This painting, generated by artificial intelligence, “using a carefully crafted prompt,” won first prize at the Colorado State Fair’s fine art competition for digital arts:
The award, which includes a $300 cash prize, was won by Jason Allen. While other candidates for the award used software like Photoshop and Illustrator to create original digital art by hand or alter photographs, Allen used an AI called Midjourney which can generate artworks or just about any kind of synthetic image from a mere line of text. The AI-generated artwork, which Allen calls “Théâtre D’opéra Spatial,” depicts a bizarre but intriguing scene from the distant future or some other world in which human figures are in awe as they stare into a huge circular viewport into a sun-drenched landscape reminiscent of Frank Herbert’s Dune.Tibi Puiu, “This stunning AI-generated picture won a fine arts competition — and human artists were not happy at all” at ZME Science (September 1, 2022)
Other digital artists, who were probably not using Midjourney, were not pleased with Allen’s win: Genel Jumalon took to Twitter to complain and, we are told,
The comments on the post range from despair and anger as artists, both digital and traditional, worry that their livelihoods might be at stake after years of believing that creative work would be safe from AI-driven automation.Shanti Escalante-De Mattei, “AI-Generated Artwork Goes Viral after Winning Award at State Fair” at ARTNews (September 1, 2022)
Game designer Allen celebrated his win with a comment:
“I feel like, right now, the art community is heading into an existential crisis if it’s not already. A big factor of that is … the disruptive technology of open AI,” Allen told the Chieftain. “A lot of people are saying, ‘AI is never going to take over creative jobs, that’s never going to be something that artists and sculptors have to worry about.’ And here we are smack in the middle of it, dealing with it right now.”Shanti Escalante-De Mattei, “AI-Generated Artwork Goes Viral after Winning Award at State Fair” at ARTNews (September 1, 2022)
Is Allen right in thinking that AI will take over creative jobs?
Thoughtful geek Eric Holloway looked at the situation for us and says, um, no:
The AI is trained on a bunch of existing artwork, so is highly likely to generate a picture that correlates with art that was previously considered good. However, the correlation is only surface deep.
The “goodness” of the art breaks down quickly on closer examination. If we look at “Théâtre D’opéra Spatial” closely, we notice the figures’ anatomy is not like any human anatomy, and with some of the human-looking figures, it’s unclear whether they are persons, or furniture. The lights on the wall also look like a cityscape.
What on first glance looks like a surrealist dreamscape loses all its mystique once we understand how the AI works. The AI is using a process called “inverting the neural network.” A neural network has been trained with human-created art to predict labels for the art.
Normally, once a neural network like this has been trained, it will be given new artwork to classify. However, in this case, the reverse is done. The neural network is given a set of labels and it goes backward to find artwork that fits the labels. Since there will be many pieces of artwork with the same labels, the neural network interpolates between all the art, so we get something in between all the pieces of human-created artwork.
And that’s why the AI generated pictures give off a surreal, dreamlike quality. Surrealism and dreams are marked by generating feelings by surface impressions, but become nonsensical on rational analysis. The same occurs with these AI-generated pictures. However, the difference is that while surrealism and dreams have coherent pieces for the most part, even if the overall work is hard to rationally pin down, the AI pictures do not even have coherent pieces.
He adds, “Intelligent design theory can make a clear case this isn’t art. If we say that good art exhibits complex specified information, then the AI generated pictures cannot be art. The complexity is very low.”
Business prof Jay Richards, author of The Human Advantage: The Future of American Work in an Age of Smart Machines, is not averse to treating the outcome as art as long as we are clear about where the art part lies:
This story, like other stories like it, is far less astounding than it seems in the headline. The impression is that a program created something that looks like art — as if the program had intentions and real creativity. But of course, the program itself was designed, and has all sorts of constraints, including many examples of actual art, but actual artists to imitate.
And in this case, there was a human giving it intentional prompts. As a result, I don’t see any reason not to treat the outcome as art, and to credit the many human agents that played a role in bringing it about. We’re not dealing with a robot artist.
Overall, “generative art,” like Allen’s Théâtre D’opéra Spatial is likely to bump the art news cycle for a bit. For one thing, it pays. Puiu reports that AI-generated Nude Portrait#7Frame#64 netted $821,000 and Edmond de Belamy netted $432,500 in 2018. But, to the extent that melding a number of artworks together requires little creativity — relative to originally creating them — it won’t likely continue to command high prices or retain high public interest for very long.
You may also wish to read: New text-to-art image generator easier to misuse, critics say. Relative to DALL-E 2, it is easier to use to generate pornography, as controversial website 4Chan has recently demonstrated. Comparison of DALL-E 2, Midjourney, and Stable Diffusion as AI text-to-art generators reveals significant differences in treatment, analyst Fabian Steller says.