CANOE's multi-agent framework boosts care plan safety with human contestability

Care plan coordination requires synthesizing clinical, functional, and psychosocial data across disciplines, but monolithic LLM pipelines can't guarantee transparency or safety. To address this, researchers from multiple institutions introduced CANOE (Contestable Argumentative Network-of-Experts), a multi-agent neuro-symbolic framework that structures decision-making as a transparent debate. Its five modules handle complexity assessment, adaptive team recruitment, argumentation via a novel Arena-based Quantitative Bipolar Argumentation Framework (A-QBAF), human-in-the-loop contestation, and final care-plan synthesis.

Role-specialized agents generate supporting and attacking arguments for candidate interventions, with conflicts resolved through arena-based clash resolution before acceptability scores propagate across the argumentation graph. Crucially, human care planners can accept, reject, edit, or add arguments, and the system deterministically recomputes the plan, ensuring every change is auditable. The team evaluated CANOE on the Discharge Me! and MedicalRAG datasets using ROUGE-L, AlignScore, MEDCON F1, FKGL, and LLM-as-a-judge. Results show medically fine-tuned models achieve the strongest clinical correctness and safety, while CANOE's argumentative structure provides faithful explanation and contestability—paving the way for safer AI-assisted healthcare.