My Wife Is a Nurse Practitioner, and She Has a Concern About AI's Usage in the Exam Room

My wife is a nurse practitioner. She wants AI to help clinicians, but she also has a concern about AI's usage in the exam room.

This concern carries weight with me because she is not anti-technology. She has three master's degrees. Her first master's degree is in English Literature. Her second is in Computer Science from the University of Florida. I received my PhD in Computer Science from the same university, and we graduated at the same time. Her third degree is the one she depends on for her living: an MS in Nursing.

She has been in the medical field for more than 10 years. She meets patients every working day. She knows how exhausting clinical documentation can be. She knows the burden of taking notes, navigating EHR screens, sending prescriptions, documenting intent, and still trying to look at the patient as a human being instead of a data entry task.

So when she says AI can help, I listen. And when she says AI can be dangerous, I listen even more carefully.

I Have Been Preaching AI Too Much

I have talked to her about AI many times. Probably too many times. I am an AI enthusiast, and when I see a new tool, my first instinct is to push it to its extreme. Can it write code? Can it design a website? Can it analyze documents? Can it automate business workflows? Can it help a one-person company do work that used to require a team?

After a while, she gets tired of hearing me preach about the goodness of AI. I understand that. From my side of the world, many AI mistakes are annoying but not catastrophic.

I work mostly on non-mission-critical software development. If there is a bug in my software, in most cases it is not the end of the world. We can fix it. We can deploy a new version. We can apologize to the customer. We can improve the next release. Software has consequences, but many of my daily consequences are reversible.

The Exam Room Is Not a Demo Room

In software demos, AI looks wonderful. It listens, summarizes, formats, recommends, and writes. In a clinical setting, the same strengths can become risks if the output is treated as more authoritative than it really is.

My wife wants AI to remove burdens. She would welcome AI that takes good notes, prepares a clean summary, helps organize information, reduces repetitive typing, and gives clinicians more time with patients. That is the good version of AI in medicine: less clerical work, more attention, less burnout, more humane care.

But the exam room is not a software playground. A prescription is not a blog draft. A diagnosis is not a marketing slogan. A clinical note is not just text. It becomes part of the patient's medical record, and the next clinician may rely on it.

That is why AI in medicine is tricky. It is not only about whether AI is impressive. It is about whether AI is reliable enough for a mission-critical environment where mistakes can move from screen to body.

The Real Issue Is the Honest Mistake

Some discussions about AI in medicine focus on whether AI will replace physician judgment. That matters, but I think the deeper practical problem is more ordinary and more dangerous: how do we handle the honest mistakes AI will make?

An AI system does not need to be malicious to be harmful. It can misunderstand a sentence. It can summarize too aggressively. It can fill in a gap with a plausible assumption. It can misread the difference between "continue this medication" and "patient stopped this medication." It can turn a cautious clinical thought into a confident statement.

In my field, that kind of mistake may create a bug. In medicine, that kind of mistake may create a wrong medication, a wrong dosage, a missed allergy, or a wrong conclusion in the chart. The consequence can be catastrophic.

AI Summaries Are a Double-Edged Sword

My wife gave me a concrete example from current EHR systems. Some systems already have AI summary functions. That sounds useful, and often it is. A patient chart can be long. A visit history can be messy. A summary can help a clinician quickly understand what happened before.

But summaries are also a double-edged sword. Sometimes AI can misinterpret the provider's intent and give the wrong conclusion. It may compress nuance into certainty. It may leave out the one sentence that mattered. It may make a patient's situation sound more settled than it really is.

In a calm environment, a clinician can double-check the output carefully. But clinical work is not always calm. Providers are busy. Patients are waiting. The schedule is full. A message needs a response. A prescription needs to be sent. A note needs to be closed. Out of hurry, the provider may not have enough time to double-check every AI-generated sentence.

That is the main problem for these new systems. AI can reduce burden, but it can also create a new burden: the burden of verifying a polished output that may contain a hidden clinical error.

What the AMA Is Warning About

This is why I appreciated the recent discussion from the American Medical Association. In June 2026, the AMA emphasized that AI has enormous potential in health care, but it cannot replace physician judgment. The AMA's language is important because it does not reject AI. It asks for transparency, accountability, meaningful physician oversight, evidence-based recommendations, and regular audits.

The AMA also specifically discussed AI-generated clinical notes. It warned that the medical record could contain an error related to AI-generated documentation, and that such an error could interfere with care. That is exactly my wife's concern. The issue is not whether AI can write a beautiful note. The issue is whether the note faithfully represents the clinician's intent.

The AMA's concern extends beyond the clinic into insurance and prior authorization. If an AI-driven or algorithmic tool is used to delay, deny, or shape care, patients and clinicians need to know what logic, evidence, data sources, and guidelines were used. A black box is not acceptable when the output affects whether a patient receives care.

That same principle belongs in the exam room. If AI helps write, summarize, recommend, prescribe, or prioritize, the clinician needs to understand what the AI did, what evidence it used, where uncertainty remains, and where human review is required.

The Right Standard Is Not Anti-AI

I still believe AI should be used in medicine. In fact, medicine may be one of the places where AI could do the most good. It can reduce documentation burden. It can help find patterns. It can support evidence review. It can improve communication. It can help clinicians spend less time fighting screens and more time caring for patients.

But the standard has to be higher than in ordinary software. Reliability matters. Auditability matters. Training matters. Clear responsibility matters. A system should not quietly turn an AI suggestion into a clinical fact. It should make uncertainty visible. It should make review easier, not harder. It should support the treating clinician instead of becoming a second hidden decision-maker.

As an AI enthusiast, I have to admit something: pushing AI to its extreme is not always the right instinct. In some professions, the right question is not "How far can AI go?" The right question is "How do we make sure AI stays inside a trustworthy human process?"

My Wife's Concern Changed My Framing

My wife is not asking for medicine to reject AI. She is asking for AI to respect medicine. That means respecting the seriousness of the medical record, the pressure of clinical workflow, the complexity of patient stories, and the fact that a clean summary can still be wrong.

I still want AI to help her. I want AI to remove repetitive work from her day. I want AI to make documentation less painful. I want AI to help clinicians think, not distract them from patients.

But I also understand her warning better now. In my world, an AI mistake may create a bad release. In her world, an AI mistake may create a bad clinical record. Those are not the same thing.

Sources

American Medical Association: With AI increasingly part of care, transparency and quality are musts

American Medical Association: AMA adds more to its game plan to fix prior authorization