A Hybrid Cognitive Architecture for AGI: Bridging Symbolic and Subsymbolic AI

This paper presents a novel hybrid cognitive architecture designed to advance Artificial General Intelligence (AGI) by integrating symbolic and subsymbolic AI approaches using graph neural networks (GNN). The architecture comprises several key components: a perception module using neural networks to process raw sensory data, a symbol grounding module to map subsymbolic features to symbolic representations, a symbolic knowledge base for storing facts and rules, a reasoning engine for logical deductions, and a learning module for updating the system based on experience. The GNN serves as an integration layer, connecting symbolic concepts and subsymbolic features to facilitate bidirectional communication, enhancing the system's ability to reason and generalize. This hybrid approach aims to combine the strengths of symbolic AI (logical reasoning) and subsymbolic AI (pattern recognition), offering a promising pathway toward achieving general intelligence, with potential applications in robotics, natural language processing, and visual reasoning.