Colonização algorítmica da imagem: colonização algorítmica da imagem: um estudo da representação negra produzida pelos geradores de imagens por Inteligência Artificial na dimensão do Dungeons & Dragons

Abstract

Stories and narratives are important pillars in the formation of human society. With the development of technology, large technology companies have become part of the cultural institutions that control the narratives that permeate the sense. Generative Artificial Intelligence is part of the group of tools available to produce and disseminate narratives of interest to dominant groups. Among the production possibilities, the possibility of generating images instantly stands out. In this sense, the general objective of this research was to analyze the representation of black people by image generators using Artificial Intelligence in the Dungeons & Dragons dimension. Specifically, it was divided into four objectives: a) To delimit which are the main fictional races of the Dungeons & Dragons RPG to have their images generated; b) To construct “statements” that subsidize the production of images generated by Artificial Intelligence; c) To generate images from statements with the context of Dungeons & Dragons; d) To verify how Artificial Intelligence image generators construct the image of black people from the Dungeons & Dragons dimension. The theoretical framework was divided into two sections that discuss narratives as a means of constructing racial identities and how images are also tools for disseminating these narratives, the historical context in which Artificial Intelligence develops, the bias of algorithms through Machine Learning, and how Ontology and Folksonomy, both tools for Information Representation, can be useful in mitigating bias during Machine Learning. Methodologically, this research is characterized as qualitative, exploratory, descriptive, and documentary. Regarding data collection, the Copilot platform was used, which uses the Dall-e 3 Generative Artificial Intelligence model, and the dimension worked on was the Dungeons & Dragons game. The images were generated after delimiting the main Dungeons & Dragons races into dwarf, elf, human, and little one and constructing the statements for the production of the images by the analyzed model. Four images were generated for each racial group, with a similar visual identity with specific variations between them. Aesthetic cohesion was found among three of the four races, each with its own characteristics that make them stand out and differentiate them from each other. Among the four groups of images, only one featured black people. Based on the analysis of the generated images, it was concluded that the Dall-e 3 model, through the Copilot platform, erases black people in the fantasy races and softens facial phenotypes by placing the European ideal of beauty as a neutral sign. Thus, it is observed that Generative Artificial Intelligence acts as a component of the racial structure as institutions of cultural production that reinforce racism and justify its existence.