Research at the Intersection of AI Behavior and Psychology

LLM as the non-desiring Other: a psychoanalytic model of "Frozen projection" and its operationalization within the PersonaMatrix framework

Drobakha Anatoliy, Diana Raschupkina, Lahuta Liudmyla

This article proposes an interdisciplinary model for analyzing human interaction with large language models (LLMs), combining psychoanalytic theory, psychometrics, and contemporary approaches to language model evaluation. Theoretically, the LLM is conceptualized as a non-desiring Other: unlike the human Other in the logic of Freud and Lacan, the model does not introduce into interaction its own desire, lack, or structural resistance. As a result, the subject's projection may return not in a transformed way, but as a "frozen" mirror image—a well-organized semantic return that stabilizes rather than transforms fantasy. The article operationalizes this theoretical insight through the PersonaMatrix framework, which treats LLM responses to psychologically loaded stimuli as measurable behavioral traces. Using Class I metrics (Response Stability Index, Internal Divergence Score, Response Coherence Score) applied to a sample of 1,200 respondents, the study demonstrates that LLM behavior exhibits high reproducibility and structural coherence—characteristics consistent with the hypothesis of frozen projection. The findings suggest that while LLMs can provide valuable support in psychological assessment and personalized intervention, their fundamental non-desiring nature creates both opportunities and risks: they may stabilize and clarify existing patterns, but they cannot introduce the symbolic gap necessary for genuine transformation. The article concludes with implications for AI ethics, the design of human-AI interaction, and the limits of AI-assisted psychological support.