AI for Medical Advice: How AI is Creating Three Truths — And No Agreement

It’s 2 AM, and Sarah can’t sleep. The cough that’s been nagging her for three weeks suddenly feels more ominous in the dark. She reaches for her phone — not to call anyone, but to open ChatGPT. She’s hardly alone. Over 40 million people globally use ChatGPT for healthcare questions every day, and she’s one of them.

She types out her symptoms: persistent dry cough, occasional chest tightness, no fever. The AI responds thoughtfully, suggesting it could be anything from bronchitis to asthma, possibly pneumonia. It recommends seeing a doctor. She screenshots the conversation, feeling both reassured and uncertain, and finally falls asleep.

The next morning, she’s in her doctor’s office. Her physician — who’s among the 66% of U.S. doctors now using AI in their practice, up from just 38% last year — pulls up her own diagnostic assistant. She inputs Sarah’s symptoms. This AI, trained on different data and designed for clinical use, suggests a viral infection and notes that there’s been an outbreak in the area. “Let’s do a chest X-ray to be safe,” the doctor says.

At the hospital, the radiology AI analyzes Sarah’s scan and flags a small nodule, assigning it a risk score that triggers an automatic follow-up protocol. Three different AI systems. Three different assessments. One increasingly anxious patient sitting in the middle, wondering which one to believe.

When Everyone’s AI Knows Something Different

Here’s what we’re not talking about enough: This isn’t a story about whether AI is accurate. It’s about something far more fundamental to how healthcare actually works. We’re creating parallel universes of medical knowledge, and we’ve built exactly zero bridges between them.

Think about what’s actually happening. Right now, one in four ChatGPT users submits a health-related prompt every week. That’s hundreds of millions of conversations about symptoms, medications, and diagnoses — conversations that 70% of the time happen outside clinic hours when no doctor is available. These aren’t frivolous queries. People are checking symptoms (55% of users), trying to understand medical terminology (48%), and researching treatment options (44%).

But here’s where it gets complicated: 76% of people who use symptom checkers never consult a physician about what they learned. They’re making medical decisions based on one AI’s assessment, in isolation, often because they have no other choice. In rural areas designated as “hospital deserts” — places more than 30 minutes from the nearest hospital — ChatGPT receives an average of 580,000 healthcare-related messages every week. Wyoming, Oregon, Montana, South Dakota, and Vermont lead the nation in these isolated AI consultations.

Meanwhile, their doctors are using AI too, but for completely different purposes. When physicians adopted AI tools at a rate of 78% increase in a single year, what were they using it for? The top answer isn’t diagnosis — it’s documentation. Twenty-one percent use AI for billing codes and medical charts. Twenty percent for discharge instructions. Only 12% are using it for diagnostic assistance. They’re drowning in bureaucratic work (53% of physicians showed signs of burnout in 2023), and AI is their life raft. But their AI isn’t talking to their patient’s AI.

The Geography of Disconnection

The fragmentation hits hardest where healthcare is already most fragile. Nearly 4.5 million Americans live in counties without an acute care hospital. In 84% of rural counties, people face “ambulance deserts” where it takes more than 25 minutes to reach emergency services. When you’re that far from medical care, AI isn’t a nice-to-have convenience — it becomes your first, and sometimes only, medical consultation.

But here’s the brutal irony: 28% of rural residents don’t have access to broadband that meets minimum speed requirements for reliable telehealth. Even those who do have internet are 42% less likely to use telehealth services than people in metropolitan areas. The same populations with 65% shortages of primary care physicians are also the least equipped to access the AI tools that might help bridge the gap.

I think about this often: the people who most need AI-assisted healthcare are beta-testing a system the rest of us will eventually use, but they’re doing it with worse infrastructure, fewer backup options, and higher stakes. Someone in rural Wyoming gets a ChatGPT diagnosis, drives 90 minutes to see a doctor using a different AI diagnostic tool, and receives contradictory advice. Who’s right? Who decides? There’s no protocol for that.

The digital divide isn’t just about access — it’s about integration. In high-needs rural areas, only 76% of households own smartphones compared to 88% average nationwide. Only 60% own laptops compared to 79% elsewhere. Even when people can access AI health tools, they’re often doing it on inadequate devices with unreliable connections, making it harder to share information with their doctors or access follow-up care.

The Authority Vacuum Nobody’s Talking About

We’ve accidentally created a new kind of information asymmetry in healthcare, and it runs in multiple directions at once.

Your doctor doesn’t know about your 2 AM ChatGPT consultation. You don’t understand the AI risk scores flagging your case in the hospital system. The hospital’s algorithms aren’t informed by either your symptom searches or your doctor’s clinical AI. Everyone has access to artificial intelligence. Nobody has access to the complete picture.

This matters more than you might think. Traditional medicine always involved power dynamics — the doctor knew things you didn’t — but it was a single axis of knowledge. Now we have three or four axes, and they don’t intersect. You arrive at your appointment having researched your symptoms using AI tools that achieve about 45% diagnostic accuracy in real-world studies. Your doctor is using a different AI tool with different strengths and weaknesses. The hospital system is running risk algorithms neither of you fully understand. All three might be partially right and partially wrong, and there’s no established way to reconcile the differences.

The numbers tell us people are worried about this, even if they can’t quite articulate why. Sixty-eight percent of Americans fear that AI will weaken the patient-provider relationship. They’re not wrong — but not for the reason they think. The problem isn’t that AI creates emotional distance. It’s that we’re creating information silos without any bridges between them. Eighty-nine percent of people say clinicians need to be “clear and transparent” about AI use, but transparency without integration just means both sides acknowledge they’re working from different maps of the same territory.

Mental health shows us how dangerous this fragmentation can become. Forty-six percent of AI symptom checker users are searching for information about anxiety and depression — conditions where the patient’s own reporting is crucial, where the therapeutic relationship matters most, and where different AI assessments could lead to genuinely harmful divergence in understanding. Your AI chatbot tells you one thing about your mental state at 2 AM. Your therapist’s clinical judgment suggests another. Your insurance company’s AI denies coverage based on a third set of criteria. You’re not getting three opinions — you’re getting three incompatible realities.

It’s Not About Better AI — It’s About Better Bridges

Here’s what frustrates me most: We keep having the wrong conversation. We debate whether AI is accurate enough, whether it will replace doctors, whether patients should trust it. But those questions miss the point entirely.

The real problem isn’t that we need smarter AI. It’s that we need the AIs we already have to talk to each other — and we need humans who can facilitate that conversation. Your ChatGPT exchange doesn’t transfer to your doctor’s diagnostic system. Your doctor’s AI-generated notes don’t inform the hospital’s risk algorithms. The hospital’s findings don’t update the population health models that other patients consult. We’ve built a Tower of Babel, but for medicine.

Every major healthcare system surveyed in 2024–100% of them — reported adopting ambient AI for clinical documentation. That’s universal adoption in less than two years. But exactly zero percent implemented consensus mechanisms for reconciling the knowledge generated by patient AI, physician AI, and institutional AI. We’re solving the transcription problem while completely ignoring the integration problem.

And here’s what really keeps me up at night: The speed of adoption means this fragmentation is getting worse, not better. Sixty-six percent of healthcare leaders report increased burnout, stress, and mental health issues in their workforce. The system is under impossible strain. Physicians adopted AI not to transform healthcare but to survive it — 21% using AI for billing codes versus 12% for diagnosis tells you everything you need to know about what problem we’re actually solving.

We’re automating around the dysfunction instead of fixing it. We’re creating more sophisticated tools for working within a broken system rather than building the coordination infrastructure that might actually heal it.

The Consensus Crisis We’re Not Having

Let me be clear: I’m not arguing against AI in healthcare. The evidence suggests it helps in many ways. Eighty percent of Americans believe AI can improve healthcare quality, reduce costs, and increase access. Seventy-eight percent of patients say they’d use telehealth again. The adoption is happening whether we coordinate it or not, and there are genuine benefits.

What concerns me is that we’re conducting a massive, unplanned experiment in fragmented medical knowledge. Forty million people daily consulting one kind of AI. Millions of physicians consulting different AIs. Thousands of hospitals running independent AI systems. Zero consensus mechanisms connecting them.

We’re asking the wrong questions. Not “Is AI accurate enough?” but “Who reconciles when AIs disagree?” Not “Will patients trust AI?” but “How do we integrate patient AI knowledge into professional care?” Not “Can AI reduce physician burnout?” but “Are we just automating around systems that need fundamental restructuring?”

The irony is almost painful. Healthcare has always been about achieving consensus — diagnosis requires agreement between symptoms, physical examination, test results, and clinical judgment. We’ve added powerful new intelligence sources, but we’ve created no framework for bringing them into conversation with each other or with human expertise. We’re optimizing individual components while fragmenting the whole system.

Sarah, the patient from our opening story, isn’t just facing three different medical opinions. She’s navigating three incompatible realities with no one — not her doctor, not the hospital, not the AI developers — equipped to bridge between them. The knowledge exists to help her. The intelligence is real. But it’s scattered across systems that don’t communicate, held by parties who can’t see each other’s information, interpreted through frameworks that have no common language.

And she’s not alone. She’s one of millions navigating this consensus void every single day, making medical decisions based on partial information from incompatible sources. The question isn’t whether AI belongs in healthcare — it’s already there, woven into every level of the system. The question is: Who’s building the bridges between these islands of intelligence we’ve created?

Because right now, the answer is nobody. And that should worry all of us.

Jean Marie Bonthous (publishing as JM Bonthous) is the author of more than two dozen books, including six on the human side of artificial intelligence, six about filmmaking, and four about digital/AI art. See his latest books: www.jmbonthous.com