In the Agnes Gund Garden Lobby at New York’s Museum of Modern Art (MoMA), a vibrant 25-foot screen displays a mesmerizing flow of words, constantly changing. Every minute, new text emerges—sometimes mundane, like “Just Keep Breathing,” and other times strikingly poetic:
Trace this poem
in mud
so rain dissolves it.
Carve it in wind on stone,
a script no hand can hold, not tightly.
This particular piece, “A Living Poem,” is also showing concurrently in South Korea and will run at MoMA until spring. It’s generated by an artificial intelligence program, trained on artist Sasha Stiles’ writing and data from MoMA’s extensive collection. The line “Keep breathing”, for instance, nods to Yoko Ono’s 1965 instruction-based artwork, “1st Day Breathe.” The captivating aspect is that the text is infinite, never repeating itself exactly.
Stiles explains the technology: “Because each 60-minute performance is generated live by code, what viewers experience is not a fixed recording, not a video loop, but a living manuscript that’s always in motion, always becoming. Improvisation happens here at algorithmic scale, so no two tellings are ever the same.”
With the rapid advancement of artificial intelligence, machines are playing an ever-larger role in the creative process. This development raises a fundamental question for the art world: What truly constitutes art when the human hand is no longer directly involved? This can be particularly disorienting for audiences, given the widespread integration of AI-generated text and imagery throughout modern culture.
In response, leading art museums are actively grappling with what, within this expansive new landscape, can be genuinely classified as art. They are committed to helping the public look beyond mere technological spectacle. Through their efforts, these institutions are beginning to forge a new canon, offering artists, collectors, and audiences a framework for understanding what matters in AI art and why.
Curators often argue that the value of AI art lies less in its surface aesthetics—some pieces, to be frank, are quite ugly—and more in the artist’s intent and the underlying process.
Paola Antonelli, a senior curator at MoMA and director of the museum’s research and development, emphasizes this distinction: “The technology is not the art. In the end, it’s about the person’s creativity and vision.”
Melanie Lenz, curator of digital art at London’s Victoria and Albert Museum, echoes this sentiment, noting that “Audiences can quickly tire of works that rely solely on technological novelty.”
This inherent tension between novelty and genuine artistic substance is a core challenge for curators. Christiane Paul, curator of digital art at the Whitney Museum of American Art in New York, points out, “There seems to be a general assumption that the term A.I. art simply means visuals created by means of a text prompt, which ignores artists’ deep engagement with the medium.” In reality, AI art is a vast and diverse field.
Even readily available tools can demand profound conceptual rigor, as Paul highlights with the work of Bennett Miller, who utilizes publicly accessible AI image generators. However, curators are often most drawn to artists who delve into the very computer code of AI systems. Many acclaimed AI artists today dedicate months or even years to meticulously collecting and curating data for custom AI models, which they actively participate in building themselves.
Artists have a long history of experimenting with computers, from Vera Molnar’s IBM mainframe drawings in 1968 to Harold Cohen’s painting system AARON in the 1970s. AI-generated art truly entered public consciousness in 2018 with the $432,500 sale of “Portrait of Edmond de Belamy” by the Parisian art collective Obvious at a Christie’s auction. Since then, artists have increasingly begun to develop their own art-generating AI systems.
Paul from the Whitney Museum asserts that the crucial factors are the sophistication of both the medium’s application and the underlying concept. The curator’s essential role is to illuminate for the public where the true artistry resides within these complex creations.
William Latham, an artist who organized the influential Creative Machine Exhibition a decade ago in London, suggests that the optimal dynamic occurs when the machine acts as a creative partner, akin to “Picasso with his assistants.” Latham and his collaborator, Stephen Todd, are currently working with Google DeepMind on an evolutionary system. This system generates images from text prompts, producing increasingly elaborate variations based on mathematical formulas. While humans can choose which variants to further develop, they also have the option to let the machine make those decisions autonomously.
“It’s sort of like survival of the aesthetic,” Latham remarked in a telephone interview. In their latest iteration, “Evolution and Foundation,” currently on display at the Oxo Gallery in London until October 26, the machine is entirely responsible for all artistic choices.
This evolution inevitably prompts significant questions regarding artistic autonomy and the extent of human involvement.
“A world in which machines write our music and make our art is a world we could lean into if we wanted to,” states Maya Ackerman, a professor, AI musician, and author of “Creative Machines: A.I., Art and Us.” She adds, “But I really think it’s probably a net loss for humanity to disengage the arts to such a complete degree.”
However, for many artists working in this innovative medium, the very act of creation is embedded in writing the code itself.
“I don’t really believe in the artist’s hand,” says Casey Reas, a pioneering artist and computer programmer in the field. “I believe in the artist’s point of view.”
Refik Anadol, a Turkish-born artist who contributed to the design of MoMA’s impressive LED screen showcasing Stiles’s work, has embarked on an ambitious project: to establish an AI art museum in Los Angeles from the ground up. This museum will be equipped with the specialized infrastructure needed to properly exhibit AI-generated art, which often requires significant high-end computing power and electricity, leading many artists to collaborate closely with technology firms.
Anadol’s museum, named Dataland, plans to feature its own bespoke AI model for artists to utilize. Developed with support from Google and Nvidia, this model is trained on half a billion images, all collected with the permission of their owners. Dataland, set to open at The Grand LA—a Frank Gehry-designed complex in Los Angeles—aims to address the fact that traditional museums currently lack the resources to fully showcase the vast and growing spectrum of AI art.
“Without museums, the art forms are lonely,” Anadol reflects. “To me, the museums are the campfire for humanity.”
Reas believes the fundamental question facing AI and art is whether this new work should be integrated into existing artistic canons or whether entirely new institutions and canons should be built around it. “I personally believe both should be happening,” he states. He has put this philosophy into practice with Feral File, an online platform dedicated exclusively to digital art.
Despite this growing momentum, not all museums are ready to fully embrace the AI art form; many are adopting a wait-and-see approach.
“We didn’t show any NFTs when they first emerged, and that seems to have kind of fizzled out,” explains Marshall N. Price, chief curator at the Nasher Museum of Art at Duke University. “We’re taking a similar tack with A.I.-generated art.” His team notably experimented with having ChatGPT curate an exhibition a few years ago, with admittedly subpar results.
Yet, while some museums remain hesitant, others are boldly forging ahead. Valentina Ravaglia, curator of London’s Tate Modern’s recently closed Electric Dreams exhibition, asserts, “The role of A.I.s in generating images is a cultural milestone that cannot be ignored.”
Further showcasing this emerging field are exhibitions like “Subject to Change” at Gazelli Art House in London, and “The Age of Creative Intelligence” at the Goodman Center Lobby at the Fashion Institute of Technology in New York City.
The current discourse surrounding AI-generated art draws parallels to the early 20th century. At the New York Armory Show in 1913, American audiences were famously unsettled by cubism, particularly Marcel Duchamp’s provocative “Nude Descending a Staircase.” Today’s AI creations evoke a similar sense of unease, prompting audiences to ponder whether new technological tools expand or diminish the very essence of art. Of course, Duchamp and the cubists are now firmly established as canonical figures in art history.
“Whenever there’s a new technology, there’s that moment of drunkenness,” says MoMA’s Antonelli, describing the initial fascination. “Then you sober up, you establish base lines.”
Ultimately, she suggests, audiences will cultivate their own discerning critical judgment and, in an organic process, lend their support to the institutions that truly resonate and matter most.