AI in Art: Exploring the Boundaries of Bias, Beauty, and Belonging
I’ve made another foray into the world of AI art, this time in décor. I’ve recently remodeled my bathroom, and used some AI art to decorate it. You can see the video here.
I like the still that YouTube chose for the cover of this; it doesn’t look like a bathroom at all.
A quick summary of the video, if you don’t have the 4 minutes to watch it. I’ve used an AI image generator (Midjourney, in particular) to generate a mural that I have used to decorate the “wet area” of the bathroom. The mural gives the trompe l’oeil effect that you are in a Greek seaside village. It transforms the cozy, but possibly claustrophobic wet area into a friendly, expanse of a touristy village. I am very happy with the result.
As I have written elsewhere, one of the things I really like about doing AI art is the personal connection. When I show off this mural on the House and Garden tour, visitors will see an image that nobody has ever seen before. And this is an image that I had a lot of say in creating. As did my husband; it was his dedication to genuine classical references, and an affinity for horses of various forms, that resulted in the final work.
This is an interesting collaboration of all three of us, i.e.., myself, my husband, and “Midge” (the affectionate anthropomorphic name that its fans give to Midjourney). I wanted the Greek seaside town (well, actually, merged a bit with Sidi Bou Said, a Tunisian seaside town, which has the distinction of being a place I have actually visited, unlike, say, Santorini). Tim wanted a Pegasus. Midge was unable to give us a Pegasus in a seaside town. There was a give-and-take, just like there is with a real-life artist. And the creative solution, of a mural-within-a-mural, was a collaboration of all of us. It wouldn’t have happened with me working on my own, or my husband on his own, nor Midge on her own. And the result is charming.
I can’t help but think of the usual criticisms of AI art. I’m going to address them here.
- It’s fake. It is common for AI art naysayers to talk about how the images aren’t of real people, or real things they’ve done, or real places.
This is certainly true of this piece; it doesn’t even try to be real. Is it Tunisia? Or Greece? I wanted the colors from Sidi Bou Said (which are famously blue-and-white). I wanted the seaside resort vibe of a Greek island. My husband wanted a mythological creature. There’s no way that this is real. if you look closely at the image, there are a bunch of more mundane unrealities; the sea level changes throughout the image, when a foreground object obscures the horizon. The tables are disconnected from any structure that could possibly serve them. The walls don’t connect in a way that could be a real building. And that empty wine glass just bugs me. But none of this matters; the fact that it isn’t real is a big part of the appeal. There isn’t a place on earth with a sapphire Pegasus mural. That doesn’t keep me from having a picture of one in my bathroom.
- You’re taking work away from real artists.
The company that supplied the wallpaper with my print on it had a few dozen stock photos that I could have chosen from, and one of them was of a Greek seaside town, on Santorini, I think. I could have had that one, for no extra charge. That’s right, the stock photo is just a way for the wallpaper company to sell their real product, which is wallpaper. That stock photo was taken once upon a time by a photographer, who sold it to a stock company, who sold it (probably as part of a subscription) to the wallpaper company. They don’t get any money from it; they are just using it to get people to buy their wallpaper printing services. Which I did. If I had taken the stock image (for no increase in cost), would that have put money in the pocket of the artist, who is at least two steps away in the value chain? Even if so, it wasn’t my money; it was the subscription money that the wallpaper company is already paying. It’s hard to come up with a scenario where having chosen the stock photo would have resulted in money going into the pockets of an artist.
And the stock photo would have come with its own headaches; the aspect ratio of the space is very specific, could I have had the ground and the tops of the towers visible in the same mural? Not to mention that there is no stock photo of a Pegasus mural; that doesn’t exist.
- The artists don’t get credit
The image I created draws on millions of images of Greek seaside towns, most of which probably adorn travel brochures for tours of the Greek islands. Again, stock photos taken by photographers who were paid for their work, according to a contract that they themselves brokered. Even if we could figure out which stock photos the AI drew upon, and which photographers took the photos (that’s a bit of a contradiction when talking about stock photos), would it make sense to credit them on this work? Would they even want their name put on an image with uneven sea levels, walls that don’t end, windows to nowhere? And not a single one of them took a photo of a mural of a Pegasus. Which photographer deserves the credit for that?
- Artwork created by AI is sh*t
In some sense, I don’t have to respond to this at all; I’m displaying this work in my home, not asking you to display it in yours, and unless I have invited you to shower at my home, I’m not even asking you to view it. If I ever am featured on a House and Garden tour, feel free to give my bathroom the lowest marks possible on your judging card.
This criticism is the one I find the oddest, in light of the others. If AI art is so low quality (a matter of taste, but let’s go with it), then why are we worried about how it threatens ‘real’ artists? We don’t usually worry about graduates of Bob Ross’ classes threatening the art market. For the most part, his students create passable works of art that they can take some pride in, and display in their homes. Most people don’t take art classes to start a new career. And if they do, they had better do a lot better than the average art student. Either the images are quality or they aren’t, you can’t have it both ways.
- AI models perpetuate social biases
They certainly do! That’s how they get any work done at all. One particular bias in this model is that horses don’t fly; this caused me a lot of trouble in early drafts, and eventually resulted in the insight to put the Pegasus into a mural. There are some other biases; a church in Greece will be Christian, and a Christian church will have a cross (this one seems to have beat out the bias that a place of worship in Tunisia would be Muslim, and certainly not have a cross). There are a bunch of more mundane biases; café tables in Greece have coffee cups or wine glasses. Greek seaside villages (and Sidi Bou Said) have blue-and-white architecture. The arches in Greece are rounded, not pointy. A street mural in Greece would be a mosaic. These biases are what give the model the ability to create a (somewhat) coherent image. And if I didn’t like those biases? There are hundreds or maybe thousands of models available on huggingface and civit.ai that have different biases; I could have chosen one of them. Or trained my own.
Do these models have any biases that stem from dangerous social stereotypes? Almost certainly so. Will these biases be used to perpetuate stereotypes? Only if someone chooses to display an image in a harmful way; but that’s on them. The AI model will generate hundreds of images if you ask it, and each image will include its own stereotypes. It’s up to you to decide which one you display.
Why am I responding to these objections? I don’t really have to; AI art isn’t illegal, and if someone doesn’t want to look at the images I create, they don’t have to. But I get some real satisfaction from the images I create. I make a special trip to just look at my shower a few times during the day. I actually look forward to spending time in that room (is that weird?). I really hope that my enjoyment of these things isn’t at someone else’s expense. I really don’t think it is, for all the reason’s I’ve outlined here.
