In 1992, the the poet Anne Carson has published a small book entitled Short interviews. This is a series of micro-essays, ranging from one sentence to one paragraph in length, on seemingly disconnected topics: orchids, rain, the mythical Andean vicuña. His “Short Talk on the Sensation of Airplane Takeoff” is what it sounds like. His “Short Talk on Trout” is mostly about the types of trout that appear in haikus. In what passes for the book’s introduction, Carson writes, with dry Canadian relatability, “I’ll do anything to keep boredom at bay. It is the task of a lifetime. Just when she posted this, the internet started to take off.
Fast forward 30 years and one of the last ways to avoid boredom, at least for me, is to stay up late and have fun with AI image generation. Tools such as DALL-E 2, Midjourney and Stable Diffusion can be tasked with text prompts to produce ersatz oil paintings of dogs wearing hats in the style of Titian, or simulated photos of models in plasticine of astronauts on horseback. When I first started playing around with Stable Diffusion, which is open source and great fun, it reminded me of Carson’s lectures. I went back to them to find out why. Pretty soon I realized the resemblance had something to do with form.
Everyone says content is king, but the secret monarch of the content economy is form: constraints, rules, minima and maxima. You grow in the form of learning. A high school essay is five paragraphs. Sitcoms leave eight minutes in the half hour for commercials. Novels are long. Tweets are limited to 280 characters.
What makes my tweet, essay or studio film different from yours? The choices that each of us makes in the form. In a word, our style. Carson’s book takes a familiar form, the little lecture, and subverts it, manipulates it, until as a reader you begin to feel like you’re inside his wondrous brain, scrolling through the history of her mental browser, joining her in hyperlinked fantasies and half… abandoned rabbit holes. Image generation is a bit like that, but instead of communing with a single brilliant Canadian brain, you commune with a giant dumb world-brain. (A less neurological way of putting it: a large number of data objects grouped together in layers, linked together to an incomprehensible degree, like the string-and-nails wall art of a multi-masted clipper but on fire with the data flow.)
In general, humans like to use machine learning to help pathologists refine a phone photo or create a better map. But AI generators bother a lot of people. These tools work by scanning images across the internet, absorbing the visual culture they contain by scanning their captions, then adding bubbly visual noise to them until they look like static. To create a new image, the AI starts with a caption and some static, then runs the process backwards, removing noise until an image appears that more or less lines up with the caption. (It’s bad for drawing hands, but so am I.)
It seems rude. It’s disgusting to see artists in the database being forgotten. It seems rude that someone could say to a computer, “I want a portrait of Alex Jones in the style of Frida Kahlo,” and the computer would do it without moral judgment. These systems bundle scenes, territories, cultures – things that people considered “theirs”, “their lives” and “their jobs” – into a 4 gigabyte open source tarball that you can download on a Mac to to make a penguin playing baseball in the style of Hayao Miyazaki. People who can use the new tools will have new power. People who were great with old tools (brushes, cameras, Adobe Illustrator) will be thanked for their service and returned to Soylent. It’s like a guy wearing Allbirds dropping into a residential area where everyone was barely holding on and saying, “I love this place, it’s so weird!” Siri, play my Quirky playlist. And open a blue bottle around the corner!
So of course people are upset. Art websites ban AI-generated work, at least for now; stock image services refuse it too. Prominent bloggers who have experimented with AI illustrating their writing have been chastised on Twitter and vowed not to do it again. AI companies talk a lot about ethics, which always makes me suspicious, and some words are banned from the image generator interface, which is a shame because I wanted to ask the bot to paint a cottage “busty ” à la Thomas Kinkade. (You have to face your deepest fears.)