Captivating algorithms: Recommender systems as traps

Abstract

Algorithmic recommender systems are a ubiquitous feature of contemporary cultural life online, suggesting music, movies, and other materials to their users. This article, drawing on fieldwork with developers of recommender systems in the US, describes a tendency among these systems’ makers to describe their purpose as ‘hooking’ people – enticing them into frequent or enduring usage. Inspired by steady references to capture in the field, the author considers recommender systems as traps, drawing on anthropological theories about animal trapping. The article charts the rise of ‘captivation metrics’ – measures of user retention – enabled by a set of transformations in recommenders’ epistemic, economic, and technical contexts. Traps prove useful for thinking about how such systems relate to broader infrastructural ecologies of knowledge and technology. As recommenders spread across online cultural infrastructures and become practically inescapable, thinking with traps offers an alternative to common ethical framings that oppose tropes of freedom and coercion.

Perspectives on algorithmic normativities: engineers, objects, activities

“…The case of collaborative filtering recommender systems deployed on commercial platforms such as YouTube or Netflix can help us ground these considerations (Seaver, 2018). In a nutshell, these systems seek to foster unprecedented interactions between items and users by looking at past interactions between similar items and users.…”

Metrics and decision-making in music streaming

“…Still, despite all of the data Spotify collects in house, it still turns to other stakeholders for access to other data points that it does not have. When managers, labels reps, or distributors are pitching new artists or music for playlists, Spotify face what are known as "cold start problems" (Seaver, 2018), because it simply lacks the data to inform its algorithms' predictions and choices. As one informant explains, "What Spotify, Apple, and others say [in pitch meetings] is 'Why should we push this particular artist?…”

Ethical Aspects of Multi-stakeholder Recommendation Systems

“…The most effective way to do so is to make recommendations more transparent. Explanations for the system's inferences and recommendations could be used to empower users and other stakeholders to use the RS as a tool to support their interests, instead of interacting with it in a passive way (Karakayali, Kostem, & Galip, 2018;Seaver, 2018).…”