The Expected-Outcome Model of Two-Player Games

The Expected-Outcome Model of Two-Player Games deals with the expected-outcome model of two-player games, in which the relative merit of game-tree nodes, rather than board positions, is considered. The ambiguity of static evaluation and the problems it generates in the search system are examined and the development of a domain-independent static evaluator is described.

Comprised of eight chapters, this book begins with an overview of the rationale for the mathematical study of games, followed by a discussion on some previous artificial intelligence (AI) research efforts on game-trees. The next section opens with the definition of a node's expected-outcome value as the expected value of the leaves beneath it. The expected-outcome model is outlined, paying particular attention to the expected-outcome value of a game-tree node. This model was implemented on some small versions of tic-tac-toe and Othello. The book also presents results that offer strong support for both the validity of the expected-outcome model and the rationality of its underlying assumptions.

This monograph is intended for specialists in AI and computer science.

I Introduction

1 Prelude: Why Study Games?

2 Overview: What Lies Ahead?

3 Background: What is Already Known?

3.1 Preliminaries

3.2 Heuristic Evaluation Functions

3.3 Control Strategies

3.4 Summary

II The Model

4 Proposal: What is the Basic Model?

5 Support: Does the Model Work?

5.1 Analytical Evidence

5.2 Empirical Evidence

III Conclusions 75

6 Contributions: What's been Accomplished?

7 Implications: Where Might the Model Lead?

8 Reprise: Why Study Games?

A Standard Evaluation Functions

B The Random Sampler