One of the most interesting areas in Artificial Intelligence (AI) is the area in which a machine is taught to play a game against an educated opponent. The basic premise in all of the reported techniques is that the machine is informed of the rules of the game, which are encoded efficiently. In this paper, we consider the scenario in which a Learning Mechanism (LM) is given the task of playing a game without being aware of the rules of the game. It is neither aware of what constitutes a valid or invalid move. The LM learns the rules of the game and the strategies with which it should play as it makes the moves. It accomplishes this by processing the responses it gets from the game Environment - which serves as an informed teacher. The entire game is modeled in a novel setting where the salient differences between the Agents and Environments are erased, but in which the players and the game board are considered as abstract entities interacting with each other. In this regard, we believe that our paper presents pioneering work.

Additional Metadata
Conference Joint 9th IFSA World Congress and 20th NAFIPS International Conference
Batalov, D.V. (Denis V.), & Oommen, J. (2001). On playing games without knowing the rules. In Annual Conference of the North American Fuzzy Information Processing Society - NAFIPS (pp. 1862–1868).