The Ethical Use of AI Art
How to Protect Artists from the AI Apocalypse
Imagine being an artist. You’ve spent years perfecting your craft and developing a style that people recognize. You’ve managed to become one of the few fortunate artists who ekes out a living creating art. Then one day you type your name into an app, and a computer starts spitting out image after image in your style.
That has to be jarring, disorienting, and perhaps even terrifying. Automation and redundancy are not new concepts, but at one time artists believed they were protected.
Artists are no longer above the flood zone.
When the ethics of AI-generated art are discussed, both artists and AI researchers tend to focus on the training data and copyright. Artists don’t understand how a model trained on their artwork isn’t obviously violating copyright. Researchers are confused by this since their models don’t copy images, they just learn to recognize concepts and use that knowledge to create new images.
What sort of protections do we need to give artists whose work is used to train AI art models?
The Power of Human Art
First of all, why are artists so special? Visual artists are a small part of the training data, but they are a very potent part. Having read through many prompts, I have seen a large portion contain the names of artists sprinkled in like seasoning. That’s because words are generic; names are specific.
When I say the word “bridge” you will imagine one of a multitude of possible bridges in a variety of shapes and sizes and even alternate meanings. When I say “The Golden Gate Bridge” all those possibilities collapse into a single bridge.
This technology is still in its infancy and understanding meaning is still hard for computers. Using an artist’s name is the equivalent of picking up an object and pointing to it in a game of charades.
So, it’s tempting to use artists this way. “Make something beautiful, like this.”
But AI art needs human artists. AI art can only explore latent spaces between concepts it recognizes. Human artists create those concepts.
The roll-out of this technology has been messy. AI companies want to move fast and break things. Well, they’ve broken the trust of the artist community. Some artists now want to protect their industries from this perceived threat by any means necessary, including law suits.
What is an AI art enthusiast to do? People using AI art generators don’t hate art, and certainly don’t want to harm artists. They want to express ideas visually. They don’t have the technical skills, but they might have a knack for product design. How can they use this paradigm-shifting technology without harming artists?
An Artist’s Identity
Artists have now joined the rest of the working class living under the threat of automation and redundancy. Not many artists have complained that their pencils and paper are being made by machines. But art is different, isn’t it? Isn’t art uniquely human and special?
Yes, watching an artist draw is magical. But the same magic applies to paper-making.
Watching an artisan make paper by hand is hypnotic and feels special. But that doesn’t mean all paper should be handmade. When paper was made by hand, artists paid a high price for their materials.
Artists who want high-quality paper for their artwork will pay a premium for a sketchbook filled with handmade lokta paper produced by monks in the Nepalese mountains. Their art deserves nothing less. Other artists will prefer the consistency and price of a precisely cut pad of 100-pound vellum paper, available at stores near you.
Where paper is cheap to produce, art flourishes.
Art will flood out into the world like never before in human history and this will profoundly affect the way artists and designers work. People will continue to hire individual artists for their skill and vision. And yes, some jobs will change or disappear. But new jobs will arise as new products become viable.
Art is cool, but what people really love are the artists themselves. This is what needs to be protected.
The Path Forward
Artists have three paths forward.
- The path of the pilgrim. Develop an identity as an artist. People will always pay a premium for a skilled artist’s work.
- The temple run. Focus on doing what the AI cannot do, and keep staying one step ahead of its steady, unstoppable progress.
- The trailblazer in the wilderness. Use both AI and the technical skills of an artist to do more than either a human artist or AI model could do alone.
There’s a problem, though, with the first path. It’s hard to develop an identity as an artist when artists’ identities are being diluted by a deluge of text prompts that use their names. If this were happening behind closed doors, it wouldn’t be as much of a problem. Unfortunately, on sites like Lexica.art, OpenArt.ai, and Midjourney, artists’ names are being associated with thousands of AI images that are but shadows of their epic art.
Honestly, many of the results are monstrous mutations with too many heads and deformed hands.
The art being tied to artists’ names is not representative of what they are able to produce. At best, the images are in the style of the artist, or inspired by their work, but very different from what they would create.
When artist Hollie Mengert was shown images created using a textual inversion technique that was trained on her images, she found the experience disturbing.
As far as the characters, I didn’t see myself in it. I didn’t personally see the AI making decisions that that I would make, so I did feel distance from the results. Some of that frustrated me because it feels like it isn’t actually mimicking my style, and yet my name is still part of the tool.
— Hollie Mengert
Imagine a potential client decides to search for “art by *name redacted*” and they end up seeing a gallery of several thousand AI images. That client might mistake the gallery for *name redacted*’s work, or believe that it is representative of their work, and decide not to hire them because the AI art is mediocre.
Worse yet, they may try to find the artist’s website and simply can’t find them.
This is why users of AI art models should do everything in their power not to harm the identity of an artist. If you do use an artist’s name in a prompt, please try to protect the integrity of that name by keeping those prompts private.
- Don’t post prompts with artist names to any public galleries or social media.
- Be aware that some apps like Midjourney publish your prompts to online galleries.
- Be aware that some sites are scraping prompts and republishing them.
- Be aware that some apps include the prompt text in the filename and metadata of the image files they create.
And as for training models, hypernetworks or embeddings, using an artist’s art and name in this case is a blatant case of identity theft.
- Don’t train models, hypernetworks, or embeddings to mimic a particular artist’s style.
- Don’t use author’s names as tokens in a model, hypernetwork, or embedding.
- Be aware that some methods of training on public sites like Huggingface will publish your results to a public library.
End users who want to design products or post images using AI art should put at least this much effort into protecting any artist names they use in their content.
Also, as a specific warning, keep in mind that Midjourney’s paid “private” mode isn’t actually private if you send a prompt through one of their public channels. To keep a prompt private it needs to be sent to the Midjourney Bot in a direct message while in private mode.
This is probably the least, best thing that users can do to protect artists. Of course, if you can generate images without using an artist’s name, you won’t have to worry about any of these protective measures. You will be creating a style that is all your own.
But if you do use an artist’s name, especially a living artist who is trying to make a living, just keep it to yourself.
… every word doth almost tell my name,
Showing their birth, and where they did proceed
— William Shakespeare, Sonnet 76
