Wayne McGregor: Training an AI on Twenty-Five Years of Dancing
On the choreographer who gave a machine his archive — and what it gave back
Context
When Google Arts & Culture Lab approached Wayne McGregor in 2019 with a proposal to train an AI system on his choreographic archive, the pitch was unusual even by the standards of an artist who had spent two decades collaborating with cognitive scientists, software developers, and neuroscientists. The idea was not to replicate his style, automate his decisions, or produce dance on his behalf. The idea was to create a dialogue — a tool that could learn the movement logic embedded in his body of work and offer it back to his dancers as a creative provocation during the process of making new work.
The result was Living Archive, performed at The Music Center in Los Angeles in July 2019. It was the first publicly staged work in which an AI system trained on a choreographer's own archive was used as an active creative partner in real time. The system, later developed into a tool called AISOMA, processed nearly four million poses extracted from hundreds of videos spanning McGregor's career, and could generate new movement sequences in response to input from dancers in the studio.
That debut marked a turning point — not only in McGregor's practice but in how the broader dance and technology field thought about what machine learning could mean for a choreographer who was still actively working.
Public Record
Wayne McGregor CBE was born in 1970 in Stockport, England. He trained at the Bretton Hall College of Higher Education and José Limón School in New York. He founded Random Dance (later Company Wayne McGregor) in 1992. His choreographic practice spans ballet commissions — he was Resident Choreographer of The Royal Ballet from 2006 to 2024 — to experimental live performance, film, opera, and installation work. He is currently artistic director of Studio Wayne McGregor and Director of Dance of the Barbican Centre.
His list of works runs to more than 60 pieces for his own company and commissions from major institutions internationally, including Chroma (2006), FAR (2010), Atomos (2013), Autobiography (2017), and Deepstaria (2023). He has received honorary degrees from the University of Surrey, University of East London, and University of the Arts London, and has been a Visiting Research Fellow at Cambridge University's Cognition and Brain Sciences Unit. His collaborations with cognitive scientists and AI researchers span more than twenty years of documented activity.
Documented Work at the AI Intersection
The Choreographic Language Agent (2009–2013)
The earliest phase of McGregor's computational research produced the Choreographic Language Agent, an intelligent software system designed to generate solutions to choreographic problems. Developed with digital artists OpenEndedGroup and software architect Cassiel, the agent could respond to symbolic choreographic instructions and propose movement sequences. The most developed iteration, Becoming (2013), was used during the creation of Atomos and was presented as an interactive object that supported live studio decision-making. There is no peer-reviewed paper describing the technical architecture — the primary documentation is through Studio Wayne McGregor's public research programme records.
Distributed Creative Cognition Research (2002–ongoing)
Since 2002, McGregor has collaborated with Professor David Kirsh at UC San Diego to investigate distributed creative cognition — how decisions, intentions, and creative ideas are distributed between the choreographer, the dancers, the studio space, and the body in motion. Kirsh's published work with McGregor includes studies of epistemic actions (how choreographers use their own bodies to think) and the role of external memory in choreographic process. These studies are documented in peer-reviewed conference proceedings and are among the most rigorous empirical investigations of choreographic cognition in the field.
Kirsh, D. (2011). How marking in dance constitutes thinking with the body. Versus: Quaderni di studi semiotici, 113–115, 179–210. Kirsh, D., Muntanyola, D., Jao, R. J., Lew, A., & Sugihara, M. (2009). Choreographic methods for creating novel, high quality dance. Proceedings of the 5th International Workshop on Design and Semantics of Form and Movement, 188–195.
Living Archive / AISOMA (2019–2026)
AISOMA (the production-ready version of the Living Archive tool) was built in collaboration with Google Arts & Culture Lab in Paris. The system uses a neural architecture trained on approximately four million poses extracted from video documentation of McGregor's 25-year archive — a process that moved from 2D analysis in the earlier prototype to full 3D pose extraction using TensorFlow 2 and MediaPipe. In the studio, dancers interact with AISOMA by submitting movement inputs; the system generates movement proposals derived from the archive's statistical patterns, which dancers can accept, transform, or refuse.
AISOMA was used in the creation of Autobiography V95 and V96, performed at Sadler's Wells in March 2024. The public-facing version of AISOMA launched online alongside McGregor's major exhibition Wayne McGregor: Infinite Bodies at Somerset House, London, which ran from October 2025 through February 2026.
The technical documentation of AISOMA is published through Google Arts & Culture Lab's project pages. No independent peer-reviewed technical paper on the AISOMA architecture has been identified — the system description remains primarily in production and exhibition documentation. 🚩
Google Arts & Culture Lab. (n.d.). Living Archive: Creating choreography with artificial intelligence — Studio Wayne McGregor. Google Arts & Culture. https://artsandculture.google.com/story/living-archive-creating-choreography-with-artificial-intelligence-studio-wayne-mcgregor/1AUBpanMqZxTiQ
Critical Engagement
The questions McGregor's AI work raises are sharper than most coverage acknowledges.
The first is the problem of what the archive contains. Four million poses from twenty-five years of documentation captures the external geometry of movement — the shape of joints in space, the skeletal configurations that cameras can see. It does not capture proprioceptive sensation, the felt texture of weight-shift, the intention preceding the visible movement. AISOMA is, in this sense, a deep archive of McGregor's movement vocabulary as seen from outside. Whether that is the same as learning how McGregor choreographs is a question the project's framing productively leaves open.
The second is the question of authorship distribution. McGregor describes AISOMA as a creative partner and provocation tool — not an autonomous generator. The decisions remain with the dancers and with him. But this position is structurally similar to how he describes his human collaborators: as sources of material that he and the company transform through editorial decisions. The AI collaboration does not so much resolve the authorship question as make it explicit in a way that the choreographer-as-sole-author model normally obscures.
The third is scalability and access. AISOMA is a purpose-built system trained on one choreographer's archive, requiring a multi-year institutional collaboration with a major technology company to develop. Most practitioners cannot replicate this model. The more interesting downstream question is whether the conceptual framework — using a practitioner's own movement history as the training data for a personalised generative tool — can be made accessible at different scales.
Field Significance
McGregor occupies a genuinely unusual position: he is one of the most commissioned choreographers working in European contemporary and classical dance, and simultaneously one of the longest-running systematic researchers into the science of choreographic cognition. The AI work is not an outlier in his practice — it is continuous with two decades of inquiry into how movement knowledge is created, stored, and transmitted.
What distinguishes the AISOMA project from most dance-and-AI initiatives is its fidelity to process rather than product. The tool was built to intervene at the moment of creation, in the studio, where movement decisions are made. Its outputs are proposals to be evaluated by human movers, not finished sequences to be performed. This positions it closer to a cognitive research tool than to an automated choreography system — a distinction that matters for how the field frames what AI is doing in creative practice.
His "Infinite Bodies" exhibition at Somerset House (2025–2026) and the concurrent public release of AISOMA mark a moment when this research became accessible to non-specialist audiences for the first time. The field will be watching what that wider exposure generates.
APA Works Referenced
Google Arts & Culture Lab. (n.d.). Living Archive: Creating choreography with artificial intelligence. https://artsandculture.google.com/story/living-archive-creating-choreography-with-artificial-intelligence-studio-wayne-mcgregor/1AUBpanMqZxTiQ
Google Arts & Culture Lab. (n.d.). When technology met Wayne McGregor. https://artsandculture.google.com/story/when-technology-met-wayne-mcgregor-studio-wayne-mcgregor/fgUhvjdg-9RMuw
Kirsh, D. (2011). How marking in dance constitutes thinking with the body. Versus: Quaderni di studi semiotici, 113–115, 179–210.
Kirsh, D., Muntanyola, D., Jao, R. J., Lew, A., & Sugihara, M. (2009). Choreographic methods for creating novel, high quality dance. Proceedings of the 5th International Workshop on Design and Semantics of Form and Movement, 188–195.
Studio Wayne McGregor. (n.d.). AISOMA. https://waynemcgregor.com/research/aisoma
Studio Wayne McGregor. (n.d.). Living Archive. https://waynemcgregor.com/productions/living-archive