This article is part of Mochi Magazine’s summer issue on Art: We invite you to expand your perspective of what art is or who an artist can be. As you check out the rest of our issue here, we hope you are inspired to find your own form of artistic expression.
Last summer, the AI art-generating website DALL-E was all the rage. I’ll admit that my friends and I were sucked in too. It was easy to put silly prompts into the website, like “Picasso Hello Kitty” or “Naruto eating McDonald’s,” and laugh at the results.
But if you look closer, there’s something off about DALL-E’s art. Faces look more like spirals or blobs of skin tones than actual humans. Hands somehow sprout six fingers. Backgrounds of elaborate architecture, lighting, or nature are jam-packed with one-point and three-point perspectives going haywire, if you just zoom in a little.
“Being an artist, you’ll never make money,” and “It’s too much of a risk,” were things I heard a lot in response to my aspirations. “You’ll never grow big; you should never take the risk.” But I can’t imagine myself doing anything but art. The first time I tried braving the steep, vertical learning curve of digital art, I forgot to even look up tutorials. I was so eager to make all the things I’d ever dreamed of, with just a laptop and tablet.
For those who feel new to art or are unable to create art on their own, these AI generators can seem like accessible ways to enter the art world without much training or practice. However, as an artist, the popularity of these models is more disheartening than the finger-wagging or head-shaking warnings, especially since many people do not consider the models’ ethics and where they came from. The “art” generated by these machines was built from the data and work of thousands of artists. Automating the most intrinsically human thing we can do — create — has more consequences than possibilities.
Where AI Started
You’ve probably encountered AI art in one form or another by now. It’s in our TikTok filters, in our workplaces, in the media we consume. These art generators exploded in popularity in the last couple of years. But these models — whether it’s Midjourney, Stable Diffusion, or DALL-E 2 — are trained to respond to users’ text prompts through data taken from millions of images without the creators’ consent.
University of Chicago’s Neubauer Professor of Computer Science Ben Zhao has helped create multiple tools to protect people against unwanted AI usage. I spoke with him to learn more about the impact and significance of these models in our world.
The biggest dataset that trained many of these AI models, Zhao says, is called LAION, a massive collection by the nonprofit OpenAI, one of whose founders was Elon Musk.
“It was completely indiscriminate. It did not care about copyright, it did not care about ownership, it did not care about the source, it did not care about watermarks, it made no difference about any sort of image,” Zhao says.
This process is referred to by some as “data laundering,” according to Zhao, in which organizations are able to bypass messy ethics and copyright issues under the guise of being nonprofits. This is because one exception under U.S. copyright law states that nonprofit, educational, and noncommercial uses, such as university research, can utilize copyrighted images if it’s not for commercial purposes. Thanks to the shield of nonprofits like OpenAI, companies could skirt around copyright liabilities, even though the LAION dataset was funded by the for-profit group Stability AI.
Say you ask AI to generate an image of a cat in a space suit, in the style of Van Gogh. AI models that have been trained on data corresponding to Van Gogh’s art style will then be able to spit up that prompt in his style — only it isn’t anything that Van Gogh made himself.
What’s the problem if Van Gogh isn’t around anymore to say a thing? Unfortunately, AI models also target living artists, many of whom rely on their art for income, are trying to build a reputation and brand, or who are simply looking for ways to share their creations.
“There’s something really perverse about the fact that artists are now finding themselves at the mercy of models that are only in existence because they have been trained on their art, and now are used to replace them,” Zhao comments.
But the impact of these models reverberates beyond individual artists. Entire generations of future artists in training might not even bother going into the field, because of how hostile it has become.
“Some of these artists who teach classes are seeing their classes empty out, because art students for the next generation are completely despondent and discouraged from trying to pursue art as a career,” Zhao notes. “They think, ‘What’s the point? I can spend 20 years becoming extremely successful, develop my own style, only to be ripped off in a matter of minutes.’”
AI Art Does Not Exist in a Vacuum
The problem of AI feels incredibly personal.
As an Asian American artist, I cannot untangle my identity from my art. I couldn’t even if I tried. Art is where I can write or paint love letters for my own experiences, while finding community in the midst of a world that’s often predominantly white.
Last year, a beloved Korean illustrator, Kim Jung Gi, passed away in October, bringing an outpouring of grief from the art community. Only a few days after his death, a new tool popped up on Twitter.
A user, @BG_5you, announced that they had created an AI model to generate images in Kim’s style, as a sort of homage to him. Notably, they asked any users that used the tool to credit @BG_5you — not Kim.
Like many artists seeing this news, it outraged me. The Kim Jung Gi model, according to writer Leo Kim of Art News, is a manifestation of a new wave of techno-Orientalism. When Asians could no longer be generalized as backwards and uncivilized barbarians in the last century, a new tactic to strip them of their humanity arose — affiliating Asianness with roboticness. Sure, Asians could be advanced. But they were uncreative. Uninspired. They could not be artists the same way a Westerner could.
We see this deprivation of humanity when artists are not allowed to rest even after their death, when their life’s work is fed through machines to create mere facsimiles of their styles, and creations they were never around to witness.
The internet was revolutionary, especially for digital art, creating a safe space spanning geographical borders for people to share their work and find community. Seeing our labor subsumed into machinery, seeing ourselves turned into forever content creators and never humans, feels lonely. I keep asking myself: What can I do? Where is the heart?
How Artists can Protect Themselves
The future is not all bleak, though. Zhao is one of the research leads on a tool called Glaze, a project started by the University of Chicago’s Security, Algorithms, Networking and Data (SAND) Lab, and a collaboration with professional artists.
Previously, SAND Lab had developed a tool in 2020 called Fawkes that protected people’s photos from getting used in unauthorized facial recognition models such as Clearview AI. Zhao says that last summer, his team began getting emails from artists asking if Fawkes could protect their work as well. When novel AI models began exploding in popularity in the news, Zhao and his team got to work contacting professional artists and developing Glaze.
“We definitely saw an obvious ethical and technical problem, and we tried our best to work on that,” Zhao says. “We thankfully got a lot of help from individual artists like Karla [Ortiz] and several others, and they helped us communicate and reach out to the art community.”
I had heard of Glaze thanks to the buzz on Twitter from fellow friends and artists. It’s a free tool, constantly being updated and improved, where artists can essentially make little tweaks to their art to prevent AI mimicry. These changes, or cloaks, are barely visible to the human eye. However, to a machine, they will make no sense at all.
Glaze is a relatively young tool, born out of an intrinsic gap between the way humans perceive art and how machines view it. So far, it’s the only option of any kind of solid protection that artists can seek out against AI models. Zhao says that future instances of protection against AI theft will likely stem from this basic foundation: taking advantage of how humans and machines consume art differently.
“If that artist’s name is invoked in an attempt to reproduce art using their style, the model will get it completely wrong and produce something that may be correct in subject matter, but the style will be completely off and not at all representative of the person that it was supposed to mimic,” Zhao explains. “If that happens, then that’s what we call a success.”
In the face of all my AI-related doomscrolling, Glaze allowed me to take a sharp breath of relief. Zhao acknowledges that it’s a work in progress to make the tool truly accessible and effective for all art styles, as well as ensure that it cannot be bypassed by AI models. But Glaze is proof that people do care. Artists are not alone in this fight.
Artists Cannot Be Replaced
Maybe it’s tempting to conjure entire spreads of color and line from just a few words. The sky’s the limit, as is our vocabulary. But the problem with AI lies precisely in how deftly it invisibilizes the humans behind its datasets and training.
“What people are missing is the fact that what got you there, what allowed these models to do what they do today, is training [with] data without artists’ consent,” Zhao says, “and relying on decades of hard work by many of these artists to create their own styles, then to actually take the fruits of their training and labor and basically subsume it into these models without permission.”
It’s not just drawings. AI will come for, and has come for, practically all creative industries. Joe Russo, who oversaw the iconic Marvel Avengers films, predicted that AI could begin creating movies in the near future. AI can generate songs that mimic voices of singers who never wrote the lyrics in the first place. Voice actors in Latin America are being paid to give their voices to the AI models training to replace them. Most recently, the Writers Guild of America has gone on strike for the first time since 2007, against the use of AI in Hollywood,
It’s a catch-22 of sorts now. What happens when there are no more new songs being released, no more new books or shows getting written, no more new artists willing to create?
“The ugly fingers, the misshapen shapes, the inability to generate new types of style that are not already well-defined into these models — all these limitations will be frozen, because there’ll be no more human creativity to push the boundaries forward,” Zhao says. “When that happens, we will truly be stuck. Because we’ll have no humans to add their creativity to the mix.”
We are far from a future where AI is intelligent enough to completely overtake human work and create without being trained, but as more and more people begin to embrace this technology and larger industries take notice, we need to acknowledge the faulty foundations on which they were built in the first place. Perhaps one day, we can teach AI to create on its own, without our prompts — but what then?
“Before we get there, AI will have destroyed the industry that it needs in order for it to survive,” Zhao warns. “That’s the real problem.”
It’s honestly a little terrifying, to picture a world where nothing we consume has been touched by a person.
Think of your favorite songs, artwork, or anything you consume regularly. Those were probably all made by people. The oldest artwork we know — from handprints to animals to figures running across cave walls — were made by human hands, the same ones these models can never seem to get right.
In the battle of humans versus machines, you cannot say that human art has lost, or that AI has won when these models are relying on our contributions in the first place. So, I will continue drawing. Not for the sake of AI’s survival, but because I have always needed to, and because spite is a great motivator.
Mostly, because no force or artificiality can keep me away from art. Unless, god forbid, my hand breaks in some freak accident — even then, I’m happy to simply wait until the day I can pick up a pen again.
For more information, check out Glaze on twitter and their project website.
