What do you think?
Rate this book
Computing Taste: Algorithms and the Makers of Music Recommendation
Meet the people who design the algorithms that capture our musical tastes.
The people who make music recommender systems have lofty they want to broaden listeners’ horizons and help obscure musicians find audiences, taking advantage of the enormous catalogs offered by companies like Spotify, Apple Music, and Pandora. But for their critics, recommender systems seem to embody all the potential harms of they flatten culture into numbers, they normalize ever-broadening data collection, and they profile their users for commercial ends. Drawing on years of ethnographic fieldwork, anthropologist Nick Seaver describes how the makers of music recommendation navigate these how product managers understand their relationship with the users they want to help and to capture; how scientists conceive of listening itself as a kind of data processing; and how engineers imagine the geography of the world of music as a space they care for and control.
Computing Taste rehumanizes the algorithmic systems that shape our world, drawing attention to the people who build and maintain them. In this vividly theorized book, Seaver brings the thinking of programmers into conversation with the discipline of anthropology, opening up the cultural world of computation in a wide-ranging exploration that travels from cosmology to calculation, myth to machine learning, and captivation to care.
The people who make music recommender systems have lofty they want to broaden listeners’ horizons and help obscure musicians find audiences, taking advantage of the enormous catalogs offered by companies like Spotify, Apple Music, and Pandora. But for their critics, recommender systems seem to embody all the potential harms of they flatten culture into numbers, they normalize ever-broadening data collection, and they profile their users for commercial ends. Drawing on years of ethnographic fieldwork, anthropologist Nick Seaver describes how the makers of music recommendation navigate these how product managers understand their relationship with the users they want to help and to capture; how scientists conceive of listening itself as a kind of data processing; and how engineers imagine the geography of the world of music as a space they care for and control.
Computing Taste rehumanizes the algorithmic systems that shape our world, drawing attention to the people who build and maintain them. In this vividly theorized book, Seaver brings the thinking of programmers into conversation with the discipline of anthropology, opening up the cultural world of computation in a wide-ranging exploration that travels from cosmology to calculation, myth to machine learning, and captivation to care.
216 pages, Hardcover
Published December 6, 2022
33 people are currently reading
623 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 15 of 15 reviews
December 29, 2022
A fascinating anthropological study of the music recommendation industry, outlining the self-conceptions of participants about what they are doing (with clear explanations of topics in recommender systems like matrix factorization, audio analysis, user embeddings, neural network methods, and so on, and the business model of the streaming music industry) and also providing a big picture outsider's view on how these relate to ways we think about music, business, and society.
Usually when an academic breaks out the Bordieu (etc) references I prepare myself for flabby prose and even more muddled thinking, but here the ties to social theory are clearly and succinctly explained and illuminate precise and concrete distinctions. In some cases the tools of anthropological analysis, like deeply interrogating a widely used metaphor, like "trapping" users or the analogy between feature space and physical space, are more or less enlightening, but overall the exercise provides a perspective that encourages both deeper and broader understanding of the role and methods of the field.
Overall, this book was a late contender for my favorite the entire year, and I would recommend it to anyone who works with user data, broadly construed.
Usually when an academic breaks out the Bordieu (etc) references I prepare myself for flabby prose and even more muddled thinking, but here the ties to social theory are clearly and succinctly explained and illuminate precise and concrete distinctions. In some cases the tools of anthropological analysis, like deeply interrogating a widely used metaphor, like "trapping" users or the analogy between feature space and physical space, are more or less enlightening, but overall the exercise provides a perspective that encourages both deeper and broader understanding of the role and methods of the field.
Overall, this book was a late contender for my favorite the entire year, and I would recommend it to anyone who works with user data, broadly construed.
April 11, 2023
When you interact with anything but the simplest website these days, you're interacting with algorithmically presented content, the output of a class of machine learning techniques called recommender systems. When people complain about the evils of the Algorithm with a capital-A, they usually mean a recommender system which is malfunctioning in some way, whether it's Youtube, Twitter, and Facebook boosting extremist viewpoints, Amazon suggesting suicide kits via 'commonly bought together', or TikTok showing conservative legislators hot twinks.
Recommender systems are messy, complex objects, and Seaver provides a close study of streaming music. As an anthropologist, he focuses on the cultural aspects of music recommendation systems, but shorn of the heated political aspects of recommender systems in general, music becomes a kind of data drosophilia, enabling us to view the many complexities and intellectual assumptions behind recommender systems in a kind of model system. Seaver ably synthesizes a survey of relevant academic theories and ethnographic work at music recommendation companies and conferences to provide a deeper and more full understanding of algorithms not as a black monolith, but as an open framework of human effort.
I need a picture here, and this monolith seems good.
Seaver begins with the early web of the 1990s, and the then heady concept that computers + networking + digital audio files could create a kind of galactic jukebox, seamless access to every piece of recorded music ever. But with great music comes the problem of information overload. When you can listen to everything, how do you decide what to listen to? Even simple cataloging becomes a problem, as anyone who remembers the glory days of Napster knows. A college experiment in databasing my dorm's shared music library lead to a host of bad file names and random deletes.
Aside from technical issues, music is also deeply personal. We all have a favorite song, which probably hasn't changed much since we were 21. Music is mood, generation, and subculture defining. And traditionally, music has been defined by gatekeepers: label executives, radio DJs, what's available at the local record store, and the constantly shifting definitions of hipness. For music aficionados, finding what you like in the galaxy of everything becomes a burden, though actual information overload is much harder to pin down as a phenomena real people experience with music.
The first suite of techniques was based around collaborative filtering. Given users, items, and rankings, simple mathematical techniques can fill in the grid, producing lists of what each user would like. Collaborative filtering isn't actually very effective, and modern services have turned to captivation metrics to see what drives people to keep listening, and what drives them to stop listening.
Seaver's ethnography closes studies some of the key imaginaries of algorithmic recommendations, including idealized listeners, who are active or passive, and much more diverse than the overwhelmingly American, white, male, and hip developers who work for these companies. Developers describe their issues with wanting to play plunderphondics for listeners who prefer K-pop.
Computing Taste also provides a solid description of the Second Good Trick of Data Science, transforming messy objects into a dense vector space. A song which can be classified in many ways: human labeling, collaborative filtering, acoustic pattern matching, becomes described as a set of coordinates in a n-dimensional space. Coordinates which are close are similar songs, and a listening session becomes a journey through this space. Spatial metaphors can be deeply misleading, but are fundamental to machine learning.
As a data scientist and Science, Technology, and Society PhD with an avid interest in music recommendation systems (top 1% of Spotify listeners by time in 2022), this book could have been written specifically for me. Match between user and item aside, I'm serious in my recommendation that this provides a nicely grounded and well executed case study of a key branch of applied machine learning.
Seaver doesn't answer all the questions I have. Having tried a bunch of streaming music services, I qualitatively believe that Spotify's secret sauce, whatever it might be, is better than the competitors. There is a fine balance between familiarity and novelty in playlists, and I do wish Spotify provided better tools to tweak the algorithm for power-users. And while the harms of algorithmic misjudgment in the case of music are pretty minimal compared to other uses of recommender systems, there's still plenty left unsaid about pitifully low artist payments, the potential for payola, the dominance of old hits over new music, and the reshaping of performance around what triggers curiosity in a playlist, like ALLCAPS band names, and what won't cause a negative captivation interaction where the user shuts off the song, in terms of unchallenging styles. The dark side of the galactic jukebox might be a million songs all the same, running off of an engine of venture capitalist money, aiming towards monopoly abuses. But until the machine breaks breaks, I'm going to turn that song up.
(Disclosure notice: I received a free copy of this book from the author, and no other compensation)
Recommender systems are messy, complex objects, and Seaver provides a close study of streaming music. As an anthropologist, he focuses on the cultural aspects of music recommendation systems, but shorn of the heated political aspects of recommender systems in general, music becomes a kind of data drosophilia, enabling us to view the many complexities and intellectual assumptions behind recommender systems in a kind of model system. Seaver ably synthesizes a survey of relevant academic theories and ethnographic work at music recommendation companies and conferences to provide a deeper and more full understanding of algorithms not as a black monolith, but as an open framework of human effort.
I need a picture here, and this monolith seems good.
Seaver begins with the early web of the 1990s, and the then heady concept that computers + networking + digital audio files could create a kind of galactic jukebox, seamless access to every piece of recorded music ever. But with great music comes the problem of information overload. When you can listen to everything, how do you decide what to listen to? Even simple cataloging becomes a problem, as anyone who remembers the glory days of Napster knows. A college experiment in databasing my dorm's shared music library lead to a host of bad file names and random deletes.
Aside from technical issues, music is also deeply personal. We all have a favorite song, which probably hasn't changed much since we were 21. Music is mood, generation, and subculture defining. And traditionally, music has been defined by gatekeepers: label executives, radio DJs, what's available at the local record store, and the constantly shifting definitions of hipness. For music aficionados, finding what you like in the galaxy of everything becomes a burden, though actual information overload is much harder to pin down as a phenomena real people experience with music.
The first suite of techniques was based around collaborative filtering. Given users, items, and rankings, simple mathematical techniques can fill in the grid, producing lists of what each user would like. Collaborative filtering isn't actually very effective, and modern services have turned to captivation metrics to see what drives people to keep listening, and what drives them to stop listening.
Seaver's ethnography closes studies some of the key imaginaries of algorithmic recommendations, including idealized listeners, who are active or passive, and much more diverse than the overwhelmingly American, white, male, and hip developers who work for these companies. Developers describe their issues with wanting to play plunderphondics for listeners who prefer K-pop.
Computing Taste also provides a solid description of the Second Good Trick of Data Science, transforming messy objects into a dense vector space. A song which can be classified in many ways: human labeling, collaborative filtering, acoustic pattern matching, becomes described as a set of coordinates in a n-dimensional space. Coordinates which are close are similar songs, and a listening session becomes a journey through this space. Spatial metaphors can be deeply misleading, but are fundamental to machine learning.
As a data scientist and Science, Technology, and Society PhD with an avid interest in music recommendation systems (top 1% of Spotify listeners by time in 2022), this book could have been written specifically for me. Match between user and item aside, I'm serious in my recommendation that this provides a nicely grounded and well executed case study of a key branch of applied machine learning.
Seaver doesn't answer all the questions I have. Having tried a bunch of streaming music services, I qualitatively believe that Spotify's secret sauce, whatever it might be, is better than the competitors. There is a fine balance between familiarity and novelty in playlists, and I do wish Spotify provided better tools to tweak the algorithm for power-users. And while the harms of algorithmic misjudgment in the case of music are pretty minimal compared to other uses of recommender systems, there's still plenty left unsaid about pitifully low artist payments, the potential for payola, the dominance of old hits over new music, and the reshaping of performance around what triggers curiosity in a playlist, like ALLCAPS band names, and what won't cause a negative captivation interaction where the user shuts off the song, in terms of unchallenging styles. The dark side of the galactic jukebox might be a million songs all the same, running off of an engine of venture capitalist money, aiming towards monopoly abuses. But until the machine breaks breaks, I'm going to turn that song up.
(Disclosure notice: I received a free copy of this book from the author, and no other compensation)
August 20, 2023
Excellent ethnographic work on who and what's powering your Spotify Weekly. Not only was it illuminating, it was very well written, carefully weaving theory to explain, contextualize, problematize his field work. Even if you're not into ethnography, information science, etc. it's a fascinating read to understand a part of your life that's probably playing in the background right now.
Hopping on my high horse for a second, this book certainly vindicated some of my intuitions of recommendation algorithms that lead me to leave Spotify and then Apple Music several years ago. Not to invoke the "wake up sheeple!" meme, but Seaver does employ/investigate an evocative pastoral metaphor that should make you lift up your head and ask yourself: just how was this music delivered to my playlist?
Hopping on my high horse for a second, this book certainly vindicated some of my intuitions of recommendation algorithms that lead me to leave Spotify and then Apple Music several years ago. Not to invoke the "wake up sheeple!" meme, but Seaver does employ/investigate an evocative pastoral metaphor that should make you lift up your head and ask yourself: just how was this music delivered to my playlist?
April 9, 2024
It’s a really good ethnography! At parts I found it hard to keep up with the various aliases and history but overall it was quite well done.
December 10, 2022
My first time being a book’s first review!! I will take it seriously -
1. kind of a bold mic drop moment to have your epilogue chapter be the interview w the guy who says a book look this would go nowhere
2. author loves using ginormous words to explain somewhat simple concepts - were they sleeping w a thesaurus or something ?
3. chapter 3 on listener personas was most interesting to me, could probably have just skimmed the rest
1. kind of a bold mic drop moment to have your epilogue chapter be the interview w the guy who says a book look this would go nowhere
2. author loves using ginormous words to explain somewhat simple concepts - were they sleeping w a thesaurus or something ?
3. chapter 3 on listener personas was most interesting to me, could probably have just skimmed the rest
May 29, 2024
I appreciate the underlying thesis to the book: algorithms are not mysterious forces of nature but are rather the result of choices made by their human creators and therefore it is useful to understand the thinking of those people. However, it felt like the book was incomplete. Most of the statements made came from quotes of individual creators with no discussion of what percentage of creators shared this opinion (especially frustrating in the epilogue when an interviewee shares an opinion that seems to suggest the rest of the book is now no longer applicable because the field has changed so much since its inception). There was no mention of how the business side affected algorithmic choices; surely the engineers don't have free rein to do whatever they want. I'm not familiar with anthropology research - maybe it is best practice to include only observational analysis without offering recommendations or drawing conclusions. The writing itself is academic and includes lots of references; it is clearly deeply considered. But I was left wondering how much any of it mattered.
Read
July 22, 2024I particularly enjoyed the chapter debunking the idea that algorithmic recommendation systems are called for in an era of unprecedented abundance in "content"; like myths of moral and linguistic decline, people have been complaining that there is too much to read and listen to and watch for about as long as records exist.
It was also fascinating, as a philosopher of language, to see that there are basically perfectly parallel debates in anthropology about whether "context" can be codified in any meaningful way, or whether it is too amorphous and subject to its own kind of context-sensitivity to play a role in systematic theorizing. And it turns out that the debate has consequences for how music recommendation algorithms are structured! Super cool stuff.
It was also fascinating, as a philosopher of language, to see that there are basically perfectly parallel debates in anthropology about whether "context" can be codified in any meaningful way, or whether it is too amorphous and subject to its own kind of context-sensitivity to play a role in systematic theorizing. And it turns out that the debate has consequences for how music recommendation algorithms are structured! Super cool stuff.
August 18, 2025
This book was immensely insightful and helpful in wrapping my head around algorithms and machine learning. As a novice to data science, but an anthropology PhD student working in the realms of data sovereignty and making AI more environmentally friendly this book helped bridge my understanding between the complex systems of thought used to build an algorithm and how much human cultural influences are embedded in these supposedly objective technologies. It helpfully gave many detailed descriptions of foundational anthropological frameworks and clearly communicated the connections to algorithms and machine learning in general. I will most certainly be referencing this book often and return to analyze as I write my dissertation.
September 12, 2024
I knew this would be somewhat academic when I picked it up, when what I really want is pop science this detailed. The information the author presented and the perspective he took on music recommendation algorithms was fascinating. As a non-specialist, I'd just do better with something with either less jargon or more definitions, especially of words that mean something different in this context than when we use them colloquially. I learned enough that I'm glad I read this and I'd do it again, but even the personal anecdotes from the author's research experience couldn't keep this from being kind of a slog.
April 14, 2024
Judging from the book, the author actually spend a lot of time learning, interning in the field of recommendation. Fully appreciate his effort of devoting enough time to understand the thing, which seems to be not-as-common in a busy world full of authors.
I like the questions arises in the book: what role are the people (e.g., engineers) playing in the recommendation system? Is it better or worse? Is it bringing prolific or extinction, maybe both ?
I like the questions arises in the book: what role are the people (e.g., engineers) playing in the recommendation system? Is it better or worse? Is it bringing prolific or extinction, maybe both ?
Read
December 31, 2024This book is at its best when it demonstrates how transformations in our relationship to the body (marked by analogies to computational processing) during the course of the 20th and into the 21st century gets built into algorithms. It's worse when it gets hyper repetitive and hand hold-y to its readers... The age of informatics that Seaver describes has grander ramifications than the music recommended which I wished was stressed more.
February 15, 2024
taste is essentially *instrumented technique*, mediated and organized through countless devices。把技术和品味放在一起看，还有启发。谈到secrecy 的那部分也很喜欢。
March 20, 2024
Refreshing in that it recognizes the limitations of data scientist's power in realistic way, rather than the slightly unhinged fear-mongering approach to algorithmic critique
Want to read
May 19, 2024Atlantic
March 21, 2025
Your late-night sad boy playlist? Your 2010 emo nostalgia binge? That one time you looped “Never Gonna Give You Up” 34 times in a row? Someone, somewhere, has a graph about it.
Displaying 1 - 15 of 15 reviews