The EMR, and AI: Are they good? Pose risks? Both?

When I worked with residents in the hospital, the electronic medical record, EMR, was relatively new. We were fortunate to work in a hospital that invested heavily in a good, well-regarded EMR, and spent quite a bit on training the doctors to use it. In the end, almost all the “stakeholders” agreed on which one was best, and the hospital bought it, and had us trained. Good for them.

The EMR wasn’t perfect though. In addition, the hospital didn’t buy all the parts. EMRs come in modules, some necessary, some elective, especially back then. It was clear that the hospital had prioritized the modules for billing, and especially for maximizing billing. Also, anything that the subspecialists who earned the hospital lots of money wanted. Other modules, particularly those that would enhance primary care, were more rudimentary or absent. Some of the things that many of us thought would be easily facilitated by a computerized database and looked forward to having were not available. Surely, once everyone is loaded into the computer, it should be simple to print out a list of all the patients with diabetes assigned to a particular doctor! That would really help us to track them, contact them, make sure they didn’t fall through any cracks. Whoops, sorry, we didn’t buy that module. The maximization of potential billing, on the other hand, was not only there but required many different screens to be filled out, effectively transferring work to physicians from someone else.* And it was inconsistent in how it treated health risks. For example, tobacco use had a who series of questions, including information the patient themselves probably forgot about how much, when, etc., but there was only one on whether they drank alcohol.

There were many things that the EMR did make easier, though, including writing long notes in the chart, since people didn’t have to write by hand. Like the Word® program I am using, and most other computer programs, cut-and-paste became easy and routine. Residents’ notes got longer because they could cut-and-paste yesterday’s note and (hopefully) update it. But sometimes they might forget the update part; it could be embarrassing if yesterday’s note said “surgery tomorrow” and it still said it in today’s note, even though the surgery had occurred that morning! The EMR also facilitated making notes longer by importing all the lab results and radiology reports. This is important information, but it is also available elsewhere (i.e., in the lab and radiology sections). A simple “Radiology exams normal” or whatever they showed would have been much better than cutting and pasting the whole report, as well as briefer. Better because it would have required the resident to read it, make an assessment (“it’s normal”, or “it shows a tumor”) and write that. It would have required thinking. Not to say that they didn’t think, but a summary in their own words would have demonstrated that in a way that cut-and-paste couldn’t.

But the biggest problem with the EMR is the amount of time that it takes to complete, especially in outpatient clinic settings, and especially for primary care clinicians who usually have a wider variety of issues to address and less money to hire others (scribes, sometimes nurses or even NPs or PAs) to do their documentation for them. It is not uncommon for primary care physicians to spend more time documenting in the EMR, frequently at home at night**, than seeing the patient! And in the inpatient setting, hospitals hire nurses to comb charts looking for ways to “upcode”, to charge more. A part of the ongoing contest between providers and insurers to see who can hit the other up for more (except when they have been vertically integrated, more common in outpatient settings, see Vertical Integration saves money. And CVS and its competitors use that to line their pockets, not provide healthcare, May 21, 2026). And potentially costing the patient more, if the insurer refuses to pay it all.

And now we have AI. Or AI is having us. The debate on AI, on whether it will create a great new world or a “Brave New World” à la Huxley, rages on, now with the Pope getting involved with a new 42,000 word encyclical. AI is happening, will continue to happen, and will continue to have effects, many untoward, and some of those resolving – but not necessarily in ways that are good for people. And there are many different people, not just in the US but in the world. Recent commentaries have suggested the benefit would be greatest for the well-off and well-educated (well, almost all things do), although what seem to be “regular” people are using it to bolster their “home brewed lawsuits” and clogging up courts (good or bad?)

I know a lot of doctors who are thrilled about AI, and see it as a vehicle for reversing, or at least slowing, the constant drain on their time that comes from more documentation being required for billing, for insurers, and even for government regulations, in some ways a counter-weight to the EMR. They have apps that record the entire encounter, and then AI drafts a progress note that covers all the essential information in the conversation for both clinical and legal/billing purposes. Then the clinician reviews, augments, and corrects the AI-generated note. Hopefully. That is a danger. AI (as well as clinicians, it should be noted) can make mistakes, and provide incorrect information. With people, we know who to blame. However, recent experiences with friends and family encounters with the health care system suggests that once something gets into the medical record, especially a digital one (indeed, all digital data collection), it is there forever and efforts to correct it do not always take.

And, back to the residents copying their notes rather than creating original ones, it is comparable (if more high-stakes than) to students using AI to write their papers. It allows the appearance of creation and completion without the thinking required to learn to do the job right. Of course, AI advocates argue that AI learns to think more reliably than do people. Maybe this is not a scary idea. A recent opinion piece in the New York Times by Dr. Helen Ouyang suggests that AI (ChatGPT, in this case) gives good, well-researched medical information, and, more important, is accessible to answer questions when the doctor isn’t. The author notes that ”Of course, as a doctor, I know when to question the chatbot and when to ignore it. Many other patients don’t.” That’s right, and that’s a concern. Most of us who have used AI know that it isn’t always right, but if it’s a topic we don’t know about, we don’t know.

The other thing that Dr. Ouyang liked about ChatGPT was, ironically, its personality, since “I had always assumed the ‘human side’ of medicine was the part A.I. couldn’t touch.” The AI was unflaggingly positive, upbeat and encouraging, and never got irritated about repeated or “stupid” questions. People miss this when dealing with – people. While some doctors, like other people, are not, by nature, always warm, positive or supportive, the circumstances in which they work, the pressure from their employers (see several previous pieces, recently Why is it so hard to get medical care? And what should we do about it?, March 15, 2026, and The problem with the US healthcare 'system': THE INSATIABLE PURSUIT OF EVER MORE MONEY BY CORPORATIONS AND WALL ST., Feb 25, 2026). We should also remember, that while being nice, and friendly, and supportive is usually good, it is also a strategy for gaining your trust that has been misused by bad actors throughout history. And AI never gets tired of doing it, never wants to go home, never misses its kids, and doesn’t have to worry about spending as much time completing the EMR as it did seeing you! (see Does AI communicate better than real doctors? If so, why is that?, Nov 20, 2025).

So, I guess that the jury is not in on AI, or its most effective and reliable and accurate utilization. When it is, it will probably be too late to change it.

  

*This is only one example of work that has been transferred to the primary user. I have long made my own travel arrangements, and like it because I know what I want, but it takes a lot of my time.

**Another example of work transferred to the clinician, at the expense of their family.