Evaluation of Human-Understandability of Global Model Explanations using Decision Tree

Global explanations outperformed local ones for non-expert users: The study found that the majority of participants preferred global (holistic) explanations over local (case-specific) ones, challenging the current XAI research focus on local explanation approaches and suggesting that one-size-fits-all explanation strategies may not serve all audiences equally well.

Healthcare requires different explanation approaches than other domains: The research highlights that healthcare applications serving non-expert end users like patients have fundamentally different needs than expert-focused AI systems, requiring explanations that are narrative-based and patient-specific rather than technical—a distinction often overlooked in current XAI development.

Individual preferences for explanations vary significantly: The study identified "strong individual preference for a specific type of explanation," meaning some users benefited more from global explanations while others preferred local ones, suggesting that responsible AI systems in healthcare should offer customizable explanation options rather than standardized formats.

Mental models reveal gaps between designer intent and user understanding: Task-based evaluation of how participants mentally understood the explanations provided "valuable feedback," indicating that testing explanations with actual non-expert users uncovers critical misunderstandings that designers might otherwise miss.

Trustworthiness requires actionable, understandable explanations: The research demonstrates that for AI systems to be both trustworthy and useful in healthcare, explanations must enable patients to understand and act on the model's reasoning, not just passively receive information.