CoPlan Interface Boosts Trustworthy AI Care Planning

AI-supported care planning holds significant promise for coordinating complex decisions across various patient needs, involving clinicians, patients, and caregivers. However, many existing AI systems present recommendations as fixed outputs, limiting the ability of stakeholders to scrutinize, challenge, or modify plans that conflict with clinical judgment, patient values, or practical feasibility. This research introduces CoPlan, a novel co-intelligent and contestable interface designed to address these limitations. CoPlan employs a multi-agent workflow where specialized AI agents generate intervention candidates and supporting or challenging arguments. Human care planners can then interact with these arguments, accepting, rejecting, modifying, or adding their own before a final plan is generated. This design fosters "co-intelligence" by combining human and AI expertise and ensures "contestability," preserving human agency and clinical accountability. Demonstrated in an aging-in-place scenario, CoPlan supports adaptive care team recruitment, role-based argument review, and practical follow-up through scheduling agents.