Why does artificial intelligence look like the cover of a 70s progressive rock album?

The good news is that I’m a journalist, so I asked some artists, researchers, and art historians what they think of the aesthetics of AI art. First, I called Amelia Winger-Bearskin, an artist and professor at the University of Florida. Winger-Berskin was cataloging various visual trends she has noticed in recent AI art. She names one trend Nightmare Corp.– often illustrated with images created by Google’s Deep Dream, an old generator released in 2015. He specializes in swirling psychedelic imagery, such as flashbacks to a particularly torturous acid trip. “Of course, prog-rock influences,” she says. The other category that Winger-Bearskin explores, which she calls Dada 3D, is very similar to the silly scenes I suggest when I play with these generators. She describes it as “something like a surreal parlor game”.

In addition to taxonomization trends, Winger-Bearskin noticed broader stylistic features in these generators. She sees the obvious influence of Western animation and Disney-style anime, as well as a tendency to treat whiteness as the default race—a result, she suspects, of training these generators on datasets that rely heavily on Western Disney-style animation, anime. and images of white people.

Lev Manovich is also watching closely. The cultural theorist and professor at the City University of New York has been hiding on the Midjourney Discord server since last year, analyzing how people use the generator. After Midjourney released an update last fall, he saw some changes to what people were suggesting the generator should do. For example, after it became better to realistically depict people, the number of requests for portraits of both men and women increased.

Digital artist Sam King first began following AI art closely in 2021. Excited by what they saw, they exchange his favorite images on social media, gaining popularity as a curator once the technology took off. They describe the earlier wave of oscillators as favoring “trippy, abstract stuff”. (These generators are known as Generative Adversarial Networks, or GANs. I have seen Little people call this view, rather uncreatively, GANism.)

King considers the latest wave of oscillators, called diffusion models, to be stylistically distinct. In the same way that oil painting and watercolor produce recognizably different effects, GAN generators and diffusion generators produce recognizably different images. If you want a more realistic visualization of, say, Tony Soprano drinking a cappuccino with Shrek, diffusion models will likely produce convincing results. “Theoretically, with these machines, you can create a wide variety of aesthetic objects,” they say. However, more realistic does not necessarily mean more stylistically diverse. Like Winger-Bearskin, King often sees Disney and anime influences as well as comic book art.

“The rhetoric of these companies is that you can do anything you can imagine. It’s about this open border. But, of course, mass culture follows certain stereotypes and tropes,” says Manovich. He sees variations on several themes over and over again: “Fantasy, fairy tale, comic book, video game.”