Trust in AI Is a Fluid Process: Building Trust of AI through Clinicians’ Needs in the Breastscreen Victoria Program. a Qualitative Study.

With the rapid growth in development of AI-enabled medical products as decision-making support tools, there has been increasing awareness that clinicians’ trust is key for their successful adoption in clinical settings. However, few study investigates empirically what factors other than their technical functionality (e.g., accuracy and reliability) can enable clinicians to enhance levels of trust in AI. This study explores this engaging with clinical professionals, including radiologists, radiographers and nurses, who work for the population-based BreastScreen Victoria. We conducted five focus groups and two face-to-face interviews, involving a total of 27 clinicians. Utilising the ArAAIT framework, this qualitative study illustrates clinicians’ trust is not a fixed psychological state but a fluid process. It is often circumstantial and volatile but can be built strategically. Our findings revealed, while the prevalent view of AI as ‘opportunity’ among them was promising in forming an initial trust formation as a fundamental requisite, it was easily debilitated by the sense of ‘uncertainty’ surrounding AI or a disappointing prior experience. Such drawback can be invigorated through organisation’s ongoing evaluation of AI’s efficacy and effectiveness and developing rigorous AI governance and policies to facilitate an ethical and effective clinician-AI collaboration. We argue trust building is a process, starting from development of AI model to its implementation. Acceptance of AI requires not only trust in AI’s technicality, but also trust in the whole ecosystem where AI and humans are autonomously co-exit. Our findings have a practical implication beyond the radiology/breast screening context, that is, gaining trust in AI does not occur naturally or passively but requires a holistic effort to build it.