HEAL
Menu

Research

Start with people.
Study what changes.

We connect questions about clinical judgment, relationships, and work to evidence from healthcare as it happens.

01 · Human-centered AI safety

When does helpful AI become harmful?

We study when clinical AI misleads, encourages inappropriate agreement, or changes trust and judgment. Safety depends on how a system behaves with people, not just how it scores on a test.

Preprint and ongoing research

Can AI disagree when it matters?

How do conversational AI systems respond when patients ask for care that is not indicated?

Why it matters

A reassuring answer can still lead someone in the wrong direction. Evaluating agreement, resistance, and communication together makes these risks visible.

How we study it

Simulated clinical conversations test model behavior under patient pressure. The published preprint examines sycophancy; the broader Health EQ Bench effort develops ways to evaluate conversational and clinical AI.

Explore the evidence and ideas

Ongoing research

What happens when AI enters the relationship?

How does AI change the relationship between patients and clinicians?

Why it matters

Clinical care depends on trust, communication, and judgment. Introducing another voice can change all three.

How we study it

A scoping review examines relational mechanisms, risks, and tensions in AI-mediated care, bringing together evidence about how AI may change patient–clinician relationships.

Explore the evidence and ideas

02 · Clinical work & care systems

What helps people deliver better care?

We examine how workload, team coordination, and technology affect the delivery of care. Clinical records reveal patterns in the work that happens between decisions, conversations, and handoffs.

Published studies and related research

Does less typing mean better clinical work?

How does ambient AI documentation change the experience and work of an emergency department shift?

Why it matters

A tool intended to save time may also change how clinicians move between tasks and create records. Adoption alone does not answer whether it helps.

How we study it

Retrospective studies examine scribe adoption, documentation time, and burden in emergency care. Related work studies note characteristics and chart-switching behavior.

Explore the evidence and ideas

Ongoing research and a pilot preprint

What does the workflow leave unseen?

How do workload, team activity, and the timing of patient arrivals relate to care?

Why it matters

Healthcare teams constantly adapt to competing demands. Understanding those demands can reveal why the same process works differently across clinical settings.

How we study it

Electronic health record and audit-log measures help examine cognitive load, team experience, and care timing. Observational analyses describe associations rather than claiming that workload causes a particular outcome.

Explore the evidence and ideas

03 · Learning & the future of medicine

How does expertise grow—and change?

We examine how physicians develop expertise and how artificial intelligence changes their work. Understanding what clinicians learn and do helps us ask where technology belongs.

Published research and accepted work

How do physicians build experience?

What can clinical records tell us about how residents learn and how their workload changes?

Why it matters

Time in training is only one part of the picture. The cases clinicians encounter and the work they perform offer another view of developing expertise.

How we study it

Natural language processing characterizes clinical case exposure across residency. Related research examines documentation workload and frameworks for supervising AI in medical education.

Explore the evidence and ideas

Published research

Which parts of a physician’s work could AI change?

How might AI augment or automate the tasks that make up emergency medicine?

Why it matters

Treating a profession as one automatable activity overlooks the different kinds of work physicians do.

How we study it

A labor-economics framework examines susceptibility at the task level. The resulting publication provides a way to discuss the future of clinical work without assuming that technical capability establishes safe replacement.

Explore the evidence and ideas

Where could our questions meet yours?

Connect with HEAL