Notes on AI and musical knowledge

These notes were developed in April 2022, in dialogue with Nikita Braguinski

“Artificial intelligence” (AI) is the catchphrase for a variety of automatic techniques for the manipulation of computational representations. It encompasses algorithms that process, create, organise, and otherwise recompose recorded culture: texts, images, sounds.

Whether AI refers to a coherent set of techniques is immaterial, since that term has passed into popular culture (and policymaking) as a sign with many referents. It is as such—an amorphous assemblage of specific techniques, not necessarily sharing any single feature—that it should be analysed.

This position lays the groundwork for the idea that to think about AI today is to reflect on the specific socio-economic conditions that lead to the discussion of such techniques in these terms (as opposed to others: the legal concept of automated decision-making (ADM), the less-fashionable language of cybernetics/self-regulation, the biological inspired language of evolution, or the mathematical language of statistical forecasting, etc.).

Because of AI’s close relationship to cognitive science, and because of long-running rhetorical comparisons between calculating machines and human reasoning, it is tempting to view the adoption of AI in music as the automatisation of creative thought.

Following this line of reasoning, both critics and supporters make claims about the transformative effect of AI on the creative subject: the liberal image of the musical genius, whose task was once to refashion musical tradition—through the judicious application of taste and expensive training—is threatened by musical works that appear without authors, amateurs and technicians who now make music with little formal enculturation in the art, and the supplementation of human creative labour with its spectres in code.

Jumping to AI’s impact on musical creativity, however, is in some way premature. Musicologists and music theorists, for all their interest in understanding art and artistic practice, begin their investigations not with speculation about creative process, but with the careful construction of vocabularies and techniques for dealing with music.

Whether that is embodied in documents that testify to the production of music, or in the documents that testify to music “itself”, this scholarship establishes its claims in technical terms: meter, genre, edition, style, timbre, ornamentation, basso continuo, operetta, etc. These terms are co-constructed along with the media that are used by scholars to represent music and writing about music: scores, spectrographs, sound recordings, journal articles, program notes etc. Taken together, each constellation of terms and technique embodies particular local ontologies of music, because they predicate the existence of objects (musicological concepts) and relations between them.

Whether these ontologies gain widespread acceptance within a broad group of scholars, or even take hold outside of disciplinary spaces, is a question of interest but it is not central here. What is important is that terminology and media stabilise in specific contexts to provide an account of music, however niche, marginal, or apparently inconsequential. This lays the groundwork for the historical study of AI and music that is less concerned with headline-grabbing and/or popular applications and more interested in the slow and gradual arrival at the contemporary moment, which requires an archaeology of pre-digital and non-computational approaches to music creation as well as a global and non-Anglophone outlook that is sorely lacking in contemporary research.