Answers are grounded in this paper and cite their sources.
Artificial intelligence (AI) is increasingly being considered or used in criminal justice activities, including predictive policing, investigative analytics, facial recognition, recidivism estimation, correctional management and forensic mental health assessment. These applications may improve information processing and support resource allocation, however, their outputs can influence surveillance, arrest, bail, sentencing, parole, detention and access to treatment. Errors and biased predictions may therefore affect liberty, dignity and equality. This critical narrative review examines AI at the intersection of criminal justice, criminology, sociology and mental health. It distinguishes clinical mental health assessment, violence risk assessment and recidivism prediction, evaluates evidence on predictive performance and validation, and considers policing, crime prevention, judicial decision making, social inequality and human rights. The reviewed literature indicates that historical justice data may reproduce unequal policing and prosecution practices, while model performance may not transfer across populations or jurisdictions. Recent systematic evidence on machine learning violence prediction reports limited external validation, incomplete calibration reporting and substantial risk of bias. Comparative examples from the United States, the United Kingdom, Europe and India demonstrate that the meaning and consequences of algorithmic systems depend on legal frameworks, institutional practices and social context. The review argues that an algorithmic score is a probabilistic estimate, not a diagnosis, proof of dangerousness or substitute for individualised legal reasoning. If used, AI should function only as a bounded decision support aid subject to independent validation, meaningful human oversight, transparent explanation, avenues to challenge decisions, data minimisation and continuous audit. In India, constitutional guarantees of equality, privacy and personal liberty make local validation and procedural safeguards essential. The review concludes that adoption should depend on demonstrated benefit, public accountability and compatibility with human rights rather than technological novelty.
No takes yet. Share an insight, caveat, or question.
Shahid Sulaiman (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: