Considerations on the Translation of Music Culture For Artificial Intelligence

Considerations on the Translation of Music Culture For Artificial Intelligence

Recently, international Go champion, Lee Sedol, failed in ‘defending humanity’ (as The Guardian put it1) during a contest against the computer program AlphaGo in scenes reminiscent of Gary Kasparov’s contest against the IBM computer, Deep Blue, in the 1990s. In that same decade, within the music community, composer/scientist David Cope created a series of computer programs designed to make Bach fugues anew. The project underwent a peer review process, which seemed to confirm that no one could tell the difference between Bach’s original fugues, and the computer’s. In as much as Cope’s work personifies a kind of musical ‘Turing Test’ it also unintentionally set an agenda for AI within music creation, the repercussions of which I think are still being felt in our current climate of digitally augmented music making.

Today, the appetite for computer scientists to test AI approaches to music creation has given us such diverse propositions as Beyond the Fence, a musical in which the plot and all the music is generated algorithmically, Iamus, an AI composer whose work has been performed and recorded by the LSO, and several commercial, multi-million pound start-up music ventures, notably; Jukedeck, which creates bespoke AI composed soundtracks aimed at YouTube style DIY videos, and Melomics, a company seeking to make commercial algorithmic music, again for use in soundtracks on TV or elsewhere.

Whilst one might argue that much of this is low hanging fruit in the broader spectrum of music culture as a whole, I nevertheless find this model of the stupefying of well-constrained paradigms in music culture using AI to be increasingly problematic. Whilst musical structure might have game like qualities, treating a musical construct as roughly analogous to a game in terms of problem solving not only trivialises the context that fostered such works, but also reinforces the sense in which simulating a quintessential archetype using AI, by inference implies a broader achievement culturally than is often the case. At their worst, such implications amplify a kind Platonism, wherein musical works are judged in light of idealised forms, whose virtual personification trivialises the many diverse real forms and unique variations routinely embraced by music culture as a whole.

I have recently written about the need for a digital ontology of music as a method of contextualising the work of artist-programmers in the domain of music creation, a proposition that seeks to address the many propositions of AI for music creation in the 21st Century. The question of how a digital ontology of music might expand established definitions of music also opens the potential for a determined critical discourse on such technologies, something that in turn may help articulate new forms of music creation in partnership with such emerging techniques, as opposed to sidelining them to the creation of tools that either reinforce or mimic existing inadequately constrained paradigms in music.

The embracing of diversity in music culture through the creative use of technology in the 21st century has, in my opinion, never been more critical. As humans, diversity through innovation and creation are some of the things we do best, something I see personified in the far reaching, adventurous and explorative forms of music created during the course of the 20th century. That music practitioners and creators should seek and call for the continuous restructuring of music is incontrovertible and a priori instantiates the many dialectics variously articulated by the history of music. It is in this light that we must harness and utilise digital technologies such as AI. Digitally, as elsewhere, we need to do human better.

1 Steven Borowiec, 2016, AlphaGo seals 4-1 victory over Go grandmaster Lee Sedol, 15 March 2016, Available

Related Posts

Leave A Comment