Merging AI and game theory in multiagent planning

An approach to reasoning about the actions that other agents are likely to pursue is outlined. This approach is based on the idea that many attempts to reason about another agent´s beliefs and actions are based on an ability to self-reflect on one´s own reasoning process and then to extrapolate to the other agent (`If I were she. . .´). It is shown how to combine knowledge-based option enumeration procedures with game-theoretic models for calculating a minimum (maximum) probability that an agent will identify and execute a specified course of action. In addition, it is shown how this approach addresses, in part, the outguessing problem in game theory