AI and the future of music composition
This is a short essay I wrote in May 2023, it was commissioned by Perfect Sound Forever, an online magazine of music commentary. The prompt for the essay is this:
The ability of AI to make text, images and sound is improving at a rapid pace. Does this affect how you think about your own music? Does it change how you view the value and purpose of making music? For example: On a practical level, do you worry about being unable to reach an audience as AI generated music continues to proliferate? Or, on a more existential level, can you imagine a point where AI gets so good that it will no longer be meaningful for you to continue making music?
In Isaac Asimov’s novella “Profession”, education is a solved problem where children undergo “taping” at certain ages and have all the relevant information recorded in their brains without the need to make an active learning effort. The protagonist, George Platen, is however deemed unfit for taping at the age of 18, and placed in “The House for the Feeble-Minded” where he, along with others of similar mental ability, is given access to all human knowledge but has to acquire it himself. He escapes the House a year later to confront the doctor who diagnosed him as feeble-minded, and finds himself at the Olympics (happening in San Francisco!) where Educated People compete to be hired by advanced Outworlds. It becomes clear to George that their problem-solving abilities are limited by the knowledge made available to them through taping. But he thinks he has a better solution to such limitations - namely reading books and talking to people who have expertise in a given field (an extraordinary claim for his time). Later, through a series of fateful encounters, George learns that the House he was earlier confined to is in fact the Institute of Higher Studies where people like himself, who have the urge to create, despite being told otherwise, work for the advancement of science and human civilization.
This novella can be seen as a parable of AI-human relationship in music composition - and any creative endeavor, really. One might say deep learning-based music-generating AI (which so far, including Google’s MusicML, have been arguably less impressive than image- and text-generating ones) are “taped” by training on the existing music libraries and don’t have a chance of ever being as creative as humans have been. Though another might counter that even in human-composed music, everything is a remix, and it’s only a matter of enough training on high-quality data for AI to reach human levels. In any case, I don’t see the AI-human relationship in music as fundamentally antagonistic. I believe the existence of AI composers will be no more threatening to the future of human composition than is the existence of fellow human composers - both peers and masters of the past.
What is meaning-making in the space of effectively infinite combinatorial possibilities of rhythm and melody and harmony? Granted, what we call music “has temporal patterns that are tuned to the [human] brain’s ability to detect them because it is another brain that generates these patterns.” [1]. (Incidentally, this is also why aleatoricism, stochastic music and the like find few appreciators.) Composition is not about arbitrarily combining rhythms and pitches - there’s often a narrative arc, and always an intention behind a piece of music. Musical ideas are not just cycled through and then forgotten, but are developed, modified, and returned to in a way that carries an emotional charge. An AI must have a model of human emotional landscape - which abstractly relates to musical gestures and textures - and an understanding of musical narratives at different scales in order to be able to write music that would be meaningful to humans. Is it possible to gain this understanding through training on the corpi of existing human-composed music? That’s what human composers do, at any rate, in addition to being attuned to their own emotional microcosm.
I don’t think any developments in AI music will make my own writing less meaningful or less fulfilling for me personally. I write music as a way of reconnecting to my roots, of expressing a longing for my native Azeri musical language and my admiration of Western classical forms, spanning centuries - which is arguably a niche crossover, in the overall musical soundscape. I don’t plan to use AI in my writing in the near future - again, because of the personal significance, I want musical ideas to be born in my heart unaided.
But there is no doubt a huge potential in using AI for experimenting with combinations of existing musical ideas and textures, irrespective of one’s level of musical education. Arguably, some musical genres are more amenable to such experimentation than others. I expect high-quality AI-generated techno music to emerge earlier than, say, an impressive example of an AI-generated Renaissance-style madrigal. Neither will stop musicians from writing and producing music in those genres or from inventing new genres and soundworlds out of whole cloth - which may for now remain a human prerogative.
[1] György Buzsáki, Rhythms of the brain, p. 123
Artwork generated with Midjourney.
Originally posted here. There are also other very thoughtful contributions to the August/September issue that I enjoyed reading.
