World Models Without Subjectivity: Indexical Cognition, Functional Objectivity, and the Limits of AI Understanding

We revisit the debate over whether contemporary large-scale artificial models (LLMs and related architectures) possess or can possess world models, and we interrogate the presupposition that a world model must be an `objective', first‑person‑independent replica of the world. Building on a synthesis of cognitive-developmental observations and functional analyses, we argue: (1) human world models are primarily indexical and inter-subjective—they arise from embodied, perspectival, and social coordination processes rather than from a pre-existing objective representation; (2) the notion of a fully objective, perspective‑free world model is a theoretical limit rather than an achievable substrate for any finite intelligence; and (3) AI systems can and do develop functional world models—structures that support prediction, counterfactual reasoning, and action coordination—without inheriting human‑style subjectivity or indexicality. We propose a rigorous typology of world models, operational criteria for identifying functional world models in artificial systems, and an empirical program to validate these criteria. The paper clarifies conceptual confusions, sketches formal desiderata for world‑model tests, and outlines implications for AI research paths toward robust predictive agents.