Independent Reading · 2025
Bayesian Persuasion with Externalities: Exploiting Agent Types
Jonathan Shaki, Jiarui Gan, and Sarit Kraus · AAAI
Bayesian PersuasionExternalitiesMulti-Agent Systems
Full text
Read the Paper
Use View PDF or Download PDF to read the full paper on this device.
Paper Information
- Authors
- Jonathan Shaki, Jiarui Gan, and Sarit Kraus
- Venue
- AAAI · 2025
Summary
This paper examines how a principal can inform and coordinate multiple agents whose utilities depend on one another’s actions.
Core Idea
Grouping agents into a small number of types provides a compact structure for persuasion problems that also involve coordination externalities.
Method
The paper characterizes public, private, and semi-private signaling schemes, gives linear-program formulations, and establishes tractability and hardness results.
Key Takeaways
- Persuasion with externalities combines information disclosure and action coordination.
- Agent types can make otherwise exponential joint-action descriptions manageable.
- Optimal computation is polynomial in bounded-deviation settings but becomes NP-hard more generally.
Questions
- How sensitive are optimal signals to errors in the assumed agent-type model?
- What approximation methods work well when many agents may deviate?