CoPlan: A Trustworthy Co-Intelligence Interface for Care Planning through Role-Based Contestable Argument Graphs
Hung Truong Thanh Nguyen, Hélène Fournier, Piper Jackson, Makoto Itoh, Shannon Freeman, Rene Richard, Hung Cao
Read on arXiv →
Key claim
CoPlan enables dynamic human-AI collaboration in care planning.
In plain English
Imagine you're a clinician trying to create a care plan for a patient who wants to age in place. You have to consider not just medical needs, but also personal values and practical realities. Currently, many AI systems provide fixed recommendations that can clash with your clinical judgment or the patient's preferences, leading to frustration and potential errors. This is what's called a lack of contestability, where the AI's suggestions can't be easily inspected or modified by the human users involved. CoPlan addresses this by introducing a multi-agent system where specialized AI agents generate various intervention options along with arguments for or against them. This allows human care planners to actively engage with the AI's suggestions, accepting, rejecting, or modifying them based on their expertise and the patient's needs. The design emphasizes co-intelligence, where both humans and AI contribute their strengths, and contestability, ensuring that recommendations can be scrutinized and adjusted. Compared to traditional AI systems that offer static outputs, CoPlan fosters a more dynamic and collaborative environment for care planning. This means that for builders in the healthcare space, integrating such a system could lead to more personalized and accountable care solutions, ultimately improving patient outcomes.
CoPlan introduces a novel multi-agent workflow for human-AI collaboration in care planning.
The approach is demonstrated in a practical scenario, showing solid applicability and user interaction.
Deep reliability assessment
The methodology supports a collaborative and contestable approach to care planning, but it may overclaim the extent to which it can ensure clinical accountability and human agency without further empirical validation.
Reproducibility
yes, the implementation is available at https://github.com/Analytics-Everywhere-Lab/CAIAiPCP/
Key figure
Figure 5 illustrates the end-to-end pipeline of CoPlan in an aging-in-place care planning scenario, showing the transition from patient information to a finalized care plan through four connected stages.