BICA Project

The challenge addressed by this proposed research project is to design a new, hybrid cognitive architecture that will possess key features of human cognition including the capability of autonomous cognitive growth, the basic kinds of human memory, and social, communicational and emotional capabilities. The proposed GMU approach to this challenge is based on a cognitive architecture that integrates symbolic components and connectionist components from the outset, rather than addressing issues of integration at a later date.

This integration is achieved through the blending and interaction of innovative concepts in both the symbolic and connectionist components. On the symbolic side the key innovative elements are:

  1. a unique, central notion of “self”, and
  2. a consistent and formal representation system based on the general notion of a “schema”.

On the connectionist side the key innovative elements are:

  1. neuromorphic cognitive maps that provide for associative indexing of symbolic memories and path-finding in modeled cognitive spaces, and
  2. a functional mapping of the cognitive components onto brain structures that would, in principle, allow future elaboration to the level of neurons.

It is our belief that the ambitious goals of achieving human-level cognition cannot be attained without a clear and central notion of Self. Consequently, the notion of Self plays a central role in this project and is defined in a unique way: as an imaginary abstraction (rather than the cognitive system per se) and is implemented via a set of axioms corresponding to fundamental beliefs that an agent holds about its own Self and via an emergent lattice of mental states representing multiple instances of the Self.

The underlying schema-based representation formalism generalizes the principles underlying state-of-the-art cognitive architectures, including Soar, Act-R and Epic, raising them to a meta-level. The formalism is specifically designed to enable cognitive growth capabilities and to allow for indexing, retrieval, and organization by the connectionist components.

The project defines three neuromorphic cognitive maps to be elaborated as part of this effort:

  1. a contextual map for handling episodic memories,
  2. a conceptual map for handling semantic memories, and
  3. a separate map for emotional memories that may involve episodes as well as semantic knowledge.

These maps collectively provide for associative indexing and retrieval of stored memories, dynamic cognitive growth, and the path-finding in memory space necessary for contextual reinstatement, strategic free recall and episodic-to-semantic conversions.

The project specifies a functional mapping of these cognitive components onto brain structures. In particular, the mapping will cover the brain’s major functional areas associated with the limbic, medial temporal and frontal lobes, including the hippocampal complex, the nuclei of the amygdala and the hypothalamus, and other structures.

These four key elements of the proposed integrated architecture (“self”, “schema”, cognitive and functional mapping) are grounded in the current state of the art of cognitive psychology, cognitive and computational neuroscience, and cognitive modeling in artificial intelligence.