‘A Digital Copy of Yourself’: Artists Concerned Over AI’s ‘Industrial-Level Identity Theft’

“You are competing with a digital copy of yourself, with a machine that does not sleep, does not rest and does not get paid.” Karla Ortiz, a San Francisco-based artist and illustrator, is suing artificial intelligence (AI) generator companies Midjourney and Stable Diffusion for what she views as copyright infringement. She told Fox News, “Somebody is able to mimic my work because a company let them. It feels like some sort of industrial-level identity theft.”

Ortiz is not alone. Artists are realizing that their artwork and signature styles are training AI art generators — the very machines that may end up taking their jobs.

AI art generators rely on massive volumes of data to generate images. These data sets include artists’ copyrighted works, with some AI-generated images even containing tracesof watermarks. Any artist with an online portfolio is at risk, having few, if any, ways of opting out of the training data. More troubling, users can train models to copy a specific artist’s style without the artist’s knowledge or consent. This has some creatives like Hollie Mengert worried.

Mengert is an illustrator and character designer, specializing in cartoon characters for children’s literature and animation. In 2022, a Reddit user utilized 32 of Mengert’s illustrations to fine-tune an AI model. People using the AI art generator Stable Diffusion could then type in Mengert’s name to create AI art imitating her style.

When asked about this new development, Mengert responded, “My initial reaction was that it felt invasive that my name was on this tool; I didn’t know anything about it and wasn’t asked about it. If I had been asked if they could do this, I wouldn’t have said yes.” She explained that some of the images used to train the model belonged to outside clients like Disney and Penguin Random House. Those companies owned the images, making it illegal for others to use them even if Mengert granted permission.

“For me, personally,” she said, “it feels like someone’s taking work that I’ve done, you know, things that I’ve learned … and is using it to create art that I didn’t consent to and didn’t give permission for. I think the biggest thing for me is just that my name is attached to it.” (The Reddit user has since renamed the art style from “hollie-mengert-artstyle” to “Illustration-Diffusion” and clarified that Mengert is not affiliated with the model).

Greg Rutkowski, a talented digital artist who creates fantasy landscapes for video games, ran into a similar issue. His name had also been turned into a popular search prompt, exceeding the names of Michelangelo and Picasso by around 91,000 searches. Rutkowski lamented, “As a digital artist, or any artist in this era, we’re focused on being recognized on the internet. Right now, when you type in my name, you see more work from the AI than work that I have done myself, which is terrifying for me. How long till the AI floods my results and is indistinguishable from my works?”

It’s an irony: artists are unwillingly training the machines that may replace them. Furthermore, once an AI system has been trained on copyrighted data, it’s impossible to remove an artist’s part without destroying the entire model. These fine-tuned generators have already disrupted careers. “[T]he damage is irreparable,” says Jon Juárez, an artist whose picture was used by an AI platform without his consent.

But the AI issue is also complex. While Ortiz, Mengert, Rutkowski, and others are concerned about the ethics governing AI image sourcing, others believe these AI data sets fall under fair use. Fair use is a legal doctrine that allows use of copyrighted work without permission in specific circumstances. Among other things, works are generally considered fair use if they have little resemblance to the copyrighted work in question. In addition, “transformative” works (pieces that add additional details, further the purpose, or dramatically change the original copy) are more likely to fall under fair use.

So, is AI art transformative? AI technology does not store images, but collects mathematical representations of patterns from data available on the web. When a user enters a written prompt, the AI tool makes pictures from millions of these patterned representations. It does not piece multiple images together, therefore creating a “new” or “transformative” work. Furthermore, while original art is protected under copyright, artists can’t copyright a style, which further blurs the line between fair use and theft.

Additionally, AI advocates observe that creatives take ideas from each other all the time. According to art director Floris Didden, “The nature of art-generating AIs doesn’t bother me as much as it seems to bother many artists. We all look at each other’s work for inspiration on style, execution, ideas, subjects, etc., and [mix] it with our own ideas in some way to hopefully create something that can stand on its own. ... I’m not saying there’s no originality but let’s not pretend we don’t massively feed off each other.”

In general, most argumentation boils down to whether artists should be compensated when their art is utilized and whether they can decline usage of their artwork. While there is much contention on whether artists should have remuneration for artwork used by AI companies, most agree that standards should be set in place.

Regardless of the solution, the challenges of the digital world and the prevalence of bad actors will make any resolution difficult to enforce. Rutkowski expresses the sentiment of many artists, “I understand how these programs use artwork and images to build their models, but there should be some protections for living artists, those of us who are still doing work and advancing our careers.”