Abstract
The explicability of artificial intelligence (AI) is often presented as essential for the adoption of these systems by physicians. However, its usual approach has two limitations: on the one hand, the lack of anchorage in concrete professional situations, leading to a mismatch between proposed solutions and user expectations, and on the other hand, an excessive focus on how AIs work, to the detriment of other aspects of their design that can also induce a lack of intelligibility of results. This talk will present an AI project in pathological anatomy and cytology, a medical specialty central to cancer management, which proposes a novel perspective on AI intelligibility, focusing on the construction of their "ground truth". Examination of the strategy implemented by this project argues in favor of a contextual approach to AI intelligibility, enabling the development of solutions that are truly aligned with the expectations of healthcare professionals.
