Proceedings of the Annual Meeting of the Cognitive Science Society

Cognitive agents are continuously faced with new problems.To facilitate adaptation, emerging theories of neural reuse pro-pose that evolution might often favor re-purposing existingbrain structures for new functions. This paper presents a novelapproach to the study of neural reuse based on the evolutionof simulated agents in an object-categorization task. We arti-ficially evolve populations of dynamic neural networks to per-form two variants of a categorization task that alternate overevolutionary time. We find that populations become increas-ingly adaptive over repeated exposures to the tasks. Analysisof evolved networks reveals two types of equally-fit solutions:one that is specialized to a given task variant and does not adaptto changes easily; and another that is more general, in that itcan adapt to the other task with minimal change to its structure.Interestingly, we find that populations exposed to alternatingtasks spontaneously locate the latter type of structures.