Debates about whether large language models (LLMs) genuinely understand the language they produce often rest on the assumption that understanding necessarily entails conscious experience. In this article, we argue that this assumption is unwarranted. By drawing a principled distinction between structural meaning and phenomenal access to meaning, we defend the thesis that genuine understanding does not require consciousness. We situate our position within the Turing tradition, while responding to classical objections such as Searle’s Chinese Room by showing that they conflate semantic competence with conscious awareness. After distinguishing symbols, information, and meaning, we argue that meaning constitutes an autonomous, abstract domain characterized by contextuality, potentiality, and non-classical composition. We then show how LLMs operate within this domain by exhibiting sensitivity to semantic structure rather than merely manipulating symbols. Drawing on results from quantum cognition, we further argue that both human and artificial semantic intelligence are naturally modeled by quantum-like structures, revealing a convergence between biological and artificial cognition at the level of meaning. We conclude that LLMs instantiate a genuine, though non-phenomenal, form of semantic intelligence, and that acknowledging this requires a revision of inherited concepts of understanding rather than an appeal to consciousness as a necessary condition.

Sassoli de Bianchi, L. and Sassoli de Bianchi, M. (2026). Understanding without Consciousness: Quantum Structures and the Autonomy of Meaning. To be published.