Since October, singers, songwriters and music writers have been uploading 100,000 new songs to services like Spotify every day. Too much music. There is no reality, variable or otherwise, in which all this can be experienced even after a thousand lifetimes. Whether you're into Japanese noise, Russian hardcore, Senegalese Afro-house, Swedish doom metal or Bay Area hip-hop, the range of available listening options is staggering. This is a huge problem that data scientist Glenn McDonald is trying to solve. In the following excerpt from Computing Taste: Algorithms and the Makers of Music Recommended , Tufts University author and anthropologist Nick Seaver explores McDonald's unique landscape methodology for identifying all the prospects you don't need.
© Powered by Engadget
Reprinted with permission from Nick Seaver's Computing Taste: Algorithms and Makers of Music Recommended , published by The University of Chicago Press. © 2022 University of Chicago. All rights reserved.
musical world
"We are now at the dawn of an era of seamlessly connected music," said Data Alchemist under the Space Needle. Glenn Macdonald chose the name himself, preferring "alchemy" with its esoteric associations to the "data science" that has become commonplace. Her job, as she described it on stage, is to "use math, typing and computers to help people understand and discover music."
McDonald's is practicing alchemy for music streaming service Spotify, where it's working to turn the foundations of big data — listener engagement logs, snippets of digital audio files and whatever else it can get its hands on — into precious gold: products that can attract and retain paying customers. .. . The extraordinary power of McDonald's alchemy is that ordinary data, when processed properly, is transformed from subtle interactive traces into dense cultural meaning.
It was 2014, and MacDonald was speaking at the Pop Conference, an annual gathering of music critics and academics held in a shabby Frank Gehry-designed building in downtown Seattle. I follow the news on the Internet all over the country. The theme of that year's conference was "Music and Mobility," and Macdonald began his talk by talking about his personal musical journey, playing along with samples. "When I was a kid," he began, "you can find music by standing still and waiting." As a child, he listened to folk music at home, which his parents played on the stereo. But as he grew older, his hearing expanded: car radios played heavy metal and new wave; The Internet has opened up an obscure new world of genres to explore. Where he had once been stuck, a passive observer of music, he finally measured his life's progress by his ever-expanding musical horizons. McDonald's has been able to turn that passion into a profession, helping others explore what he calls "the world of music," which the on-demand streaming service has made more accessible than ever.
Elsewhere, MacDonald (2013) describes the world of music as if it were a landscape: “Follow any path, however unusual and unexplored it may seem, and you will discover hidden valleys with hundreds of groups that once lived there… reconstructing the music scene for years has been in methodical and idiosyncratic miniatures, such as Australian hip-hop, Hungarian pop, microhouse or viking metal.
Musical travelers will find both the familiar and the unexpected: sounds they never imagined and songs they love. McDonald's was impressed with this new opportunity to hear music from all over the world, be it Scotland, Australia or Malawi. "The perfect music for you might come from the other side of the planet," he says, but that's okay: "In music, we have teleporters." On-demand streaming offers a kind of musical mobility that allows listeners to explore the world of music instantly.
However, he suggested, repeating the same refrain, that the scale of these worlds could be vast and difficult to navigate. "For this new world to really be seen," Macdonald said, "we need to find a way to map that space and then create machines that will guide you through it in interesting ways." Recommendation systems offered by companies like Spotify are engines. MacDonald's recent work has focused on maps, or as he described it in another lecture: "something like a thin layer of faintly understood sequences on top of the beast of spatial information that swirls, spins and voraciously spreads out from all the world's music."
Although his language can be very poetic, Macdonald expresses an understanding of musical diversity that is widely shared by music advocates: Music exists in space. This space is, in a sense, quite ordinary, like a landscape that one can walk through, discovering new things along the way. But in another sense, this space was fantastic: beyond the valleys and hills writhing, raging, ever-growing beasts connected the points of space with each other, connected without borders. A music room can look as natural as the mountains seen from the top of the Space Needle; but it can also look like artificial topological disorder at its core. It's organic and intuitive; it is technological and chaotic.
Spatial metaphors provide a dominant language for thinking about differences between music recommendations, as in machine learning, and between Euro-American cultures more generally. In this context, it is easy to imagine that some similar things are gathered here , and other, different things are grouped together. In conversations with engineers, it is very common to find that the music chamber is actuated by a movement that surrounds the loudspeakers in an imaginary environment filled with short breaths of air and is controlled by hand movements. One gender on the left, the other on the right. In the panels and windows scattered around the office, we find a music room presented in two dimensions, containing many points grouped and scattered around the floor plan.
In the music room, similar music is nearby. When you're in a space like this, you should be surrounded by music that you like. To learn more, look and move. In the music space, genres are like regions, playlists are like journeys, and flavors are like islands of drifting territory. Your new favorite song could be just around the corner.
But despite the familiarity, such spaces are strange: similarities can be found everywhere, and points that seemed far away can suddenly become closer. If you ask, you will find that all these spatial representations are simply contractions of something much more complex, a space that contains not two or three dimensions, but potentially thousands of them. It's McDonald's info-space beast, a mathematical abstraction that pushes human spatial intuition to its limits.
These spaces, commonly referred to as "similarity spaces," are symbolic regions where a lot of machine learning works. To classify data points or recommend items, machine learning systems typically place them in space, group them into groups, measure the distance between them, and draw boundaries between them. Machine learning, as cultural theorist Adrian McKenzie (2017, 63) puts it, “takes into account all differences such as distance and direction of movement”. Thus, while musical space is, on the one hand, an informal metaphor (a landscape of musical variation), on the other, it is a highly technical, formal object (the mathematical basis of algorithmic recommendations).
Spatial understanding of the movement of data through technical infrastructure and everyday conversations; both are forms of metaphorical expression and concrete computational practices. In other words, the “space” here is formalism—a narrow technical concept that favors precision over abstraction—and what anthropologist Stefan Gelmreich (2016, 468) calls informalism—the lack of disciplined metaphor mixed with formal methods. In practice, it is often difficult or even impossible to separate technical specifications from metaphorical accompaniment. When music advocates talk about space, they're talking figuratively as well as technically.
For many critics, this “geometric rationality” (Blanke 2018) of machine learning makes it anathema to such “culture”: it measures quality, rationalizes desires, and daily plucks cultural objects from their social context to place them in sterile isolation. settlement network. For example, traditional cultural anthropology has long resisted such formalisms, which seem to lack the depth, sensitivity, or relevance to lived experience that we seek in ethnography. As political theorists Louise Amour and Olga Piotukh (2015, 361) argue, such an analysis “reduces heterogeneous data and forms of life into a homogenous computational space”.
Using geographer Henri Lefebvre's (1992) terminology, the space of similitude is the first example of an "abstract space," a kind of space of representation where everything can be measured and measured by the control of a central authority serving the capital. Applying Lefebvre's concepts to streaming music, media theorist Robert Pray (2015, 16) suggests that people like Macdonald—“data analysts, programmers, and engineers”—are primarily concerned with the calculation and measurement of hidden and abstract spaces. According to Lefebvre, space is conceived as a parasite of social and living space, which Prey associates with an audience that rejects and reinterprets the work of technologists. The expansion of abstract space under capitalism heralds in this picture “the crushing conquest of that which lives by that which is conceived” (Wilson 2013).
But to the people who work with it, the music room does not look like an empty box, even in the most mathematical way. Music recommendation generators are not limited to thin abstractions of the considered space. During training, they learn to perceive the music room as ordinary and suitable for life, despite its hidden features. Musical spaces are as intuitive as landscapes to traverse, and as alien as complex multidimensional engineering objects. Using a distinction often contested by cultural geography, they treat "space" as "place," as if an abstract, homogeneous grid were some kind of livable local environment.
The similarity space is the result of many decisions; they're just not "natural," and people like MacDonald know that the choices they make can change them forever. However, spatial metaphors that move through speech, gesture, illustration, and calculation help make the patterns of cultural data real. The confusion between map and territory – between soft representation and objective landscape – benefits people who are interested in creating objective knowledge and want to consider their own subjective influence on the process. This spatial understanding transforms the meaning of musical concepts such as genres or social phenomena such as tastes into forms of grouping.
