Artificial Intelligence-Assisted Criminal Profiling: A New Perspective On Positivist Criminology

Abstract

Abstract

The theme addresses the structural continuity among positivist criminology, actuarial criminal policy, and AI profiling, focusing on the transition from physical to algorithmic control. It questions how AI-assisted profiling reproduces positivist deterministic logic and which mechanisms conceal this continuity under technical neutrality. The hypothesis states that algorithmic surveillance modernized the positivist paradigm, shifting biological determinism to data determinism. The objective is to demonstrate the evolution of penal determinism, analyze facial surveillance and algorithmic racism in Brazil, and propose treating AI output as mere circumstantial evidence, per constitutional guarantees and MJSP Ordinance 1,122/2026. Using the hypothetical-deductive method, this study conducts a literature and documentary review, analyzing CESeC/DPU reports, NIST technical studies, and STJ case law. Results proved that facial recognition error rates are up to 100 times higher for Black, Indigenous, and Asian individuals. In Brazil, 90% of arrests via this technology in 2019 involved Black people. Furthermore, the STJ established a precedent (HC 1,059,475/SP) prohibiting unsupervised generative AI reports as criminal evidence due to a lack of epistemic reliability.  Therefore, AI use in the penal system must be restricted to the investigative phase as circumstantial evidence. This approach demands transparency, auditability, and the preservation of the digital chain of custody to prevent violations of the adversarial principle and the presumption of innocence.