Learning with Conflicts of Interest
Published in arXiv cs.LG/cs.AI, 2026
Recommended citation: Aryal, N., Termehchy, A., Vakilian, A., & Winslett, M. (2026). Learning with Conflicts of Interest. arXiv:2605.15504 [cs.LG]. https://arxiv.org/pdf/2605.15504
Learning with Conflicts of Interest
Overview
ML systems increasingly mediate decisions, recommendations, and information access, but the interests of system owners and users are not always aligned. When incentives diverge, models may shape user decisions through biased or manipulative information.
This work frames the interaction between users and ML systems as a game with conflicts of interest. It develops scalable algorithms with theoretical guarantees that increase desired information and actions while limiting biased or manipulative actions.
Research Themes
- Game-theoretic modeling of user and ML-system interactions.
- Learning under misaligned incentives.
- Algorithms for reducing biased and manipulative actions.
- User-centered safeguards that do not assume voluntary platform cooperation.