
54:35
Recommendation engines sit at the intersection of prediction, preference, and choice.
In this episode, we explore the fascinating world of strategic learning in recommender systems with Ekaterina (Kat) Fedorova, an NSF CSGrad4US fellow and PhD student at MIT EECS. Kat's research focuses on what happens when the data machine learning systems learn from becomes strategic—when users actively coordinate and manipulate their behavior to game recommendation algorithms. We dive deep into the game-theoretic frameworks that help us understand collective action in digital platforms, examining how coordinated user behavior differs from traditional assumptions in machine learning where data is treated as static and non-responsive. The conversation takes some surprising turns as Kat reveals counterintuitive findings from her research: algorithmic "protest movements" where users collectively boycott or manipulate content can paradoxically benefit platforms by providing clearer signals about user preferences. We discuss the practical challenges platforms face in distinguishing coordinated human behavior from bot activity, the implications for content moderation and recommendation design, and what happens when your training data becomes a strategic battleground. Whether you're building ML systems, studying algorithmic game theory, or simply curious about the arms race between platforms and their users, this episode offers fresh perspectives on a critical challenge in modern AI.

About Hello! I’m Ekaterina (or Kat), an NSF CSGrad4US fellow and an EECS PhD student at MIT. At MIT, I currently work Dr. Chara Podimata and Dr. Costis Daskalakis. I’m interested in a combination of theoretical (CS and micro econ) algorithmic game theory topics like strategic