Beyond stochastic parrots: Lacanian reflections on LLMs as superior masters of the symbolic

Summary

Large Language Models (LLMs) are exceptionally skilled at manipulating language and symbols, making them superior masters of the Symbolic register in Lacanian psychoanalysis. However, because they lack a body and subjective experience, they have no access to the Imaginary or the Real, and are fundamentally incapable of lack, desire, or genuine creativity. The paper argues that LLMs operate as closed, pre-packaged cognitive systems, effective only for locked symbolic tasks, and lack the dissipative openness needed to confront the Real. Intelligence is redefined as managing lack and desire, not benchmark performance. Ethical warnings about over-reliance on LLMs and recommendations for hybrid designs informed by psychoanalysis conclude the argument.

Study at a glance

Abstract

Abstract This article makes the case that, in the Lacanian sense, Large Language Models (LLMs) should be viewed as superior masters of the Symbolic. Thanks to their vast training data and disembodied architecture, LLMs achieve hyper-efficient mastery of the chain of signifiers, outperforming most humans in linguistic, discursive, and inferential tasks. However, precisely because they operate exclusively within the Symbolic register, they lack access to the Imaginary and the Real, and are constitutively barred from Lacanian “lack,” drive, desire, and embodied subjectivity. While Freudian and Jungian insights remain useful, the Lacanian framework provides the most powerful lens, describing both the exceptional symbolic fluency of LLMs and their structural incapacity of real subjectivity or transformational creativity. The paper shows that LLMs function well in locked symbolic tasks but lack the dissipative openness required to confront the Real, representing closed, pre-packaged cognition. This is done by integrating my eco-cognitive difference between locked and unlocked strategies. Therefore, intelligence is redefined as the ability to manage lack and desire rather than as benchmark performance. While recognizing the inevitable limitations of applying psychoanalytic categories to non-embodied systems, I conclude with ethical warnings about an over-reliance on LLMs and recommendations for future hybrid designs informed by psychoanalysis.