Algorithms on Trial: Does Evaluative Probabilistic Reporting of Forensic Evidence Infringe the Presumption of Innocence?

Scientific evidence plays an important role in criminal justice. Recent technological developments including the use of AI and advanced computational forensic software have made possible forensic examinations and expert opinions that previously would have been impossible. Alongside benefits, the use in criminal trials of forensic evidence based on computational technologies such as Probabilistic Genotyping (PG) DNA, are posing difficult problems for courts and have been met with controversy. This study focuses on one important aspect of the criticism surrounding the use of PG DNA evidence, which relates to the probabilistic reporting of the forensic evidence results. It explores whether the use of likelihood ratios to report evaluative expert opinions infringes the presumption of innocence. This is a fundamental question that concerns not only evidence based on advanced computational technologies such as PG DNA but all forensic disciplines where the use of likelihood ratios and probabilistic assessments of the evidence are being actively promoted. This article posits that the criticism on the use of probabilistic methods for evidence evaluation encountered in legal practice, scholarly debate, policy and legal reform documents, is founded on misunderstandings of the role and limitations of the forensic evidence, of the processes involved in arriving at an evaluative expert opinion, as well as of the meaning and scope of the presumption of innocence itself. This study aims to contribute to an enhanced understanding of these fundamental issues which will lead towards a better regulation of AI and forensic algorithms across jurisdictions, without diminishing the impact of the scientific evidence in criminal proceedings and beyond.