Artificial Intelligence Replacing Physicians in the Foreseeable Future: NO CHANCE!

Countless news outlets, bloggers, AI pundits, etc., have talked about how AI will replace physicians in the immediate future. 

They cite ChatGPT’s recent success passing the USMLE step 3 physician board exam as proof that AI will soon replace physicians.

Accordingly, it’s time to dispel the myth that AI will replace doctors in the relative future right now!

The people who are saying this have no clue about true medicine, and that includes all of the physicians who converted to healthtech for any one of a number of reasons, often involving money/lifestyle.

In fact, I don’t care who is saying it. Whether it’s tech people, pharmacists, “inadequate” physicians in the tech industry, etc.

People are ALWAYS trying to capitalize on the money in health care, and often have no idea what they are talking about.

They think they are the ones playing chess and everyone else is playing checkers, when it is the exact opposite.

Indeed, although a physician like me can learn the tech without formal training tech people cannot learn the medicine without formal training.

And just because you practiced medicine some time ago and transitioned to healthtech, or pharmaceuticals for that matter, doesn’t mean you are intimately familiar with modern day medicine or that you were every truly good at it. 

But that never stops these people from talking about what they have no idea about.

Indeed, that’s the world we live in.

People learn about a single drop of water on the surface of the ocean and suddenly think they’re intimately familiar with the ocean.

NOT ONLY DO THEY HAVE NO IDEA WHAT THE REST OF THE OCEAN SURFACE LOOKS LIKE THEY HAVE NO IDEA WHAT RESIDES BENEATH IT.

YES, AI will be good at providing answers to multiple-choice questions based on textbook patient presentations. In fact, it will probably best a lot of physicians at it.

The fact ChatGPT passed the medical boards means almost nothing to me.

I could have a human use the internet, including uptodate.com, right now and pass the medical boards if no time limit was involved.

You only need ChatGPT to automate and simplify the process, which is admittedly impressive for what it is (this is not an attack on ChatGPT which is truly mind-blowing and I use every day).

YES, AI will be able to follow guideline-based algorithms and provide treatment recommendations based on various inputs regarding patient characteristics, symptoms, etc.  Indeed, one of my companies is in this space, and others are already doing this at a somewhat suboptimal level.

These are simple decision trees. There is nothing to this process and one doesn’t actually NEED AI FOR THIS. Just go to NCCN guidelines for cancer and you can walk through all the cancer-based algorithms. I literally talked about this here (https://www.youtube.com/watch?v=AsaclBWksWo).

BUT HERE’S WHAT AI CANNOT DO.

AI will require decades to emulate the nuances of patient care intrinsic to the "stuff" you don't see in textbooks.

It will take forever before it will ever be me.

Initially, the amount of information I get about a patient just by looking at them is something AI cannot touch right now. 

Sometimes I will admit a patient to the hospital or hold chemotherapy based solely on how a patient looks. 

And that may not be reflected in their vital signs, lab values, etc.

But if it’s missed a patient can die.

Beyond that, there are countless times where patients don’t present like they do in the textbooks. Where you have to decide which symptoms are important and which are not. Which symptoms are red herrings and which ones aren’t. 

In AI and machine learning algorithms, weights are ascribed to different parameters to provide an output. These weights are adjusted constantly through reinforcement learning to yield an optimal accuracy, F1-score, precision, recall, etc. (although, many AI companies avoid talking about F1-score, precision, and recall, because they can hide behind accuracy which often is an “inaccurate” measure of an AI’s efficacy).

But here’s the thing about experience. Here is why experience matters.

Think of experience as a human’s reinforcement learning process. It’s how humans learn the appropriate weights to give to different facets of information. 

And that will separate human physicians from AI for a long time to come.

And before you say we can input that into an AI and let the AI learn, good luck even knowing what the thousands of different permutations one can face in this context actually are, and then getting the data required to adequately train an AI in that regard.

And before you say AI can learn to do it better than physicians. Good luck proving AI can do better than I can at seeing all the pure medicine, cancer-related, social-related, etc., issues pertaining to a patient, and make a decision based on that.

Yes, for simple chemotherapy decisions that are algorithmic AI can do well. Hell, we could automate that without AI with basic rule-based decision trees. Just because companies like Flatiron Health, IBM, etc., can fool people into thinking this is a huge deal doesn’t mean it is. But practicing hematology/oncology and making treatment decisions goes way beyond that.

People are quoting a study that showed that AI performed better than NCCN guidelines. 

Let’s see such a trial ACTUALLY PERFORMED ON PATIENTS WHERE THE AI DECIDES EVERYTHING, NOT JUST THE CHEMOTHERAPY A PATIENT GETS, BUT WHEN TO ADMIT THEM, WHICH OTHER MEDICATIONS TO CHANGE, WHEN TO IGNORE MEDICATION-MEDICATION INTERACTIONS, HOW TO ADJUST FOR SUPPLEMENTS, ETC.

AI companies, contrary to anything they might say, have no shot at anything close to me for a very long time.

Indeed, you even have to know when to dismiss a conventional thought, which AI is going to have a really hard time with in medicine. 

The best example of this in drug-drug interactions that pharmaceutical algorithms/AI spit out that are best ignored. I can’t tell you how many of these aren’t clinically relevant and it’s annoying to have to constantly resolve them before a patient can get a medicine.

Not to mention all the times the guidelines don’t reflect current practice, are too small minded, etc.  AI will not be able to know when to go off script because they will train on the textbook, not the art.

AI will not be able to account for all the layers that constitute a patient. It’s not just about what medications their taking or the labs and imaging you see. It’s about how much and what they’re eating, how much they’re exercising, where they live, what’s going on with them socially.

Consider this example…

AI will be used to employ pharmacogenomics related information. 

But think of a patient on coumadin whose dose is determined by their genetic composition.

Well… that’s fine and dandy, but it all falls apart if they choose to eat a massive salad or forget to take a dose of their coumadin. 

No AI is accounting for this properly in the context of everything else going on.

Indeed, I can list thousands of reasons why it’s going to take a very long time for an AI to replace a physician like me.

“Inadequate” doctors, of which there are A TON, who don’t work to stay current, care mostly about money, etc., will be effectively replaced by AI, but not in the ways you think.

The good, excellent, and exceptional doctors cannot adequately be replaced by AI without losing a ton.

Even the average doctors can’t be replaced by AI any time in the foreseeable future.

If you really want to talk about AI replacing doctors, YOU NEED TO READ THESE TWO ARTICLES FIRST.

https://www.biopharmatrend.com/post/505-playing-chess-against-cancer-a-pharmaceutical-biotechnological-and-clinical-guide-to-modern-day-oncologic-treatment-cartography/

https://www.biopharmatrend.com/post/510-quarterbacking-a-patients-cancer-care-for-submission/

After reading these two articles you will realize first-hand that the AI pundits are playing checkers, not chess.

You will see it’s going to take forever for AI to catch good physicians because good physicians see a forest AI cannot touch.

Anyone else telling you otherwise has no idea what they’re talking about, and that includes all the healthcare tech pundits who think they know all there is to know about medicine despite the fact they have never taken care of a patient or been educated in medicine.

Practicing EXCEPTIONAL medicine isn’t IBM Watson playing chess, DeepMind playing Alpha Go, ChatGPT writing a poem once prompted, etc. 

Oh, and as much as I love nurse practitioners (NP) and physician assistants (PA), if you think there is a single one that can remotely emulate what I can do you are on crack. 

Just look at the body of content I have on LinkedIn, biopharmatrend.com (https://www.biopharmatrend.com/topic/the-insiders-guide-to-translational-medicine/posts/), revolutioncancer.com, rcimtb.com, and the Revolution Cancer youtube.com channel (https://www.youtube.com/channel/UCtzupz9dOvIeKKTXFKqAI9A).

No physician is matching that, let alone an NP or PA.

Again, knowing some facet of the surface of the ocean doesn’t mean you have any real understanding of the ocean.

There is NO chance for a PA or NP, or for AI replacing GOOD physicians at least not right now.

And if you think otherwise good luck when you get cancer and people like me are nowhere to be found.

• Image taken from: https://padhamhealthnews.org/artificial-intelligence-in-healthcare-can-it-replace-the-doctor/