Computing Taste: Algorithms and the Makers of Music Recommendation by Nick Seaver (review)

Reviewed by: Computing Taste: Algorithms and the Makers of Music Recommendation by Nick Seaver Siel Agugliaro Computing Taste: Algorithms and the Makers of Music Recommendation. By Nick Seaver. Chicago: University of Chicago Press, 2022. xv, 203 p. ISBN 9780226702261 (hardcover), 99; ISBN 9780226822976 (paperback), 20; ISBN 9780226822969 (ebook), 19. 99. Illustrations, bibliography, index. The recent expansion of music scholarship on artificial intelligence (AI) reflects the growing concern in and outside academic circles over the impact of machine learning on every aspect of the production, distribution, and fruition of music. While some researchers, such as Eric Drott, Ross Cole, Melissa Avdeef, and Charles Hiroshi Garrett have investigated the ethical and aesthetic implications of the use of AI in connection with copyright issues and creative and listening practices, other scholarly efforts, such as the recent European Research Council–funded project Music and Artificial Intelligence: Building Critical Interdisciplinary Studies, led by anthropologist and musicologist Georgina Born, aim at bridging longstanding ontological and disciplinary gaps currently separating music studies from the AI industry. In Computing Taste: Algorithms and the Makers of Music Recommendation, Nick Seaver continues to shed light on the relationship between machine-learning systems and the human actors who design and supervise them. He based his book on several ethnographic studies that he conducted in the 2010s as well End Page 645 as on dozens of interviews with CEOs, software engineers, and other IT professionals and academics involved in or researching about the making and the functioning of music recommendation systems. At the outset of Computing Taste, Seaver questions the human vs. machine dichotomy that is often envisioned by critics of algorithms' increasing pervasiveness in contemporary life. In the introduction, "Technology with Humanity, " Seaver argues that music recommendation systems, far from existing beyond the realm of human intervention, "are full of people who make choices and change things up" (p. 7). As the book's title itself suggests, developers of music recommendation systems continuously mediate between culturally ingrained notions of musical taste and the informatic means on which they operate "by drawing on their own experiences from worlds of calculation and feeling, by programming computers and exercising their cultural intuitions about taste and music" (p. 8). In the book's six chapters, Seaver describes how "the makers of music recommendation" (p. 19) think about music, their own work, and the digital environment in which they operate. In the first chapter, "Too Much Music, " he describes one of algorithm developers' perceived main tasks—namely, to provide users with effective tools to sift through an otherwise overwhelming accumulation of musical choices. Seaver argues that the idea of music overload is a "myth" (p. 30), largely fueled by the capitalistic need to create new desires to be fulfilled by ever-new services and goods. But even as such, he explains, overload is experienced as a "real" and "new" problem (p. 29) by those working in the music recommendation industry. Here he examines why this is the case from a historical standpoint. According to Seaver, the rise of cybernetics in the immediate post–World War II years was responsible for the emergence of an "informatic cosmology" (p. 42) in which computational machines were seen in ideal continuity with the human brain. This cosmology, he argues, informs to this day the mindset of the developers of recommendation algorithms. Just as a computer may "crash" as it attempts to process an excessive amount of information, so too is the human brain believed to be vulnerable to overload. Moreover, because people, both in the online environment and in their daily lives, are imagined to be part of a network, their experiences are necessarily biased, depending on their individual position in it. As such, music recommender systems are intended as "good" filters that can replace existing ones and expand the horizon of those who make use of them (p. 48). While the image of the overloaded listener has deep roots in the history of informatics, recent developments in the business model of streaming companies have generated a new notion of music listeners as the target of increasingly sophisticated strategies to retain them as customers of music online services. In chapter 2, "Captivating Algorithms, " Seaver describes how the growing availability of. . .