AI as Meaning-Constructor: Syntax, Structure, Coherence, and the Role of the Interpreter

Recent advances in artificial intelligence, particularly large language models, have renewed debates about whether AI systems genuinely understand meaning or merely simulate it. This paper offers a novel conceptual framework for clarifying this debate by distinguishing between meaning construction and meaning possession. Drawing on and extending pattern-theoretic insights associated with Ulf Grenander, the paper introduces a three-circle model—syntax, structure, and coherence—to explain how contemporary AI systems generate outputs that are widely experienced as meaningful. The central claim is that modern AI statistically superimposes syntactic order, structural constraint, and global coherence into a single probabilistic surface, thereby producing stable Gestalts of intelligibility. These Gestalts are not random artifacts but converge on a 'sweet spot' shaped by human language use, feedback, and normative expectations. However, while this intersection suffices to construct forms that invite interpretation, it does not suffice for meaning as a lived, owned, or situated phenomenon. The paper argues that meaning emerges only at the level of interpretation, through agents who already possess prior Gestalts grounded in embodiment, practice, and historical continuity. AI systems therefore function as meaning-constructors, not meaning-givers, while humans remain the locus of semantic attribution. This distinction explains both the remarkable power of AI to simulate understanding and the persistent intuition that something essential is missing. By relocating meaning at the interface between statistical construction and human interpretation, the paper provides a non-reductive, non-mystical account of AI's capabilities and limits, with implications for ongoing discussions of intelligence, understanding, and artificial agency.