Tignum 2 is a computer system for declarer play at the game of contract bridge. Tignum 2 currently performs better at declarer play than the strongest commercially available program. ' On 5000 randomly generated deals (including both suit contracts and notrump contracts), *This material is based on work supported in part by an AT&T Ph.D. scholarship to Stephen J. J. Smith, by Maryland Industrial Partnerships (MIPS) grant 501.15, by Great Game Products, by ARPA grant DABT 63-95-C-0037, and by the National Science Foundation under Grants NSF EEC 94-02384 and IRI-9306580. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the funders. 'It is probably safe to say that the Bridge Baron is the best program in the world for declarer play at contract bridge.
The latest world-championship competition for computer bridge programs was the Baron Barclay World Bridge Computer Challenge, hosted in July 1997 by the American Contract Bridge League. As reported in The New York Times and The Washington Post, the competition's winner was a new version of Great Game Products' Bridge Baron program. This version, Bridge Baron 8, has since gone on the market; and during the last three months of 1997 it was purchased by more than 1000 customers. The Bridge Baron's success also represents a significant success for research on AI planning systems, because Bridge Baron 8 uses Hierarchical Task-Network (HTN) planning techniques to plan its declarer play. This paper gives an overview of those techniques and how they are used.
A computer program that uses AI planning techniques is now the world champion computer program in the game of Contract Bridge. As reported in The New York Times and The Washington Post, this program--a new version of Great Game Products' The classical approach used in AI programs for games of strategy is to do a game tree search using the well-known minimax formula (eq. 1) The minimax computation is basically a bruteforce search: If implemented as formulated here, it would examine every node in the game tree. In practical implementations of minimax game tree searching, a number of techniques are used to improve the efficiency of this computation: putting a bound on the depth of the search, using alpha-beta pruning, doing transposition-table lookup, and so on. However, even with enhancements such as these, minimax computations often involve examining huge numbers of nodes in the game tree. Because a Bridge hand is typically played in just a few minutes, there is not enough time for a game tree search to search enough of this tree to make good decisions.
A computer program that uses AI planning techniques is now the world champion computer program in the game of Contract Bridge. As reported in The New York Times and The Washington Post, this program -- a new version of Great Game Products' BRIDGE BARON program -- won the Baron Barclay World Bridge Computer Challenge, an international competition hosted in July 1997 by the American Contract Bridge League. It is well known that the game tree search techniques used in computer programs for games such as Chess and Checkers work differently from how humans think about such games. In contrast, our new version of the BRIDGE BARON emulates the way in which a human might plan declarer play in Bridge by using an adaptation of hierarchical task network planning. This article gives an overview of the planning techniques that we have incorporated into the BRIDGE BARON and discusses what the program's victory signifies for research on AI planning and game playing.
AI planning techniques are beginning to find use in a number of practical planning domains. However, the backward-chaining and partial-order-planning control strategies traditionally used in AI planning systems are not necessarily the best ones to use for practical planning problems. In this paper, we discuss some of the difficulties that can result from the use of backward chaining and partial-order planning, and we describe how these difficulties can be overcome by adapting Hierarchical Task-Network (HTN) planning to use a total-order control strategy that generates the steps of a plan in the same order that those steps will be executed. We also examine how introducing the total-order restriction into HTN planning affects its expressive power, and propose a way to relax the total-order restriction to increase its expressive power and range of applicability.