This blog is not (yet) written by AI

© photo by Bertie

I

I remember when I first heard about ChatGPT, the AI platform that does your writing for you. 

I was standing around the kitchen table with some family members, including my 9th-grade daughter, Georgia, who had just handed in an essay for social studies class. The assignment was, “What would John Locke and Thomas Hobbes think about mandatory Covid testing?”

The week before she had shown me a first draft. Let’s just say it needed some work. We talked over her ideas and I helped her with an outline. She dove back in and produced a second, and then a third draft. 

She wrote a lot of it in my presence - at the dining room table or in the kitchen. Every so often she would pipe up and ask about a word choice or whether a train of thought or argument made sense.

By the time it was done, she had a solid grasp of the topic, and I thought the final product was pretty good. “Guaranteed 9th grade A!” I told her. Rookie error. She got an A- (so unfair - Locke and Hobbes must be spinning in their graves). 

The whole thing, from start to finish, took a few days. 

Someone had sent me an article about ChatGPT, which talked about how high school and college students were already using the platform to write essay assignments for them. Each time the program would come up with a slightly different version, so it was hard to prove plagiarism.

I read the article aloud. Georgia and I looked at each other. “Let’s take her for a spin,” I said. 

I went online and registered for an account. I typed in the prompt: Essay on what John Locke and Thomas Hobbes would think about mandatory Covid testing, written at a 9th-grade level. Instantly text started to appear and seconds later we had a six-paragraph essay with essentially the same arguments that she had labored over for days.

As we read it through, I made some editing suggestions. Georgia demurred. “It’s better like this,” she said. “Don’t be offended, Daddy, but this one would have gotten an A.”

II

Medicine was one of the last fields to go digital. 

When I was a resident, and then years into being an attending, we used paper charts. All the notes were written by hand - some legibly, some illegibly - and organized in a thick binder containing the entire hospital record, from vitals signs to lab and radiology reports to consult notes. 

If you wanted to know what was going on with the patient, you had to locate the binder (often a challenge), leaf through it manually until you found the information you were looking for, and then try to decipher the handwritten scrawls and abbreviations.

Call them archaic but handwritten notes had a definite appeal. 

Since each one had to be written from scratch,  you had to be concise and economical. Otherwise, you would be there all day. Notes often included hand-drawn illustrations and diagrams. You could recognize a doctor by their handwriting, and the writing style - angular, blockish, cursive, loopy - often seemed to reflect their personality.

Is it going too far to say that progress notes were more like poetry than prose, with rules and conventions as formal as a sonnet? 

Maybe, but some of the lines from these notes have stuck with me longer than most poetry. I remember, for example, a particular attending neurologist’s entire terse assessment of a young woman with traumatic brain injury, printed in clean, blue letters at the bottom of his note: prognosis grave.

Now, of course, we all use electronic medical records - known as EMRs - and they are the bane of our existence. 

Does this make me sound like a Luddite?

Well, first of all, I freely admit that I kind of am a Luddite, but in any case, it’s true. I have never met a doctor who likes their EMR (please don’t think that is a figure of speech, by the way - I literally have never met anyone who does).

There are so many reasons for this that I could go on listing them for the duration of the post, so I’ll limit myself to three.

First, pointing, clicking, and pecking. 

Every patient has experienced the frustration of sitting in the room with a doctor who is more focused on the computer than on them. It’s just as frustrating for the doctor. 

Most EMRs were designed by computer programmers motivated by data collection and manipulation, not by user experience (UX in the parlance). To be hunting for the right field to use, the right box to click, or the right template to choose - a task which demands attention (think of airline check-in staff, for example - it would be comical how much staring at a screen and typing goes on if it didn’t trigger my EMR PTSD) - instead of making normal eye contact with patient, is a misery.

Second, cutting, pasting, and carrying forward. 

In theory - like all technology - it seemed like such a convenience at first. Instead of writing a note from scratch, you could easily copy and paste entire sections from previous notes and edit them. Not only that, but the EMR would allow you to carry forward an entire previous note and use it as a template - keeping what you want, like med lists or previous history, and adding, deleting, or editing the rest. 

In practice - like so much technology - the result is a total mess. Gone is the concision and economy of style that writing a full note demands. Instead, progress notes now tend to be bloated, confusing, inaccurate pastiches of old information. 

It would not be an exaggeration to say that most progress notes these days contain less than one part in twenty of useful information - which is buried in so much junk that finding it has become an art in itself. 

I’ve seen progress notes with pages of radiology reports and lab results pasted in with no indication why; medications listed that the patient has not taken for years; social history that is totally inaccurate; physical exams from template models that have no relation to the actual patient (are the patient’s pupils really reactive to light, Dr. Resident? That’s amazing because on the Head CT he has two artificial eyes! True story…).

At its best, reading a progress note used to feel like being pricked by a needle of pertinent information. Now it feels like hunting for a needle in a haystack.

Third, the inbox. 

EMRs have their own inboxes, generally containing lab results, scanned reports, and messages from patients or staff. 

Sometimes these various types of messages get split up into different inboxes. What’s worse, many of them are flagged as urgent or critical, especially abnormal labs. Sometimes they really are, and sometimes they are totally not - whatever algorithm is doing the flagging has no genuine ability to distinguish between abnormal/critical and abnormal/trivial.

This is in addition to regular email, texts, faxes, and whatever other platform you have to check. The result is a flood of information that is impossible to manage - at least if you want to get anything else done in life - coupled with low-grade anxiety that you might be missing something important. 

There’s an Israeli expression that perfectly captures how most doctors feel about their EMR inbox: “delete complete”.

III

Like many evils, EMRs are not all bad. 

It’s nice to have electronic prescribing. You wouldn’t believe how similar can look Ondansetron (an anti-nausea drug) and Olanzapine (an antipsychotic), or Amoxicillin (the antibiotic) and Atomoxetine (an ADD drug), when written in the doctor’s notorious scrawl (guilty as charged).

And let’s not forget the promise of a convenient, portable, accurate, and easily transmissible electronic medical record. How great would that be? Unfortunately, none of the dozens of different EMR programs that exist know how to talk to each other. 

New patients still come into my office with armfuls of paper printed out from their prior doctor’s EMR that I have to sift through in order to find and scan in one or two pages of useful information. Then, in a perfect example of how the “paperless office” can appear environmentally friendly without actually being so, the whole pile is dumped into a box to be shredded.

There is one thing I love about EMRs - they have generously provided us with a delightful historical irony: scribes.

Because of all the headaches and frustrations they entail, EMRs have given birth to an entire industry of EMR scribes - people whose job it is to follow the doctor around from room to room, watching and listening to what transpires while transcribing it all into the medical record. They can even go through your inbox and separate out the wheat from the chaff.

Scribes are a truly brilliant hack. 

Whoever thought of scribes is like Gerard Depardieu in one of my favorite scenes from the movie Danton, where he meets with Robespierre across a long wooden dinner table piled high with delicacies to determine the course of the French Revolution. The two men reach an impasse and Danton stands up, leans forward, looks his adversary in the eyes, and wipes the whole table clean of food and cutlery and candelabras and porcelain, everything crashing to the floor with one sweep of a burly forearm. 

Scribes make the EMR disappear. Now you can focus on what matters while the distracting and onerous task of electronic record-keeping is relegated to the background.

The thing is, you know who else realized that having a scribe do the record-keeping while you focus on the important stuff was a great idea? 

Babylonians.

So however we feel about EMRs, let’s at least thank them for demonstrating so elegantly how everything that is old can be new again; and for the irony that for all the futuristic Utopianism of Silicon Valley, users on the ground can be left with problems whose only solution is to go back to exactly the way it was done thousands of years ago.

Maybe now you can understand why my immediate reaction to ChatGPT was not tech worship. It was not trepidation about how this could affect my child’s education. It was not even existential dread over the potential cheapening of human intelligence and accomplishment.

Rather, the minute I saw what ChatGPT could do, my first reaction was selfish excitement: could this be what we have all been waiting for? Could ChatGPT be an AI scribe?

How would it work? I don’t know - however it works! 

You feed the algorithm a gazillion clinical notes, let it machine-learn how they are supposed to look, and let it listen in while you see the patient. Presto: a coherent clinical note, complete with prescriptions, orders, and referrals. 

In fact, since that day with Georgia, I have been approached by a couple of AI start-ups who had the same idea. I am thinking of working with one on development.

Am I being short-sighted here? Collaborating in my own destruction? After all, why should AI stop at scribing? Many human scribes are post-Bacc pre-med students - it’s the perfect gig while you wait to hear from medical schools. 

Why not take the next step and send AI to med school? The same process described above could just as easily be applied to writing the note as to scribing it, cutting the doctor out of the picture entirely (this just in: AI passes the USMLE medical boards).

You could say that to adopt AI I would be to violate one of my golden rules of parenthood - I like to pull this one out when parents ask me about pacifiers: whatever makes your life easier today will come back to haunt you tomorrow!

But you know what? I just don’t care. I guess my experience with EMRs has taught me something about myself: I’m not above making a deal with the devil. 

As far as I’m concerned, if AI will take the EMR off my hands today, then it (he? she? they?) can have my job tomorrow.