On the Ethics and Practicalities of Artificial Intelligence, Risk Assessment, and Race.

Abstract

Artificial intelligence (AI) has been put forth as a potential means of improving and expediting violence risk assessment in forensic psychiatry. Furthermore, it has been proffered as a means of mitigating bias by replacing subjective human judgements with unadulterated data-driven predictions. A recent ethics analysis of AI-informed violence risk assessment enumerated some potential benefits, ethics concerns, and recommendations for further discussion. The current review builds on this previous work by highlighting additional important practical and ethics considerations. These include extant technology for violence risk assessment, paradigmatic concerns with the application of AI to risk assessment and management, and empirical evidence of racial bias in the criminal justice system. Emphasis is given to problems of informed consent, maleficence (e.g., the known iatrogenic effects of overly punitive sanctions), and justice (particularly racial justice). AI appears well suited to certain medical applications, such as the interpretation of diagnostic images, and may well surpass human judgement in accuracy or efficiency with respect to some important tasks. Caution is necessary, however, when applying AI to processes like violence risk assessment that do not conform clearly to simple classification paradigms.