The visual representation of Amazonian peoples. A study on the biases of Generative Artificial Intelligence

Mostajo Pantoja A.; Gomero G.; Barredo Ibáñez D. · Revista Austral de Ciencias Sociales (2026) · DOI: 10.4206/rev.austral.cienc.soc.2026.n50-09

Resumen

Generative Artificial Intelligence (GAI) can create works by constantly learning from documents, data and alternative representations. However, some critics point out that its training starts from accessible materials that, in themselves, present numerous elements of subjectivity. Following in the wake of studies critical of the IAG, this research analyzes the images generated by the DALL-E and Runway platforms on the representation of Amazonian indigenous peoples to identify bias. Four hundred images were analyzed in terms of number of people, gender, age, geographic location and composition characteristics. The main results show that more than half of the images from both platforms presented images in which the geographic space could not be determined, and no image depicted an urban environment. These findings evidence biases towards Amazonian peoples, which may perpetuate symbolic constructions determined by the simplification, stereotyping and distortion of these people.

Instituciones: Universidad Peruana de Ciencias Aplicadas