A Human-Computer Interaction Approach to Medical Explainable AI (XAI)

As artificial intelligence (AI) systems become integral to clinical decision-making, the demand for explainability has shifted from algorithmic transparency to human-centered usability. In high-stakes domains such as medicine, opaque AI outputs can erode trust, introduce biases, and lead to harmful decisions. This conceptual contribution examines how integrating Human-Computer Interaction (HCI) principles and methods into the development and evaluation of Explainable AI (XAI) can improve understanding, trust, and clinical relevance. We analyse key dimensions of XAI, including model-agnostic vs. model-specific methods, local vs. global explanations, and post-hoc vs. ante-hoc strategies. Challenges such as interaction blindness, inconsistent feature attributions, and the absence of ground truth are critically assessed. We propose a multi-dimensional evaluation framework grounded in cognitive alignment, user satisfaction, and contextual effectiveness, supported by empirical UX methods. We outline several methods, including surveys, usability testing, contextual inquiries and focus groups. Their role and practical value are examined in the context of developing user-centric XAI systems. The findings underline that effective explanations must adapt to users’ mental models, roles, and environments. Achieving trustworthy AI requires not only interpretable algorithms but also transparent data pipelines, fair modelling choices, and interactive explanation interfaces tailored to the needs of diverse stakeholders.