Artificial Intelligence in the Interrogation Room

Abstract

Introduction: Interrogation techniques have undergone significant evolution from inquisitorial methods to modern rapport-based approaches, with the 2021 Mendez Principles establishing new ethical standards. However, the emergence of artificial intelligence raises new questions: how do AI-based emotion analysis systems align with criminal procedural guarantees and the human rights framework?

Objectives: The study aims to provide a comprehensive legal and technological analysis of the INTU-AI system in light of the international and Hungarian regulatory framework.

Methodology: The author employs an interdisciplinary approach combining legal analysis (UNCAT, ECHR, Hungarian Criminal Procedure Code), technological evaluation (examination of Garcia et al.’s INTU-AI system), and critical review of psychological literature (false confessions, limitations of emotion analysis). The analysis extends to the differing procedural positions of witnesses, victims, and suspects, the heightened risks for vulnerable groups, and the validation results of the system.

Results: The analysis revealed that the INTU-AI system in its current form does not meet the scientific, legal, and ethical standards required by the Mendez Principles. The 67% F1-score and testing on simulated data raise questions about scientific validity. The system cannot distinguish between interviewer and interviewee voices, does not operate in real-time, and the reliability of the emotion analysis component is limited. False confession statistics (29% of DNA-based exonerations in the USA involved false confessions) and heightened risks for vulnerable groups (25% of false confessors have intellectual disabilities) raise serious concerns.

Conclusion: The author proposes a moratorium on the use of AI-based emotion analysis in suspect interrogations until the technology reaches appropriate validation levels and an adequate legal framework is established. However, in the context of complainant/victim interviews, experimental implementation of the system could be considered. Future research should focus on validation using real interrogation data, protection of vulnerable groups, and development of appropriate legal safeguards.