Collective Altruism in Recommender Systems

February 27, 2026·55 min·5 clips

Kat reveals how users collectively manipulating recommender systems could actually help the platforms being gamed.

Recommendations look individual. The host starts with the usual setup: one person tells a platform what they like, and the platform sends back suggestions. Kat, a PhD student at MIT studying strategic learning, pushes on that picture. Users can act together and deliberately manipulate recommender algorithms, and that manipulation is not always bad for the platform. The episode keeps circling the word altruism. When the host brings up a friend's channel or underrepresented communities trying to get seen, Kat says those examples fit the paper's definition. Utility gets handled with care. People may not run a utility function in their heads, but the model still needs a way to represent concern for someone else's outcome. The examples stay easy to picture: liking, watching, or commenting because a friend needs views, or because suppressed users and suppressed items deserve better treatment from the algorithm. Acting as a group changes the whole thing. One person can nudge; a coordinated group can turn recommendation into strategy. The conversation ties that to game theory and matrix completion, then brings in survey evidence suggesting this kind of algorithmic activism may be less rare than listeners think. The tone stays careful and research minded. Kat closes by pointing listeners to the online paper and hoping the work leads to more questions.

As heard by us

A clear look at how coordinated users can push recommender systems toward overlooked people and content.

Data Skeptic treats recommender systems as social machinery, not just engines for personal taste. Kat, a PhD student at MIT EECS studying strategic learning, argues that users can coordinate likes, watches, comments, and similar signals to help underrepresented content reach the…

Read the full review in PlayNext →

Why you'd press play

Press play when you want a recommendation episode about users working together, not just using a feed alone.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Data Skeptic