Subsymbolic Versus Symbolic Data Flow in the Meaningful-Based Cognitive Architecture

Abstract

The biologically inspired Meaningful-Based Cognitive Architecture (MBCA) integrates the subsymbolic sensory processing abilities found in neural networks with many of the symbolic logical abilities found in human cognition. The basic unit of the MBCA is a reconfigurable Hopfield-like Network unit (HLN). Some of the HLNs are configured for hierarchical sensory processing, and these groups subsymbolically process the sensory inputs. Other HLNs are organized as causal memory (including holding of multiple world views) and as logic/working memory units, and can symbolically process input vectors as well as vectors from other parts of the MBCA, in accordance with intuitive physics, intuitive psychology, intuitive scheduling and intuitive world views stored in the instinctual core goals module, and similar learned views stored in causal memory. The separation of data flow into the subsymbolic and symbolic streams, and the subsequent re-integration in the resultant actions, are explored. The integration of logical processing in the MBCA predisposes it to a psychotic-like behavior, and predicts that in Homo sapiens psychosis should occur for a wide variety of mechanisms.