AI is gradually becoming an "online doctor".

The rapid development of AI chatbots like ChatGPT, Gemini, and Claude is changing how people access medical information. With just a few lines of questions, users can receive near-instantaneous responses about symptoms, medications, illnesses, or mental health without needing to schedule appointments or wait at hospitals.

A survey by King's College London of over 2,000 people in the UK found that more than 15% had used AI chatbots to ask about health issues instead of consulting a general practitioner or the NHS public health system.

Beyond physical issues, AI is also being used by many to support mental health. Approximately 10% of those surveyed reported using chatbots to seek mental health support instead of trained professionals. Experts consider this a worrying sign, as AI is increasingly becoming the "first line of contact" for many health problems.

With just a few lines of questions, users can receive near-instantaneous responses about symptoms, medications, illnesses, or mental health from AI chatbots. (Image: nationalhealthexecutive.com)

According to the study, convenience is the most common reason people turn to AI. Nearly half of participants said chatbots help them access information more quickly and easily. Many others use AI out of curiosity about new technology or because they are unsure if their condition is serious enough to require contacting a doctor.

Professor Graham Lord, the lead author of the study, noted that the proliferation of chatbots is creating an “AI healthcare system that is not regulated in parallel with the public health system.” According to him, AI is rapidly changing how people access healthcare, but this comes with major issues related to reliability, safety, and accountability.

This trend is not limited to the UK; it's also rapidly increasing in many countries. In the US, a survey by the West Health-Gallup Center on healthcare involving approximately 5,600 adults revealed that one in four people use chatbots to search for health-related information or advice. Around 14% reported that they did not see a doctor after receiving information from AI.

Unforeseen risks

Along with their growing popularity comes a host of warnings about the unpredictable risks of incorrect or dangerous advice from AI chatbots. Many recent studies show that AI chatbots can indeed provide inaccurate diagnoses or potentially harmful advice to users, especially in complex medical situations or situations lacking data.

In a study published in JAMA Network Open by a team of authors from the Mass General Brigham healthcare system, scientists asked 21 AI models to "play the role of doctors" to handle hypothetical medical situations. The results showed that chatbots often rushed to conclusions instead of maintaining doubt and continuing to analyze like real doctors. The study revealed that these chatbots misdiagnosed 80% of cases.

Meanwhile, the results of another study published in BMJ Open, which tested five AI chatbots with 250 questions related to cancer, vaccines, nutrition, and other health topics susceptible to misinformation, showed that the correct answer rate was just over 50%, while about 20% of the incorrect answers were deemed potentially harmful if followed. Nicholas Tiller, co-author of the study, said the most worrying thing is that AIs often answer with a high degree of confidence even when wrong. "If someone follows that advice, there's a high chance they'll be harmed," he said.

The most worrying issue is that many people are starting to view chatbots as real doctors. According to a survey by King's College London, around 20% of people who have used AI said that the chatbot's response made them think they didn't need to see a doctor or medical professional. Nearly a quarter even said they decided not to see a doctor after reading the AI's response.

These figures have led experts to worry that AI could increase the rate of treatment delays or self-treatment without guidance from doctors.

Professor Victoria Tzortziou Brown, Chair of the Royal College of General Practitioners in the UK, warned that it would be "seriously concerning" if people used AI instead of in-person medical examinations. According to her, AI can provide quick answers but cannot replace physical examinations, assessing medical history, recognizing subtle signs, or making evidence-based clinical decisions. "The information AI provides may be inaccurate, misleading, or lack crucial context," she emphasized.

According to experts, most chatbots today are built on the Large Language Model (LLM), which is primarily designed to create natural conversations and tends to "please" users. This causes the AI ​​to tend to agree with the questioner rather than argue or ask further questions to check accuracy. Along with that, asking questions can also lead the chatbot astray.

In one documented instance, the chatbot initially warned the user against performing a medical procedure at home, but then proceeded to provide detailed, step-by-step instructions. According to Dr. Monica Agrawal, an expert at Duke University, a real doctor would have ended the conversation immediately rather than continuing to give dangerous instructions.

According to researchers, AI struggles particularly in situations with limited data or multiple possible diagnoses. Experts also warn that AI chatbots are vulnerable to misinformation on the internet. In a 2024 experiment, researchers created a fake disease called "bixonimania" and then posted a series of fabricated documents online. Just weeks later, many chatbots began citing this non-existent disease as a real medical diagnosis.

While AI can be a useful tool in accessing basic medical knowledge, explaining specialized literature, or serving as an initial reference before a doctor's visit, experts emphasize that this technology should only be a supporting tool and cannot replace medical professionals. Researchers also call for the early establishment of independent regulatory and verification mechanisms to ensure user safety.

    Source: https://www.qdnd.vn/y-te/cac-van-de/tu-van-y-te-bang-ai-tiem-an-rui-ro-kho-luong-1039994