CoPlan: A Trustworthy Co-Intelligence Interface for Care Planning through Role-Based Contestable Argument Graphs

Care planning for complex, aging-in-place scenarios requires integrating clinical evidence, patient preferences, and practical constraints. Existing AI systems often present recommendations as fixed outputs, which can obscure the reasoning process and limit the ability of clinicians to challenge or adapt plans to specific patient contexts. The authors introduce CoPlan, a system designed to shift the paradigm from automated recommendation to collaborative negotiation.

CoPlan employs a multi-agent framework where specialized AI agents (representing roles like nurses, pharmacists, and therapists) generate candidate interventions along with supporting and challenging arguments. These arguments are structured using a Quantitative Bipolar Argumentation Framework (QBAF), which mathematically models the strength of competing viewpoints. The system then presents these arguments through an interactive interface, allowing human stakeholders to inspect the reasoning, add their own insights, or reject specific arguments before a final care plan is generated.

The study demonstrates that by treating AI recommendations as contestable proposals rather than final decisions, the system preserves human agency and clinical accountability. The workflow consists of three stages:

1. Complexity Assessment: The system uses standardized interRAI data to assess patient needs and recruit an appropriate interdisciplinary team of AI agents.
2. Argumentative Discussion: Specialized agents generate evidence-based support and challenge arguments for each proposed intervention, which are then weighted using QBAF to provide a confidence score.
3. Human-in-the-loop Contestation: A visual interface allows human planners to review the argument graph, edit nodes, and resolve conflicts, ensuring the final plan reflects both AI-driven synthesis and human clinical judgment.

This research addresses a critical gap in high-stakes AI deployment: the need for systems that are not just explainable, but contestable. By formalizing the process of disagreement and negotiation, CoPlan provides a blueprint for building AI tools that support rather than replace human expertise. This approach is particularly vital in healthcare, where the "correct" decision is often subjective, context-dependent, and requires the alignment of multiple stakeholders.