This paper introduces METAGAMER, the first program designed within the paradigm of Meta-Game Playing (Metagame) (Pell 1992a). This program plays Metagame in the class of symmetric chess-like games (Pell 1992b), which includes chess, Chinese-chess, checkers, draughts, and Shogi. METAGAMER takes as input the rules of a specific game and analyses those rules to construct for that game an efficient representation and an evaluation function, for use by a generic search engine. The strategic analysis performed by METAGAMER relates a set of general knowledge sources to the details of the particular game. Among other properties, this analysis determines the relative value of the different pieces in a given game. Although METAGAMER does not learn from experience, the values resulting from its analysis are qualitatively similar to values used by experts on known games, and are sufficient to produce competitive performance the first time METAGAMER actually plays each new game. Besides being the first Metagame-playing program, this is the first program to have derived useful piece values directly from analysis of the rules of different games. This paper describes the knowledge implemented in METAGAMER, illustrates the piece values METAGAMER derives for chess and checkers, and discusses experiments with METAGAMER on both existing and newly generated games.

/pdf/a-strategic-metagame-player-for-general-chess-like-games-x9396b62f1.pdf

A strategic metagame player for general chess-like games

Estimating sentence types in computer related new product bulletins using a decision tree

A heuristic approach to optimum experimental design

: The authors describe the context and the constituent modules of a large-scale programming system, the Quasi-Optimizer. Its objectives are (a) to observe and measure adversaries' behavior in a competitive environment, to infer their strategies and to construct a computer model, a descriptive theory of each; (b) to identify strategy components, evaluate their effectiveness and to select the most satisfactory ones from a set of descriptive theories; (c) to combine these components in a quasi-optimum strategy that represents a normative theory in the statistical sense. Also discussed are certain properties of decision trees which are the primary representational structures of strategies in the computer. The verification of these properties, such as identity, equivalence and similarity between two decision subtrees, enable us to eliminate redundancies in the decision trees.

/pdf/on-some-issues-concerning-optimization-and-decision-trees-3pumz247cs.pdf

On Some Issues Concerning Optimization and Decision Trees.

The task of credit assignment is interpreted as the ability of a system to: identify and distinguish strategy components; associated with such components, in different regions of the domain of confrontation, good and poor outcomes of a sequence of actions prescribed by the strategy; improve the boundaries of these regions in minimizing the errors of misclassification of strategy actions; estimate the overall quality of a strategy; and provide information for a metastrategy that shifts the domain of confrontation to those regions in which the strategy studied is most proficient and thereby raises the effective quality of the strategy The program QO-4, the fourth module of the large system QA (the quasi-optimizer) which can accomplish the above tasks within certain limitations, is described

A conceptual framework and a heuristic program for the credit-assignment problem

A note on computing the asymptotic form of a limited sequence of decision trees

: The authors, together with a large and varying number of collaborators, worked on a long-term project aimed at how decisions are and should be made under uncertainty and risk. Learniing programs of different types have come to the center of their attention, both in the course of trying to simulate human recognitive behavior and in constructing wholly machine intelligence-oriented competitive strategies. They describe an interactive environment in which an arbitrary number of human and machine strategies, up to a total of eight, can be pitted against each other. The game of draw poker was selected as the vehicle of these studies since it shares many characteristics with real-life decision-making tasks. This project has come to an end. The authors also describe their current work on automatic analysis and synthesis of strategies. Although the systems being developed are highly context-independent, they are able to utilize their experience gained in the previous project. (Author)

http://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA115380&Location=U2&doc=GetTRDoc.pdf

Bede McCall

Papers

A note on computing the asymptotic form of a limited sequence of decision trees

A conceptual framework and a heuristic program for the credit-assignment problem

A Multi-Strategy Gaming Environment.