Predicting is not judging: Algorithmic systems of criminal risk and presumption of innocence in Latin America

Abstract

In 2023, a criminal court judge in Bogotá relied, in a precautionary ruling, on the output produced by PRISMA, a risk assessment algorithm used by the Office of the Attorney General, without the defense being able to examine either the variables considered or the weight assigned to them. This situation, recorded in Judgment T529/2023 of the Colombian Constitutional Court, serves as the starting point for this article. The paper argues that the incorporation of algorithmic risk assessment tools into Latin American criminal justice has not led to a rightsbased modernization, but rather to the automated reproduction of preexisting inequalities. The study follows a qualitative, doctrinal, and comparative approach, grounded in the analysis of case law, statutory sources, and specialized literature. Based on a review of Argentina, Colombia, Brazil, and Chile, the findings identify the functions assigned to these systems, their levels of transparency, and the tensions they generate in relation to due process guarantees.