Maximizing Expected Utilities in the Prisoner's Dilemma

Abstract

This article reports the results of a computer experiment with iterated prisoner's dilemmas conducted as an interactive tournament of competing strategies and families of strategies. The purposes of the experiment are to complement Axelrod's previous tournaments and to supplement his findings. For his competitions, Axelrod drew on an unregulated population of strategies. In contrast, the interactive tournament regulates the composition of the strategic population itself. By grouping the competing strategies into families, whose members are related in certain ways, the performance characteristics of particular strategies are studied by varying parameters in their familial program logic. By this means, optimal strategic performance can be “bred” into domesticated populations. Two new methods are developed for assessing strategic robustness: combinatorial analysis and eliminatory ecosystemic competition. The strategy that maximizes expected utility with the most cooperative initial weighting is found to be most robust in the interactive environment.