Unsupervised — Machine Hallucinations

Art That Dreams About Art

In November 2022, the Museum of Modern Art unveiled Unsupervised, a large-scale installation by the Turkish-American artist Refik Anadol that occupied the museum’s ground-floor lobby gallery. A massive LED screen — roughly 24 feet wide — displayed constantly shifting, fluid forms that resembled nothing so much as the museum’s own collection dissolving and reforming in real time. Colors pooled and separated. Shapes that hinted at recognizable artworks emerged and dissolved before they could be fully grasped. The effect was hypnotic, beautiful, and deeply strange.

The piece was trained on metadata from over 138,000 artworks in MoMA’s collection — data about the images, their compositions, colors, textures, and contextual information spanning more than a century of modern art. Anadol’s custom machine learning models processed this data and generated an endlessly evolving visual output that the artist describes as the machine’s “hallucination” or “dream” of what art looks like.

The Artist Behind the Machine

What distinguishes Anadol from many AI artists is the depth of his involvement at every technical and aesthetic layer. He did not use Midjourney or Stable Diffusion. He built custom machine learning architectures with his studio team. He designed the physical installation, including the screen specifications, the room acoustics, and the environmental sensors that allowed the piece to respond to the movement and density of viewers in the gallery.

Anadol’s background spans architecture, media arts, and computer science. He holds an MFA from UCLA’s Design Media Arts program and has been working with data-driven art since long before generative AI became a mainstream phenomenon. His practice is rooted in a specific artistic question: what does it look like when a machine processes vast amounts of human cultural data and produces its own interpretation?

This is not a prompt typed into a text box. It is a multi-year artistic research program with custom-built tools, and the sophistication of the technical and creative infrastructure is evident in the work.

The MoMA Context

The significance of Unsupervised is inseparable from its location. MoMA is not just any museum — it is the institution that has defined what counts as modern art for nearly a century. Its collection includes Picasso’s Les Demoiselles d’Avignon, Warhol’s Campbell’s Soup Cans, and Starry Night. To place an AI-generated installation in the lobby gallery — the first thing visitors see when they enter — was a statement of institutional endorsement that reverberated through the art world.

MoMA’s curatorial team was deliberate about the placement. The lobby gallery is a high-visibility, high-foot-traffic space. The choice to give it to an AI artwork signaled that the museum views AI-generated art not as a curiosity or a sideline but as a significant contemporary practice worthy of the institution’s most prominent platform.

The exhibition ran for over a year, far exceeding the typical duration of a lobby installation, suggesting that the museum viewed it as both a critical and popular success.

The Debate

Not everyone was convinced. Jerry Saltz, the Pulitzer Prize-winning art critic for New York Magazine, was among the most prominent skeptics. He described the work as visually impressive but intellectually vacant — “a lava lamp” for the Instagram age. Other critics questioned whether the work’s beauty was merely surface-level, lacking the conceptual depth that the art world demands of work displayed in its most prestigious institutions.

Defenders countered that Unsupervised achieves something genuinely unprecedented: it creates a visual conversation between 138,000 human artworks, mediated and reinterpreted by machine intelligence. The result is not just pretty — it is a new form of art criticism, a visual essay about what modern art looks like when seen through non-human eyes.

The debate itself became part of the work’s significance. By generating passionate disagreement among serious critics, Unsupervised demonstrated that AI art had moved beyond the novelty phase and into the territory of genuine aesthetic and intellectual engagement.

Technical Innovation as Artistic Practice

Anadol’s approach challenges the assumption that AI art is inherently lazy or derivative. His studio employs engineers, data scientists, and designers alongside traditional artists. The development of Unsupervised involved:

• Custom data pipelines to process and structure MoMA’s collection metadata into formats suitable for machine learning training
• Novel neural network architectures designed to produce specific types of visual output — fluid, organic, and continuously evolving rather than static or photo-realistic
• Real-time rendering systems capable of generating unique visual content indefinitely without repetition
• Environmental sensors that allowed the installation to respond to the presence and movement of viewers, making each experience subtly different

This is closer to the practice of an architect or a filmmaker than to the typical prompt-based AI art workflow. Anadol is not using a tool — he is building tools, and the act of building is itself part of the artistic practice.

Legacy and Influence

Unsupervised has become a reference point in virtually every serious discussion about AI art’s place in the institutional art world. It demonstrated that AI-generated work could command major museum space, attract large audiences, and generate substantive critical discourse.

More importantly, it offered a model for how AI art can move beyond the limitations of prompt-based generation. Anadol’s work suggests that the most artistically significant AI art may come not from users of existing tools but from artists who build new ones — who treat the technology itself as a creative medium to be shaped, customized, and directed toward genuinely novel aesthetic experiences.

The piece has also influenced how museums think about AI art more broadly. Since Unsupervised, major institutions including the Serpentine Gallery in London, the Centre Pompidou in Paris, and the Smithsonian in Washington have expanded their engagement with AI-generated and AI-assisted art, citing Anadol’s work as a precedent.