Algorithmic Game Theory Meets Explainable AI: Uncovering Strategic Feature Manipulation in Machine Learning-Driven Platforms

Building on the fusion of machine learning and game theory in works like Li et al. (2024) and Hardt et al. (2023), this project would explore how users or organizations game feature-based ML systems—think resume padding for job platforms, or content manipulation for social media algorithms. By integrating explainable AI methods (like SHAP, as in Li et al.) with game-theoretic models of strategic input manipulation, this research would identify vulnerabilities in ML-driven platforms and propose robust, interpretable defenses. This approach is novel in its synthesis of algorithmic game theory, explainable AI, and adversarial ML, going beyond simple adversarial robustness to address the real-world strategic adaptation of agents in complex, platform-driven ecosystems.

References:

1. Concrete Creep Prediction Based on Improved Machine Learning and Game Theory: Modeling and Analysis Methods. Wenchao Li, Houmin Li, Cai Liu, Kai Min (2024). Buildings.
2. Algorithmic Collective Action in Machine Learning. Moritz Hardt, Eric Mazumdar, Celestine Mendler-Dunner, Tijana Zrnic (2023). International Conference on Machine Learning.