Explaining explainable AI for healthcare: a Q-methodology study

Résumé fourni par la source

Technological innovations are being developed and introduced at a rapid pace to manage increasing demand on the healthcare system. Artificial intelligence (AI) tools promise to improve patient access to quality care and reduce work burden for staff. While healthcare stakeholders have shown increasing interest in AI, the complexity of healthcare provision makes it difficult to integrate these ‘black box’ technologies. Understanding and trusting these technologies is essential to address technical, medical, organizational, ethical, and legal issues, and allow stakeholders to collaborate. Explanations that increase the transparency of black-box tools can improve understanding and trust. Policymakers, regulators and technology developers therefore increasingly turn to explainable AI (XAI) to aid acceptance and adoption of AI technologies. However, there is still disagreement about what XAI for healthcare means in practice and what kind of transparency will improve understanding of and trust in AI technologies. We explored the meaning and importance of XAI through a Q-methodology study involving 37 professional healthcare stakeholders. Our participants included medical professionals, researchers, hospital administrators, policy advisors, domain experts, and developers of AI healthcare tools. We used mixed methods to produce a typology of four perspectives on XAI. We explore each perspective to clarify the rationale and needs of stakeholders who hold each perspective. Each perspective represents different values and meanings placed on XAI for healthcare, potentially impacting the development, acceptance, adoption, and implementation of AI technologies. We discuss the implications of these different perspectives for both future research and practical provision of explanations for healthcare.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.