? They aren’t direct copying, but the popularity of prompts “in the style of X” show that this remains a strong element. Especially since most prompters do not bother to analyze the techniques that contribute to the desired style. Most users cannot explain why the Simpsons looks like the Simpsons, or Ghibli looks like Ghibli, they just know they want a profile pic that looks “like that.”

AI art does takes creative input from the user: a description of something imagined. In this, it is similar to traditional art. While some artists, especially abstract artists, may start with a technique or a set of colors or a mood, most will start with some idea of concrete subject.

Many non-artists assume the process of creating art is all about getting that imagined thing from brain to paper with as little deviation as possible. This, I think, is simply not true. The process of generating a piece of art can be broken down into hundreds or thousands of tiny micro-expressions.

Each line and brushstroke in a painting is an atomic particle of artistic expression. Each word in a novel is a choice the artist had to make. The canvas is ten thousand interconnected empty spaces, and the artist chooses how to fill each one, accounting for the spaces around it that have already been filled.

AI proponents will say an AI prompt is constructed of many tokens, and LLM users will iterate repeatedly, tweaking their prompt. But how do brushstrokes in a painting compare to tweaking words in a prompt?

Determinism and Intentionality

LLMs are non-deterministic. With the exception of some very tightly constrained outputs, the exact same prompt to an AI will generate different outputs on each attempt. The user’s input is only partially responsible for the output. The rest is dictated by the ineffable statistical noise of AI inference. Some elements of the resulting work are inherently external to the user’s prompt.

We could argue about whether the same brushstrokes by the same artist over multiple paintings can result in identical paintings, but if they do not, then it’s due to small differences in environment, in tools, or inconsistency of skill. The artist, in the process of making ten thousand micro-expressions of their art, responds to these tiny, incremental outputs and adjusts the rest of the piece.

Differences between outputs are an expression of the artist, an interplay between the intentional and the accidental. Differences between nondeterministic AI outputs are not an expression of the user-patron. They are a mindless side-effect of AI generation. They are noise, not signal.

Art is Process, Not Product

By being directly involved in every micro-expressive decision and adapting and adjusting to the results with every output, the process of “traditional art” is a conversation between the artist and the art in progress. Even if two processes result in nearly identical pieces, the traditional process provides many more opportunities for making decisions, and adapting to the work as it’s created. These are direct expressions of skill and artistic intent.

Even if we treat generative AI as an artist’s tool, it is a blunt and wildly inaccurate instrument. It allows for relatively little input, and that input affects the output in highly variable and nondeterministic ways. Despite allowing for much faster iteration on “fully completed” pieces, there is scant mechanism for interplay between artist and art within the process of creation. It remains a black box.

In short, you don’t create a piece of art in collaboration with gen AI. You ask as best you can, then evaluate a finished output and decide if you need to ask again, trying to come up with the magical incantation that provides something approximating the desired result.

Vibe Check

Many of the most vocal advocates of prompting as artistic expression were not making art before widespread availability of LLMs. Most professional and long-practicing artists are disdainful of AI art. Yes, this is a broad generalization, and admittedly anecdotal to my own personal experience. But all I’ve seen from those in the pro-AI camp is equally anecdotal, so I figure my personal experience is as valid as someone else’s.

Why are artists not embracing AI wholesale? I believe it is because practicing artists are already used to exercising their creativity and skill constantly. They tend to recognize that prompting an AI is a fundamentally different thing from painting or drawing or writing fiction, even if painting is mediated through technological tools like touchscreens and tablets, and writing is done in a word-processor will spelling and grammar check.

Artists recognize that prompting doesn’t allow for the depth of creativity that they need and expect. They feel the way it takes away granular control and intimate feedback.

Gatekeeping

There is an argument, frequently deployed by the pro-AI crowd, that artists are gatekeeping art and resisting its democratization. A parallel argument is that AI is an artistic accessibility tool for the disabled and, weirdly, the unskilled.

This argument seems easily debunked by the many incredibly accomplished disabled artists in the world, many of whom have developed remarkable skills despite the challenges they face. Again anecdotally, it also seems like it’s rarely disabled artists actually making this argument, and much more frequently someone making it on their behalf.

There are examples of artists suffering disease or injury that made it impossible to continue creating art in the same way. Some are able to work around this. For some, using AI is an opportunity to gain some expressive ability back. But it is of a substantially different kind.

However, the frequent discussion of “lack of artistic skill” in these contexts is absurd to me. It implies that skill is inherent and not learned. It suggests that lack of effort or determination is on par with real disability, and that we should have pity on the would-be painter who hasn’t picked up a brush since grade-school and is offended that artists would dare deny them the tools to express themselves.

What these would-be artists fail to understand is that they are cheating themselves. They have not experienced the complex interplay between artist and art that is fundamental to the act of practiced creation. They believe that their AI-prompted art is giving them the same sensation, the same creative outlet, that the practiced artist gets from drawing or painting by hand. But it is only a shallow facsimile.

The artists are not trying to keep these people out of the walled garden of art. They are trying to get the prompters to understand that they are depriving themselves of a far richer experience by choosing the tool that seems more approachable.

Making Art

The sudden advent of tools that allow us to speak a few words and generate a painting is interesting and worrying and weird; embedded in complex and often problematic cultural and technological contexts. It’s that shocking sensation of science fiction suddenly become real.

Prompting AI clearly allows for some creative expression that influences the output, and that output can look quite competently rendered. But it is a fundamentally different and less expressive activity than traditional, “manual” forms of creation. It is less human. It should not be classed as the same thing as making art.

It also reveals broad cultural misconceptions about the artistic process, where only the final output is valued, and the process is not. That may sound cliché, but it is a meaningful cliché nonetheless. The process is valuable, and ignoring it is crass materialism and commodification of art; indifference to the meaning and mysticism of human creation.

Art can be sold, so corporations care about the output. They see artists and process only as a cost center. They have trained us on their viewpoint, to care only about the product (and it is purely a product to them). They call it “content,” so indifferent to the actual thing that it’s given this bland, meaningless label. The plain cardboard box of art.

I think I will have more to say on this subject, but I’ll stop here for now. I’d love to hear what other creatives think. Am I misrepresenting you? Do you see value in gen AI as a tool for artists? How much of the value of art is in the output vs. the process of creation? Let me know in the comments.