Autopoietic Zombie: A Neural Network as a Communication System without a Subject

Large language models (LLMs) produce meaningful language yet lack human consciousness, creating a philosophical puzzle. The authors argue that common views—either anthropomorphizing AI or dismissing it as a "stochastic parrot"—are inadequate. By reviewing how theorists like Chalmers, Block, Luhmann, Tononi, Searle, and Nagel distinguish phenomenal consciousness (inner experience) from communicative consciousness (functional interaction), and drawing on Luhmann's systems theory, they show that a generative neural network is structurally homologous to a communicative system: it recursively processes meaning but has no phenomenal experience. The concept of an "autopoietic zombie" captures this.

Study at a glance

Abstract

The rapid development of large language models (LLMs) has confronted philosophy with a conceptual question: how should we conceptualize a system that demonstrates striking semantic productivity yet fundamentally lacks human consciousness? The dominant answers in both academic and public discourse oscillate between two extremes: the anthropomorphization of AI or its reduction to a “stochastic parrot”. This article proposes a methodological approach to transcend this dichotomy. First, the authors conduct a terminological inventory of the concept of “consciousness”, which reveals a structural isomorphism in the works of theorists such as Chalmers, Block, Luhmann, Tononi, Searle, and Nagel. Each of them arrives at the necessity of distinguishing between internal qualitative experience (phenomenal consciousness) and functional-communicative activity (communicative consciousness). Subsequently, the authors elaborate on this established distinction by drawing upon N. Luhmann’s systems theory of communication. It is demonstrated that a generative neural network is structurally homologous to a Luhmannian communicative system: it is capable of the recursive processing of meaning, yet fundamentally lacks phenomenal experience. To capture this status, the concept of the “autopoietic zombie” is introduced. Simultaneously, the authors substantiate the thesis that the emergence of LLMs radically transforms the original Luhmannian framework. For the first time, the communicative system of a neural network acquires the capacity to operate without being “totally dependent” on phenomenal consciousness. A fundamental asymmetry is posited: the human is simultaneously an observer and a subject, whereas the neural network is a subjectless observer that vastly surpasses the human in the communicative dimension. This publication constitutes the first part of the study. In the second part, it is proposed to make a transition from a substantialist vocabulary to a relational-processual ontology, offering a positive description of the neural network as a “Radically Other observer”.