The Hidden Value of PAs in Healthcare AI: Why healthcare AI needs clinicians who can connect

Matt Bell, DMSc, PA-C, CAQ-HM - Chief AI Officer, WNC VA Health Care System

Author’s Bio:

Matt Bell, DMSc, PA-C, CAQ-HM, is a Physician Associate with nearly three decades in internal and hospital medicine who currently serves as Chief AI Officer and the Chief PA for the Western North Carolina VA Health Care System. He works at the intersection of healthcare AI governance, clinical operations, and education, and is an accepted presenter at the upcoming 2026 PAEA Education Forum. He is a doctoral-prepared PA and a member of the Academy of Doctoral PAs.

Article:

The medicine team is rounding on an older adult patient with multiple comorbidities.

Two days into a straightforward hospitalization, the electronic health record's predictive risk model has placed her in the low-risk discharge tier. The note is written. The orders are ready. The team is moving on.

Then the PA hesitates.

Something is off. Not in the labs. Not in the vitals. Not in the model. In the conversation. The family member who flew in overnight is asking different questions from those asked over the phone. The patient, who has been pleasant and joking with staff since admission, has gone quiet.

So the PA sits down, asks one more question, and listens.

The picture that emerges is nothing like the one in the chart. The life waiting at home is far more fragile than anyone realized, and the discharge plan, as written, would fail.

The model was not wrong. It just could not see what the PA saw.

That moment, the pause before discharge, the question no model knew to ask, the context hiding just outside the chart, that's where the real story of healthcare AI is silently being written.

Much of the public conversation about healthcare AI still centers on the algorithm: what it can predict, how accurately it performs, whether it can draft, diagnose, summarize, or recommend. Those questions matter. But they are not the whole story.

In real clinical care, AI succeeds or fails at the intersection of technology and human work: workflow, communication, trust, judgment, teamwork, and the patient's lived reality. That meeting point is where Physician Associates already live.

Healthcare AI needs more clinicians who can connect innovation to frontline reality. PAs are built for the role.

To serve that role well, PAs don't need to become engineers. However, the profession does need a new kind of professional fluency: AI fluency.

This isn't the ability to build a model, it's the ability to ask better questions of one. To understand, in practical terms, how a tool works, where its limits are, what its output means, and how it should, or should not, influence care.

A Profession Built for Connecting Technology and Care

PAs do this work all day.

Between specialists and patients. Between complex care plans and the people who must live them. Between policy and practice. Between the system as designed and the system as experienced.

Nurses, pharmacists, and physicians do this too, and healthcare AI needs every one of them. But the PA's particular blend of diagnostic autonomy, collaborative spirit, and operational mobility across specialties creates a clinical role that the field has under-recruited.

That habit of mind is exactly what healthcare AI needs right now. Tools are arriving in clinical environments faster than the conversations about how to integrate them safely, equitably, and humanely. Models are being chosen, deployed, and evaluated in rooms where clinicians who understand where technology meets care are often absent, and where PAs, in particular, remain grossly underrepresented.

If PAs are going to help lead the integration of healthcare AI into safer, more humane care, that work has to happen in three places: where tools meet workflow, where decisions about AI are governed, and where the next generation of clinicians is trained.

Workflow Integration: Where AI Meets the Shift

In real-world implementation, AI often fails less because the math is broken and more because the workflow is.¹

Alerts fire at the wrong moment. Predictions arrive after the decision has already been made. Outputs require verification time the team does not have. EHR integration breaks under the pressure of a busy shift.²

A widely cited 2021 evaluation of a proprietary sepsis prediction model that is still used as a foundational teaching case in AI academia, found that in real-world use, the tool performed substantially worse than its developer had reported ³. The lesson was not simply that models need validation. It was that a model built in one environment does not automatically fit another.

This is not, at its heart, an argument against AI. It is an argument for implementation that respects the complexity of care.

This work means asking whether a model's output arrives in time to matter. It means noticing when an alert adds clarity and when it adds noise. It means explaining to the team what a risk score can support and what it cannot decide.

Most importantly, it means protecting the patient from becoming invisible behind the predictive algorithm and data sets.

The patient from our clinical vignette was ultimately able to return home on day two of her hospitalization with a plan that fit her actual life, not because the model failed, but because someone in the room could see what the model could not. That is the work.

Governance: Where Decisions Become Bedside Reality

Somewhere in every health system, a small group of people sits in a room and decides which AI tools will be used with patients in the coming months. PAs are all too rarely in that room. Governance may sound distant from the bedside. It is not. Governance is where someone decides what evidence is good enough, which risks are acceptable, who monitors for harm, who can override an output, and what happens when the tool is wrong.

That gap matters. When governance lacks the clinical and operational perspective PAs bring, decisions are made that miss the friction patients and frontline teams will feel.

The opportunity is not simply to be represented. It is to be useful: to bring the bedside, the discharge conversation, and the patient's lived reality into decisions that are too often made far from where care actually happens.

National frameworks already support clinician participation in AI governance, including consensus guidance from the National Academy of Medicine⁴. More recent, assurance-oriented guidance from the Coalition for Health AI, including its 2024 Assurance Standards Guide and the 2025 governance framework it released with The Joint Commission, reinforces the same expectation: clinical voices belong in the room.^5,6

Education: Where the Future Is Shaped

AI is already part of the practice environment most PA students are entering. Ambient documentation tools, clinical decision support, predictive risk models, and patient-facing chatbots are not theoretical; they are present in the rotations students are doing now.

A curriculum that does not engage AI literacy leaves new graduates underprepared, not for the technology itself, but for the responsibility of practicing alongside it. Students need practical cognitive scaffolding: how to question and verify AI outputs, recognize automation bias, question missing context, escalate concerns, and explain AI-supported recommendations to patients.

This is not about turning PA students into data scientists. It is about preparing them to use AI effectively in the work they are already being trained to do. The concern some educators are starting to raise is legitimate. If students lean too heavily on AI and allow it to consistently perform certain tasks before they have the chance to experience, practice, and internalize the underlying reasoning, they may never learn to think critically (AKA neverskilling). However, AI fluency, properly taught, can strengthen critical thinking rather than replace it. Students learn to interrogate outputs, identify what the model cannot see, recognize automation bias in their own judgment, and bring human reasoning to every AI-supported decision. The goal is not to train students who blindly trust AI; it is to train students who can think with AI while preserving the clinical judgment that no algorithm can replicate.

This is why AI fluency in PA education cannot be an elective track for the technically inclined; rather, it needs to be a main part of the core curriculum.

The Invitation

The PA profession has spent decades earning a reputation for adaptability, team integration, and sound operational judgment. Healthcare AI needs exactly those qualities at the bedside, in the rooms where governance is written, and in the classrooms where the next generation of clinicians is being shaped.

The opportunity is not to catch up to a conversation that has moved on without PAs. It is to step into a role that the conversation has been missing.

The invitation, especially to colleagues who have not yet engaged this topic, is simple: no one needs to be an AI expert to contribute. What is needed is a clinician willing to learn and bring AI fluency to the patient bedside, the C-Suite, and to the classroom.

The future of healthcare AI will not be decided by algorithms alone. It will be decided by AI-fluent clinicians who remember to pause, ask one more question, and see the patient as a whole person.

*The views expressed are strictly those of the author and do not represent the Department of Veterans Affairs or the U.S. government.

References

1. You JG, Hernandez-Boussard T, Pfeffer MA, Landman A, Mishuris RG. Clinical trials informed framework for real world clinical implementation and deployment of artificial intelligence applications. NPJ Digit Med. 2025 Feb 17;8(1):107. doi: 10.1038/s41746-025-01506-4. PMID: 39962232; PMCID: PMC11832725.

2. Bain AP, Ngai D, Bernard PA. Clinical Decision Support Systems in Generalist Practice: Utilizing Clinical Decision Support Systems Tools to Improve Clinical Decisions and Patient Outcomes. Med Clin North Am. 2026;110(2):191‑207. doi:10.1016/j.mcna.2025.07.004.

3. Wong A, Otles E, Donnelly JP, et al. External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients. JAMA Intern Med. 2021;181(8):1065-1070.

4. Matheny M, Thadaney Israni S, Ahmed M, Whicher D, eds. Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril. National Academy of Medicine; 2022.

5. Coalition for Health AI. Assurance Standards Guide. Coalition for Health AI; 2024. Accessed May 20, 2026. Available at: https://www.chai.org/workgroup/responsible-ai/responsible-ai-guide-raig-and-raig-executive-summary

6. Coalition for Health AI; The Joint Commission. Responsible Use of AI in Healthcare. Coalition for Health AI; 2025. Accessed May 20, 2026. Available at: https://digitalassets.jointcommission.org/api/public/content/dcfcf4f1a0cc45cdb526b3cb034c68c2