In lieu of an abstract, here is a brief excerpt of the content:

  • AI Comes of Age
  • Ellen Pearlman (bio)

Not since Philip Glass and Robert Wilson premiered Einstein on the Beach at the Avignon Festival in France in 1976 has a new work highlighted such a profound shift in the zeitgeist—in this case, the twenty-first century’s Discrete Figures. Glass’s hypnotic minimalism soundscape complimented Wilson’s spare stage sets, while Einstein’s opening chorus intoned the sequential numbers 1-2-3-4-5-6-7-8. That combination, an unintentional presaging of the rise of 8-bit binary computer code accompanied by Lucinda Childs’s repetitive, linear choreography, mimicked the signal routing of then relatively unknown computer circuit boards.

Discrete Figures, an interactive dance piece, explores the relationship between performing arts, math, the human body, and its simulated body. It is a line in the sand from which there is no turning back in terms of the use of machine learning (ML) and artificial intelligence (AI) in performance. Created by Tokyo-based artist, interaction designer, programmer, and DJ, Daito Manabe (Rhizomatics Research), and Kyle McDonald, an artist and coder, the piece was choreographed by “intrinsically Japanese” (her words) MIKIKO, with dancers from the Japan-based dance troupe ELEVENPLAY. After having its U.S. premiere at San Francisco’s Gray Area/Grand Theater on April 19–21, 2018, the performance moved to New York City, where it opened at the New York Live Arts theatre on May 8–11, 2019.

Manabe had e-mailed McDonald, asking if he wanted to work on an new idea— “Dance x Math (ML).” The objective was to use computers to incorporate computer vision algorithms that are typically used for surveillance of the human body. McDonald agreed, saying he wanted to “explore the possibilities of training a machine learning system to generate dances” that complemented an individual dancer’s unique style. He focused on creating two sections of Discrete Figures, nicknamed the “debug” and “AI dancer” scenes. [End Page 55]


Artificial Intelligence is a complex term: it’s so vast, it can even include the idea of consciousness, or sentience, in machines. For the purposes of this article, it is simpler to emphasize the term machine learning when thinking about AI. In machine learning, an algorithm, or set of commands in computer code, learns to build upon a well-defined task by repeating that task again and again. It does this based on a complex mathematical calculation, or algorithm. Most machine learning currently uses more nuanced sets of commands called “deep learning.” Deep learning draws upon vast sets of electronic data sorted into hierarchies. Some of these hierarchies mimic the flow of information in a manner that resembles how neurons function in the brain. Because of deep learning’s network structure, these hierarchies are also referred to as “neural nets.” The information in these neural nets flows at very rapid rates. If you imagine the human brain as containing vast data banks (memory, both cognitive and muscle), and algorithms as the predetermined structure or characteristic of the brain (different parts of the brain), then you can think of deep learning as the types of things the brain does based on a hierarchy of what is most important at any one moment. In a computer, this means a superfast automation of many tasks that were traditionally not automated. For Discrete Figures, five dancers interact with cubes of images and light, drones, cameras, and machine learning algorithms that emulate them in live time.


Discrete Figures incorporated a Vicon motion-capture system that collected two-and-a-half hours of data during forty separate recording sessions, a type of intensive setup traditionally used for creating complex gaming applications. Dancers improvised, and instead of using the movements to generate an end-product, like characters in a game, the motion capture was used to explore differences between the dancers and their own unique movement styles. The coding team created a special network called dance2dance. It was based on neural network architecture for sequential modeling from motion capture, modified to handle complex mathematical data. The data was based on predictive patterns. A predictive pattern explores the probability that a specific motion...


Additional Information

Print ISSN
pp. 55-62
Launched on MUSE
Open Access
Back To Top

This website uses cookies to ensure you get the best experience on our website. Without cookies your experience may not be seamless.