Abstract This series of papers builds on the cognitive (including thinking) principles proposed by Jianyu Duan to conduct a series of studies on various cognitive activities. The theory suggests that cognitive processes are the formation and transformation of meanings, achieved through a series of operator operations. From this basis, a mathematical model of neural networks can be conceptually inferred theoretically, leading to the successful development and construction of specific parameters and structures of neurons. This includes various neural mapping operator circuits capable of recognition, perception, language, and reasoning, which have been successfully simulated on a computer. For the first time in human history, the specific parameters and circuit structures of neurons provide a comprehensible and clear explanation of the core cognitive mechanisms underlying why large language models (LLMs) and the human brain can achieve the transformation of meaning (including reasoning). The first paper will focus on the neural circuits and mechanisms of reasoning. The series will further reveal the mechanisms by which artificial and biological neural networks engage in multimodal cognitive activities (such as hybrid image-language tasks), demonstrating that artificial intelligence and human brain intelligence share common cognitive soft mechanisms. This provides a unified cognitive theory and research foundation for both. The series demonstrates that AI will catch up with and surpass human creativity, uncovering the black box of cognitive thinking functions in neural networks, providing a theoretical basis for developing AI with more advanced cognitive functions, and promoting the safe and widespread application of AI. This research will drive innovation and interdisciplinary integration in neuroscience, cognitive science, artificial intelligence, and philosophy.
Duan et al. (2025) studied this question.