Reinforcement Learning in 20Q Game with Generic Knowledge Bases

Chun Hei Lo, Luyang Lin

Abstract: 20Q, invented by Robin Burgener in 1988, is a computerized game of twenty questions that began as a test in artificial intelligence. The game asks the player to think of something and the system will then try to guess what they are thinking of with twenty yes-or-no questions. Due to the highly flexible and volatile game setting, designing a best questioning strategy for the system is not trivial. In the proposal, we hope to provide possible research directions, particularly under the formulation of the problem as reinforcement learning. We would also investigate on methods and potential challenges of incorporating the use of a knowledge base for the game. The presentation can be found at https://drive.google.com/file/d/1lZ7tTvTFfspJ1P_TEpJvG5TPu0lFSVW6/view?us=sharing.