AI Network-of-Experts Improves Care Plan Coordination with Human Contestation.

The paper presents CANOE, a novel multi-agent framework aimed at enhancing the coordination of complex care plans, particularly in healthcare settings. This system integrates various types of information—clinical, functional, and psychosocial—from multiple professional disciplines. Unlike monolithic large language model (LLM) approaches, CANOE prioritizes transparency and safety through its neuro-symbolic architecture.The framework operates by recruiting specialized agents for different roles, who then engage in an argumentative process to propose and critique potential interventions. Conflicts are resolved using an arena-based system, and the acceptability of arguments propagates through a graph. A crucial feature is the human-in-the-loop contestation, allowing care planners to accept, reject, edit, or add arguments, which then triggers a deterministic recomputation of the care plan.Evaluations show that while medically fine-tuned models offer strong clinical correctness, CANOE's argumentative structure provides superior explainability and enables human oversight, addressing key limitations in current AI applications for sensitive domains like healthcare.