Ontological Folding: A Paradigmatic Foundation for the Six-Path Unified Cognitive Operating System of AGI

Contemporary research on artificial general intelligence (AGI) encompasses six major research lines: neural connectionism, symbolic axiomatics, dynamical evolution paradigms, metacognitive introspection, ontological cross-domain frameworks, and low-compute lightweight paradigms, which often develop alongside embodied intelligence and neuro-symbolic integration. Mainstream treatments regard these lines as mutually competing, mutually exclusive technical alternatives, or adopt shallow modular assembly strategies for integration. This paper adopts ontological folding as its core operational framework, arguing that these six branches are not independent and contradictory schemes, but local partial manifestations of a single cognitive ontology under different imposed constraints. On this basis, we construct a sixpath unified cognitive operating system paradigm: neural networks serve as the cognitive curtain field, a top-level meta-axiom set provides rigid reasoning constraints, a fourthorder self-referential metacognitive closed loop acts as the governance layer, and the tripartite ontological substrate — the world prefers parsimony, all stable structures are mathematizable, topological unification — underpins the whole architecture. Real physical observational data function as the external calibration anchor. This framework incorporates the respective strengths of all six paradigms and establishes a novel inference system distinct from probabilistic generative models. Dominant autoregressive and diffusion language models optimise for text distribution fitting and token generation, amounting to incremental improvements at the generative surface level. By contrast, the sixpath unified paradigm realises reconstruction at the ontological layer: the core of inference is no longer probabilistic random walk, but topological path pre-checking, retrieval and convergence within cognitive space. The bottleneck of this paradigm lies not in computational power or matrix scale, but in ontological engineering: the self-consistent construction, boundary demarcation and hierarchical differentiation of the top-level meta-axiom spectrum. It is a representative highdimensional, lowcompute approach.