Use of Artificial Intelligence in Forensic Medicine and Criminal Investigation: A Critical Analysis

Abstract

Applications include (but are not limited to) the use of artificial intelligence to screen post mortem images, classifing histopathology, prioritising digital evidence, identifying a person, detection of manipulated media and to support decision making in investigations. While these applications offer the potential for increased speed, consistency and access to specialist expertise, they also involve unique dangers: forensic products may impact on determinations of cause of and manner of death, and affect a person's liberty and reputation. This paper aims to critically evaluate the technical, ethical, and legal factors of AI in forensic medicine and criminal investigation. It separates assistive measurement from autonomous inference and assesses systems on scientific validity, on the performances of subgroups, on explainability, on provenance, on chain of custody, on privacy and contestability. Several factors are revealed in the analysis, none of which can be resolved by high accuracy of the aggregates: base rates, domain shift, feedback loops, adversarial manipulation, undocumented software changes, and automation bias can turn small technical errors into big miscarriages of justice. A risk leveled assurance model is proposed that may use a range of low-risk tools on a periodic basis, but a system of biometric identification, predictive policing, cause of death suggestion and evidence-authentication will need to be tested independently, authorised by humans, validated, disclosed, monitored and provide actionable opportunities for challenges. AI should be used to complement and inform, not supplant, informed forensic decision-making.