The Effect of Tag Recognition on Non-local Adaptation
Submitted to CEC2004

Daniel Ashlock and Brad Powers

Abstract PDF eprint

Populations of agents are evolved to perform noisy iterated prisoner s dilemma on a toroidal grid. The agents consist of a finite state machine specialized for playing iterated prisoner s dilemma with a simple tag recognition capability. The populations are allowed to evolve for 10,000 generations and the world is stored every 500 generations. Populations from these samples are placed in competition with populations from generation 10,000. This procedure is repeated for varying levels of overall mutation rate, with and without tags, and varying frequencies of tag related mutations. Non-localized adaptation is seen in these populations, however, tags seem to slow the acquisition of non-localized adaptation. Although the concept of non-localized adaptation is not a widely accepted phenomenon in biology, these results suggest that it does happen and that the effect is persistent in the face of changes in mutation rate and in the face of increased task complexity. Also, the study shows patterns of tag space usage by populations with tag recognition enabled. The population tends to have a predominant tag most of the time with punctuated periods of increased tag space usage that most likely correspond to invasion of the population by an opportunistic agent with a new tag identifier. This study serves to provide more evidence for and give a more detailed view of non-localized adaptation.