Ƶ

Doctors Should Think of AI as a Medical Student

— A new approach to the tech-physician partnership

Ƶ MedicalToday
A photo of a male physician shaking hands with a robot.
Shan is a medical student with experience in healthcare AI. Millstein is an attending physician.

Since the development of artificial intelligence (AI) solutions in healthcare, a compelling question looms: "Will AI eventually replace your doctor?" AI can often outperform physicians in , , and but faces significant , , and that, at least for now, categorize this technology as more of a companion solution.

A more practical and urgent question is: What should the AI-physician partnership actually look like?

We believe AI should be thought of as a medical student in training. While rewarding, working with actual medical students can also be burdensome for attending physicians in our busy clinical environments. The AI medical student will require preparation and ongoing iterative adaptations, but it learns so quickly that it can meaningfully lighten the physician's workload.

With a growing wealth of medical knowledge, AI can help complete tasks, gather information, and tailor communications. Since it can still make mistakes, miss the full picture, and have gaps in its clinical thinking, there should be a fully licensed attending physician supervising the work. Moreover, as in actual teacher-student collaborations, learning can be bidirectional. In the case of AI, the "student" can point out their mentor's errors without the constraint of hierarchy.

We hope this framework for thinking about medical AI will shift the conversation and development efforts away from approximating the tasks of human physicians and towards improving physician excellence. While there's so much excitement about AI in healthcare, we also need to recognize these solutions are still early in their application, and like medical students, they need time to grow and prove themselves.

What Would This Look Like in Practice?

Imagine the start of a future clinic visit or hospitalization: AI takes over the medical student's task of digging through a patient's chart to understand their prior medical course, focusing on how their health has changed since they were last seen. It is the first "provider" on the care team to engage directly with the patient, gather preliminary information, and help the supervising physician get a sense of what's going on before they enter the patient's room. Because many patients come in with extensive medical histories, and physicians only have so much time to see each patient, the AI medical student helps focus the visit and make the most of the time the physician spends with the patient.

The AI medical student will also: perform a review of all home medications; summarize the patient's case; craft a succinct consult question for another specialty when needed; put together a condensed discharge summary; query prior notes to answer ad hoc questions about the patient's history; and engage the patient's family with updates. In addition, it will perform literature searches to help broaden the clinical team's list of possible diagnoses and ground each care plan with an evidence-based approach. As a result, physicians can reprioritize tasks more directly relevant to improving quality of care while giving more complete attention to the patient.

After the clinic visit or hospitalization, the AI student will take a first pass at writing a documentation note that the supervising physician reviews and co-signs with relevant edits. It also provides tailored information to all the stakeholders in the patient's care, from a set of discharge instructions written at the patient's level of understanding to relevant updates for all their outpatient providers. Finally, AI continues to follow the patient in a low-touch manner. For instance, a patient presenting with poorly controlled hypertension will receive automated post-visit communication asking about their blood pressure, and any concerning updates could be flagged to human doctors. In this setting, AI can help create seamless transitions of care while making sure patients continue to be supported beyond the acute care setting.

Throughout these examples, the AI medical student is responsible for important but low-stakes tasks. These drive better care, but they are also easy to double-check, and mistakes can quickly be corrected. The current capabilities of AI are well-suited to these needs. AI tools have already been built to natural human language, parse and connect massive amounts of information, and scale across a sizable panel of patients.

The Impact: Integrating AI Into Care the Right Way

Thinking of AI as a medical student can help healthcare leaders and physicians better understand how to cohesively integrate these solutions into the workflow of a typical clinical encounter. This can encourage us to temper our expectations with these innovations. Instead of talking about whether a technology is going to replace human cognitive capabilities, we should instead be talking about finding the most helpful way to collaborate with it -- in the best interest of physicians and patients.

As AI increasingly takes on tasks that medical students currently have, we envision a concurrent shift in what medical training will look like. We anticipate a greater emphasis on engaging with AI tools and on refining core clinical decision-making skills. Additionally, prompt engineering (i.e., determining the best way to ask AI a question so that it outputs the most helpful clinical information) will be a core component of medical training.

All of this begs the ultimate question: medical students eventually learn enough to become fully-fledged doctors -- will this also hold true for AI? We remain open to this possibility. But just like medical training, we should slowly advance AI in healthcare by giving it increasing clinical responsibilities and subsequently evaluating how it performs at each step.

is a third-year medical student at the University of Pennsylvania's Perelman School of Medicine in Philadelphia. is an attending physician at Penn Medicine Woodbury Heights and a clinical assistant professor of medicine at the Perelman School of Medicine.