20th June 2023
Precedes the 24th ACM Conference on Economics & Computation (EC 2023), London, UK.
Schedule (all times are ET):
20:30-21:15 Session 1 (algorithmic fundamentals by Konstantin Zabarnyi) [slides]
21:15-22:00 Virtual coffee break
22:00-22:45 Session 2 (new frontiers by Haifeng Xu) [slides]
Format: Fully virtual, live presentations
Organizers: Haifeng Xu and Konstantin Zabarnyi
Abstract
The tutorial will survey the algorithmic results of one of the burning topics of recent years – information design – and its recent applications. It aims at introducing the participants to the fundamental model of Bayesian persuasion by Kamenica and Gentzkow (2011) and its various variants and extensions. The organizers will survey the evolution of the computational results in the original model and its extensions. They will especially focus on the connection of information design to other subfields of game theory – including mechanism design, data/information market, and contract design. They will further present many recent variations of the Bayesian persuasion model adapted to capture real-world problems, and discuss several robustness paradigms.
Motivation
Given the tremendous access to massive amount of information today, it is not surprising that information design is a timely and increasingly important topic. Indeed, as suggested by the increase of the estimated share of persuasion in the US GDP from 25% in 1995 (McCloskey-Klamer) to 30% in 2013 (Antioch), information design has evolved to become one of the most important subjects in the economy of the 21st century. Recently, there has been a rapidly growing interest in Bayesian persuasion both from the algorithmic and the economic aspect. The tutorial will focus on the former aspect – both in classic and in cutting-edge variations of the Bayesian persuasion model. The organizers will discuss the motivation behind different Bayesian persuasion variations and the known important computational results in these settings. They will further list open questions of a potential interest to the tutorial participants.
Overview
The first session will start by introducing the Bayesian persuasion setting and presenting the two basic approaches for tackling it – concavification and straightforward policies. It will proceed by listing the most classic computational results in a single-receiver, public and private Bayesian persuasion settings. Then the discussion will go on to further variations of multi-receiver Bayesian persuasion, including persuasion in networks and persuasion under constraints originating from privacy or rational inattention considerations. The session will conclude by discussing several notions of robustness, including robustness to receiver’s prior distribution and to unknown receiver’s incentives – both in an online and in an offline setting.
The second session will cover the following three lines of recent advances or applications of information design, the relevant computational results and (really) many open problems. The first line is generally about information design in sequential decision making problems, such as Pandora’s box and secretary problems. This goes beyond the classic static decision making problems as in the standard Bayesian persuasion model, and exhibits much richer problem structures in both single-sender and multi-sender setups. The second line is about the valuation of information to a decision maker, which is tightly related to information design. Based on the valuation of information, the organizers will formulate the interesting problem of optimal pricing of information and survey many recent works on this topic. Finally, the organizers will discuss the marriage between information design with two popular topics in EC community: information structure in contract theory and information elicitation in auction design. Many of these models are new and rich, and thus will leave significant amount of open problems to the audience.
