How to Evaluate a Recommendation System

In the digital economy, recommendation algorithms get…a LOT of attention. To some, they’re the special sauce behind everything from Spotify’s personalized playlists to Tik Tok’s “For You” page. For others, they represent a dark, vibe-generating demiurge slowly sapping music’s social power. But for all the discussion of how these programs are transforming our world(s), there’s surprisingly little analysis of what—exactly—they are, or how they’re meant to work.

Answering these seemingly simple questions is the goal of Nick Seaver’s new book “Computing Taste,” which explores the identities, goals, and practices of the programmers behind these technologies. Far from Machiavellian manipulators, the coders he describes are surprisingly idealistic music-lovers, desperately trying to analyze an almost infinitely complex cultural practice. Their failures to do so—and the ideologies they adopted as a result—would have enormous implications for the development of digital music, remaking genres, redefining listening, and shaping the platforms at the heart of the modern industry. Put it this way—we’ll definitely never look at a "Discover Weekly" playlist the same way again.