Neural Network of a Cognitive Crow: An interacting map based architecture

Studies on animal reasoning can serve as a very relevant guide in building artificial systems that autonomously evolve their cognition and develop new skills in structural/functional coupling with their environments. New Caledonian crows display sophisticated abilities in fashioning tools and using them to acquire otherwise unobtainable food rewards. We may consider this an act of creativity, during which a cognitive agent reasons about things that do not yet exist but could exist as a result of his actions. However, the computational basis of such cognitive processes have remained elusive. Taking the example of tool manufacture by Caledonian crows, we explore an interacting self organizing map based architecture for virtually manipulating neural activity so as to achieve ones goals in novel situations. Simulation results demonstrate how interactions between sensorimotor space, action space and work space can be exploited in new situations to find action/action sequences that result in known intermediate states (sub goals) that in turn lead to the overall goal.