Breakthrough Research Reveals Neural Operator Mechanisms Behind AI Reasoning

For years, the scientific community has grappled with one of the most formidable challenges in artificial intelligence: understanding exactly how artificial neural networks develop the capacity for reasoning.

Recently, a research team led by Principal Scientist Research Fellow Jianyu Duan—formerly of Central South University and now with the AEQ AI Research Institute in New Zealand—reported a major breakthrough. Following 16,800 simulation experiments and multiple rounds of neural network training and validation, the team demonstrated cognitive neural operator structures and operational laws predicted by Duan's theory of neural thinking mechanisms, which was developed on the basis of his 1987 paper, A Preliminary Study on the Principles of Thinking in Meaning Activity.

The work reports a transition from raw computation to interpretable cognitive operation at the level of neuron activation functions, weight parameters, and functional neural operator structures. The experiments map language reasoning circuits and provide a parameter-level route for explaining mechanisms behind neural network reasoning.

According to the report, the findings have also been reproduced by a team at the Central South Institute of Applied Science and Technology. To support further verification and international collaboration, related source code, experimental results, and analytical reports have been made available through an open GitHub repository.

Dual Experimental Approaches

Programmatic mechanism simulation. The team constructed a transparent, explainable neural network reasoning system based on theoretically proposed language-reasoning neural operator structures. The system demonstrated complete natural language reasoning chains through structured neural operators.

Neural network training experiments. Multiple explainable neural network models were trained and analyzed across tasks involving propositional logic, predicate logic axioms, syllogistic reasoning, fuzzy linguistic reasoning, abstract concept formation, and analogical reasoning. The trained networks showed functional neural operator structures consistent with the theoretical framework, often appearing as distributed or topological variants rather than manually assigned nodes.

Toward More Interpretable AI

The research emphasizes that cognitive neural operators may represent fundamental functional structures involved in neural network comprehension, reasoning, and cognition. The experimental results suggest that neural reasoning can be studied not only as black-box input-output behavior, but also through explicit operator-level mechanisms, including recognition, classification, comparison, gating, constraint propagation, and conclusion release.

From the theory's early formulation in 1987 to experimental validation in 2026, this work represents a long-term research effort to explain how meaning transformation and reasoning mechanisms can arise inside neural networks.