The Collective Pet Unconscious: Balancing Intelligence and Individuality in Populations of Learning-Enabled Virtual Pets

A conceptual architecture is presented in which a population of learning agents share a dynamic collective knowledge base but also retain individual memories, biases and learning systems. Knowledge learned by agents goes into the collective base, but each agent also has a “personality filter” that controls how the collective base affects its individual base. An application in virtual worlds is discussed, in which behaviors are taught by users to virtual dogs and the shared knowledge is used to speed up learning.