The Measured Listener | Los Angeles Review of BooksThe Measured Listener<br>For the Legacies of Eugenics series, Daniel Shanahan shows how algorithmically mediated music recommendations have a dark backstory.<br>By Daniel ShanahanMay 31, 2026<br>Science & Technology
History
Music
Did you know LARB is a reader-supported nonprofit?<br>LARB publishes daily without a paywall as part of our mission to make rigorous, incisive, and engaging writing on every aspect of literature, culture, and the arts freely accessible to the public. Help us continue this work with your tax-deductible donation today!
Donate
This is the 15th installment in the Legacies of Eugenics series, which features essays by leading thinkers devoted to exploring the history of eugenics and the ways it shapes our present. You can read the first part here. The series is organized by Osagie K. Obasogie in collaboration with the Los Angeles Review of Books, and supported by the Center for Genetics and Society, the Othering & Belonging Institute, and Berkeley Public Health.
A CENTRAL PREMISE of High Fidelity—Nick Hornby’s 1995 novel, which was turned into a film in 2000 and then a Zoë Kravitz–starring TV series in 2020—is that its main character, Rob Fleming, owner of a failing record store, cannot understand himself without first measuring and ranking his experiences. “You are what you like,” as the line goes. For this protagonist, the act of ranking every aspect of his life is both a coping mechanism and a pathology. Rob Fleming can only discuss his feelings and emotions in list form—“they have opinions and I have lists,” he says.
In the last several years, models of algorithmic recommendation have one-upped Rob’s lists and human curators in general. In part, they have done so by hitting the sweet spot between surprise and not too much of it—by going beyond the titular list. In my case, they have learned what kind of listener I am. Beyond knowing my favorite artists and genres (Pavement, Courtney Barnett, Lucy Dacus, and other artists falling under the broad umbrella of “indie rock”), they know which music makes me feel most nostalgic (Elliott Smith, Aimee Mann, Wilco, and other “indie folk” artists) and which songs I listen to for myself (these days, Phoebe Bridgers), for my teaching at a school of music (everything from Chopin’s Mazurkas to J Dilla to Katy Perry), and for my kids (the KPop Demon Hunters and Percy Jackson soundtracks). They know I prefer certain kinds of music only at certain hours, know that I prefer my favorites during the day but am more likely to listen to news podcasts in the morning. Their recommendations invariably feel intimate, even inevitable.
They are neither. The algorithms are built upon decades of research that asked not what music is, nor what it means to listen to music, but how a human might be measured based on their musical preferences. Like IQ tests, this research has a dark history. We’ve inherited methodologies born of a desire to associate musical preference with inherited “fitness” in a eugenicist sense. The original goal was diagnostic: to use what people listen to as a window into their selves, including their personality disorders and any mental illnesses they might have. Algorithmically mediated music has absorbed these methodologies.
In a January 2024 interview, Spotify co-founder Daniel Ek announced what music platforms would do with the multitude of tunes now at people’s fingertips: “[W]e’re going to become that trusted friend where we’re going to introduce you to things that you probably thought, ‘No way in hell am I going to be interested in this,’ and you’re going to be totally open to it.” He went on to claim that the company would “do a better job” choosing your music than you possibly could. It would transcend any list you could make yourself. “[E]ven if you spent a whole working day trying to figure out what you wanted to listen to,” he bragged, “we will be able to create a playlist that is so much better than any of that.” Spotify alone has more than 700 million users, and by the company’s own accounting, a third of all new artist discoveries happen through algorithmic recommendations rather than active search—a measure, in every sense, of how thoroughly the listening experience has been quantified.
The algorithms are based on certain assumptions about music and the listener. The first is that music can be understood as discrete streams of information—made up of tempo, key, and lyrics—which become identifying parts of a larger musical style or genre. This approach descends from a long history of stylometrics (or stylistics)—the use of quantitative methods to ascribe authorship and style markers. Pandora’s Music Genome Project, for example, looks at 450 musical features, which the company regularly refers to with terms such as “genes” or “DNA.” This project became pivotal to how, in the 2010s, streaming platforms grouped artists and songs into genres and microgenres, a prerequisite for...