To advance a procedural approach to trustworthiness in clinical AI, emphasizing the importance of embedding ethical and regulatory principles throughout the AI lifecycle, while providing practical implementation strategies.
Key Findings:
Trust in clinical AI is a multidimensional construct that includes transparency, interpretability, accountability, and ethical alignment, with implications for user trust and system adoption.
There is a persistent implementation gap between ethical guidelines and practical application in AI systems, highlighting the need for actionable strategies.
A procedural approach to trustworthiness can help bridge the gap by embedding safeguards throughout the AI lifecycle, ensuring compliance and fostering user confidence.
Interpretation:
Trust in AI systems is shaped by dynamic interactions among users, institutions, and sociotechnical environments, necessitating a lifecycle-spanning approach to embed normative safeguards and enhance user trust.
Limitations:
Existing frameworks often focus on outcomes rather than mechanisms, leading to a lack of practical guidance for trustworthy AI design, as seen in several high-profile AI failures.
The complexity of clinical environments may challenge the straightforward application of procedural frameworks, requiring tailored approaches.
Conclusion:
Embedding trust-oriented safeguards in the design and deployment of AI systems is essential for fostering public trust and ensuring ethical compliance in health care, with active user engagement being a critical component.
by Ali Saad, Sofia B Dias, Ghada Alhussein, David Lyreskog, Ioannis Gerasimou, Beatriz Alves, Μaarten de Vos, Ioannis Drivas, John Zaras, Andreas Stergioulas, Iskanter Bensenousi, Leontios Hadjileontiadis, Christos Chatzichristos, Stelios Hadjidimitriou
