Intrinsic Motivation for Exploration in Corrupted Games

Tsuchiya et al. (2024) focus on bounding regret under corruption but don’t address how agents detect or respond to deviations. Meanwhile, Merrick and Shafi (2013) show intrinsic motivation (e.g., curiosity) can guide exploration in games. This idea proposes intrinsic rewards as corruption detectors: agents reward themselves for discovering strategies that deviate from prescribed algorithms, then use these signals to adapt their learning dynamics. For example, in a security game (Clempner, 2025), a defender might use surprise rewards to identify anomalous attacker behavior. We’d formalize this using Bayesian inference (à la Clempner) to estimate corruption levels in real time. This bridges psychology (intrinsic motivation) and robust learning, creating systems that actively identify and mitigate corruption rather than passively absorbing it.

References:

1. Corrupted Learning Dynamics in Games. Taira Tsuchiya, Shinji Ito, Haipeng Luo (2024). Annual Conference Computational Learning Theory.
2. A game theoretic framework for incentive-based models of intrinsic motivation in artificial systems. K. Merrick, Kamran Shafi (2013). Frontiers in Psychology.
3. Learning Deceptive Tactics for Defense and Attack in Bayesian–Markov Stackelberg Security Games. J. Clempner (2025). Mathematical and Computational Applications.