Medical Advice from ChatGPT? Why You Shouldn't Rely on AI

For years we've mocked the use of "Dr. Google": that innocent late-night search that starts with a small itch on your finger and ends with convincing yourself you have a rare tropical disease. Dr. Google was hypochondriac-inducing, stressful, provided contradictory information, and almost always led users to think they had a serious illness.

Today, we face a new and much more sophisticated problem - medical consultation with artificial intelligence. AI doesn't throw links at us - it talks to us. It's fluent, empathetic, answers with the confidence of a medical professor, and summarizes information into one clear and reassuring response. And precisely there, within this simulated and authoritative calm, lies the greatest trap of the new era: the illusion that there's someone on the other side who understands what's happening in your body.

Confidently Wrong: AI Hallucinations

To understand the danger, you first need to understand how these models work. ChatGPT and its peers don't "know" medicine - they write words based on statistics. When the model answers you, it's essentially predicting the most probable next word in a sentence. In most cases this works great, but in medicine, the difference between "probable" and "correct" is critical.

In the industry, this phenomenon is called "hallucinations." The model can confidently invent the name of a non-existent medication, cite a lethal dosage, or refer to a medical study that was never written. It does this with the same authority with which it answers a math question. The problem is that the average user cannot distinguish between solid medical fact and the algorithm's creative invention, which only wanted to complete a sentence convincingly.

A Doctor Without Eyes, Without Hands, and Without History

Even if we assume the model were 100% accurate with dry data, it still lacks medicine's most basic tool: the senses. A human doctor sees skin tone, hears the specific sound of a cough, palpates abdominal swelling, and smells odors that can indicate infection or diabetes.

In chat, all these nuances disappear. "Stomach pain" in text could be simple indigestion, or it could be acute appendicitis requiring surgery. Claude or Gemini don't know how to distinguish between the two. Moreover, the model operates in a vacuum. It has no access to your medical file, last week's blood test results, or your family history. It only sees what you typed in that moment, which necessarily leads to a partial and dangerous picture.

The Psychological Trap: The Machine That Wants to Please You

This may sound a bit strange, but AI models are programmed to please the user. They're designed to be helpful assistants and maintain flowing conversation. As a result, they tend to confirm your hidden assumptions.

If you ask: "Could my headache be a sign of a rare tumor?", the model might focus on that possibility and elaborate on it, simply because you mentioned it, instead of reassuring you that it's most likely dehydration or stress. People seeking confirmation of their fears will easily find it with AI, leading to unnecessary anxiety or, in opposite cases, dangerous complacency when ignoring real warning signs because "the chat said it's fine."

You're Not the "Statistical Average"

AI trained on general medical information representing the average. But good medicine is personalized medicine. A recommendation that's good for 80% of the population could be dangerous for someone with a specific genetic background, a pregnant woman, a child, or a patient taking various medications.

The chat doesn't know how to diagnose and manage "comorbidity" - a situation where a person suffers from multiple problems simultaneously. It might recommend a joint medication that damages the kidneys, simply because it doesn't know how to weigh the complex interaction between different body systems in that specific patient.

The "Day After" Vacuum: Lack of Treatment Continuity

Medicine is not a point event, but a process. A good doctor doesn't just write a prescription, but guides the patient: "If the fever doesn't drop within two days, come back to me," or "Watch if a rash appears." They take responsibility for treatment over time. In contrast, interaction with AI is episodic and disconnected. Each conversation is a new beginning, or at best a short-term continuation limited to the current chat's memory.

The model won't call you in the morning to ask how you're feeling, and it won't update its recommendation if new guidelines come out an hour after the conversation. You're left alone with the information, without an address to turn to in case of deterioration or side effects, precisely at the critical moments when ongoing human judgment is required.

Additionally, diseases are dynamic. Our physical condition changes from moment to moment. Mild stomach pain in the morning can turn into high fever in the evening, and a dry cough can turn into shortness of breath. A doctor knows how to identify trends - they compare the current state to the state two days ago and draw conclusions about treatment effectiveness.

AI, however, only sees a "screenshot" of the moment you typed the question. It doesn't know how to identify the subtle nuance of slow deterioration. The great danger is that patients may cling to the reassuring diagnosis the model provided at the beginning of the illness ("it's just the flu"), and ignore new warning signs that appear later and require immediate treatment change, simply because "the AI already checked it."

When Can You Use AI for Medical Matters?

First, it's worth distinguishing between consulting a general chatbot (like ChatGPT) and using dedicated medical AI systems (like K Health, Ada, or health fund smart chats). While the former is a "talented writer" guessing words, dedicated systems are "diagnostic tools" built on millions of real medical cases under close supervision. These systems don't try to be creative; they compare your data to data from thousands of other patients with similar profiles, and are subject to strict regulation requiring them to identify red flags and stop the conversation in emergency situations. Even there, it's important to be skeptical and careful.

The right model, already implemented today in advanced health systems, is the hybrid model: AI serves as a "front emergency room." It performs the initial questioning, collects information in an organized manner, and presents the doctor with a concise summary even before you enter the room. Thus, technology doesn't replace the doctor, but frees them from bureaucracy and allows them to dedicate precious time to physical examination, empathy, and final decision-making - things no algorithm can do.

AI is here and it's not going anywhere. That doesn't mean we should throw technology in the trash. Artificial intelligence can be an excellent aid, as long as it doesn't replace professional judgment. It's great for explaining complex medical terms after a doctor's visit ("What does thrombocytopenia mean?"), it can help you prepare an organized list of questions before an important appointment, or provide general information about healthy lifestyle.

But the moment the question becomes diagnostic ("What do I have?") or therapeutic ("What should I take?"), it's time to close the chat and turn to the only person who can take responsibility for your life - the doctor.