Paesaggi latenti, territori sintetici. In viaggio nell’Intelligenza Artificiale generativa

Pictures of territories produced by Generative Artificial Intelligence are increasingly widespread, contributing to novel forms of (possibly distorted) geographic and touristic representation. Humanistic research usually focuses on the phenomenal and pragmatic level but is excluded from the analysis of the technical processes which generate these pictures, given the field’s limited access to such expertise. Understanding these processes, however, is often necessary to explain a variety of effects which can be observed. The goal of this paper is to provide a bridge between the technical details on how the machine concretely works and some possible theoretical insights from the humanities, and chiefly semiotics, to analyze these operations. To this end, Baudrillard’s description of “simulation” and the Borgesian idea of a map which is co-extensive with its territory are shown to provide useful ways to conceptualize the technical working of these machines; special attention is paid to the construction of latent spaces that occur in latent diffusion models, which exemplify such hybrid map-territory entities, and to the denoising process which enables to “travel” within them.