Misty Copeland performing in Swan Lake. Andrea Mohin/The New York Times
Read here my other piece on the Math and Physics of ballet
Of all the fields AI can infiltrate, I never would have thought ballet to be one of them. Of course, AI might be used for administrative work in ballet studios/companies. But ballet itself? An art form performed live that has been around for centuries? Apparently, it is.
The Body as a Tool
Dance is a creative expression that builds on traditions of instruction to help student master the movement of their body. Digital technology has infiltrated the ballet environment to aide students in learning; recording one’s performances digitally, wearable devices to monitor a dancer’s movements, and software management tools for stage production and choreography. Digital tools have been introduced to improve understanding of the physiology of movements and new approaches for production design and scores.
Dance choreography is an incredibly complex skill which involves multiple movements of several bodies, spatial directions, timings of such movements, all synchronized to music (Check out Mary Nora’s Substack who writes in detail about different types of ballet!). It is very much an art from and requires innovation and artistic vision on the choreographer’s part. Coding such complexity into a computer system so it can “generate” choreography is a difficult task, especially given that ballet notation, names, and positions can vary within different schools and cultures. Often it requires uploading video recordings which require large computational memory and fast processing power,
AI is infiltrating ballet choreography and instruction. Even more problematic is that most major choreographers and company leaders are men. The upper levels of the professional ballet world are still very much controlled by men, and profit off the movements of women’s bodies. The structure of ballet classes is fairly rigid, and instructors don’t accept much pushback from students, creating disturbing power dynamics. There are rigid gender roles in traditional ballet companies and ballet performances. In a sense, the ballet dancer’s body is a conduit to reflect a choreographer’s/director’s vision.
In the United States, there is a surplus of aspiring female ballet dancers, and a far smaller pool of aspiring male ballet dancers. Because there are fewer men at lower levels, there is a greater demand for male dancers in companies. Given the large number of aspiring female ballerinas, competition is rife. Male ballet dancers are more likely to move up the ladder to become choreographers and directors. There are reports that talented female choreographers are passed up for male choreographers with less talent and of course, paid less. By 2019, among the 50 biggest ballet companies in the U.S:
Fewer than one third of artistic directors were women.
Female artistic directors made 68 cents for every dollar men made
Only 43% of executive directors were women.
The average U.S. ballet dancer only make about $30,000 a year. Misty Copeland, the first black principal dancer at the American Ballet Theater only earned $100,000 a year. Yet, they will shell out thousands of dollars for AI technology when most ballet dancers are criminally underpaid, and many women are excluded from higher positions. This is not unlike STEM fields, where lower levels of labor are populated by women, and they struggled to advance up the career ladder as well as salary-wise.
“Fusion” has been declared the first “AI ballet”, performed by the Leipzig Ballet in 2023 at the Leipzig Opera House. Generative AI was used to develop the music score (a blend of contemporary classical, electronica, and “beatbox” techniques which included AI augmentation of human voices) and choreography. Composed and directed by Harry Yeff, Gadi Sassoon, Teddy Riley with the “AI-infused choreography” developed by Mario Schroder. The sets and costumes (Paul Zoller) were designed to imitate Plato's concept of the divided self.
Within the system of ballet though, it is the women’s bodies that are heavily scrutinized to fit an aesthetic. And now they are using AI, instead of giving an unknown choreographer a chance. And as we know, AI and algorithms perpetuate bias.
Computers & Dance: A History
Surprisingly, computers have been integrated into the dance world for several decades now. Ballet moves can be described by basic mathematical and physical principles which can be modelled by computer programs.
Choreography
Merce Cunningham working on choreography on the computer in 1993.
Computers were first used for choreography in 1964 where Jeanne Beaman and Paul Le Vasseur fed a computer different dance movements, directions, and times; the computer generated a list of different combinations of movements. In 1986, the computer program Lifeforms was developed in the Computer Graphics Research Lab at Simon Fraser University, directed by Dr. Thomas W. Calvert. The program had a Graphical User Interface (GUI) in which the user can position and arrange virtual dancers and their movements. The programmers worked with Merce Cunningham, an American dancer and choreographer and major proponent of the integration of technology with dance. In March 1991, Cunningham premiered a dance piece entitled “Trackers’, in which part of the choreography was planned in Lifeforms. (Lifeforms has been improved and is currently known as DanceForms). These early programs, however, were more for ‘working out’ and visualizing choreography rather than generating it.
Generative AI can now create custom choreography based on the input data. This was debuted with Royal Ballet choreographer Wayne McGregor, and the Google Arts and Culture Lab. Damien Henry, Google Arts and Culture Lab technical program manager, created an AI program to generate independent choreography based on hundreds of hours of video footage from the choreographer’s archives, and dancers in McGregor’s company. Living Archive: An AI Performance Experiment premiered in 2019 and even won a prize. Wayne McGregor believed that “the most fascinating aspects of the technology is that it can learn and recreate the particular style of a dancer”.
Living Archive’s algorithms detect the dancer’s movements and “learn”. The unique movements of ballet are rendered to computer data.
The Georgia Institute of Technology developed a project entitled LuminAI. LuminAI has a computer-controlled dancer/avatar who views the dancer and them improvises new movements based on prior movements. The project was “housed inside a 15-foot-tall geodesic dome […] lined with custom-made projection panels for dome projection mapping”, which was “designed and constructed by Georgia Tech digital media master’s student Jessica Anderson”. This program relies on motion capture technology to gain input data.
Reactions and adoption of such AI tools are mixed. Damien Henry, Google Arts and Culture Lab technical program manager, believe it enhances creativity. Mikhail Jacob, the lead developer of the LuminAI technology, said that it “gives autonomy back to the computer”, it forces a person to create something new — potentially something better” because the dancers have to take their virtual “partner’s actions into consideration”. However, Michael Klien abandoned the use of a “ChoreoGraph” (“a sequencing tool that let[s] the choreographer set up digital variations on a timeline” and acts “as a cue-sheet for dancers during performance”) because he felt it was ultimately limiting of creativity and that it is prone to human assumption in inputs. There is also concern about the ownership of the choreography.
Virtual Dancers
Image of Royal Ballet dancers Sergie Polunin and Natalia Osipova (considered one of the best ballet dancers of all time) performing in front of a greenscreen; their images were turned into holograms for “Symphony to a Lost Generation” (the first fully holographic ballet performance ever) which featured projections of a philharmonic choir and symphony orchestra in 2016. The Guardian gave the performance 2 stars.
Computers, however, can become the entertainment via virtual reality, augmented reality, and holograms. Computationally, replicating dancer’s movements on a computer are very difficult given that the program must generate continuous motion with high kinematic complexity to music. This requires very fast Internet speed, as well as high memory and processing capacity. The programs can generate a ‘virtual partner’ known in online spaces as an ‘avatar’. This avatar can be used to generate new movement sequences or analyze current sequences. They can reflect a dancer’s movement in real-time. However, a 2021 study about a neural network trained on motion capture from dancers showed that “AI dance examples receive generally lower ratings compared with human dance examples, [though] some traits of human dance movement is reflected in the AI dance.”
Teaching
Given the highly detailed and complicated nature of dance routines, recording one’s performance can greatly enhance critique as the recording can be paused, forwarded, zoomed in and out, and slowed down. AI was first introduced as a tool to improve the recording of dancer’s movements. Since then, AI has been used to analyze the recordings of dancers and develop personalized training programs.
AI programs have been pioneered to teach students, to critique their dance movements, and provide feedback. Several programs combine “mixed reality” and motion capture technology. The OutsideMe is a program in which dancers can review their movements using a head-mounted display (HMD); they can train with virtual personal trainers and an additional dance partner. Companies like “MoveMYway” act as ‘instructors’ for dancers. They have an “AI tutor” they watch, and a motion capture system evaluates their performances and generates evaluations. Videos of their practice are also uploaded and analyzed by the system.
However, the question of what is “good” or not is not simple in the ballet world. In the ballet world, achievement is often based on a degree of conformity to prescribed movements (relevé, plié, and grand jeté). However, other components such as artistic expression, creativity, and emotional engagement are also assessed. Feedback is continuous through the ballet semester. Moreover, there are different ballet styles and teaching methods. Developing feedback based on AI is difficult. Furthermore, such systems are likely trained on thin, white, dancer’s movements and nearly impossible standards of flexibly and athleticism. Ballet standards change all the time. Often, choices for moves are aesthetic choices and evidenced by changing ballet exams over the years:
Ballet technique has increasingly valued increased flexibility; in the 1960s an arabesque was ideally slightly above 90 and now is ideally at a full 180 degrees.
Conclusion
The movement of technological tools from recording (passive) to creating (active) poses ethical issues in the ballet world. Such tools can add a level of impersonality between the teacher and student. “AI” developed ballet can also devalue emotional expression, musicality, and minimize diverse dance forms. They can supplant jobs in ballet. Such tools should be treated as aids, not replacement of the artists or the entertainment that dance provides. When a system, like ballet, has such a rigid hierarchy, and history of exploitation, this facilitates replacement by AI—hence the disturbingly little commentary on the ethics of AI in the dance/ballet world.