The generative illusion of Large Language Models: Language between human subjectivity and algorithmic computation
Author(s)
Date Issued
2025
Type
article
Volume
12
Issue
02
Start Page
97
End Page
122
Abstract
In recent years, Large Language Models (LLMs) have raised radical issues regarding generativity. This article examines the relationship among language, natural intelligence, and artificial intelligence (AI) in artistic practices such as writing, with regard to the interplay between generativity and creativity. It investigates whether – and to what extent – such systems can participate in the generative processes that characterise human language, thus bringing the classical notions of generativity into dialogue with the operational dynamics of LLMs. AI’s “meaning simulation” strategies will be examined through a playwriting experiment carried out with ChatGPT. We will show that AI generativity remains at the level of surface structures, without fully accessing the semionarrative transformation that defines the parcours génératif, while LLMs yield a “generative illusion,” keeping within the discursive level only. The impression of creativity arising at the intersection of computational recombination and human semantic investment is a hybrid artefact that reveals our interpretive habits and the machine’s generative constraints. As a result, a distributed but asymmetrical model of generativity emerges. Machines provide statistically significant discursive forms; humans provide semantic depth, narrative transformation, and cultural resonance. The article traces the tension between two registers of meaning-generation: one formal, combinatorial, and distributive; the other temporal, intentional, and transformative. We propose to name this tension the double register of generativity. This double register exposes why comparisons between human and artificial modes of meaning-making often founder on category mismatch. The machine’s outputs are not deficient because of a mere lack of competence, but because they operate according to a different ontological logic, namely, distributional semantics and contextual attention rather than chronogenetic and enunciative operations. The result, as Claudio Paolucci suggests, is that the machine compels us to confront the machinic essence of human beings themselves: we too are generative machines, though endowed with temporality, intentionality, and value-oriented transformations.