When asked about his upcoming exhibition, History Painting: Jason Salavon, the artist doesn’t launch into technical jargon about neural networks or machine learning models. Instead, Salavon muses, “What if we just put the model on acid?”
This is a refreshingly direct way to -explain how Salavon is pushing artificial -intelligence beyond its usual boundaries.
“I think one of the issues, aesthetically and artistically, is people really get used to what these models can make, and they become kind of boring,” he continues. “There’s a blandness and a generic quality to the outputs of AI.”
Salavon’s solution? Break the system—hack into open-source image models and make them produce works that break free of conventional systems.
“We kind of selectively modify the weights and biases of a model to make it do weird stuff,” Salavon explains. In doing so, he has created hundreds of what he calls “presets” that force the AI to trip out in different ways. The resulting exhibition features four digitally dense and massive panels: “Origins,” “Emergence,” “Sapiens” and “Modernity,” in which Salavon attempts to tell the story of the universe. The panels are accompanied by 800 looping animations made from the same prompts.
“It’s sort of a macro piece of encyclopedic storytelling, but with a kind of subjectivity, and lots of factual errors,” he notes. “I’m having a lot of fun making this show. Even though it’s a bit of a struggle, it’s a productive struggle.”
The pieces make use of what Salavon calls “Salavon’s pathology-inducing machine”—a system that shocks standard AI models into producing unexpected results, complete with a user interface he plans to make publicly available at some point.
“The work has a real ambivalent sort of posture toward mainstream AI,” he explains, hinting at an embedded critique of the field.
History Painting represents a significant shift from Salavon’s previous work. While proud of earlier projects like Totem (a 100,000-pixel GIF employing stock images), he wanted to move beyond what he says felt like an “advanced collage.” His new approach allows for more distinctive imagery while maintaining the conceptual depth that marks his practice.
“I’m able to feel personally more of a sense of authorship of the content,” he tells SFR, “of the images being used themselves.”
Salavon’s background straddles both artistic and technical worlds. The son of a painter, he grew up surrounded by fine art and art history books. While naturally drawn to drawing and painting, he was equally -interested in math and science. By 1991, as an undergraduate at the University of Texas, he had already decided to make art with computers, specifically by writing software to create what he then called “little weird machines that make art autonomously.”
Now, as an associate professor in the Department of Visual Arts at the University of Chicago, Salavon teaches a class in Art Machine Intelligence. He has observed how new AI tools have transformed student work over the past couple of years.
“The floor for what people can make has really raised,” he notes. “I’m having kids make 50-page graphic novels combining these tools with like 20 or 30 hours of labor.”
But such accessibility comes with concerns. There’s what Salavon calls, “a natural tendency to lean into these tools and to not think about how they could be combined with something else.”
“Your creative pathways are kind of greased by these AIs,” he adds.
Thus, as an educator, Salavon pushes students to complicate their various approaches.
“Don’t just ask it to make you an image and then paint that image,” he says. “If you want to use AI for painting, why don’t we think about how to use it to really fuck that up?”
The challenge, then, is maintaining artistic depth while engaging with new technologies.
“I’m like a conceptual artist who truly cares what things look like,” Salavon says, “but I’m interested in objects that can stand by themselves.”
He aims to create what he calls “hyperobjects,” or works that operate simultaneously on multiple levels and engage with both contemporary technologies and art histories—while delivering compelling aesthetic experiences.
“I want objects that can do a lot,” he explains. “Pieter Bruegel was making hyperobjects in the 16th century. I think that fine artists have let other forms like video games take the mantle of these scopes, but pictures and shorter videos can do a kind of work in that space that is different.”
Looking ahead, Salavon sees both challenges and opportunities. While current AI tools lack precise control and struggle with time-based imagery, he expects significant advances are coming.
“We’ll look back at this time as a pretty primitive moment,” he predicts.
History Painting: Jason Salavon Opening: 5 pm Friday, Nov. 29. Free
Jason Salavon Artist Talk: 2 pm Saturday, Nov. 30. Free
TAI Modern, 1601 Paseo de Peralta, (505) 984-1387
