Computing Taste: The Making of Algorithmic Music Recommendation
- Author(s): Seaver, Nicholas Patrick
- Advisor(s): Maurer, Bill
- et al.
This dissertation reports on several years of multi-sited ethnographic fieldwork with the developers of algorithmic music recommendation systems in the US. It identifies and contributes to a nascent, transdisciplinary body of scholarship in “critical algorithm studies”—studies of algorithms’ sociocultural lives by scholars outside of mathematics or computer science. It argues that critics should concern themselves not with “algorithms” narrowly defined, but with sociotechnical “algorithmic systems,” of which humans are an integral part. It proposes that ethnography is a useful method for apprehending the cultural features of algorithmic systems and that these cultural features play a crucial role in the functioning of algorithms and how they change over time. Recommender systems provide a case in which to investigate these cultural concerns as they play out in the development of “preferential technics”—the intermingling of circulatory infrastructures with theories about taste. Arguing that theories of taste are embedded in algorithmic systems, the dissertation examines three areas that demonstrate this intermingling: listeners, music, and listening. The chapter on listeners describes how recommender systems have come to be used as tools for capturing users, bringing the anthropological literature on trapping to bear on the question of how imagined listeners inform the design of systems for captivating them. The chapter on music investigates how developers imagine music to occupy a “similarity space,” through which recommenders help listeners travel; theories about the nature of that space and the influence of developers on it mediate between understandings of space as a constructed or as a discovered order. The chapter on listening examines the changing techniques through which computers are taught to “hear” musical sound, arguing that the quantification of music is not simply a rationalization, but the establishment of a resonance between auditory and quantitative phenomena with unanticipated consequences. The conclusion explores the similarity between ethnographic methods and big data analytics, understood through the frame of “attention.” Thinking of algorithmic systems and critical research methods as techniques for organizing attention offers new, fruitful avenues for critical algorithm studies.