Captivating algorithms: recommender systems as traps
Nicholas Seaver (Tufts University)
Paper short abstract:
Internet platforms use recommender systems to "hook" users, anticipating their preferences to keep their attention. Anthropological theories about traps clarify key features of this relationship, and these captivating infrastructures suggest comparative approaches for an anthropology of algorithms.
Paper long abstract:
Contemporary life online is marked by the presence of algorithms that recommend materials to users, from movies to songs to newspaper articles. These recommender systems are designed to "hook" users, anticipating their preferences in order to keep their attention. Critics argue that these systems trap users in "filter bubbles," shutting out serendipity, and that they embody the biases of their creators, narrowly contouring cultural worlds in their image. In this paper, I consider algorithmic recommender systems as captivating infrastructures, drawing on fieldwork with developers of music recommendation and the anthropological literature on traps. The long, sporadic history of anthropological entanglement with traps and capture — from Otis Mason's turn-of-the-century psychological speculations to Clifford Geertz's webs of significance to Alfred Gell's material-semiotic analyses — illuminates features of algorithmic captivation that existing critiques neglect. Thinking of algorithms as traps foregrounds the role of anticipation, casts computation as a "nexus of intentionalities" instead of crude force, and highlights how algorithms produce environments for human experience. Recommender systems (like other traps) are not merely bits of technical ingenuity, but rather tangles of agency and anticipation that unfold over time. Though they do not project the all-purpose dazzle of a Trobriand canoe prow or the blunt materiality of a deadfall, they are dynamically tailored, behaviorism-inflected attentional traps. As algorithms capture the attention of social scientists, this comparative anthropological theorizing promises an understanding of algorithms that escapes the lure of novelty and locates them within the realm of pragmatic, technical, semiotic interaction.