Data Skeptic Collective Altruism in Recommender Systems
8 snips
Feb 27, 2026 Ekaterina (Kat) Filadova, a PhD student in MIT EECS studying strategic learning and algorithmic game theory. She explores how users can coordinate to influence recommender systems. Topics include framing recommenders as multi‑agent games, collective altruism that helps underrepresented content, challenges in distinguishing coordinated behavior from bots, and empirical tests showing collectives can improve minority items.
AI Snips
Chapters
Transcript
Episode notes
Collaborative Filtering Turns Recommendation Into A Multiagent Game
- Recommender systems with collaborative filtering create a multi-agent game because users' interactions influence each other's recommendations through similarity signals.
- Kat models users as latent vectors and shows one user's intentional interaction can propagate via matrix completion to affect similar users' feeds.
Sequential User Actions Make Strategic Effects Cascade
- Collective, coordinated user actions can deliberately alter the observed entries the recommender learns from, creating inter-user strategic propagation over rounds.
- This propagation is complex because interactions happen asynchronously and one user's choice influences later users' observed signals.
Account For Collective Altruism When Designing Or Testing Recommenders
- Consider collective altruism as users deliberately interacting to surface underrepresented content they care about, not just individual actions.
- Design studies or defenses assuming users may coordinate to like/comment to boost niche items via collaborative signals.