Structural Cognition, Thought-nets, and the Limits of Large Language Models An Axiomatic Theory of Understanding and Thinking

Human cognition appears to be simultaneously stable and revisable, local and global, fast and reflective. Perception delivers rich sensory input; concepts carve that input into linguistic units; judgments form propositional structures; understanding integrates these structures into coherent patterns; and thinking sometimes restructures them altogether. Yet contemporary philosophy of mind and cognitive science offer no unified account of how these components fit together into a single architecture. This paper develops an axiomatic theory of cognitive structure that treats cognition as the construction, stabilization, and transformation of relational structures. Six axioms define perception, concepts, judgment, understanding, thinking, and a multi-layer “Thought-net” that encodes long-term worldviews. On this basis, I derive a seven-stage transformation engine that explains how new information is parsed, mapped, tension-tested, transformed, evaluated, and reintegrated across layers. The resulting architecture yields a three-level Thought-net: L1 immediate structures, L2 operational structures, and L3 worldview structures. I argue that this structural framework resolves several standing puzzles about understanding, belief revision, conceptual change, and cognitive stability. It also provides a principled lens on current debates about large language models (LLMs). In contrast to recent Synthese literature that explores the possibility of linguistic intentionality, semantic content, or understanding in LLMs, I argue that current LLMs lack the mechanisms required for structural cognition: they do not stabilize relational structures over time, detect internal structural tension, or engage in structure-transforming operations analogous to human thinking. The theory thus yields both a unified model of human cognition and a diagnostic framework for evaluating artificial systems.