Beyond black-box AI: Interpretable hybrid systems for dementia care

Foundation models (FMs) are entering Alzheimer's disease and related dementias (ADRD), yet bedside impact remains limited. We analyze the interpretability and reliability gaps that impede adoption, including opaque inference, hallucinations in generative models, and weak causal grounding. We argue for hybrid artificial intelligence (AI) that pairs statistical learning with computable clinical knowledge and clinician oversight. We propose a three-level framework for hybrid AI integration: (1) knowledge retrieval with linked citations; (2) contextualized decision support that combines predictive models with actionable plans derived from expert rules; and (3) adaptive optimization through continuous feedback. We demonstrate the potential of hybrid AI using clinical examples, including individual-specific interpretation of novel biomarkers, integration of multimodal data including speech and text, as well as patient-centered digital therapeutics. Finally, we outline pragmatic evaluation aligned with reporting standards, prioritizing adoption, safety, equity, workload, and patient outcomes. This roadmap aims to convert benchmark gains into accountable, interpretable tools for ADRD care.

Map artificial intelligence (AI) applications across clinical fields and reveal key findings in current literature.Identify how AI supports and advances next-generation technologies in dementia care.Present real-world cases where AI assists clinicians in context-specific scenarios.Examine major challenges and future opportunities in clinical AI adoption.