Non-localAdaptation ofArtificial Predators and Prey
Daniel Ashlock · 2005
Non-local adaptation istheacquisition ofgeneral skill at acompetitive task. General skill isdefined asskill against a broadspectrum ofopponents rather thanjustthose anagentencountered during itstraining orevolution. Biological dogmasuggests that evolved creatures should be adapted onlytotheir environment andtheopponents they encounter during evolution. Ifthis dogmaapplies todigital evolution, thenweshould notobserve non-local adaptation inagents trained withevolutionary computation toperformsomecompetitive task. A numberofprevious studies havefoundnon-local adaptation inprisoner's dilemma, in amodelofcompetitive exclusion inplants, andinavirtual robotics task. Thispaperexamines non-local adaptation ina virtual predator-prey system. Onehundred distinct predator-prey lineages areevolved for250,000 timesteps, saving anintermediate population attimestep100,000. Predators andpreyfromdistinct lineages areplaced in competition. Forthefourpossible comparisons inwhich thetypeofonecompetitor isheldconstant andtheother is varied fromless tomoreevolved, astatistically significant increase inability toacquire foodisseeninthemoreevolved agents.