Curating A Personal Data Set
Tiffany Calvert is a Louisville-based painter and Associate Professor at the Hite Institute. She has a particular interest in still life paintings from the Netherlands from the late 16th to the early 18th centuries.
The Dutch interest in objects and their symbolic meaning informed her own painting when, in the late 2010s, she started repainting still life images and then painting abstractly on top of them. "I only did that a few times before I realized, 'this is silly.' That takes forever." So she began to make reproductions of paintings and paint on top of them. This was her way of engaging with some of the most representational paintings in Western art history as if they were abstraction: "to literally paint abstractly into them and try to confuse the two things."
A problem developed when she enlarged images and could see the remnants of the reproduction. Anyone who has looked closely at a vintage book or magazine will recognize Ben-Day dots, the small dots of cyan, magenta, yellow, and black used to reproduce color in ink on paper. "When you blow it up, you can see them, so that got me thinking about reproduction and digital media. And I started glitching those images."
During this exploratory time, Calvert had a conversation with a digital media artist who challenged her to try AI. At first, she was unsure if she knew enough coding to be able to play with AI, but the artist introduced her to beta software. "This was before DALL-E came out to the world," she says. When she realized she had already been manipulating images, only with a different method, she was inspired to create her own data set for the AI, "which was as many still life paintings as I could gather and ask it to try to learn how to make a new one." Those were her first steps into her current body of work. She has now been using that software for four years.
Many platforms claim that a user can input one to 15 images and the AI will learn, but what actually happens is that the AI has already been trained in an algorithm, and users are simply adding a style on top of it — not entirely unlike what Calvert had been doing manually. "When I was putting in over a thousand images, I had more influence over the data set." So the next logical step for her was to work with a computer engineer to develop her own machine-learning algorithm without anyone else's data in it, only hers, and to see what she could do.
A Uniquely 21st Century Approach to Painting
Describing both her process and her finished paintings, Calvert says, "As much as I can abstract them and mutate them, the general public will still recognize the still life-ness of them," Calvert says. "I've always wanted my work to draw someone in and then repulse them, that it be sort of beautiful and enticing and attractive and then its wrongness would come second. That's what painting has always done."
Artists have continually looked to paintings like Édouard Manet's "Le Déjeuner sur l'herbe," the picnicking painting with a female nude and a half-dressed female bather among two fully-dressed men. Generations of painters have returned to this painting to remake it, Calvert says, "recognizing the abstraction inherent in the original… then abstracting it further." Painters understand this idea, and then intentionally "mess with space, mess with color, mess with painting, mess with abstraction."
Calvert's still life paintings are "quite distorted, and kind of mutant or gothic," which she hopes will point out the ugly side of AI. "Obviously, the biological content is also in my work. These are organic forms that look mutated." Her use of AI to produce these abstractions is deliberate and it will be continual: "I want more people to try using AI because to have a critical eye on it, you need lots of artists' perspectives."
The Question Of Ethics
"I'm hugely concerned with the ethics of AI," Calvert says. "But not in the way you would think." Many artists, from comic book creators to editorial illustrators, are concerned that AI platforms like Leonardo.Ai and Midjourney scrape data from their work without any consideration for their copyrights.
Commercial AI platforms have been unapologetic about their use of the intellectual property of multitudinous creatives for their own data sets, which they claim are proprietary, and do not disclose. This practice is not only unethical but also illegal. "They design packets of millions of images and that's what they train their machine algorithm on, and then they don't reveal to the public where that imagery came from." But copyright issues do not bother Calvert much. "I like the idea of my images going back into the feed. I don't really mind that."
Her concern is that the general public might not completely understand what is in the data sets that companies are not sharing. If entire social groups and their histories are not included, the data sets that teach AI will begin engaging a new kind of natural selection. She says "the data sets overall skew overwhelmingly toward Western white culture. If you're going to use those data sets to, for example, solve medical problems, then you are skewed towards solving that medical problem for a white Western genetic makeup."
The possibility that AI might privilege the health and survival of the dominant culture, to the detriment to others, is a bigger concern to Calvert than intrusion on intellectual property rights because, she says, "I think there's much bigger implications.”
AI As An Unreliable Narrator
Calvert's husband Josh Azzarella is also a digital media artist. A recent recipient of the Artist Professional Development Grant from the Great Meadows Foundation, he works especially within the medium of photography. The introduction of AI into his image-making has already presented him with new complexities.
"Photography, since its inception, has often been perceived as a reliable index of reality," he says. "However … photography has always been subject to interpretation, bias, manipulation, and alteration." The same can be said about the rapid expansion of AI image generation. Azzarella wants to examine the ways AI might change the criteria of authenticity that have long been associated with photography.
His piece Untitled #310 was conceived during "a confluence of personal circumstances and response to societal unrest." Azzarella is immunocompromised, so physical participation in the Breonna Taylor protests in Louisville during the early stages of the COVID-19 pandemic was not possible for him. Nevertheless, he did want to contribute to the discourse around institutional anti-Blackness, police violence, and social justice. The result was his interpretation of gunshot detection technology, a response to "not only the events that unfolded in Louisville but also a resonant commentary on the pervasive gun violence in the U.S."
Unlike conventional security cameras that passively record their surroundings, Untitled #310 is coded to listen, interpret, and make decisions based on pre-specified auditory stimuli. The AI is trained to recognize the sound of a gunshot, and when it is identified, the AI communicates with a camera to capture an image of the sky directly above it — all within 0.25 seconds. "This positions the AI not as the unflinching eye of a surveillance camera," Azzarella says, "but as an active agent in the act of recording and indexing."
The AI was trained on a Google AudioSet of more than two million sounds across more than 500 acoustic categories. The AI runs on a computer processor Raspberry Pi 3B+ with Azzarella’s own custom audio processor and a single microphone, all powered by a rechargeable battery pack maintained by a second battery pack with a solar cell. The electronic components of his piece are housed in a weatherproof case.
While the images produced by Untitled #310 are poetic snapshots of the skies above violent scenes, they raise questions about how AI and technology can amplify human error. AI does more than create imagery—it also crafts narratives. "Because AI is driven by algorithmic selections," Azzarella says it is "poised to curate our personal memory, emphasizing certain events while potentially overshadowing others." This selective narration of the human social sphere could eventually alter "the way collective experiences are remembered and understood."
Art reflects the customs and values of the culture that creates it. Technology reflects not only our aesthetic preferences, but also our belief systems. The incorporation of emergent technology within traditions of visual art has the ability to reinforce existing power structures or dismantle them. If the artists who use AI understand this idea, and then — as Calvert says, "mess with it" — the possibilities are as constricting or as liberating as our prompts.