Resumo:
The Northeast region of Brazil is often underrepresented or marked by negative stereotypes in audiovisual content and literature. With the rise of generative artificial intelligences (GAIs), it remains unclear whether these tools can generate a representation of the unique characteristics of each state in the region or if they produce images that reinforce stereotypes. In this context, the present study aimed to analyze the representation of the Brazilian Northeast in images generated by GAIs. The research addressed concepts of artificial intelligence, machine learning, semiotics, algorithmic racism, and cultural identity. Images were generated using the AI tools Leonardo AI and Dall-E 3 for each of the nine states in the region. The signs present in these images were then analyzed and compared to data about the states. A total of 72 images were analyzed during the experiment, in which 92% contained generic and interchangeable signs across the states in the region. Leonardo AI portrayed the Northeast as predominantly rural in 64% of the generated images, while Dall-E 3 represented the Northeast as a beach destination in 78% of the images. Both tools simplified and reduced the cultural, economic, and social diversity of the northeastern states, neglecting their unique characteristics. The results highlighted the need to improve machine learning algorithms to ensure a deeper and more contextualized understanding of the signs that reference the specificities of each region.