Rethinking Health Agents: From Siloed AI to Collaborative Decision Mediators

A team of researchers from leading institutions has published a paper challenging the current paradigm of AI in healthcare. Ray-Yuan Chung, Xuhai Xu, and Ari Pollack argue that most large language model (LLM) health agents, like those based on GPT-4 or Claude, are designed as standalone assistants for individual users. This siloed approach fails to address the core collaborative nature of healthcare, where decisions involve patients, caregivers, and clinicians. Their analysis of a fictional but clinically validated pediatric chronic kidney disease case reveals that breakdowns in treatment adherence often stem from fragmented situational awareness and misaligned goals among these parties, problems that a single-user AI chatbot cannot solve.

Instead, the researchers propose a fundamental reframing: AI should act as a "collaborative decision mediator" embedded within multi-party interactions. Their conceptual framework outlines how future AI systems could be designed to actively surface relevant contextual information to all stakeholders, help reconcile differing mental models about the illness and treatment, and scaffold a shared understanding. Crucially, this model preserves human decision authority, positioning AI as a facilitator rather than an autonomous oracle. The work, accepted for a workshop at the prestigious CHI '26 conference, provides a critical roadmap for developers to move beyond building better diagnostic chatbots and towards creating tools that genuinely strengthen the human relationships at the heart of effective care.