Premium
Interdependent decision strategies in zero‐sum games: A computer‐controlled study
Author(s) -
Messick David M.
Publication year - 1967
Publication title -
behavioral science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.371
H-Index - 45
eISSN - 1099-1743
pISSN - 0005-7940
DOI - 10.1002/bs.3830120105
Subject(s) - interdependence , minimax , computer science , task (project management) , construct (python library) , zero (linguistics) , decision theory , control (management) , game theory , artificial intelligence , operations research , mathematical optimization , mathematical economics , mathematics , statistics , economics , linguistics , philosophy , political science , law , programming language , management
While the theory of games has stimulated research into the nature of interdependent decision strategies it leaves much to be desired as a psychological theory. Furthermore, experiments which have been aimed at evaluating the predictive power of this theory have failed to use methods which permit the detailed analysis of interdependent decision processes. It is suggested that digital computers can be used to solve this methodological problem by having the machine interact, according to predetermined strategies, with human subjects. An experiment was conducted to investigate the structure of interdependent strategies in an effort to illuminate the psychological processes involved. In order to construct an experimental task in which both interdependence and experimental control of the form of the interdependence would be possible, a small digital computer was programmed to play zero‐sum games against human subjects. The machine was programmed to play the game according to three different strategies. The first of these was the minimax strategy. The second program employed a strategy which kept a record of the number of times the subject chose each response and used this information to compute, on each trial, an estimate of the expected loss associated with each of its choices. On each trial, the program chose the response having the least expected loss. The third program generated its choices in the same manner as the previous one, except that it only “remembers” the five most recent choices of the subject to use as estimates of choice probabilities in computing the expected losses. This program thus had a limited memory. Fourteen human subjects played a zero‐sum game against each of these programs for a total of 150 trials. The sequence of choices for each subject was analyzed in an effort to determine how the different strategies used by the machine affected the strategies used by the subjects. The analyses indicated that the decision strategies adopted by the subjects depended critically on the strategy used by the computer. The subjects' strategies were effective, although not strictly optimal, against those of the machine. The results reveal a level of complexity not previously observed in the study of interdependent decision processes and some implications of these results for theory construction are discussed.