Unguided Categorization, Direct and Symbolic Representation, and Evolution of Cognition in a Modified Connectionist Theory

Severe constraints apply to the architecture of any system which uses large numbers of components to deliver complex functionality. If not taken into account, these constraints can invalidate conclusions drawn from simulation of cognitive subsystems. The pattern extraction hierarchy is a connectionist architecture which can satisfy these constraints. Within the architecture, functionality is partitioned at a high level in a manner which is consistent with the range of qualitative functionality available from neurons. A system with this architecture can perform unguided categorization resulting in internal representations experienced as mental images. These internal representations function to generate alternative behaviors and select the most appropriate. Functional descriptions of cognition can therefore be created which make use of grounded representations supported within a connectionist architecture. A scenario for the step by step development of human cognition is proposed on the basis that biological brains have the pattern extraction hierarchy architecture. This scenario demonstrates how the ability to generate behaviors appropriate for very complex situations uses a constant succession of mental images and has developed through the interaction of verbal and tool making skills.