A Bridge Tutoring System

Proposer: Ian Frank, ianf@aisb

Suggested Supervisors: Alan Bundy, Helen Pain, Ian Frank

Principal goal of the project: To produce a system capable of tutoring the skill of playing card combinations in the game of Bridge.

Description:

The Finesse system for forming Bridge declarer play plans has been under development at DAI since 1991. This system uses the approach (developed as `proof-planning' in the Mathematical Reasoning Group) of restricting the possible actions considered by the planner to a set of high level actions, or tactics, which capture the techniques used by human experts when playing the game. To select the best move in any situation, a tree of the possible sequences of tactics is generated and then `interpreted' by an algorithm which decides which of the branches at the root of the tree is most likely to produce the best results.

A tutoring system based on Finesse would allow a user to test their ability to play card combinations. The first stage of the project would therefore involve writing a shell which uses Finesse's plans to generate responses to the user's moves as they select cards to play. The next stage would add the ability to compare the user's line of play to the best line of play selected by Finesse, and to critique the user's performance.

The exact design of the tutoring system will need to be decided after evaluating a number of alternative interface issues. For example, for each card play the user could either be presented with a list of sensible possibilities, or be allowed a free choice. The list of possible plays could also be in the form of tactics or in the form of individual card plays. Further, comments could be made about each of the user's moves, or alternatively only when they deviate from the optimal line of play. Other useful features would be to check the user's estimation of the chances of the success of their line of play against that produced by the planner, and to help beginners by describing their chosen card plays in the higher level language of tactics. Explanations of some of the basic rules and concepts of Bridge such as following suit and the notion of finessing, might also be profitably incorporated.

As an extension to the basic system, it would be useful to enhance the system's ability to justify its choice of best plan. Finesse can already produce textual explanations of why one line of play is better than another, such as `Finesse the Queen: this leads to two tricks when the King is with West,' but in a tutoring system, it would be useful to allow the user to query such explanations with the simple question `Why?' (e.g., by clicking on a button). The system should then use the plans generated by Finesse to demonstrate why the chosen line of play is optimal. For example, the explanation above would be illustrated by generating one situation where the King is with West and one where the King is with East, and allowing the user to examine the subsequent play with all four hands visible. The cards which are of no consequence to the outcome of the play should be replaced with `x's to concentrate attention on the significant features. An important part of this extension would be the automatic generation of card distributions between the defenders to be used as exemplars when explaining the plan.

Resources Required: A workstation and Prolog. Access to the Finesse planning system.

Degree of Difficulty: The initial stage is of average difficulty, but the extensions allow scope for challenging work.

Background Needed: Attendence at AIED module. Knowledge of the Bridge game in general would be advantageous. Some proficiency in Prolog will also be required, for understanding and interfacing to the existing planning system.

Degree Programmes Suitable: MSC AI/CS4 AI/M4

References: