Generative AI and the death of the artist

Hello! If you’re a reader of Dirt, the digital culture newsletter I co-founded, you know that I have some personal news: I’m departing Dirt to focus on my new position as staff writer at The New Yorker and finish my upcoming book Filterworld: How Algorithms Flattened Culture. For Dirt readers who are new here, welcome to my personal newsletter. Please stick around for more short essays on tech & culture and teases from the book.

I recently wrote a New Yorker column on a group of artists who are suing generative AI tools like Midjourney and Stable Diffusion for stealing their work and replicating their style. The case is fascinating: Kelly McKernan, one of the artists in the lawsuit, feels like AI is already taking away their livelihood, as book publishers and other potential clients turn to AI tools to make cover art rather than commissioning a human artist. AI databases vacuumed up McKernan’s art (along with millions of other creators), trained it into a machine-learning algorithm, and allowed random users to replicate it by using their name. “I can see my hand in this stuff, see how my work was analyzed and mixed up with some others’ to produce these images,” McKernan told me.

In the column, I wrote, “You could say that artists are losing their monopoly on being artists.” This line caused some consternation: No one has a monopoly on being an artist, because anyone can make art, and art can only be “monopolistic” if we consider it to only exist as a commodity (which it kind of does, at least in the capitalist public marketplace). I wanted to deepen that argument here.

The phrase “the death of the artist in the age of AI” keeps sticking in my head. It’s not that artists are going to die out; it’s that, when generative AI tools promise the ability for anyone to make “art,” then the definition of artist is going to radically change. We can’t define an artist as someone who is skilled enough to produce a unique thing on their own — the kind of hard-won skill that takes years or decades to personally develop, on top of a natural gift. AI generators suck up that human talent and creativity (in the form of images, sound, and data), turning it into grist for the mill of technology that entrepreneurs and investors are positioning as the next great innovation. They seem to say: We won’t need artists to make art, you can just make it on your own, and it can be whatever you want. But will that be satisfying to anyone?

It’s already leading to some bizarre phenomena, cases in which the human artist is outmoded except for the role of fuel. As The Information reported (paywall), one 19-year-old used open-source AI tools to build a trained model of the pop star Ariana Grande’s voice and then generate cover songs as if Grande had sung them: Ariana sings SZA’s song “Kill Bill,” which she never did in reality. But it doesn’t sound terrible, perhaps because the source material is so good: a human voice and a song that other humans already wrote. It’s the same as the viral “Lisa Frank Lloyd Wright” images made using Midjourney: two artists’ work mashed up together.

In these cases, is the artist the person who writes the prompt, who had the idea to try it out in the first place, even if they couldn’t execute it without software? Plenty of artists in art history work with prompts that they don’t execute by hand, including Sol LeWitt and Donald Judd. But those artists’ distance from their process was part of the conceptual strategy of the work, whereas “prompters,” as they’re sometimes derisively called, are only interested in the end product, the instant generation of the image.

People like the teen who replicated Ariana Grande don’t hate the artist. In fact, they love them: That’s why they wanted to generate infinitely more copies, to take Grande’s (literal) voice into their own hands and play around with it. This has only recently become possible, with the creation of accessible AI tools. I remember sitting around on forums two decades ago imagining what it would be like if Dave Matthews Band covered a particular song by another artist that I also liked. (I’m getting old, and very uncool to begin with.) I couldn’t hit a button and hear what it would sound like; the closest I could get was playing guitar myself, learning to replicate the songs in a much more arduous, human way.

With generative AI, the artist becomes a template, a brand-name, a signifier of some specific style. It reminds me of those labels on some Old Master paintings: “From the school of Rembrandt von Rijn.” (Though in that case, a trainee painter still had to learn to work in the style of Rembrandt, perpetuating the human creative process — fanfiction works the same way.) We appreciate those paintings less because they’re somewhat derivative. But anything generated by AI and instantly made public is by its nature utterly derivative. Fans generate art (or music) in the style of the artist, and the figure of the artist becomes emptier and less meaningful, something to be ignored the way we ignore factory workers on an assembly line, or allow algorithmic feeds to do the jobs editors and curators once did.

I don’t care if people enjoy AI-generated culture. It might be a fun solo game to play. Yet I worry about its consequences, the way it suggests we overlook where the data that the tools run on came from and the people who made it. The Surrealists used the word “automatic” to describe art or writing that came directly from the personal subconscious, unmediated. This new kind of automatic culture instead avoids messy, interesting humanity entirely.

— The Supreme Court Probably Won’t Break the Internet—At Least for Now: I wrote another New Yorker column on two recent Supreme Court hearings that questioned Section 230, the law that protects digital platforms from being responsible for what users publish on them. Do algorithmic recommendations actually mean that platforms are responsible, since they drive the user to particular content? It was interesting to hear the justices discuss just how much algorithms influence our digital lives.

— Jess has had a great run of stories lately as national politics reporter for the Boston Globe here in DC. It was fun to watch her report and write this profile of New Hampshire’s Repulican governor Chris Sununu, for which she went skiing with the governor!! Skiing journalism! Sununu is trying to chart a “normal” path for his party and pushing for a presidential campaign. “I like steep and fast, I’m not going to lie,” he said, about halfway down the run, impressed and possibly relieved that this reporter had managed to stay on her skis. “But I try to stay out of the bumps.”

— I’ve been enjoying this City Pop mixtape on Spotify. It has a few familiar names (to readers of this newsletter) like Hiroshi Sato, but plenty of new-old stuff. Or, you could try out “Indonesian City Pop,” which was apparently known as Pop Kreatif. The sonic palette is very similar, but a bit hazier and more tropical, very cool.