Audiences at this weekend’s Louisville Orchestra performance will be treated to one live soloist, and one appearing in the form of a hologram – both given sound by cellist Yves Dhar . The hologram represents a participant called AGNES, which is an acronym for Automatic Generator Network for Excellent Songs. AGNES is artificial intelligence.
The piece, titled “Automation,” is composed by Dhar’s fellow Juilliard graduate Adam Schoenberg . According to Dhar, the two set out to explore society’s relationship with technology.
Surreal first day with the #LouisvilleOrchestra @LouOrch: Face-to-face with AGNES #onstage before the #heavyweight #fight. 🥊 #Man vs. #Machine. Who’s your 💰 on? pic.twitter.com/9rz62Epijh
“We can’t talk about these subjects without having some sort of artificial intelligence,” Dhar stated. Dhar pointed to how ubiquitous technology is now, in everything from chess playing to virtual smartphone assistants.
AGNES is a creation of mathematicians Kathryn Leonard and Ghassan Sarkis. Leonard had approached Schoenberg, her fellow professor at Occidental College, with an offer to build a program that could learn his music. Their goal was to create a tool that could collaborate with a composer. Leonard was emphatic that the program would never replace a composer, pointing out that one of the most important factors of music was “humans communicating with humans.”
Despite no desire to challenge the role of the human composer, the type of program that Leonard and Sarkis created does have a fairly combative name. It’s called a generative adversarial network .
Meet AGNES who will battle cellist @YvesDhar on stage with the @LouOrch and @TeddyConducts. #testfootage
World premiere of #Automation on May 13-14
VFX: Ryan Wise, DASYSTEM AGNES (AI learning Algorithm): Kathryn Leonard @Occidental and Ghassan Sarkis @PomonaCollege pic.twitter.com/rkj7vMLigr
First, AGNES ingests information about music, via a MIDI file . She is taught what is music, and what isn’t (sounds such as static). Then, she tries composing. But, just like a human composition student, she makes mistakes.
In order to train her on what music is, they gave her work from her collaborator Adam Schoenberg. When she first tried to generate her imitation, results were mixed. An early result sounded like one key being repeated at the piano.
Schoenberg wasn’t sure what he was listening to. But Leonard was pleasantly surprised that, as they fed back her so-called mistakes, labeling them “not-music,” AGNES improved rapidly.
Despite having learned from an American composer, AGNES isn’t tied to the conventions of western music. So her results crossed the line into unplayable, but Dhar embraced the unpredictability.
“It’s that nexus: it’s that symbiosis of digital and human.”