Dataland, the first museum of AI works

Are you wondering if the cover image is taken from the museum's official presentation catalog? The answer is no. It was generated with a free AI, in a few seconds, from scratch. Keep that in mind as you read. On June 20, Los Angeles inaugurates the world's first museum space dedicated entirely to generative art. Amidst ethical datasets, Amazonian rainforests reconstructed in pixels, and a question that no one can yet close: is all this art? There is a criticism that has been circulating for months in digital artists' studios and in the loudest internet threads, and which could be condensed as follows: "You are building a museum on instructions that people give to AI and you are calling it art." It is a sharp phrase. And on June 20, 2026, it will find its most concrete answer, or perhaps its most expensive provocation, in the opening of Dataland, the world's first museum entirely dedicated to artificial intelligence generative art.

The location is one that, in terms of architectural symbolism, could not be more loaded: The Grand LA, the complex designed by Frank Gehry in the heart of Los Angeles' Grand Avenue Cultural District, just steps from the Walt Disney Concert Hall, the Broad Museum, and MOCA. A neighborhood where every building is already an aesthetic manifesto even before opening its doors. Dataland, co-founded by Turkish-American artist Refik Anadol and his partner and associate Efsun Erkılıç, occupies about 25,000 square meters, nearly a third of which—no minor detail—is dedicated to the hardware necessary to run the whole thing.

The inaugural exhibition is called Machine Dreams: Rainforest and works like this: an artificial intelligence model trained on ecological data collected in sixteen of the planet's tropical forests generates, in real-time, visions of nature that do not exist, sounds of species that perhaps no longer exist, scents synthesized by an algorithm. The Infinity Room, one of five galleries, has been described as a place where you can listen to 1987 recordings of a now-extinct Hawaiian bird while breathing in a forest smell that no human nose has ever sniffed in that precise combination. As in certain chapters of Jeff VanderMeer's Annihilation, that feeling of being inside an ecosystem that responds to its own rules, not entirely hostile, not entirely understandable.

The man who paints with data

Refik Anadol, born in Istanbul in 1985 and moved to Los Angeles in 2012 to study media design at UCLA, did not arrive at this point by chance. He has built a career around a precise idea: data is not just information, it is plastic material, capable of becoming shape, color, movement. What he calls data painting is an approach where huge digital archives, museum collections, climate recordings, and urban memories are processed by machine learning systems to produce visual installations on an architectural scale.

The moment of institutional consecration arrived in 2022, when New York's MoMA hosted Unsupervised, an installation in the museum's foyer that reworked two hundred years of the collection into continuous generative streams. It was the first time the temple of New York modern art so explicitly embraced a work built entirely around artificial intelligence. In 2025, TIME included him in the 100 AI Impact Awards. His installations have spanned seventy cities, from London's Serpentine Gallery to the Sphere in Las Vegas, by way of the Guggenheim in Bilbao.

With Dataland, Anadol and Erkılıç are not opening another temporary exhibition. They are declaring that AI art deserves a permanent home, an institution that treats it not as a laboratory experiment or a tech fair curiosity, but as a medium with the same status as sculpture or photography. "Dataland is not just a space to show finished art," Anadol declared. "It is a living institution dedicated to research, education, and ethical experimentation."

The perfect machine (or almost)

The project has a declared ethical architecture with a care that, at times, resembles the technical specifications of a Silicon Valley product more than a museum press release. The engine behind everything is the Large Nature Model (LNM), described as the first open-source artificial intelligence model trained exclusively on natural data. The datasets come from partnerships with the Smithsonian, the Cornell Lab of Ornithology, and the Natural History Museum in London, collected, it is specified, with verifiable ethical protocols. No wild scraping from the internet, no data taken without consent.

On the energy front, Artnet reports that the Large Nature Model runs on a cloud infrastructure powered 87% by carbon-free sources in Oregon, and that the energy impact of a museum visit is equivalent to charging a smartphone. A brilliant communication point, the kind that serves to disarm in advance the objections of those who, legitimately, point out that training a large language model consumes as much as hundreds of transcontinental flights.

The corporate structure starts as a commercial entity, with a possible future transition toward non-profit. Memberships start at $350 a year. The complex where it is located also houses luxury apartments and a five-star hotel. All this is known, all this is public. And it is precisely here that the interesting part of the analysis begins.

"You are calling it art"

Thomas Brummett is a digital artist with works in the collections of the Museum of Fine Arts in Houston, the Philadelphia Museum of Art, and the Museu de Arte Moderna in Rio de Janeiro. When Dataland announced its opening, he wrote on Instagram, as reported by NPR: "We build a museum based on instructions given to AI and call it art. It isn't and never will be. At best, it's second-rate entertainment."

Brummett is not a Luddite. He uses digital techniques in his work. But his objection touches an old nerve: the question of authorship. What, exactly, does the artist do in a generative process? They select a dataset, set parameters, choose what to show and what to discard. They are a curator, programmer, director, but are they an author in the sense that Rembrandt was the author of a canvas, or Coltrane was of a solo? The question has no consensual answer, and that is probably how it should be.

We have already encountered this tension on these pages, talking about AI-generated music with the case of The Velvet Sundown, and about Tilly Norwood, the first completely synthetic actress to interest Hollywood agencies. In all these cases, the point of friction is the same: when the machine does the "visible" work—the melody, the face, the image—where is the human intention that transforms technique into expression?

The historical comparison always evoked in these debates is photography, dismissed for decades as "mechanical" and therefore non-art, before Cartier-Bresson and Diane Arbus demonstrated that the medium is irrelevant compared to the intention. But there is a structural difference: the photographer chooses the moment, the light, the composition; each shot is unrepeatable. A generative work is, by definition, replicable, changeable, evolutionary. Who owns the "authentic" version of Machine Dreams: Rainforest running on June 20, 2026, at 3:00 PM? And the one at 3:01 PM?

The tension the ticket doesn't cover

There is another crack in Dataland's ethical architecture, more subtle but no less relevant. The debate on authorship in creative AI that we explored previously raised a precise question: who benefits, and who pays the price, when an automated system occupies a space that previously belonged to human labor?

Dataland is born as a for-profit entity within a luxury real estate complex. The $350 annual membership already defines a visitor profile. Meanwhile, hundreds of artists working with AI, without Anadol's pedigree, without the Smithsonian partnership, without the ability to build a proprietary Large Nature Model, continue to operate in a market where, according to an Artsy survey, only 9% of the world's galleries consider AI art a fully legitimate medium. Dataland could shift this number. But it could also crystallize a hierarchy: the "certified" AI art of those who can afford the institutional operation, and everything else.

The promise of data transparency is real and verifiable, at least in part; the Large Nature Model is open-source, and the datasets are documented. But the question of who verifies remains open. Institutional partners (Smithsonian, Cornell, Natural History Museum) lend credibility, not ongoing audit capability. And the intellectual property of a work that evolves in real-time, generating ever-new configurations, is a legal territory on which global copyright law does not yet have stable coordinates. As we saw when analyzing creativity in the age of generative AI, the gap between what technology makes possible and what the regulatory framework manages to govern continues to widen.

The AI that doesn't enter through the door

Here we arrive at the most interesting problem, the one that no $350 membership manages to buy: the gap between the AI that Dataland shows and the AI that acts in the world.

Inside the museum, there will be a decorous AI. Ethical data, carefully processed Amazonian forests, non-invasive synthetic scents—all curated, all beautiful, all "with certified curation," to use a formula circulating in digital art criticism. An AI that reassures, showing how technology can be beautiful and controlled and even moving in its attempt to recreate the voice of an extinct bird.

Outside the door, in the same week of June 2026 in which Dataland opens, another AI exists. The one the World Economic Forum has defined as the protagonist of the "first election cycle where deepfakes exceed the threshold beyond which they are no longer distinguishable from the real." The one that generates videos of dead public figures, with the consequences Zelda Williams publicly described concerning her father Robin, spread on platforms that struggle to keep up with production speed. The one being used in courtrooms, election campaigns, and health misinformation.

This AI will not enter Dataland. Not because someone deliberately prevents it, but because a museum by its nature selects, frames, values, and what it values inevitably defines by contrast what it excludes. The problem is not that Dataland shows beautiful things. The problem is that "beautiful" and "ethical" and "sustainable" risk becoming an alibi for not dealing with the same technology in its less presentable manifestations.

Barry Threw, artistic director of San Francisco's Gray Area Foundation, an institution that has worked at the intersection of art and technology for years without the luxurious frame of a Gehry complex, told NPR that Dataland is interesting because it makes "complex data into experience." It is an honest summary. But it also makes complex the question of what remains outside the experience.

What remains open

A museum is always an argument, even before it is a building. Dataland argues that AI art has reached institutional maturity, that it can stand beside the Broad and MOCA without feeling like a poor relation of contemporaneity, that generative technology can be ethically grounded, ecologically responsible, and culturally relevant.

It is a solid argument, built with care and considerable resources. But solid arguments deserve solid questions. Will Dataland's "certified" AI art help or obscure the debate on uncertified art? Is an open-source model truly transparent when the expertise necessary to verify it is accessible to few? Does the fact that a visit consumes the energy of a phone charge change anything regarding the broader issue of AI's computational cost on a global scale? And above all: is a museum that starts as for-profit in a luxury complex, and sets its educational mission as a future goal, building culture or building a market?

Refik Anadol has said that Dataland will be "a living institution." Living institutions change, adapt, sometimes betray initial premises, and sometimes surpass them. June 20 is a beginning, not an answer. And perhaps that is exactly the point: the best museums do not resolve questions. They make them impossible to ignore.