Authors: Milan Verhoeven, Arjun Kale, Sofia Mendes
Abstract: Artificial General Intelligence (AGI) aims to develop systems capable of performing intellectual tasks across diverse domains at human-level or beyond. A key challenge in this pursuit is designing universal knowledge models and cognitive architectures that can represent, integrate, and reason over heterogeneous information in a coherent and scalable way. This paper reviews current approaches to universal knowledge representation, including symbolic systems, neural networks, probabilistic models, and hybrid frameworks. We examine prominent cognitive architectures such as SOAR, ACT-R, OpenCog, and contemporary transformer-based models, analyzing their structural assumptions and learning capabilities. The paper discusses how universal knowledge models may bridge perception, reasoning, planning, and language understanding within a unified architecture. Limitations, open research problems, and possible integration pathways are explored. We argue that future AGI systems will likely require hybrid cognitive architectures that combine structured symbolic reasoning with self-organizing neural representations and memory-centric models.
Keywords: Artificial General Intelligence, Universal Knowledge Models, Cognitive Architectures, Symbolic AI, Neural Networks, Hybrid Systems, Knowledge Representation
Full Issue
| View or download the full issue | PDF 153-171 |