Textual Instances of Algorithmic Racism

Abstract

In this article, I examine artificial intelligence (AI) algorithms as structuring texts that process information, shape narratives, and reproduce racial stereotypes, thereby sustaining dynamics of exclusion. I present a theoretical model composed of three interconnected textual instances: the Prompt-Text, which expresses the user’s intent; the Response-Text, which materializes the generated output; and the Algorithm-Text, the invisible layer that defines the system’s internal rules. I analyze a case in which an AI-generated image depicted a Black woman holding a gun in a favela, despite the original prompt specifying a positive figure. To explore this bias, I developed a Python simulation that demonstrates how the algorithm automatically associates racialized terms with violence, revealing the systematic reproduction of stereotypes. I also examine social media comments that naturalize such distortions, shifting responsibility to the user while obscuring the algorithm’s central role in constructing representations. I argue that these systems operate as discursive agents that reinforce historical inequalities and hierarchies. I defend algorithmic rewriting as an ethical and political act, supported by strategies such as diversifying datasets, applying bias filters, ensuring greater transparency, and promoting regulation. I conclude by reaffirming the thesis that guides this work: the algorithm is a text, and, as such, it can – and must – be rewritten to dismantle exclusionary logics.