Mitigated Deployment Strategy for Ethical AI in Clinical Settings

This paper is a perspective piece that discusses a mitigated deployment strategy for ethical artificial intelligence (AI) in clinical settings, particularly focusing on diagnostic tools. It addresses the issue of poor model generalisability that can disadvantage underrepresented subgroups due to unrepresentative training data. The authors propose a selective deployment approach that excludes poorly performing subgroups, contrasting it with their advocated mitigated deployment strategy, which incorporates safety nets within clinical workflows. This strategy emphasizes human-AI collaboration and mandates postmarket evaluation to enhance model performance across diverse subgroups. The paper also examines the current regulatory landscape, highlighting the role of the Medicines and Healthcare products Regulatory Agency (MHRA) in ensuring safety and efficacy in AI deployment. A case study on the deployment of AI in NHS dermatology illustrates the practical application of mitigated deployment, showcasing its alignment with ethical principles and regulatory requirements. Overall, the paper contributes to the discourse on AI fairness in healthcare.