Reasoning by Agents in Computer Bridge Bidding

Abstract

The authors propose an agent oriented model for a bidder in the auctioning stage of bridge. Each agent selects a bid according to the criteria: Cooperate with the partner to get maximum profit and compete against opponents to minimize loss. Since bridge auction is a task of imperfect information, each agent has hypothetical reasoning ability and generates images of other players’ hands by abduction from the observed bidding sequence. This paper shows a framework for reasoning about each others’ knowledge and the details of analysis on typical examples. It is shown that the difference between one’s own real hand and its image in a partner’s knowledge motivates an agent to continue bidding. We also analyze an example of reasoning by an agent to select a sacrifice bid where the expected score of the bid is better than the score of an opponent’s possible contract. Experimental results show that the reasoning by the agent is flexible enough to play with a human partner and other computer bridge programs.

Acknowledgments

The authors thank the faculty, staff and students of Tokyo University of Technology for the opportunity to perform this research, especially Professors M. Shioya, W. Miyao, K. Ibuki and K.Fuchi, and students who participated in the project as graduate thesis research. The authors are grateful for advice from Mr. F. Nishino, Fujitsu Laboratories Limited, and Dr. K.Sato, Hokkaido University, as well as the help in using CHIP provided by Dr.W. O’Riordan and Mr.M. Rigg of ICL, and for assistance with ECLiPSe provided by PARC of Imperial College.