Key Takeaway
Make AI care recommendations inspectable and contestable so clinicians and caregivers can correct, reject, or improve plans before they reach the patient.
ON THIS PAGE
What They Found
CoPlan turns AI recommendations into editable proposals by having multiple specialized AI agents generate supporting and opposing arguments for each care option. A human reviewer (clinician or care planner) inspects those arguments, can accept, modify, or add counterarguments, and then a validation step weights arguments to produce a final plan. In a home-care demonstration, the system prioritized practical, patient-centered actions while keeping humans in control. Blackboard Pattern
By the Numbers
14-stage pipeline from patient data to final plan: complexity assessment and team recruitment → candidate options → agent support/challenge arguments → validated plan synthesis
23 highest-priority recommendations in the demo: home safety assessment and modification, social engagement/support groups, and regular health monitoring and coordination
31 public code repository and 1 demonstration video available for practitioners to inspect and reproduce the system
Implications
Clinical informatics teams, product managers building AI-assisted care planning tools, and engineers building multi-agent systems should care because CoPlan shows a practical interface pattern for preserving clinician authority while getting AI help. Care teams and policy leads can use the contestability mechanism to document and audit how recommendations were formed and challenged. LLM-as-Judge Pattern
Explore evaluation patternsSee how to apply these findings
Key Figures
Fig 1: Figure 1: Human-AI Co-Intelligence Mechanism through Contestability for Agentic AI in Care Planning Systems.
Fig 2: Figure 2: Our proposed Role-based Contestable Argument Graph Computation Workflow for CoPlan Interface in Aging-in-Place Context.
Fig 3: Figure 3: Interactive Graph Review Panel for role-based human-in-the-loop contestation. The interface allows reviewers to inspect support and challenge arguments, add new arguments, edit existing ones, remove inappropriate nodes, and confirm the revised argumentative structure before proceeding.
Fig 4: Figure 4: Final Care Plan Viewer (right panel) and Chatbot Assistant (left panel) for further discussion.
Ready to evaluate your AI agents?
Learn how ReputAgent helps teams build trustworthy AI through systematic evaluation.
Learn MoreLimitations
The work presents a demonstrative pipeline rather than a clinical trial; no user studies or quantitative outcome comparisons are reported. Validity and confidence scores are intended to support human review, not to enable autonomous decision-making. Further evaluation is needed on how different aggregation methods affect transparency, fairness, and the risk that users over-rely on the system. Emergence-Aware Monitoring Pattern
Deep Dive
CoPlan is an interface and framework that treats AI suggestions as collaborative proposals rather than final decisions. It uses multiple role-specialized AI agents to generate both supporting and challenging arguments for candidate care options, represents those arguments in a structured, role-based graph, and exposes that graph to a human reviewer. The reviewer can accept, edit, remove, or add arguments; a validation operator then assigns acceptability weights and a synthesis operator produces a recommended care plan. The design emphasizes two principles: co-intelligence (humans and AI contribute distinct, complementary knowledge) and contestability (stakeholders can inspect and dispute AI reasoning).
The paper demonstrates CoPlan in an aging-in-place scenario where the system identifies relevant care roles, generates candidate interventions, and collects arguments to make reasoning explicit. In the demo, the workflow highlighted three top priorities (home safety, social engagement, monitoring) and provided implementation steps, evidence summaries, and argument validity. The authors released code and a demo video. Major next steps include formal user studies, quantitative evaluation of aggregation and scoring methods, and exploring whether clinicians, patients, and caregivers can effectively express value-based concerns through the contestation interface. For teams building AI-supported care tools, CoPlan offers a concrete interface pattern to keep humans central, improve auditability, and reduce the risk of opaque AI-driven decisions. Consensus-Based Decision Pattern Agent Service Mesh Pattern
Explore evaluation patternsSee how to apply these findings
Credibility Assessment:
Authors show low h-indexes (1–3) and no affiliations; arXiv preprint with no citations indicates limited established reputation.