On the strategic transmission of information to imperfect agents and crowds

In both single- and multi-agent systems, the information an agent has access to plays a crucial role in his decision-making process. This fact creates an opportunity for a central planner or an omniscient player outside of the game to reveal some information to the others to steer their behavior towards more favorable outcomes for her. For instance, a central planner observing the current state of a road network can leverage her informational advantage to influence drivers, a softer approach that contrasts with usual pricing methods. Information design is the field dedicated to analyzing the ways and the extent to which agents can be nudged by information alone. The revelation principle, which replaces somewhat arbitrary messages sent to agents with the action they would have played given the message, proves to be an instrumental tool in computing optimal policies. This procedure relies on the central tenet that each agent follows their recommendation, granted all others do, and that recommendations are conditionally optimal. In addition to the cooperation of the agents (when other actions than the one recommended are equally appealing), this requires knowing or eliciting their private information and, more importantly, that all agents process data in a Bayesian way, agree on a prior, are rational and perfectly characterized by the sender. If any of these conditions fail—and they experimentally do for human agents—, a signaling policy designed with a nominal model in mind can lose its merit or even entail worse outcomes than not having communicated at all. This issue notoriously arises in the most simplistic class of games, where a single agent faces a finite choice. In the presence of model uncertainty regarding the agents, the sender should consider a robust information design program, assuming the worst could happen. One of the appeals of this approach is that it makes no specific assumption on the decision-making model of agents. Instead, it aims at guarding against a generic uncertainty only parametrized by its magnitude. In this thesis, we broadly tackle three different scenarios: quadratic persuasion, where the utilities of a sender and a single receiver are quadratic; finite persuasion, where the action space of a single receiver is finite; and information design for routing games, where a crowd of drivers take paths over a road network and cause congestion. We investigate conditions under which the revelation principle holds or can be extended, how an optimal robust policy can be devised or approximated, and the consequences for the optimal robust policy and the utility of the agents. In doing so, we incidentally derive a few theoretical byproducts that find applications in routing games and information design more largely.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Aerospace Engineering

Grantor

University of Illinois Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Massicot, Olivier

Contributors dc:contributor

• Langbort, Cedric
• Basar, Tamer
• de Clippel, Geoffroy
• Shamma, Jeff S
• Xu, Haifeng

• game theory
• computer science
• economics
• information design
• bounded rationality
• routing games
• congestion games
• Bayesian persuasion
• human agent

Statement dc:rights

• Copyright 2025 Olivier Massicot

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/129456

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F129456