The philosophical issue raised by the digital revolution is that of a possible dehumanization of agency. The progress of artificial intelligence could lead to a new non-human form of agency. The opacity that surrounds the use of deep learning techniques would then run the risk of an incommensurability between human agency and algorithmic agency, and consequently lead to the implosion of the very concept of agency. An alternative hypothesis is that the massive use of artificial intelligences does not lead to a dehumanization of agency, but rather to new forms of distribution of agency, between human and non-human agents with a predominant role for informational artifacts. The novelty would not be so much a dehumanization, as a further distribution of agency.
Musical AI provides a collection of more or less spectacular examples of algorithms capable of composing or improvising music as (or almost as) human composers and improvisers do. These kinds of examples often play a central role in those who mobilize the theme of dehumanization, either to warn against the threat of our replacement by algorithms, or to reassure us that algorithms cannot be "authentic artists". It is thus a particularly useful field to test the opposite hypothesis, which is that of redistribution. As the sociology of art has made clear, artistic creation is an eminently collective enterprise, so the idea of an algorithm competing with Ludwig van Beethoven as an individual and autonomous subject of music creation is based on the mistaken premise that the activity of composing music is a homogeneous activity of a single agent. On the contrary, many heterogeneous abilities and resources are involved, and it is trivial to note that some of them (performing an accompaniment part given a figured bass, for example) can be entrusted to a student or, why not, to a suitably programmed algorithm. What is much less clear, from this point of view, is the way in which the complex of activities underlying a creative process can be decomposed and redistributed between human agents and informational artifacts.
The project is divided into a theoretical side and an empirical side. On the theoretical side, the difficulty is to manage to think the composition of a creative agency from non-agential or non-creative components. The guiding assumption is that this problem can be solved within a theory of distributed creative agency. Empirically, the aim is to open the black box of creative algorithms, through field studies focused on musical AI projects, including in particular the OMax project and its derivatives at IRCAM.
IRCAM's team : Analysis of Musical Practices