Unruly

There is no such thing as music. Or, rather: there is no such thing as music in the universal, self-evident sense the word implies. The idea that music is a singular human practice — a language we all speak — is itself a culturally specific belief, and a relatively recent Western one. Ethnomusicology has wrangled with this for over a century: indeed, one of the discipline's first moves was to pluralize the term, to insist on musics. Its more provocative assertion, though, was to ask whether the word itself makes any sense at all in parts of the world distant from the European concert tradition. Many of the world's sonic practices are not thought of as "music" by those who perform them, but instead something more deeply intertwined with ritual, labour, devotion, or speech.

And so, when we say music, we are not naming a universal. In a historical sense, we are describing a (mostly Western) tradition organized around a set of discrete pitches of equal temperament, harmony, the work-concept, the author (see also Authorship below), etc. The unruliness of "music", then, happens when these particularities begin to masquerade as generalities, especially within technical infrastructures. When a streaming platform categorizes a monsoon raga as "E minor," for example, or a dataset recognizes only key signature and tempo, Western music theory is being invoked less as a way to describe music than to define it, globally, in epistemologically domineering ways. Music, in this sense, becomes a universal marker not of what we hear, but of a longstanding process of infrastructural universalization.

An unruly definition, by Thomas Hodgson and Emmie Head

Intermediaries are ubiquitous and mercurial. Deriving from inter- (between or among) and medius (in the middle), but distinct from the declarative mediator (to intervene or reconcile between two sides), the intermediary inhabits ambiguity. Its digital origin story is in the junctures of the material — nodes, routers, switches, proxies, gateways. Here intermediaries are both situational and propulsive, industriously shuttling suitcases of digital information across notched networks; an enterprise prided as mercantile, not mercenary. From this inheritance comes the core identity crisis of the modern intermediary. It aspires to clerical neutrality to the point of invisibility, but it does so with endemic self-importance. In between, everything is politics.

As intermediaries scaffold to platforms, the stickiest and best-known become monstrous assemblages of information exchange, attention-hacking, vanity, and profiteering. In the policy world, intermediaries are flashpoints. They entice and frustrate, thwarting all but the most determined with their aspirational camouflage, magnificent inscrutability, colossal power and, just as readily, wafer-thin responsibility. "Nothing to see here," the machine throbs.

An unruly definition, by Julia Powles

To classify music is in a sense to draw borders around sound — to decide what belongs with what, and, by implication, what does not belong at all. "Genre" is one of the most familiar of these epistemic borders, and one of the most contested. Musicology, and ethnomusicology in particular, have long been critical of the term: the names of genres are rarely neutral indicators of musical content, but rather carefully curated labels, motivated by corporate histories of marketing, taste, race, and geography. The World Music debate of the 1990s, in particular, showed how virtually an entire planet's worth of (non-Western) musical difference came to be lumped together under a single label — a label that revealed much more about the industry doing the sorting than the musics being sorted.

One of the big questions for our moment is whether streaming and AI dissolve these anxieties or magnify them. The corporate promise is one of dissolution: platforms speak of a "borderless" ecosystem, of infinite micro-genres, of recommendation unshackled from physical spaces (record stores, yes, but also local sites of music making). The reality, though, is that classification has not disappeared but rather become ever more automated, hidden within platform infrastructures. Genre tags are today assigned at scale — sometimes by experts (self-appointed or otherwise), increasingly by algorithms, and at least until recently, as one Spotify engineer admitted, simply invented in the name of algorithmic efficiency. Classification, then, collapses music's plural ontologies through technical infrastructures' requirement that everything be sortable.

An unruly definition, by Thomas Hodgson

Authorship, sometimes correlated with "ownership," associates a creator or a group of creators with their creations, a composer with their compositions. In the Western sense, authorship comprises an individual person (an "author") who originates or creates an intellectual or creative work. The humanities, including studies of Western music, romanticize the notion of the single author, establishing that the individual authorship of original ideas is accompanied by a level of genius or mastery over the subject matter or art form. This sense of authorship, while frequently referenced, is only a singular point in a web of diverse social and cultural understandings of authorship.

Authorship is culturally and socially dependent, not unlike music (see Music above). Beyond Western notions, authorship can be unknowable, unattributable, communal, collective, or unapplicable. Authorship does not always equate to ownership. Authors may choose to remain anonymous, develop pseudonyms, or even make their work freely accessible. Authors may create derivative works, transform pre-existing works, or parody. Provenance, or the lineage of custody of a product of someone's (or some people's) creativity is often associated with authorship should the author retain authority over their creative work or seek to assert it. Exchanges of intellectual property rights, works-for-hire, fair use, or Creative Commons may introduce novel complications to provenance. Yet, it is important to remember that even provenance can be falsified, debated, uncertain, or entirely unknown.

As music making artificial intelligence advances, authorship and provenance become increasingly complicated. Such Platforms (above) blur and obfuscate lineages of authors and creators by curating an anonymized front-end for users and a back-end black box within which answers to questions of authorship and provenance are housed but left invisible to users.

An unruly definition, by Emmie Head

Music, as a medium, is ephemeral, atmospheric, and embodied — all qualities that resist any neat translation into a "dataset."

Broadly, a dataset is defined as a collection of materials used to train, test, or evaluate an AI model. In the case of music, this might include: recordings (such as MP3, WAV, video files); computational representations derived from those recordings (spectrograms); symbolic representations (MIDI files, scores); textual data (lyrics, reviews); descriptive metadata (genre, tempo, key, instrumentation); contextual metadata (release year, record label, language, country of origin); or listener behavior data (streams, skips, likes, etc). None of these categories capture music — they are merely attempts to find small footholds in an expansive atmosphere of flowing sonic mist.

Yet while these data points cannot wholly capture music, they frame what counts as music, especially in processes of classification and recommendation. In the dataset, presence is often viewed as power. Models are trained on a corpus of music and go on to prioritize and recommend those same types of music. Music datasets largely skew towards Western English-language songs, reflecting larger systemic inequities, including who has access to recording and digitization. In a pattern of algorithmic bias, this loop becomes recursive, as models trained on Western English-language music continue to elevate music marked as similar. The way forward is murky. A more diverse or intentionally counterhegemonic dataset could help decenter this privilege — but to say that more data and more extraction is the tidy solution would also miss the central heartbeat of music as a medium.

Instead, we need to also consider all that slips beyond the dataset. In performance studies, Diana Taylor distinguishes between the archive and the repertoire.¹ The archive is made of the tangible artifacts or materials that can be housed by an institution, such as the text, scores, recordings, which is similar to the corpus of a dataset. By contrast, the repertoire is the corporeal knowledge or experience that transfers from body-to-body, which is resistant to the archive. In Taylor's model the archive is often curated and maintained by institutions of hegemonic power. And yet, the repertoire which dances around and between these artifacts can be counterhegemonic, passed below the eye (or ear) of the institution. From this perspective, we provoke questions about the dataset and its limits: What elements of music are beyond the reach of the dataset entirely? In what ways does eluding capture protect music from being reduced to data in dominating systems?

An unruly definition, by Devon Baur

¹ Taylor, Diana. The Archive and the Repertoire: Performing Cultural Memory in the Americas. Duke University Press, 2003.