Refik Anadol — Data as Pigment, Light as Material
Refik Anadol turns collective memory into sensory experience. Using machine learning, custom algorithms, and Nvidia supercomputers, his studio transforms billions of data points into living, breathing architecture-scale installations that challenge what art can be.
From Istanbul to the Edge of AI Art
Born in Istanbul in 1985, Refik Anadol taught himself to program on a Commodore 64 at age eight. Watching Blade Runner around the same time planted a question that would drive his entire career: what can a machine do with someone else's memories?
He studied photography and visual communication at Istanbul Bilgi University before earning an MFA in Design Media Arts at UCLA in 2014, where he was mentored by generative art pioneer Casey Reas. In 2014, Anadol and Efsun Erkılıç co-founded Refik Anadol Studio (RAS) in a 1960s warehouse in the Frogtown neighbourhood of Los Angeles.
"I use data as my pigment and light as my material. The machine is my collaborator, not my tool."
Refik Anadol Studio
RAS is not a conventional artist's studio — it is a multidisciplinary collective that brings together artists, architects, data scientists, researchers, neuroscientists, philosophers, biologists, medical doctors, composers, and environmental scientists. This cross-disciplinary model is central to Anadol's practice: no single domain owns the work.
The studio's process typically begins with a massive publicly available dataset — photographic archives, historical documents, audio recordings, scientific data — and ends with an immersive environment that gives audiences a physical, embodied experience of that data as a living material.
The Technology Stack
Anadol's work is distinguished by the scale and rigour of its underlying computation. Key components include:
- Generative Adversarial Networks (GANs) — particularly StyleGAN, used to train on large image datasets and generate continuous, ever-changing visual outputs that the studio describes as "machine dreams."
- UMAP (Uniform Manifold Approximation and Projection) — a dimensionality reduction algorithm used to map high-dimensional data (millions of records) into navigable latent spaces that can be explored in real time.
- Diffusion models and custom ML algorithms — later works incorporate diffusion-based generation and bespoke neural architectures designed around the specific dataset and artistic intent of each project.
- Nvidia A100 supercomputers — used for large-scale real-time rendering. Machine Hallucinations: Sphere on the Las Vegas Sphere relied directly on A100 clusters to drive 580,000 sq ft of LED surface.
- Google AI Quantum — Quantum Memories (2020) used Google's quantum computing software in combination with a supercomputer to 3D map a parallel world from a 300-million-image nature dataset.
- 3D projection mapping — the studio builds precise 3D models of each architectural surface (Anadol worked from Frank Gehry's original digital files for Walt Disney Concert Hall) so that projections align with the building's exact geometry.
- Open-source algorithms and public datasets — Anadol is intentional about using publicly available data and open-source tools where possible, framing it as a democratic, collective act.
The workflow often culminates in real-time, generative output — the installation is never the same twice.
Selected Works
Archive Dreaming (2017 — SALT Research, Istanbul)
Inspired by Borges' Library of Babel, Archive Dreaming ingested 1.7 million documents — 40,000 publications covering Turkish contemporary and modern art, architecture, and economics from 1997 to 2010 — and trained an AI to discover hidden correlations and unexpected connections between them. When idle, the installation "dreamed" new associations autonomously; visitors could also navigate the latent space using a joystick. The piece was one of the first large-scale public demonstrations of AI as an archival and creative agent.
WDCH Dreams (2018 — Walt Disney Concert Hall, Los Angeles)
Commissioned for the Los Angeles Philharmonic's 100th anniversary, WDCH Dreams is one of the most technically ambitious projection mapping installations ever realised. Using 42 large-format projectors, 50K visual resolution, 8-channel surround sound, and 44.5 terabytes of archival data — 587,763 images, 1,880 videos, 1,483 metadata files, and 17,773 audio files — the studio painted the stainless-steel curves of Gehry's concert hall with the orchestra's own collective memory. The New York Times described the result as "a sort of combinatorial Fantasia."
Because Anadol had access to Gehry's original 3D architectural model, the projection could respond precisely to every surface fold — a technical prerequisite that distinguished the work from standard projection mapping.
Machine Hallucinations: NYC (2019 — Artechouse, New York)
The first major work in Anadol's Machine Hallucinations series drew from 300 million publicly available photographs of New York City plus 113 million additional data points — subway sounds, radio snippets, traffic noise. A StyleGAN algorithm processed the visual data while a recurrent neural network absorbed the audio, generating a continuously evolving city that existed nowhere on earth. Art in America wrote that the work mimicked the quality of human memory "almost perfectly."
Machine Hallucinations: Unsupervised (2022–2023 — MoMA, New York)
Arguably Anadol's most culturally significant work to date. A 24×24-foot LED data sculpture installed in MoMA's main atrium, Unsupervised was trained on 138,151 freely available records from MoMA's permanent collection — uploaded to GitHub in 2016 — containing title, maker, medium, dimensions, and acquisition date, but no categories. The absence of human categorisation was a deliberate artistic choice: without labels, the AI found its own form. Visitors spent an average of 38 minutes with it. Unsupervised opened in November 2022, was extended four times, ran for almost a year, and was acquired by MoMA in October 2023 — the first generative artwork to enter the museum's permanent collection.
Machine Hallucinations: Sphere (2023 — The Sphere, Las Vegas)
Commissioned for the opening of the Sphere — a 366-foot-tall globe with a 580,000 sq ft programmable LED exterior — Anadol created the largest AI artwork in the world at the time. Two versions ran in succession: the first drew on 1.1 million public images from the International Space Station, Hubble, and other NASA spacecraft; the second used 300 million images from national parks combined with real-time wind sensor data from Las Vegas. Nvidia A100 supercomputers drove the real-time AI animations. Anadol described the premiere as "one of the most Blade Runner moments ever."
Living Architecture: Casa Batlló (2022 — Barcelona)
After his artist-in-residence year at the UNESCO World Heritage site, Anadol used the facade of Gaudí's Casa Batlló as a canvas for a 360-degree projection trained on approximately one billion images: Gaudí's sketches, the building's visual archive, academic materials, and publicly available photographs — combined with real-time climate data from sensors placed around the building. The live premiere drew 65,000 spectators. The work sold at Christie's for $1.38 million.
Echoes of the Earth: Living Archive (2024 — Serpentine Galleries, London)
Introduced at the World Economic Forum in Davos, the Large Nature Model (LNM) is a generative AI trained exclusively on nature: images, sounds, and academic research from institutions including the Natural History Museum London and the Smithsonian Institution. An early output, Living Archive, drew from 135 million publicly available coral images combined with openly available marine data. Shown at the Serpentine Galleries in January 2024, the work attracted 66,000 visitors in 47 days and later travelled to Futura Seoul for Anadol's first Asian solo exhibition.
Philosophy
Anadol's conceptual framework rests on a small number of core ideas, stated with remarkable consistency across a decade of interviews:
- Data as pigment, light as material. Dating back to his first outdoor projection in Istanbul in 2008, Anadol reframed data — previously invisible — as a sensory substance that can be shaped and rendered visible.
- Machine dreaming and hallucination. The moment he encountered DeepDream at Google's AMI residency in 2016 convinced him that if a machine could learn, it could also remember, dream, and hallucinate. His work since has deliberately explored those states.
- Collective memory. Rather than personal memory, Anadol is drawn to the memory of cities, institutions, ecosystems, and civilisations — data that no individual could perceive but that a machine can hold and re-present whole.
- Open data, public art. Most of his datasets are publicly available. He is consistent that AI art made on proprietary, inaccessible data is a different ethical proposition to work made on open commons.
"Since that moment — watching Blade Runner — one of my inspirations has been that question: 'What can a machine do with someone else's memories?'"
TED Talks
Anadol has delivered two TED talks that serve as the best available introductions to his practice and thinking:
- Art in the Age of Machine Intelligence (TED2020, Aug 2020 · 11 min)Covers the early WDCH Dreams and Machine Hallucinations works; explains how he conceptualises the machine as a dreaming, remembering entity.ted.com — Art in the Age of Machine Intelligence
- How AI Art Could Enhance Humanity's Collective Memory (TED2023, Aug 2023 · 5 min)Presents Unsupervised at MoMA and the Large Nature Model; argues that AI can act as a keeper and amplifier of collective memory.ted.com — How AI Art Could Enhance Humanity's Collective Memory
Dataland — The World's First AI Art Museum
Announced in September 2024 and scheduled to open in spring 2026, Dataland is a 20,000 sq ft museum dedicated to AI art, co-founded by Anadol and Erkılıç. It will anchor the Grand LA, a Gehry-designed $1 billion mixed-use development adjacent to Walt Disney Concert Hall in downtown Los Angeles — the same building that first drew Anadol's imagination when he arrived in the city in 2012.
Google is partnering with the studio to power Dataland's AI tools through a sustainable energy park in Oregon. Anadol describes the museum as a space "where the invisible becomes visible" — a permanent institution for an art form that has, until now, only ever been temporary.
Awards & Recognition
- Time 100 Impact Award (2025)
- UCLA Edward A. Dickson Alumnus of the Year (2024)
- Lumen Prize Gold Award — Melting Memories (2019)
- iF Design Award — Quantum Memories (2022)
- Google Artists and Machine Intelligence Residency Award (2016)
- Microsoft Research Best Vision Award
- D&AD Pencil Award, German Design Award, SEGD Global Design Award
- Lorenzo il Magnifico Lifetime Achievement Award for New Media Art
- Columbia University Breakthrough in Storytelling Award — WDCH Dreams
Published May 2026 · Eugene Murray