Abstract
Information Design studies how a sender strategically communicates information to influence the decisions of receivers. Classical models assume that receivers interpret signals in a Bayesian way. This talk introduces two models in which receivers update beliefs non‑Bayesianly.
The first model features a learning receiver who responds to signals by running a contextual no‑regret multi‑armed bandit algorithm, treating signals as contexts. We characterize the range of payoffs that are achievable by the sender against such learning receivers. In particular, when the receiver does no‑swap‑regret learning, the sender’s achievable payoff concentrates around the optimal payoff in the classical Bayesian persuasion problem.
The second model captures framing effects, where the language used to present information affects the receiver’s beliefs. We model the framing‑to‑belief mapping using a large language model (LLM), and employ another LLM to optimize the sender’s framing strategy for persuasion.
Together, these works represent initial steps toward bridging abstract information design theory and practical persuasion by incorporating ML theory and AI technologies. Links to papers: https://arxiv.org/pdf/2402.09721 (ICLR 2025 & Quantitative Economics) https://arxiv.org/pdf/2509.25565
Time
Wednesday, Jan. 7, 14:00--15:00
Speaker
Tao Lin is a postdoc in the Economics and Computation group at Microsoft Research (New England). He will be an assistant professor in the School of Data Science at the Chinese University of Hong Kong, Shenzhen, in 2026. He obtained his PhD in Computer Science from Harvard University in 2025 and BSc from Peking University in 2020. Tao’s research focuses on algorithmic game theory, mechanism design, information design, and their connections with machine learning and theoretical computer science.**
Room
Room 602
