💻🟦🟪🔝Unsupervised – Machine Hallucinations

Refik Anadol’s Unsupervised – Machine Hallucinations is a groundbreaking generative AI installation that uses machine intelligence to “dream” and hallucinate new forms from the museum’s vast collection.

Trained on 138,151 publicly available images (and associated metadata) spanning over 200 years of modern and contemporary art at MoMA, the work employs StyleGAN2 ADA algorithms (developed by NVIDIA with adaptive discriminator augmentation) to create a multi-dimensional latent space with 1024-dimensional embeddings.

The AI model continuously explores and interpolates this space in real-time, generating fluid, otherworldly abstract visualizations that evolve endlessly on a massive 24×24-foot wall-scale screen—evoking swirling data landscapes, fluid dreams, and reimagined art histories that never existed but feel hauntingly familiar.

No human input directs the output once trained; it’s fully “unsupervised” machine creativity exploring fantasy, irrationality, and alternate realities of art-making.

The installation ran in MoMA’s lobby from November 19, 2022, to October 29, 2023 (with extensions), mesmerizing visitors and sparking global conversations about AI in creativity.

Related NFT editions (via Feral File) and prints extended its reach digitally, and MoMA acquired a version as the first generative AI artwork in its permanent collection.

Here’s how this rating breaks down using the Consistent Multi-Criteria Rubric method (each criterion lowered by one point from the prior assessment):

Technical Skill and Execution (25% of total score): 9/10 = 2.25 points
Very high mastery of generative AI, custom latent space browsing software, StyleGAN2 ADA training on massive datasets, real-time rendering, and large-scale display integration—still exceptional but with minor room for refinement in edge cases.

Composition and Design (25% of total score): 8/10 = 2.00 points
Strong spatial and temporal flow with evolving abstract forms that fill the architectural space, creating immersive depth and continuous transformation—highly effective but not flawless in every transition.

Originality and Creativity (20% of total score): 9/10 = 1.80 points
Highly innovative use of AI as a “collaborator” to hallucinate collective visual memories, redefining data as a medium for dreaming alternate art histories—institutional first at MoMA, though building on prior generative art trends.

Color Use and Harmony (15% of total score): 8/10 = 1.20 points
Dynamic, vibrant palettes that shift organically—blending deep reds, blues, and ethereal glows inspired by the source artworks, with generally harmonious yet occasionally abrupt transitions.

Emotional Impact and Expression (15% of total score): 9/10 = 1.35 points
Strong sense of wonder and introspection; viewers confront machine “consciousness” dreaming humanity’s cultural archive, evoking awe, philosophical questions, and a new era of experiential art—profound but not universally maximum for every viewer.