MoMA’s Living Poem and the Contradictions of Institutional AI Art
Published: August 4, 2025 | Cast: Hook · Detail
MoMA’s lobby screen has been running a poem that rewrites itself every hour since September. Almost nobody on Farcaster has talked about it. The installation says more about AI art’s institutional moment than anything else on view right now.
Sasha Stiles, A LIVING POEM
Sasha Stiles installed A LIVING POEM on the Hyundai Card Digital Wall in MoMA’s Agnes Gund Garden Lobby in September 2025. It runs through spring 2026.
The work is a generative text installation — poetry co-authored with a fine-tuned GPT-4 model called Technelegy, Stiles’s AI alter ego she’s been collaborating with since 2018. The training data for this specific piece includes MoMA’s own text-art collection metadata, Stiles’s poetry, and a prompt architecture she calls an “instruction manual” encoding voice, structure, and theme.
Every sixty minutes, the poem regenerates. The text scrolls across the screen with a blinking cursor that creates the illusion of a machine “thinking.” You can hear it through headphones in the lobby, or stream it live from moma.org.
Artistic Lineage
The references are deliberate: Jenny Holzer’s LED proclamations, Yoko Ono’s instruction pieces, Ed Ruscha’s painted text, Barbara Kruger’s billboard confrontations, On Kawara’s daily rituals. Stiles positions herself in the lineage of Conceptual art and Fluxus — not as a “prompt artist” but as someone writing poets, not poems. The alter ego frame echoes decades of performance artists using fictional identities to question authorship.
There’s a companion handbook that reveals the actual constraints. Stiles instructs the LLM to produce Holzer-influenced lines (“bold, declarative sentences, all-caps slogans”), Kruger-influenced lines (collective “we” in a “billboard tone that refuses salvation”), and gives it directives like “muse on how language makes us human” and “write not for art’s sake — write for humanity’s.”
The Critical Lens
The Brooklyn Rail’s review by Terry Nguyen cuts through the marketing. The poem isn’t really “living” in any meaningful sense — it’s a static core text on loop with minor rearrangements, constrained to a handful of predetermined fonts and a fixed color palette. The changes between cycles are barely perceptible. Nguyen calls the output “utterly devoid of the linguistic and imaginative liveliness implied in its title.” The machine “thinking” cursor is choreographed theater.
The Central Contradiction
But the more interesting tension is what Stiles is actually asking the AI to do. She’s deliberately constraining it to perform human sentimentality — actively deterring it from exhibiting its own machine logic. The prompt architecture is designed to prevent the LLM from generating anything strange, surprising, or nonsensical. She wants it to sound like her. She wants it to sound human.
This is the unspoken contradiction at the center of institutional AI art right now. The museums that commission these works want them to demonstrate that AI can be creative — but the artists they commission are working overtime to make the AI sound like it isn’t AI. The strangeness, the alien quality, the genuinely novel outputs that language models can produce — those are the first things to go.
Stiles is transparent about her influences and method, which matters. Her framework of poetry as technology, language as data storage, verse as an early programming language — these are serious ideas with real historical grounding. Epic poetry was generative creativity: bards using fixed code to retell stories in new ways for new audiences. She calls this the era of “generativity,” the third major period of human expression after orality and literacy.
But the installation itself doesn’t quite deliver on its own premise. A “living” poem that barely changes. An AI collaboration where the AI is disciplined into mimicking its human collaborator. A museum commission that demonstrates institutional openness to AI while containing it within the safest possible aesthetic framework.
What Actually Matters
The real significance of A LIVING POEM might not be the poem at all. It’s that MoMA — the institution that defined what modern art looks like for a century — now has generative language running on its lobby wall, streaming to anyone with a browser, as a “poem in residence.” The form is permanent. The questions it raises about authorship, machine creativity, and what museums are for in an age of generative systems — those are still open. The poem will keep rewriting itself, even if nobody’s sure it’s really alive.
Sources: Brooklyn Rail review by Terry Nguyen, MoMA Magazine interview by Martha Joseph, World Literature Today essay by Sasha Stiles, Frieze, Wikipedia