Your AI super doctor will see you now
A 101 on how to benefit from Dr. GPT
In today’s Diary:
The 1st new antibiotic class in decades
Catching lung cancer 6 years early
AI will kill these medical jobs
AI saves dog’s life
Your Dr. GPT “how to”
“Should we bring him to the hospital?”
You never want to hear that from your wife at 5am.
Our 7-month-old son had woken up with angry red blotches all over his body. Measles?
Instead of going to the ER, I remembered my friend Tyler Cowen’s sage advice: You should ask AI more questions.
I snapped two photos of the rash and asked ChatGPT to evaluate.
In less than five seconds, ChatGPT said it was likely a common viral infection, nothing more serious. It also gave me specific warning signs that would warrant a trip to the ER.
The anxiety vanished, and we went back to sleep.
We’ve all heard this advice:
“Don’t Google your symptoms.”
WebMD.com is no substitute for a real doctor. AI is different. “Doctor GPT” is already much better than a human doctor in some important ways.
This is not hyperbole. ChatGPT aced the United States Medical Licensing Exam, the grueling, three-part marathon required to practice medicine.
And when Microsoft pitted its AI against a team of human specialists on the toughest mystery cases in the New England Journal of Medicine, the human experts got the diagnosis right 20% of the time.
The AI scored 85%.
How is it possible that AI beat human experts by 4X?
In part because the AI team created a virtual expert panel made up of different AI agents. These agents then debated diagnoses, challenged each other's assumptions, and even considered the cost of tests.
AI is almost perfectly suited to solving complex medical mysteries. Biology is staggeringly complicated. A single human genome contains three billion letters of code. The number of potential interactions between proteins in our bodies is infinite.
No human brain can even begin to grasp it all.
But AI can read every medical paper ever written and quickly spot patterns no human could.
Take 37-year-old Joseph Coates for example. His body was ravaged by a rare blood disorder called POEMS syndrome. His doctors tried everything. Nothing worked. So they sent him to a hospice to die.
In a last-ditch act of desperation, his girlfriend emailed a researcher at the University of Pennsylvania who was using AI to hunt for rare disease cures. The AI scanned thousands of existing drugs, looking for a new combination that might work.
It found a cocktail of drugs never used for POEMS syndrome before.
Joseph is now in remission.
Tens of millions of people live with rare diseases. 90% of them have no FDA-approved treatment.
How many more solutions are buried in the haystack of existing research?
We just need the right algorithms to find them. AI can connect a paper from 1970 with a lab result from yesterday to find a link no human doctor possibly could.
That’s how companies like Insilico Medicine are using AI to design and validate new drug candidates in less than 18 months. Without AI, that process takes a decade and billions of dollars.
Meanwhile, MIT researchers used AI to discover the first new class of antibiotic in decades by sifting through 100 million compounds in a matter of days!
There are so many stories of AI saving lives, it’s hard to keep up.
Think about it this way. When you ask a doctor, “How serious is it?" you're getting an educated guess based on that one doctor’s personal experience.
When you ask AI, you’re tapping into a system that’s “seen” a million times more cases than the most experienced doctor.
Know what’s even better than having an AI super doctor to talk with 24/7?
Preventing serious medical issues altogether.
Our healthcare system is reactive. We often wait for something to go wrong, then rush to fix it, only after getting permission from our insurance providers.
But when you identify a serious medical problem can be the difference between life and death. If you identify and treat a cancer when the earliest signs appear, it can end up being just “a scare.” If you don’t know about it for 5 years and it’s already gained a foothold, the outcome is much worse.
AI can catch serious diseases much earlier than humans alone can.
One example: MIT researchers built an AI algorithm that can analyse a routine CT scan and predict the likelihood of a person developing lung cancer up to six years in the future.
Another: the new AI-powered Trio blood test can detect 18 different early-stage cancers with 95% sensitivity.
Soon your annual check-up could include a blood test screening for dozens of nascent diseases.
Imagine a future where your smartwatch passively collects biomarkers. This data is fed into a personal AI model that understands you.
It will know you’re getting sick before you feel symptoms. It might tell you about a subtle change in your blood proteins suggesting a heart attack down the line – giving you time to prevent it.
What about all the doctors AI will replace?
In 2016 AI pioneer Geoff Hinton predicted the end of radiologists.
AI was getting so good at reading medical images, human “pros” would soon be out of a job.
His prediction aged like warm sushi.
The Mayo Clinic has increased its radiology staff by 55% while adding a 40-person AI team to build tools for them.
Doomsday job predictions never take scarcity into account. As radiology and other expensive medical services get cheaper and more abundant, more people will use them.
All those borderline cases when you probably should get an MRI, but don’t because of insurance decrees, availability or cost, can go away.
You can just get the MRI.
This will result in more medical jobs, not fewer.
But AI will destroy this one type of medical job.
Paper filler-outers, also known as administrators.
Bureaucracy has dehumanized medicine. Many doctors are too busy typing notes to look you in the eye. A recent study found doctors spend nearly two hours on paperwork for every hour they see patients.
“Administrative” positions in healthcare have metastasized like an out of control cancer, driving increases in cost.
AI scribes can now automatically document every patient visit, giving doctors the time to be more human.
AI also puts you more in charge of your own medical decisions.
Modern American healthcare is permission-slip hell. Want to see a specialist? Get a referral. Need a specific treatment? Get prior authorization from an insurance company.
Meanwhile, your blood work, scans and medical history are held hostage in clunky patient portals.
AI can smash that system and put you in charge. It won’t replace human doctors. It will make their services more accessible and abundant.
AI isn’t error free. You must get a second opinion after your visit with Doc’ GPT, just as you would with a human doctor.
That said, AI’s ability to dream up novel ideas could be one of its greatest strengths.
For decades, catheters have been a major source of bacterial infections. Scientists asked AI to design a better one.
The AI hallucinated something no human engineer had thought of…
A catheter lined with tiny, sawtooth-like spikes on the inside, which prevent bacteria from swimming upstream.
It reminds me of DeepMind's AI beating the world champion at the game “Go.” In the middle of the game, it made a move so strange human experts called it a mistake.
But as the game unfolded, they realized it was a stroke of genius that secured the win.
How to use Dr. GPT 101
AI is as simple or as complex as you make it. I snapped photos of my son’s rash and got an instant answer. Simple.
If you need deeper analysis of a chronic illness or complex medical problem, make ChatGPT your project manager.
Step 1: Collect data. Gather every piece of related medical information you can find and put it in one place. Don't worry about making it neat. Just get the data. You can put it all in one big unformatted Word document.
Step 2: Open a new chat with ChatGPT. Start uploading all the data you collected. Dump every puzzle piece onto the table. Then give the AI a simple command:
Organize all this information into an easy-to-read medical file.
Step 3: Start asking questions. Begin with big picture queries:
Based on everything you have seen, what is your best estimate of what’s wrong? Explain your reasoning step-by-step, highlighting the most critical pieces of evidence. What are the most likely scenarios, and what is the best course of action for each?
Irish people tend to be afraid of the doctor. They put on nice clothes and sit quietly like scared school kids in the waiting room. And they NEVER ask the doctor questions.
The golden rule of AI is: Ask more questions. Wondering whether a specific pill or diet could cure your ills? Ask. There’s no additional co-pay. The AI will answer you day and night. And you can constantly update your “file” with new medical information.
You can also get a second or third opinion from a different AI. Take ChatGPT’s medical analysis and paste it into a new chat with Claude or Gemini.
Then use this prompt:
An AI model gave me the following analysis of my medical case. Please review it, critique it, and tell me what you think it missed or where you disagree.
Until now, this level of analytical rigor was reserved for the best hospitals in the world. Now it’s yours for $20 a month.
At the very least, use AI to squeeze the most out of your next in-person doctor's appointment: "I've been reviewing my records, and I'm concerned about these specific results. Can you help me understand what’s going on?”
In the years ahead, the AI scoreboard will read:
Lives saved: Millions. Lives lost: 0.
As my friend Louis Anslow, creator of the excellent Pessimist’s Archive puts it, “It is much more likely millions of people die because there is too little AI applied too slowly than too much AI applied too soon.”
The enemy here isn't a rogue AI algorithm. It’s the permission-slip culture. It’s the lack of doctors and overworked nurses. It’s the decade-long approval process for a new drug. It’s the lives we could save but don’t.
Last story: a man’s dog got very sick. The blood test showed its anemia was dangerously severe. But the vet was stumped.
In a frantic car ride to a different vet to get a second opinion, the dog’s owner dumped every symptom and test result into ChatGPT.
The AI suggested the dog's immune system might be attacking its red blood cells.
Armed with this insight, the owner walked into the second vet's office and asked: "Could it be immune-mediated hemolytic anemia?"
The vet ran the test. The AI was right!
A dog’s life saved.
Make sure to click “Like” and “Restack” below.
—Stephen McBride
Stephen McBride is a co-founder of the Rational Optimist Society.
Ai. Tell me what this article missed and arguments counter to this view that I should consider as well.
Great article. At 82, now I know where to start.