Thuistezien 221 — 31.03.2021
Music history is usually expressed linearly, often thought of as a gradual form of evolution. Evidence about early rudimentary forms of music are seen as stepping stones that gradually, bit by bit, through a continual process of musical evolution, lead to the music we know today. And although sometimes we might hear casual talk of how music has nowhere else to go now, the 20th and 21st century have shown us repeatedly that music is constantly evolving and will keep doing so. It’s just difficult to imagine what the future’s as-of-yet unthought of music forms will be like.
So, if we think of music in this way, would it then be possible to build software that could generate music independently, while functioning within a principle of continuous evolutionary progress? Who knows, maybe if it were able to recreate an evolutionary process similar to that of music’s history, it could then start creating work that chimes with the vanguard of today’s and tomorrow’s new musical directions, or even possibly predict it. And although in her talk – documented in the above video – Dr. Róisín Loughran doesn’t quite speak about creating a machine to digest all of mankind’s musical output and learn how to output the natural evolution of it, she has created several evolutionary music generative systems that are exploring many of these thoughts. Drawing inspiration from the Darwinian Theory of Evolution, Dr. Loughran’s general idea seems to be as follows: if you create within a continuously-running music-generating-system a process that evaluates each of its outputs, and then through a principal of “survival of the fittest” teach it to retain traits from its successful outputs to then use them for the subsequent and hopefully improved outputs, the system should then theoretically be able to keep producing better and better music on its own. Although this is a simplification of her system, this seems to be the general idea behind it.
For some time now, this has been one of the principal questions that Dr. Róisín Loughran explores in her research work, as she reveals to us in her talk. As a researcher at the Natural Computing Research and Applications Group (NCRA) at the University College Dublin, her research finds ground in the intersection between music and science, focusing primarily on computational musical creativity. As she shows us, she seems to be quite successful at creating a computational musical evolutionary process that ever so gradually creates ‘fitter’ and ‘fitter’ musical melodies. But quickly the thorny question that comes up is: What would “fitness” be in the context of music. Or even more simplified: What is ‘good’ music? Following the idea of the ‘survival of the fittest'’, how do you then teach A.I. to decide between several melodies and pick out the ‘fittest’ one which should then pass on its metaphorical genetic material? But also, outside of these questions, her work inevitably brings up a more general debate: can A.I. even create ‘good’ music? And can it be ‘creative’ in its output.
As Dr. Loughran points out, several A.I. systems have been successful at creating music and melodies that follow the principles of folk music or classical music to create competent iterations of such genres. But music is definitely not limited to such forms. Can a self-generating program create meaningful music within the larger image of an ever-evolving lineage of global millennia-old lineage of music creation? Especially when people have such varied musical tastes, and responses from critics and audiences to new works of music and new works of art can prove so very contentious to say the very least? Dr. Loughtan then wonders sometimes if machines generating music should even really bother with the question of ‘good’ music… In her presentation she reveals some of the inner workings of her music generating systems, and many of the questions that arise from her development of them.
Text: James Alexandropoulos - McEwan