Explainable Artificial Intelligence as a Link in Dementia Patient Education: Bridging Transparency, Trust and...

Dementia is a growing global health burden, and artificial intelligence (AI) is now embedded across dementia diagnosis, prognosis, and monitoring. Because many of these systems are opaque, explainable artificial intelligence (XAI) methods, such as Shapley additive explanations (SHAP), local interpretable model-agnostic explanations (LIME), and gradient-weighted class activation mapping (Grad-CAM), have been developed to make model reasoning inspectable, and they have been applied extensively to dementia diagnosis. Far less attention has been given to whether, and how, such explanations should reach the people whose lives the predictions describe. We advance this as a conceptual proposition and an underexplored opportunity, not as a finding of a systematic review. In this commentary, XAI in patient education means the third of three related but distinct things: model-level explainability, which makes a system's computation inspectable; clinician-facing interpretability, which supports professional judgment; and patient-facing explanation, which is written to be understood by the person the prediction is about. Greater model interpretability does not by itself produce greater patient understanding. We examine what would have to be true for the proposition to hold. Explanations are only as truthful as the data that feed them, so patient-facing XAI presupposes sound data governance. Explanations can mislead, increasing trust without increasing truth. Dementia is heterogeneous, so explanation must be matched to an individual's cognitive and communicative profile and retained abilities rather than to a general cognitive level. Explainability may support comprehension, but it cannot establish decision-making capacity, which is decision-specific. We therefore frame dementia education as a relational process among the person living with dementia, care partners, and clinicians, with AI in a supporting role, and outline what a deployable, governed, workflow-integrated system would require.