If we assume that a semiotic theory is composed of two interacting faces—i.e. a conceptual face and a topological face—, the confrontation between them becomes an instrument of validation, falsification, control, and advancement of such a theory. Our epistemological hypothesis is that the possibility and practice of a confrontation between the conceptual form of a theory and its topological form represents a crucial feature of any critical science. To validate our hypothesis, we engage in a methodological exercise. More specifically, we draw on Kalevi Kull’s work, and particularly on his articles and books devoted to the different semio-topological models of biosemiotics and general semiotics: the semiosphere, the Umwelt, the niche, and so on. In search of the internal coherence of his conceptual system, we test several topological varieties, eventually escaping what seemed to us to be an aporia, i.e. to invent a final topological form, by means of a hybridization of the semiospheres and the Umwelten.
What do interactions between users and machines reveal about our semiotic and linguistic relationship to the world? What new observables do the productions of generative editorial AIs offer compared to human productions? What does a generative editorial AI like ChatGPT ultimately do? In this article, we will explore an example of co-enunciation between user and machine that highlights the dynamics of co-reference: are we talking about the same thing when we speak with a machine? To understand the linguistic and semiotic issues of generative editorial AIs, like ChatGPT, for example, we must question both the modalities of text generation and the characteristics of the corpus processed by the machines. What we call LLM (Large Language Models) is a machine learning model capable of analyzing and generating texts. As with ChatGPT, training the algorithm involves positioning the processed texts relative to each other based on the differences identified within the corpus. The relevant criteria that allow us to identify differences then become the parameters through which the corpus is organized. By analyzing these criteria, we can then understand the linguistic and semiotic issues of human-machine interaction.
By becoming capable of interpreting and generating text, images, or videos about complex notions—without having been given their logical structure—so-called “generative” artificial intelligence algorithms (ChatGPT, Claude, Gemini, Stable Diffusion, Midjourney, DALL·E, Gen-2, or Sora…) pose a scientific challenge: what is intelligibility, if it can be produced or simulated by such systems? Theorists, artists, and users have attempted to account for this phenomenon through the notion of “latent spaces,” which refers to the grids through which algorithms “interpret” cultural patterns. These “grids” comprise tens of thousands of dimensions (each dimension serving as an axis of differentiation between patterns), in the service of an “interpretation” that can be both hermeneutic (recognizing a motif or, more generally, locating it within a network of proximities and distances with other motifs) and productive. To better assess what they are and what they can do, latent spaces have been compared to the world of Platonic forms, raising both epistemological and political issues. We propose an alternative comparison with Aby Warburg’s library project, since latent spaces share with it a method of classifying cultural motifs by similarity or “affinity”, allowing them to be positioned within a network of analogies capable of connecting all motifs—transcending medial or disciplinary boundaries, and in particular the distinction between scientific knowledge and fiction. This sheds light on how the intelligibility of a cultural motif may stem less from its logical articulation than from its position within a web of analogies.
The Paris School of Semiotics has the merit of having highlighted the importance of the theme of affectivity and the need for a semiotics of passions. The seeds of this type of reflection were found (among others) in Hjelmslev's theory and in particular in certain considerations relating to substance, which were further developed by Zilberberg in the direction of the tensive model. Our hypothesis is that developing these considerations by insisting on form would allow us to better grasp how structuralim conceptualized the issue of affectivity, and perhaps to reframe the differences between Hjelmslev’s and Zilberberg’s models.
Pockets Full of Memories premiered at the Centre Pompidou in 2001 as an art installation that invited the public to digitally scan an object in their possession and describe it through a touch-screen questionnaire. Images of the objects were projected on a large screen, spatially organized by an Artificial Intelligence neural-network algorithm so that each object’s distance from every other was determined by the semantic information collected through the questionnaire. The algorithm repeatedly reorganized the data, eventually arriving at an ordered state in which each object in the collection was spatially positioned relative to the others based on their metadata. Objects with similar descriptions were grouped closely together, while those with more distinct descriptions were placed farther apart. This spatial arrangement allowed viewers to observe how descriptive metadata influenced the classification and positioning of objects relative to one another. This paper describes the conception, design and implementation of the installation.
