What's in a game: game-theoretic analysis for third party planning

摘要

Game-theoretic analysis allows for systematic analysis of complex adversarial situations modelled as extensive form games. Counterfactual regret minimization, the leading game-theoretic framework used to solve extensive form games, is generally used to develop improved and unexploitable decision agents. This paper focuses on the third-party planner - an agent with a stake in the outcome of a game yet have no agency within the game. Third-party planners include game organizers trying to anticipate game situations for broadcast and planning, those providing logistical support to the players, or anyone else who might interact with the game environment. While projecting game flow might in general be quite difficult, the problem becomes tractable if the players are sufficiently sophisticated as to follow an approximate equilibrium solution. This paper demonstrates how counterfactual regret minimization can assist third party planners under these circumstances.