Toward Unified Cognitive Architectures for Artificial General Intelligence: Integrating Symbolic Reasoning, Deep Learning, and Adaptive Memory
Abstract
The relentless evolution of artificial intelligence has sparked intense debate regarding the optimal paradigm for achieving Artificial General Intelligence (AGI). While deep learning has attained unprecedented success in perception, pattern recognition, and natural language generation, it continues to suffer from fundamental limitations, including algorithmic opacity, poor data efficiency, fragile out-of-distribution generalization, and hallucination. Conversely, classic symbolic artificial intelligence excels at exact deductive logic, compositional abstractions, and verifiable dynamic reasoning, yet remains hindered by brittleness, the symbol grounding problem, and an inability to scale to noisy real-world sensory inputs. Recent paradigm shifts emphasize that neither sub-symbolic neural connectionism nor pure symbolic manipulation alone is sufficient for AGI. This review paper provides a rigorous, comprehensive analysis of unified cognitive architectures that systematically integrate deep learning sensory-perceptual modules, symbolic reasoning engines, and dynamic, adaptive memory structures (encompassing episodic, semantic, and working vector buffers). We critically evaluate historical cognitive architectures such as SOAR, ACT-R, and LIDA, alongside modern neuro-symbolic hybrids, neuro-vectorial memory networks, and Global Workspace Theory (GWT) meta-controllers. By examining state-of-the art mechanisms for neural-symbolic translation, vector-symbolic architectures (VSA), continuous hyperdimensional computing, and plastic memory consolidate routines, this work maps out the architectural taxonomy, structural integration bottlenecks, and algorithmic tradeoffs governing modern hybrid AI systems. Ultimately, we propose a general unified framework for AGI design, detailing its theoretical significance, engineering requirements, real-world application domains, and critical future research directions.
KEYWORDS: Artificial General Intelligence, Neuro-Symbolic AI, Cognitive Architectures, Adaptive Memory Systems, Hyperdimensional Computing, Global Workspace Theory.
Full Text:
PDF 109-122Refbacks
- There are currently no refbacks.