Does Learning Imply a Decrease in the Entropy of Behavior

摘要: Shannon's information entropy measures of the uncertainty an event's outcome. If learning about a system reflects decrease in uncertainty, then plausible intuition is that should be accompanied by organism's actions and/or perceptual states. To address whether this valid, I examined artificial organism -- simple robot learned to navigate arena and analyzed outcome variables action, state, reward. Entropy did indeed initial stages learning, but two factors complicated scenario: (1) introduction new options discovered during process (2) shifting patterns environmental states resulting from changes robot's movement strategies. These lead subsequent increase as agent learned. end with discussion utility information-based characterizations learning.