Artificial Intelligence, Epistemology, and Race

Semiotic Issues

Abstract

This article challenges the discourse of rupture often associated with artificial intelligence, arguing that image generation models tend to recycle forms and meanings that predate the technologies themselves. It examines the mechanisms through which racial bias manifests in images produced by DALL·E. To this end, a series of tests were conducted linking the figurativization of human bodies to professions socially invested with varying degrees of prestige—such as doctor, university professor, civil engineer, hospital cleaner, construction worker, and waste picker—observing the regularities that emerged in the figurative distribution. The results indicate that, through statistical operations, the system organizes and hierarchizes recurrent combinations of figures and visual traits, associating socially devalued professions with Black bodies and specific plastic markers, such as concealed curly hair. Such markers tend to be absent from representations associated with higher-prestige professions. A relevant aspectual distinction is also observed: while Black bodies linked to devalued professions are figurativized under the modality of doing, amid labor activity, bodies associated with prestigious occupations are predominantly modalized through being, in static bodily positions. This configuration resonates with historically sedimented patterns in Western iconography, contributing to the stabilization of a racist semiotic norm. The article therefore argues for the need to develop a semiotics of artificial intelligence capable of critically analyzing the meanings generated by these systems and contributing to the development of more plural technologies grounded in epistemological perspectives from the Global South.