I have had a chance to use Chat GPT, and want to offer these preliminary thoughts for how it will affect medicine. Broadly, I will divide my comments into how it will affect medical writing and then medical practice.
Of course, if you haven’t played around with Chat GPT (March 14th edition) you probably will not understand what I saying. I encourage you to engage with the software.
Here is my quick summary: Chat GPT responds to questions or prompts with coherent and *mostly* accurate answers in complete sentences and full paragraphs. It can draw upon a deep well of data, and it can make arguments and comparisons. It can even draft statements or documents. It can formulate interesting, but not highly original arguments. If you want some ideas for how to get started ask it to tell you about minor celebrities or imitate the style of writers or other noted people. Ask it to draft a tweet in the style of <insert person>. The responses may be uncanny.
Now let us turn to my thoughts on its future.
Chat GPT will surely be used to be an affordable scribe or medical writer. Integration with EMR means Chat GPT will draft your clinic notes at the start of the encounter. The quality will quickly exceed that of medical students, residents and likely even fellows.
This point has been made by others, but the deeper point I want to make is just how silly it is we spend so much time documenting. This is largely done for billing purposes, and these documents are mostly read by no one. This entire billing system needs reformation.
A better system for documentation would be to have a single wikipedia page summarizing a patients key medical history— linking to documents, reports, and scans. Likely, Chat GPT will draft this, and synthesize a sea of Chat GPT notes into a single 2 page summary. Oncology is complicated, but I have never met a patient I could not summarize in 1000 carefully chosen words. Chat GPT will change the very way the chart exists and is read. (A-B reinforcement will accelerate this.)
When it comes to academic writing, Chat GPT will soon become the affordable medical writer. Medical writers will supervise Chat GPTs— and only the ones with the highest levels of content knowledge will survive. Most will need new jobs.
Most review articles will be written by Chat GPT. They will also be mostly read by Chat GPT, and users will ask the software to provide shorter summaries. Ironically then, Chat GPT will be used both to generate writing and parse it into digestible pieces.
Practically, this means that scribe jobs are over, and even many academics will be obliterated. The only ones that can survive are the ones with highly original thoughts, who can develop original arguments and insights. These are few. Most suffer from groupthink, or copy ideas they heard.
In the world of Chat GPT— it will be harder to distinguish oneself this way. Currently, original thinkers are limited by the time it takes to draft one’s ideas. With Chat GPT this restriction will be removed. Copying will be hard as original thinkers will have a head start and unlimited drafting potential. My team published 61 articles in the last year with tremendous effort. Meanwhile we had many more unique ideas we did not have time to draft. With Chat GPT we will be unchained and will easily hit 100-200 articles per annum. Research teams that borrow our ideas will be at severe disadvantage.
Chat GPT will permit more mediocre people to become doctors. There are many important aspects of being a doctor. Speaking compassionately & empathetically, performing physical labor, and making sound medical decisions. These are all equally important, but Chat GPT will dramatically change the last.
Currently, a mediocre thinker can rely on algorithms or flow charts to make medical decisions, but these can be in error, and the person may have doubt. Chat GPT will be used to make these decisions going forward. Currently, you can type the details of a case and Chat GPT can give treatment ideas. In the future, Chat GPT will extract details from the chart and make recommendations by itself.
Again, only very clever doctors will be able to exceed the quality of these recommendations. Most doctors will actually do better — in the long run— by accepting Chat GPT recommendations (likely this will be shown in RCT powered for all cause death).
Once this happens, and given other trends in America, very likely more mediocre thinkers will be recruited into medicine. I am not using mediocre as an insult here. I am just trying to say that when it comes to critical medical thinking there is a spectrum, and Chat GPT will level this tremendously. Of course, the very brightest doctors will outperform the software, but these will be few and far between.
Already we are eliminating grades and test scores as barriers to medicine, and Chat GPT will ‘validate’ this choice. Clinical medicine will likely improve with Chat GPT, but negative implications for advancement of knowledge may occur. Meanwhile, midlevel providers will be empowered by ChatGPT and be used to fill massive gaps in health care delivery.
The old joke is that the doctor at the dinner party is the expert on all topics— politics, business and medicine. Chat GPT will undermine this stereotype. Recruiting the very smartest into medicine will become a larger issue (IMO: it is already an issue).
Disclosure: This essay is entirely speculation; Imagination plus experience of the software. I think I will be vindicated in these predictions, or I would have kept them to myself, but time will tell. Mostly, this essay is meant to be a fun mental exploration (fun for me, at least!). If you agree or disagree, subscribe, and put it in the comments. I hope to write more essays on this topic as I get ideas going forward.
This essay was written by Chat GPT.
Kidding, I wrote every word my damn self. -vp
Vinay, This is my area of expertise on which I have been publishing since the very first article on Medical Informatics I wrote for Science many decades ago. ChatGPT, on which I have pretty deep experience, is a language engine, not an all-knowing "artificial intelligence". As you play with it more, you will discover that we have yet to achieve "artificial stupidity"....and it is a long slog from there to anything "intelligent".
I am shocked to see you endorsing "cookbook medicine" as the better default for the future. Without an ontologic understanding of what is happening with each patient, ONE BY ONE, there can be words that sound good but that will likely be wrong. None of that is today possible (or even being pursued) in large language engines. The cookbook will be often wrong in unpredictable ways. I know that the government (witness the cookbook response to covid) would love this kind of health care...but it is wrong on every level.
Half a century ago there was a program called Eliza that, similarly (but much more simply) did psychiatric interviews with users. It was very persuasive (and people would interact with it for hours) but, in actuality, had nothing to contribute. Perhaps large language engines may facilitate writing notes and replace scribes...I agree that much medical documentation is nonsense. But the leap to "doctoring" and the default that doctors will just be dumber (yes, happening now but that is LCME's fault and the failure of schools like UCSF (and mine) to push back) but cookbook language processing will take over is a leap way too far.
“This point has been made by others, but the deeper point I want to make is just how silly it is we spend so much time documenting. This is largely done for billing purposes, and these documents are mostly read by no one. This entire billing system needs reformation.”
This is so true. Imagine if lawyers had to document this thoroughly for every hour they bill us, or if your mechanic had to write a paragraph justifying each oil change… it’s just insane and I can’t believe this is real life