As the COVID class, entering medical school in 2020, we have seen our fair share of paradigm-shifting moments in medicine.
- The pandemic and response
- The overturn of Roe v. Wade
- The transition of Step 1 from scored, and largely dictative of what future specialty one can match into, to pass/fail
Even with that last transition, It still took me a couple of months of studying to consistently pass Step 1. Less than a year later, it took ChatGPT almost no time to pass the test. Is it dramatic to say that the didactic portion of our medical school experience has quickly become replaceable?
I don’t know but the cat is out of the bag. AI has been quietly disrupting physicians’ workflow for years. There hasn’t been a more obvious and terrifying display of how it will change our profession. Or at least what will define our profession.
Answers
The defining feature of a doctor in the 20th century was (s)he who had the answers. Patients would come to doctors seeking answers for what ailed them. Doctors spent close to 1.5 decades of higher education to be worthy of having those answers.
This is rapidly changing. Patients, if not already, will quickly go straight to the generative AI for their answers. Even in the early version, the computer likely has as close to as good of answers as we do. In subsequent iterations, the computer will undoubtedly have better answers.
As health care providers, hanging our hats on having the answers will no longer suffice.
Questions
Before medical school, I worked in Silicon Valley health tech. I saw how machine learning could change the workflow of physicians. I got to spend time with a few physicians of the future and learn some technical skills like coding.
Paraphrasing a mentor, I learned that the hardest thing about coding was googling the right question for your technical situation. Learning to translate your problem into coding language. Asking the computer the right question.
Back to generative AI, Charlie Warzel points out in his article, “The Most Important Job Skill of This Century,” there is already a niche industry of prompt engineering developing. That is an industry of asking the right questions to get AI to do the work.
A good prompt engineer teases out the desired content from a generative AI in an iterative fashion, manipulating one variable at a time. Then lets the prompter perform quality control on the generative AI output.
This already is not much different than a good attending, prompting us in the Socratic method to a differential diagnosis or physiological mechanism. They iterate on their questions, in the fashion of the prompt engineer, to lead us to the answer they seek.
On a similar note, as I have written in “How to Fail 13* Classes & Get Into A Top Medical School,” no one tells you that your job in the third and fourth year of medical school is hearing a story from a patient and translating into one for your attending. The key to these transitional stories is the right questions. We ask the patient the right one, and it opens up their story to the diagnosis.
From there, on an operational note, the medical student writes the bulk of the documentation of the encounter. The intern performs quality control and tweaks it. The attending provides a further layer of quality control. In this regard, in some ways, the third or fourth-year medical student is really no different than the premed scribe, who is just as good, if not better, at writing the first iteration of these notes. It won’t take long for the computer to replace both.
What currently sets the medical student apart from the scribe is we have the now replaceable two years of didactic work under our belts and both the privilege and the ability to ask the patient questions.
Asking the right questions will forever be an essential skill. Arguably, the essential skill. Sid Mukherjee masterfully demonstrated this in 2017 in his article A.I. versus M.D.
What can and should we do?
The students focused on pumping out notes and memorizing answers to pimp questions may be left in the dust. Both of these tasks can and should be outsourced sooner than later.
The proper question can unlock a diagnosis from a patient in ways pan scanning never will. The proper question can also let us utilize generative AI in a way that makes us superhuman instead of replaceable.
Doing this a la Sid Mukherjee’s what, how, and why paradigm with a particular emphasis on the why questions seems protective.
Our education hasn’t caught up to the question-asking paradigm and won’t for some time. It is on us if we want to stick around. We’ll need to move into this brave new world with humility, curiosity, and an aim to cave our own path. But as the COVID class, alongside providers working today, we have demonstrated our resilience and ability to embrace change. If any generation of medicine is up to the task, I would bet on my classmates.
Jackson J. McCue is a medical student.
