The Co-PI, Nicholas Seaver, under the supervision of the PI, William Maurer, will explore the factors that influence the design of algorithmic recommendation systems. The researcher asks how algorithmic classification is influenced by sociocultural, legal, and economic contexts, as well as how the cultural theories of engineers are formed and operationalized in code. This project seeks to answer these questions through a case study in music recommendation, one of the more active research and development areas in algorithmic recommender systems. In particular, it investigates how the designers of recommender systems mediate between ideas about the formalized, quantitative nature of algorithms and the subjective, "cultural" nature of preference. As domains once thought to be ineffably human become more entwined with logics of computation, anthropologists are uniquely positioned to consider how these tensions are defined, mediated, and resolved, particularly within expert knowledge communities which build specific resolutions into widely influential technical systems.
Data will be collected through ethnographic fieldwork at several academic and industry sites in the US, where techniques for representing data and producing recommendations are developed and implemented. This fieldwork includes participant-observation and in-depth, semi-structured interviews with engineers, researchers, managers and interns involved in the production of these systems.
Findings from this research will contribute to public and academic debates about the functions and effects of algorithmic filtering, as well as to the anthropological understanding of computation, work, and classificatory practice.
The project contributes to the training of a graduate student in anthropology. The findings will inform the design of new algorithmic systems by helping engineers be cognizant of the cultural paradigms they build into technical infrastructures.
The contemporary internet is distinguished by large databases of material: streaming services offer access to vast catalogs of music and movies, social networks present countless posts and updates, and the web itself is composed of untold quantities of pages. The rapid growth of these collections and systems that offer access to them has led to the development of algorithmic tools for aiding navigation, partially automating processes of finding what one is looking for or what one did not know existed. These tools — recommender systems, search engines, and the like — have been critiqued by scholars in the humanities and social sciences especially for the biases that they introduce into processes of navigation that pretend to be "neutral." Technologists inevitably build their preconceptions into technologies, and despite their protestations, these tools cannot be neutral. However, this critical work rarely engages directly with the builders of algorithmic systems and instead treats them and their work as hidden in "black boxes," relying on company press releases, user experiences, and deduction to understand the motivations and processes that produce specific algorithmic systems with specific qualities. This project aimed to contribute empirical specificity to a growing body of critical work on algorithmic systems in anthropology and allied fields such as communication, law, and media studies. It took as a case study the development of music recommendation systems, and the co-PI spent several years (one of which was funded by the NSF) interviewing engineers, scientists, and entrepreneurs in academic labs and corporate offices and conducting participant observation at conferences, "hackathons," and in a "music intelligence" company. Fieldwork focused on the US, but the co-PI also attended international conferences for academic researchers and had the opportunity to interview people working in Europe and East Asia. The research focused on a central paradox of music recommendation and algorithmic systems more generally: how can "objective" technologies like algorithms parse "subjective" phenomena like taste? Through ethnographic fieldwork, the co-PI explored how practitioners imagine and manage this commonsense distinction — how they account for taste, and then how they account for this accounting. He found that, contrary to popular critiques that suppose technologists to be culturally incompetent, the builders of music recommendation systems possess a variety of cultural theories about taste and music. These theories are closely linked to the technologies they work on — researchers on machine listening techniques were likely to claim that taste was about sound, while those working on "social" recommenders were likely to claim it was about what your friends like — but they were often ironic, tentative and malleable, evolving alongside the systems they influenced and were influenced by. These theories and techniques, though typically ad hoc, often resembled or descended from theories and techniques that originated in the social sciences, particularly post-war American formalisms in anthropology and cognitive science. The co-PI introduced the idea of the "kinship of methods" as an anthropological approach to studying these relationships and to trouble the common idea among critics that algorithmic techniques represent novel or dramatic departures from the better-informed practices of the social sciences. This project contributes to the anthropology of technology and the study of algorithmic systems in allied fields by drawing into question some of the core presumptions of this work regarding the distribution of critical expertise. Recognizing the cultural theorizing of engineers as such and identifying its close relationship with technical infrastructures, critical social scientists and humanists should reconsider the limits and dependencies of our own theories. This has ramifications for critique: engineers often know and agree with common critiques, but they respond to them in unanticipated ways, making sense of them in their own techno-theoretical milieu. Rather than treating technologists as abject others who fail to understand culture, we should understand that their theories are themselves cultural and constitutive of culture, and all the more powerful because of it. As these systems grow in influence, it is important that outside critics engage with their makers and their cultural understandings more closely.
