Qualitative Information and the Limits of Syntactic Engines: Why LLMs Lack Semantic Understanding

Read full text Summarise with VIRAT
Authors: Hankyeol Kim
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This paper investigates the epistemic limits of artificial intelligence regarding semantic understanding by proposing a rigorous conceptual distinction between ‘syntactic manipulation’ and ‘semantic understanding’. While Large Language Models (LLMs) demonstrate advanced syntactic proficiency, I argue that genuine understanding requires a foundational coupling with ‘qualitative information’ rooted in phenomenal experience (qualia). Moving beyond the classical critiques of Searle and Harnad, this study introduces a novel mechanistic framework: semantic understanding as the ‘synchronization’ of linguistic forms with subjective, qualitative states. To demonstrate this, I present the original ‘xozif paxs toqyz’ thought experiment, which reveals that purely formal manipulation of symbols—no matter how statistically sophisticated—cannot bridge the gap to intentional meaning without the integration of qualitative dimensions. The analysis further utilizes the ‘Octopus Test’ to illustrate how current AI architectures are inherently desynchronized from the ontological grounding of language. I conclude that the boundary between simulating understanding and possessing it remains firm, as genuine semantic grounding is inextricably tied to the qualitative aspects of conscious experience.

Fields of Study & Subjects

Philosophy and Theoretical Science Cybernetics and Technology in Society Embodied and Extended Cognition

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
30 March 2026
Language
EN
Document type
Preprint