Recursive Echo Chambers, Memory Mirroring, and Semantically Unresolvable Pivoting may be Genuine Phenomena that Result when a Human Person interacts with a Large Language Model (LLM)

Abstract

Contemporary discussions of large language models (LLMs) frequently conflate linguistic coherence with semantic understanding or logical consistency. This paper challenges that assumption by developing a semiotic account of coherence in human–LLM interaction. Drawing on relational and hierarchical meta-models of sign activity, it argues that coherence is not a property of shared truth conditions or formal logic. Rather coherence is an emergent effect of relational alignment across heterogeneous semiotic systems. Human cognition and LLMs operate under distinct semiotic logics—embodied, indexical, and norm-governed in the former; distributional, non-indexical, and pattern-ruled in the latter—yet dialogical coherence can arise through recursive interaction without semantic convergence or shared interpretation. The paper introduces the phenomena of recursive echo chambers, memory mirroring, and semantically unresolvable pivoting as interaction-level effects that illustrate how coherence may be sustained locally while remaining globally indeterminate. It further argues that such coherence, precisely because it is experientially compelling while lacking shared semantic grounding, can give rise to cognitive traps. By locating coherence at the level of semiotic coupling, the analysis clarifies how fluent, meaningful-seeming dialogue can occur, and also break down, in the absence of machine understanding. The paper argues that because LLMs respond smoothly and mirror user intention, users can feel as if they are a partner engaged in meaning-making with the LLM, despite the absence of a real other person, creating a risk that fluent dialogue is mistaken for genuine, ethically-mediated mutual understanding.

Keywords