Date: Monday, November 3rd, 2025
9:00 am – 10:00 am Pacific Time
12:00 pm – 1:00 pm Eastern Time
Location: Weekly Seminar, Zoom
Title: From Strategic Consumption to Altruistic Engagement: A Game-Theoretic View on Recommendation Systems
Abstract:
Recommendation systems are omnipresent in our lives. An important assumption in many deployed systems is that user consumption reflects user preferences in a static sense: users consume the content they like with no other considerations in mind. However, as many recent HCI studies (including some from our group) have documented that this is by far not the case; users engage with content dynamically in a variety of ways always hoping to achieve certain “outcomes’’ beyond static consumption.
This talk examines two behavioral extremes that reveal how user adaptation interacts with algorithmic fairness: selfish/strategic consumption and altruistic engagement. Using Stackelberg games, a classical framework from game theory, to model user system interactions, we first show that strategic consumption amplifies disparities for minority users (i.e., users with niche or underrepresented interests). Out of fear of losing exposure for their preferred content, such users may avoid consuming mainstream material, reinforcing echo chambers and self-siloing.
At the opposite end, we study (what we term) altruistic engagement, a behavior inspired by large-scale grassroots movements where users intentionally interact with algorithmically suppressed content to “boost” its visibility. We compare social welfare under truthful preference reporting and under a family of altruistic strategies. Our theoretical analysis provides sufficient conditions guaranteeing strict increases in social welfare under altruism and introduces an algorithm to compute effective altruistic strategies. Remarkably, for commonly used recommender utility functions, such strategies also improve the platform’s own utility. These findings are robust to several model misspecifications, reinforcing their generality.
Overall, this talk will serve as a proof-of-concept of the (theoretical) reasons behind why traditional RecSys may incentivize
users to adapt to them and for quantifying the downstream effects of said adaptation.
Bio:
Chara Podimata is a Class of 1942 Career Development Assistant Professor of OR/Stat at MIT and a Lead Researcher at Archimedes/Athena RC. Her research interests lie mostly at the intersection of Theoretical Computer Science, Economics and Machine Learning, and specifically on incentive-aware machine learning, social computing, online learning, and mechanism design. Recently, Chara has started thinking about policy questions related to AI and recommendation systems.
Before MIT, she was a FODSI postdoctoral fellow at UC Berkeley. She obtained her PhD in CS, and was a member of the EconCS group at Harvard, where she was advised by Professor Yiling Chen. During her PhD, Chara’s research was generously supported by a Microsoft Dissertation Grant and a Siebel Scholarship. During the summer of 2019 and spring of 2020, she was an intern at Microsoft Research in New York City, mentored by Jennifer Wortman Vaughan and Alex Slivkins respectively. During the summer of 2021, she was an intern at Google in New York City, hosted by Renato Paes Leme. Before joining Harvard, she was an intern for Google in Athens, Greece. She received her Diploma from National Technical University of Athens, where she was advised by Professor Dimitris Fotakis.