Modeling the Dynamics of Trust in Digital Pathology Using Explainable AI

Abstract

The adoption of artificial intelligence (AI) in medical diagnostics demands rigorous examination and a robust understanding of human trust dynamics in explainable artificial intelligence (XAI) frameworks. This research presents a quantitative model for evaluating the evolution of trust among pathologists based on their sequential experiences with AI-driven diagnostic recommendations and explanations. By monitoring and analysing interactions characterised by AI’s false positives and false negatives, this study captures nuanced shifts in trust. An empirical evaluation using digital pathology scenarios reveals that initial trust levels remain stable, independent of AI accuracy, but subsequent interactions significantly adjust trust based on accumulated experience. Moreover, diagnostic performance improves notably when pathologists collaborate with XAI systems, underscoring the utility of such integrations. The study identifies limitations, including restricted sample sizes and homogeneous case diversity, advocating for future research involving broader participation and personalized trust modelling to encapsulate variability among pathologists.

Ähnliche Arbeiten