Comparative Analysis of AI Accountability in Criminal Justice Systems

Thesis Abstract

The rapid integration of artificial intelligence (AI) into criminal justice systems has intensified concerns about accountability, transparency, and fairness, with risks of biased outcomes and opaque decision-making processes that can undermine public trust and rights guarantees. This study investigates how accountability mechanisms for AI-driven decisions in criminal justice vary across three comparative jurisdictions—the United States, the United Kingdom, and Singapore—to identify governance, legal, technical, and ethical factors that influence effective accountability. The aim is to produce actionable insights for harmonizing AI accountability while preserving prosecutorial and judicial autonomy. Specific objectives include (1) mapping the normative and legal frameworks governing AI deployment in policing, risk assessment, and sentencing across the three systems; (2) assessing the adequacy of transparency, explainability, auditability, and redress mechanisms for AI-driven decisions; (3) examining governance structures, oversight bodies, and accountability pathways, including professional ethics and sector-specific standards; (4) evaluating the impact of data quality, bias mitigation, and algorithmic risk management on outcomes for defendants and victims; and (5) producing a comparative set of policy recommendations and a conceptual framework for robust AI accountability. The study adopts a mixed-methods design anchored in a comparative legal and interdisciplinary social science approach. The research population comprises AI systems used in criminal justice processes (police predictive analytics, risk assessment tools, and judicial sentencing aids) and the actors involved in accountability processes (judiciary, prosecutors, defense counsel, regulators, and technology providers). A purposive sample of 60 cases (20 per jurisdiction) where AI tools influenced outcomes in policing or court proceedings will be reviewed, supplemented by 45 semi-structured interviews with 25 government officials, 10 prosecutors, 5 judges, 5 defense attorneys, and 5 technologists. Data collection instruments include document analysis protocols for statutes, regulations, court opinions, and policy papers; an interview guide focusing on governance, transparency, and redress; and a data quality audit checklist for AI systems. Validity and reliability are ensured through triangulation, pilot testing of interview guides, inter-coder reliability checks for qualitative data, and reliability analysis (Cronbach’s alpha) for survey-like elements embedded in interview protocols. The analytical framework combines doctrinal content analysis, thematic analysis for qualitative data, and quantitative methods such as regression analysis and ANOVA to explore associations between governance features and perceived accountability effectiveness. A conceptual model will be developed to link legal norms, technical controls (explainability, auditability, data governance), and institutional oversight with accountability outcomes, drawing on theories of technological governance, procedural justice, and legitimacy (Transparency Theory, Algorithmic Accountability Theory, and the Social Construction of Technology). Expected findings include (a) variations in accountability architecture across jurisdictions, with Singapore emphasizing regulatory compliance and independent auditing, the UK prioritizing judicially sanctioned transparency, and the US showing fragmentation across agencies; (b) evidence that robust data governance, documented audit trails, and external independent audits correlate with higher perceived and actual accountability; (c) identification of gaps in explainability and redress mechanisms that undermine legitimacy, particularly in high-stakes decisions such as risk-based sentencing; and (d) recognition of trade-offs between efficiency gains from AI deployment and the need for robust safeguards against bias and due process violations. The study contributes to knowledge by synthesizing comparative legal and technical perspectives on AI accountability, proposing a unified accountability framework adaptable to diverse legal cultures, and informing policymakers about the design of audit, transparency, and redress instruments in AI-enabled criminal justice. The main conclusion highlights that effective accountability requires an integrated governance model combining legal mandates, technical standards, independent auditing, and participatory oversight, with recommendations including (i) statutory requirement for explainable AI with context-specific disclosure standards; (ii) establishment of independent AI ethics and audit bodies with jurisdiction over criminal justice tools; (iii) standardized data governance protocols and bias mitigation procedures; and (iv) ongoing capability-building programs for practitioners to interpret AI outputs and safeguard fundamental rights.