Algorithmic racism: Racial perception and socioeconomic dimensions in digital image banks

Abstract

Algorithmic racism can be understood as the reproduction of racist stereotypes and practices in computational mechanisms, such as search results and advertisements, due to the biased training of algorithms and databases that replicate and intensify structural racism. Recent studies show how digital image banks reproduce several of these stereotypes, contributing to the perpetuation of oppression and microaggression against subalternate groups. Intending to understand this process, the aim of this study was to investigate the representation of racial and socioeconomic perceptions in digital image banks. Searches were carried out in Freepik, Pexels, and Pixabay banks, using the keywords Poverty, Misery, Wealth, and Money, which are indicators of low and high socioeconomic status, respectively. These words were validated in a survey carried out by the researchers to find which words are most associated with socioeconomic indicators. Free image banks and keywords in Portuguese were chosen to bring the research method closer to the reality of the behavior of most Brazilians. The searches on the three platforms totaled 6200 images, independently evaluated by three judges who also assigned a valence (positive, negative, or neutral) to each one of them. In the preliminary analyses, although the words Poverty and Misery are considered indicators of low socioeconomic status, a significant amount of images (about 40%) were evaluated by the judges as medium and high status. In the results of the Wealth and Money indicators, 60% of the images are illustrations or photos of material goods, and among the few people that appear, most are white (78%). There is still a long way to go against the structural problems of society and their reflexes on algorithms, but studies like this are important to raise questions about the content we produce and consume, as well as the way we browse the internet.