Toward a Responsible Platform Economy: Strategic Behavior, Learning Dynamics, and Social Good

Modern online platforms—such as recommender systems, social media, and content creation ecosystems—operate at the intersection of human behavior and algorithmic decision-making, shaping not only individual experiences but also broader societal outcomes. As these platforms continue to scale, concerns surrounding strategic manipulation, efficiency and fairness trade-offs, and the long-term societal impact of algorithmic interventions have become increasingly urgent. In this dissertation, I present a series of studies for understanding and optimizing machine learning (ML)-driven platform economies through multi-agent modeling—a framework that explicitly accounts for strategic user behavior, long-term learning dynamics, and alignment with societal well-being. My work centers on two core research directions: (1) investigating how the volatility and strategic nature of human behavior affect the design and reliability of ML systems, and developing robust algorithms to address these challenges; and (2) examining how ML-driven platforms and emerging AI technologies influence individual behavior and broader societal structures, and identifying principled approaches to ensure that these systems remain aligned with human values in the long term. A central application of my research lies in on online content recommendation platforms, where I demonstrate a scalable and effective mechanism for enhancing user welfare by designing incentive structures for content creators. Overall, this thesis contributes a new perspective on building socially responsible human-AI interaction systems through the lens of multi-agent and game-theoretic modeling.