Signaling Games without Common Priors: Dynamic Learning and Regret-Minimization Approaches

Traditional signaling models rest on the common prior assumption, which is often unrealistic. Camara et al. (2020) push beyond this by using no-regret dynamics, but mainly in simple principal-agent settings. This idea extends the approach to full-fledged signaling games: agents start with arbitrary (possibly conflicting) beliefs and update them based on observed outcomes using regret-minimization or reinforcement learning. What new equilibrium concepts emerge when players lack shared beliefs but learn and adapt over time? This could uncover novel types of stable signaling or information transmission, and illuminate real-world contexts like online marketplaces or political communication, where priors are not common and learning is ongoing. Such models would build a bridge between mechanism design and machine learning (Hu et al. 2018; Jaradat et al. 2025), and challenge the foundational assumptions of classic information economics.

References:

1. Inference Aided Reinforcement Learning for Incentive Mechanism Design in Crowdsourcing. Zehong Hu, Yitao Liang, Yang Liu, Jie Zhang (2018). Neural Information Processing Systems.
2. Mechanisms for a No-Regret Agent: Beyond the Common Prior. Modibo K. Camara, Jason D. Hartline, Aleck C. Johnsen (2020). IEEE Annual Symposium on Foundations of Computer Science.
3. Exploring the Intersection of Information Theory and Machine Learning. Yousef Jaradat, M. Masoud, Ahmad A. Manasrah, M. Alia, Khalid Suwais, Sally Almanasra (2025). ˜The œinternational Arab journal of information technology.