Refik Anadol's Dataland, billed as the world's first museum of AI arts, opened in downtown Los Angeles on June 20 and pulled more than 10,000 visitors in its first two weeks. The gallery, cofounded with studio partner Efsun Erkılıç, is anchored by Machine Dreams: Rainforest, an immersive installation that responds in real time to visitors' movement and biometric data through wearable sensors. It is the highest-profile attempt yet to argue that generative AI can produce something more substantive than eight-second video clips and prompt-engineered stock imagery.
The exhibit runs on Anadol's Large Nature Model, an AI system built over three years using natural science archives from institutions including the Smithsonian, whose Encyclopedia of Life alone contributed data on more than 2 million species. Anadol's team collected an additional 5 petabytes of raw material on trips to the Amazon and other rainforests. The gallery runs on Google Cloud, with Google DeepMind providing what Anadol describes as experimental low-energy compute — an unusual disclosure at a moment when data center power draw is a live political issue.
The provenance of the training data is the point. Silicon Valley's largest AI companies are defending copyright suits from writers, artists, publishers, and record labels over scraped material used without licensing. Dataland's pitch is the inverse: every source institution consented, participated, and is credited. Anadol has been working with Google since becoming the first recipient of the Google Artists and Machine Intelligence Artist Residency in 2016, and the sourcing model borrows more from natural-history museums than from web-scale crawlers.
“For three years, we started from scratch and trained our own AI models, and we worked with our own data sets.”— Refik Anadol, Dataland cofounder
Key facts
- 01Dataland opened in downtown Los Angeles on June 20, 2026, drawing more than 10,000 visitors in its first two weeks.
- 02The debut show runs on a proprietary Large Nature Model trained on 5 petabytes of raw data collected by Anadol's team.
- 03The Smithsonian's Encyclopedia of Life supplied training data on more than 2 million species.
- 04Google DeepMind provided experimental low-energy compute, with the gallery running on Google Cloud.
- 05The core exhibit, Machine Dreams: Rainforest, runs on a 40-minute cycle that responds to wearable biosensor data.
Inside the main gallery, visitors receive a smartwatch and a U-shaped shoulder collar that emits scents keyed to what is happening on the walls. A 40-minute cycle rotates through environments — during one segment, a simulated thunderstorm places a watery circle at each visitor's feet that streaks when they walk. Waving a hand redirects the fall of the rain. The shoulder harness releases the smell of a summer storm. The images and soundscapes shift based on where people stand and how fast they move.
A second room, the Infinity Room, follows a hummingbird through a neon forest in a first-person flight sequence. The Latent Gallery, adjacent, is closer to a research exhibit than an art installation: consoles let visitors scroll through the actual training data by category, so clicking Frogs turns a wall into a grid of every frog image in the model. It is a deliberate inversion of the black-box framing that dominates commercial AI product design, where the training set is the trade secret.
Anadol is explicit that most public skepticism of AI art is warranted. He told Wired the majority of critics are right when they associate the phrase with prompt engineering and generative slop. Dataland's argument is not that those critics are wrong — it is that the category is larger than the worst examples in it. Whether that argument holds up outside a $10-million-plus custom-built gallery is a separate question.
“The reason for Dataland is to understand and explain and explore—and tell the world that there is not only one option.”— Refik Anadol, Dataland cofounder
The final space, the Sanctuary, compiles biometric data from everyone in the room — heart rate, skin temperature, walking speed, path through the gallery — into a single abstract 3D visualization that is generated once and never repeated. Anadol says the museum discards visitor data on exit, though guests can retain a personal token of their own readings. In a sector where the default posture is maximum data retention, the deletion policy is a design statement.
The counterweight is scale. Dataland is one gallery with one show, drawing an audience measured in the thousands per week, running on a bespoke model that took three years and 5 petabytes of custom data to build. It is not a template that transfers to a smartphone app or a stock-image platform. The commercial generative-AI products that have shaped public opinion about AI art operate under entirely different unit economics, training regimes, and licensing assumptions. Whether Dataland changes minds about those products, rather than about Anadol's, remains open.
Still, the exhibit lands a point that the debate over training data and model licensing has largely avoided: what AI-generated media looks like when the underlying data is treated as a curated, credited, and finite resource rather than an unbounded scrape. The AI art conversation for the past three years has been dominated by lawsuits and platform policies. Dataland is one of the first credible attempts to argue the case on aesthetic and institutional grounds, which is the terrain the technology's defenders will eventually need to win on if the medium is going to move past its current reputation.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
