AI-ASSISTED CRIMINAL INTERROGATIONS: A HUMAN-IN-THE-LOOP FRAMEWORK FOR TRANSCRIPTION, COMPLIANCE MONITORING AND POST-INTERVIEW ANALYSIS

The growing use of artificial intelligence in criminal justice has intensified debate around reliability, transparency, and procedural fairness, particularly in the highly sensitive context of criminal interrogations. Although recent AI advances enable analysis of speech, behavior, and interaction patterns, prior research consistently shows that inferential applications—such as deception detection, emotion recognition, or predictive judgments of credibility— remain scientifically contested and risk introducing bias, automation dependence, and undue influence into investigative decision-making. In contrast, established interrogation research emphasizes accurate documentation, procedural compliance, and non-coercive interviewing as foundational to investigative integrity and evidentiary reliability. This paper proposes a human-in-the-loop framework for AI-assisted criminal interrogation rooms that intentionally excludes predictive or judgment-oriented functions. Instead, the framework prioritizes non-inferential, supportive AI capabilities designed to enhance transparency and structured review while preserving human authority. These include automated transcription and structured summarization of interviews, contextual visualization of interactional trends without evaluative labeling, consistency and timeline analysis across multiple interviews, and automated indicators for monitoring rights notification and procedural compliance. The framework introduces a modular system architecture and dashboard-oriented review layer that separates interview recording from post-interview analysis. This design mitigates automation bias, reduces the risk of coercive influence during questioning, and supports accountable review by investigators, courts, and defense stakeholders. Overall, the proposed approach offers a practical and ethically grounded model for responsible AI integration in criminal interrogations.