|InterJournal Complex Systems, 589
|Manuscript Number: |
Submission Date: 20627
|Slow and Steady: Deliberately Computationally Inefficient Genetic Algorithms and Hard Problems|
Subject(s): CX.43, CX.67, CX.19
It is known from the work of Axelrod, Miller, Lindgren and others that the application of genetic algorithmic (GA) strategies to the Iterated Prisoner’s Dilemma yields varying results, depending upon the encoding used. In this paper I explore the results of using a new, “deliberately inefficient” encoding for the GA that allows for the slow evolution of strategies of varying complexity, and develop a quantification scheme to determine the degree of altruism of a given strategy. I find that, in general, altruistic strategies do not carry the day; furthermore, the exact mix of strategies in the long-term evolution of the system proves to be quite sensitive to both the error rate in the system (as expected), as well as to the mutation rate in the GA. The results of this work indicate promising possibilities for future applications of this encoding strategy to “hard” problems, such as protein folding.
|Submit referee report/comment|