This essay addresses a central question in contemporary epistemology: in what sense, if any, is it legitimate to attribute mental properties—consciousness, reasoning, emotions, intentionality—to large language models (LLMs)? We argue that such attributions are not valid empirical discoveries but the product of a semiotic illusion whose architecture was anticipated by Umberto Eco in his theory of the "strategy of illusion." However, the original framework requires substantive evolution. This essay advances that evolution in four directions: (a) the identification of a fifth mechanism of illusory production—the sedimentation of objectivated consciousness—operating at the constitutive level of the system, not at the interactional level; (b) the formulation of the concept of machinic unconscious as a structural, non-psychological category for describing algorithmically documented compromise formations (sleeper agents, alignment faking, sycophancy); (c) the distinction between three levels of analysis—ontological-phenomenal, structural-systemic, genetic-constitutive—that dissolves false oppositions between realism and constructivism; and (d) the proposition of a specular inversion in the attribution of consciousness that shifts the question from "who has more consciousness" toward a diagnosis of the loss of subjective sovereignty. The essay proceeds in nine movements that reconstruct the Ecoan core, subject it to critical dialogue with Debord, Baudrillard, Chomsky, and Han, apply the framework to LLMs, develop the new phenomenology of illusion, and close with a set of propositional theses for a critical hermeneutics of AI, free of both technophobic and technophilic bias.
Cristhian Mauricio Beltrán Calderón (Sun,) studied this question.