AI Seminar

Using Empirical Game Theory to Analyze Four-Player Chaturanga

Chris Kiekintveld

Chaturanga is an ancient variant of chess. We analyze a 4-player version of this game using methods from empirical game theory. Like chess and similar games, chaturanga is challenging due to the extremely large strategy space that arises from the combinatorics of the possible board positions. The other key feature of this game is that it has more than 2 players, which allows for a range of strategic phenomena that do not arise in the restricted class of 2-player, constant-sum games.

I will introduce the basic motivations and techniques of empirical game theory. The major challenge in applying any sort of game-theoretic analysis to a game of this size is to identify a tractable subset of the game that captures interesting strategic interactions. I argue that strategic independence is likely to be a key part of the solution to this problem, and suggest several alternatives for exploiting it. Finally, I will present data from preliminary experiments in chaturanga that suggest the presence of both strategic independence and strategic interactions.

Sponsored by

AI Lab