Crisis de inteligibilidad probatoria y debido proceso en el Derecho penal: límites epistémicos de la IA en la imposición de sanciones

This article is grounded in the proposal of dogmatic criminal-law criteria for evaluating algorithmic evidence, based on the premise that the legitimacy of a conviction requires that inferences be subjected to meaningful adversarial scrutiny and to fact-based reasoning that can be externally reviewed. Through critical analysis, the study exposes the epistemic limits of artificial intelligence capable of producing, transforming or interpreting information that is incorporated as evidence in a criminal trial. The article also defines the crisis of evidentiary intelligibility as a deficit that prevents technical processes from being translated into reasons that are reviewable by the parties and by the court. To address this deficit, it proposes a minimum threshold of intelligibility structured around three dimensions: traceability, an objectively characterizable account of the system, and contestability. It concludes that opaque and materially decisive inferences lack epistemic efficacy to operate as an admissibility limit and as a determinative basis for conviction.