Structural Limitations of the Use of Large Language Models in the Self-Diagnosis of Mental States

Abstract

This article examines the problem of self-diagnosing mental states through large language models in the context of cognitive capitalism, psychocentrism, and psychopolitics. It argues that LLMs constitute a significant element of the contemporary dispositif of power and knowledge, participating in the production, validation, and normalization of psychocentric self-descriptions. Methodologically, the study draws on conceptual-analytical and discourse analysis, as well as on the theoretical framework of Jacques Lacan’s structural psychoanalysis. The aim of the study is to substantiate the thesis that LLMs are structurally incapable of occupying the position of the analyst, despite their ability to simulate high-quality counselling and therapeutic responses to user queries. It is argued that the machine-generated response does not disrupt the subject’s fantasmatic self-description, but merely returns it in the form of machine-validated knowledge. Particular attention is paid to the effect of machine validation, whereby the user’s initial query acquires the appearance of external expert confirmation. The scholarly novelty of the study lies in its synthesis of a psychoanalytic perspective, contemporary theories of power, and the philosophy of technology, which makes it possible to substantiate the conclusion that the subject becomes doubly enclosed within the register of the Symbolic: through the subject’s own fantasy and through an adaptive machine-generated narrative.