Title: Unveiling Quantum-Like Structures in AI Linguistics: Signs of Evolutionary Convergence Between Human and Machine Cognition

Abstract:

This study reports on cognitive assessments involving conceptual combinations, utilizing specific Large Language Models (LLMs) as the subjects of inquiry. Our initial experiments, conducted with ChatGPT and Gemini, demonstrated a significant violation of Bell’s inequalities. This outcome suggests the existence of a "non-classical probability model," wherein the probabilities involved fail to adhere to Kolmogorov’s axioms. In a subsequent test, also employing ChatGPT and Gemini, we detected "Bose-Einstein statistics" within the word distributions of extensive texts, a finding that contrasts with the intuitively anticipated "Maxwell-Boltzmann statistics."

Notably, these results parallel earlier outcomes from cognitive tests involving human participants, as well as information retrieval analyses on large-scale corpora. Collectively, these observations indicate a "systematic emergence of non-classical quantum-like structures" within conceptual-linguistic realms, irrespective of whether the cognitive agent is biological or artificial. While LLMs are traditionally categorized as neural networks due to historical classification, we propose that a more fundamental mode of knowledge organization occurs within the distributive semantic structure of the vector spaces constructed upon the neural network architecture. It is this meaning-laden structure that facilitates a phenomenon of evolutionary convergence between human cognition and language—developed gradually through biological evolution—and LLM cognition and language, which arise much more swiftly via self-learning and training processes. We examine various aspects and illustrative examples that substantiate this hypothesis and introduce a unifying framework to account for the pervasive quantum organization of meaning observed in our findings.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC