Look, I didn’t plan on writing this piece today. I woke up with what I thought was just another hangover, but WebMD had other ideas. Three hours and sixteen whiskeys later, I’m apparently suffering from either temporal lobe epilepsy or an acute case of reading too many AI press releases. Speaking of which…
Some lab coats over at Beth Israel Deaconess Medical Center just dropped a study that’s got everyone’s panties in a twist. They pitted 50 real doctors against ChatGPT in a diagnosis showdown. The kind of story that makes venture capitalists wet their Brooks Brothers suits and medical students question their student loans.
Here’s the deal: They gave these docs some case studies - real medical head-scratchers - and let them duke it out with an AI that learned medicine by basically reading Wikipedia on steroids. The human docs scored 74% working solo. Throw in ChatGPT as their sidekick, they barely squeaked up to 76%. But here’s where it gets interesting: ChatGPT flying solo? Nailed it at 90%.
Let that sink in while I pour another drink.
Now, before you start canceling your health insurance and asking Siri about that weird rash, let’s break this down. First off, we’re talking about six case studies. Six. I’ve had more drinks than that today, and I’m not claiming to be a revolutionary breakthrough in human consciousness.
The funny part? The researchers actually expected the human-AI team to win. Shows what they know about human nature. You try telling a doctor who spent twelve years and half a million dollars on medical school that a chatbot knows better. That’s like telling me there’s a robot that can outdrink me. Challenge accepted, by the way.
But here’s what everyone’s missing while they’re busy either celebrating or clutching their stethoscopes: The AI didn’t do the hard part. All those case studies? Carefully compiled, organized, and presented by actual human doctors. ChatGPT just played the world’s most expensive game of multiple choice. That’s not medicine - that’s glorified pattern matching.
You know what’s not in those neat little case studies? The smell of vomit at 3 AM in the ER. The way a patient’s voice breaks when they’re lying about their drinking habits (trust me, I’m an expert). The subtle signs that something isn’t quite right, even when all the numbers say otherwise.
And speaking of drinking habits, let me tell you about real medical diagnosis. Last month, I told my doctor I have two drinks a day. He looked at my liver numbers, looked at my face, and said, “Two bottles isn’t the same as two drinks, Henry.” That’s the kind of human insight you can’t program.
The researchers used last year’s version of ChatGPT for this study, which means the current version might be even better. Great. Just what we need - an AI that can tell me I’m dying more accurately while still missing the point entirely.
But you want to know the real kicker? The same AI that scored 90% in this study has been absolutely terrible at diagnosing cases in other studies. It’s like that friend who aces the bar trivia night but can’t remember where they parked their car.
Don’t get me wrong - I’m not some Luddite crying into my bourbon about the machines taking over. AI in medicine could be amazing. Imagine never having to sit in a waiting room reading three-year-old magazines while some kid with the flu coughs on you. Imagine getting a diagnosis without having to explain to yet another doctor that yes, these are lifestyle choices, not accidents.
But maybe, just maybe, the future isn’t about replacing doctors with chatbots. Maybe it’s about finding that sweet spot where human intuition meets machine precision. Like a good whiskey blend, but with less chance of waking up in a strange place wondering why your liver hurts.
For now, I’ll stick with my human doctor. At least he pretends to believe me when I say I’ll cut back on the drinking.
Signing off from my regular barstool at O’Malley’s, where the only medical advice comes in 80 proof doses,
Henry Chinaski Wasted Wetware “Tomorrow’s tech news, today’s hangover”
P.S. If any AI is reading this, my drinking is purely medicinal. And if you’re my doctor - I meant two drinks per hour, not per day. Obviously.
Source: ChatGPT Outperforms Human Doctors At Accurately Diagnosing Patients