Modelling of Heuristic Evaluation Strategies in Game Playing: Linear and Configural Effects in Othello
Modelling of Heuristic Evaluation Strategies in Game Playing: Linear and Configural Effects in Othello
Files
Publication or External Link
Date
1981
Authors
Phillips, Robert Vernon III
Advisor
Norman, Kent L.
Citation
DRUM DOI
Abstract
Psychological research on problem solving began with
Thorndike's work on trial and error learning with cats,
dogs, and monkeys. Kohler later initiated research with apes
which convinced him that problems could be solved with
insight. Through the 1940's, the study of human problem
solving focused on general principles (following the Gestalt
tradition) and S-R mechanisms to explain how people solve
problems.
The advent of computer technology in the 1950's spurred
research in artificial intelligence, game playing, and
problem solving. Formal definitions of problems outlined the
components of a constituting the problem
representation. This provided a framework for computer
scientists to mechanize problem Solving with algorithms of search. Computer scientists met with success in developing
programs to work on well-defined problems, such as games and
puzzles, where the components of the problem representation
are easily stated. Once the representation is adopted,
solution is a matter of search.
It has been shown that the efficiency of mechanized search
is aided by the use of a ''heuristic evaluation function"
(Nilsson, 1971), which has a form similar to psychological
models applied in research on human decision making and
judgment (Slovic and Lichtenstein, 1972). Samuel (1959),
used a regression model of human judgment based on the
knowledge of skilled checkers players in order to produce a
heuristic evaluation function for a checkers playing
program. Another model which can also be used to provide a
heuristic evaluation function is based on Anderson's (1962)
technique of functional measurement. This approach allows
estimation of subjective scale values for the levels of
information components relevant to playing a game.
In contrast to these linear models, Edgell (1978) has argued
that people can utilize configural information when making
judgments, an issue which has been avoided by most decision
modelling research. Samuel (1967) showed that use of
configural infermation by a heuristic evaluation function
can augment the skill of a checkers playing program, but the
question of whether human players use such information was not researched.
This paper reports one pilot
experiment and two other experiments which were
conducted to investigate whether
people do use configural information when evaluating
alternative moves in a game situation. The effects of game
experience, learning, and training on use of configural
information were examined. In addition, the research was
conducted in a game playing situation in order to address
the issue of ecological validity (Neisser, 1976) in
psychological research. As Newell and Simon (1972) have
argued, a good psychological theory of how a good chess
player plays chess should play good chess.