In the Middle, Not on Top: AI-Mediated Communication for Patient-Provider Care Relationships
New research shows AI positioned 'in the middle' of conversations reduces friction and improves care.
A team of researchers from Stanford and collaborating institutions has published a groundbreaking paper titled 'In the Middle, Not on Top: AI-Mediated Communication for Patient-Provider Care Relationships.' The work introduces a novel framework positioning AI as a mediator within clinical communication rather than as an autonomous decision-maker on top of it. This approach directly challenges the common narrative of AI replacing human judgment, instead focusing on how AI can support the trust and meaningful connection essential to relationship-centered care.
The research centers on CLEAR, an asynchronous messaging system designed to operate between patients and healthcare providers. Through practical studies, the team demonstrated how this 'middle' configuration tackles real-world healthcare constraints like severe time pressure and uneven health literacy among patients. The AI's mediator affordances—such as constant availability and perceived neutrality—help redistribute the interpretive labor of communication. For example, it can clarify patient questions for busy doctors or rephrase complex medical jargon for patients, thereby reducing relational friction and misunderstandings.
The paper ultimately frames AI mediation as 'relational infrastructure,' highlighting its potential to strengthen, rather than disrupt, care relationships. However, the authors also critically examine the design tensions this creates, particularly around the AI's framing power (how it shapes messages) and patient privacy. The work was presented at the CHI 2026 workshop 'Toward Relationship-Centered Care with AI,' signaling a significant shift in how the HCI and medical AI communities conceptualize technology's role in sensitive human interactions.
- Proposes 'middle, not top' AI framework where technology mediates communication without usurping human control or judgment.
- Tests the CLEAR asynchronous messaging system, showing it reduces relational friction by redistributing interpretive work between parties.
- Identifies critical design tensions for mediator AI, including framing power and privacy, positioning it as relational infrastructure.
Why It Matters
This framework could make AI adoption in sensitive fields like healthcare more ethical and effective by preserving human relationships.