The issue of persuasion games, where one player, the agent, tries to influence the behavior of the other, the user, is considered in this paper. In the cases considered here, the agent is assumed to have knowledge that the user does not. However, the agent does not have complete information either about the world or about the user. In particular, the agent does not have full knowledge of any strategy the user may have. The agent is assumed to create a strategy for itself that makes a selection of what information to provide the user based on some information about the world state. Two applications are considered.

In one, the agent advises the user about the route to take to arrive at a meeting on time. In the other, the advice concerns the number of sandwiches to order for a conference lunch. The user behavior is modeled in two ways: one uses fully rational behavior, and the other a linear weighted utility quantal response (LUQ) human model. This latter is one of the original contributions of the paper. In the travel advisory case, the LUQ model is preferred; in the sandwich case, the fully rational one works better.

The authors consider the question of choosing which model to use for a given problem. The models were the subject of considerable experimentation, which is described in detail by the authors. The results will be of interest to workers interested in the creation of information-providing systems that have a persuasive component.