Do Transformers Dream of Real Sheep？Exploring the Unconscious of LLMs through Žižek’s Psychoanalytic Semiotic Lens

Abstract

In his 2020 work, Hegel in a Wired Brain, Žižek explores whether digital machines can comprehend the unconscious as the surplus of language. According to him, even if future digital machines could decode and comprehend all human thoughts and discourse, they would remain incapable of capturing the unconscious dimension that is retroactively constituted within the chain of signifiers. In recent years, with the remarkable advancements achieved by large language models (LLMs)—such as ChatGPT (Generative Pre-trained Transformer)—in the field of natural language processing, these models have emerged as the most promising approaches in artificial intelligence and have become significant subjects of philosophical reflection. Consequently, the pressing question now arises: Can LLMs comprehend, or even possess, their own unconscious?To address these questions,this paper will first provide a brief technical overview of LLMs, then outline Žižek’s interpretation of the unconscious via Lacanian psychoanalytic semiotics. Based on several intriguing experiments—such as questioning LLMs about their understanding of the coffee joke and requesting them to generate similar jokes—this study will analyze the unconscious of LLMs. The conclusion will be cautionary: while LLMs do not yet possess a human-equivalent unconscious, they can comprehend and partially access this paradoxical space.Rather than focusing on the traditional issue of whether AI can understand semantics as opposed to syntax, this paper centers on the unconscious third dimension of “undead” that lies between semantics and syntax, and whether LLMs can understand this retroactively generated paradoxical viod within the chain of signifiers.