A Double-Edged Sword: Explainable Artificial Intelligence for Medical Applications

This dissertation explores the complex role of explainability in AI-driven medical decision-making. While AI offers significant benefits in improving diagnostic accuracy and support, its opaque nature challenges user trust—particularly among medical professionals and patients. Explainability is seen as a key solution, but its impact is not straightforward. Through one conceptual and three empirical studies, the research shows that explanations can increase transparency, trust, and user understanding, encouraging greater adoption of AI systems. They also help users maintain confidence after AI errors and can reduce privacy concerns by clarifying data use. However, explanations can also have negative effects, such as increasing cognitive load, overwhelming users in high-stakes settings, and even heightening privacy concerns. To address these mixed outcomes, the dissertation proposes a procedural model for integrating explanations during the pre-use, use, and post-use phases. This model helps balance transparency, cognitive demands, and privacy considerations in practical applications like clinical decision support tools and symptom checkers. Overall, the work contributes to Human-AI interaction literature by offering theoretical, empirical, and practical guidance on designing explainable AI systems that are both trustworthy and user-centered.