Before you turn to AI chatbots like ChatGPT when you're feeling sick or want to understand a symptom, think twice. A new study warns that while AI can pass medical exams with top scores, it's still not a reliable doctor in real-world scenarios. British researchers tested nearly 1,300 people using AI tools including GPT-4o and Llama 3, and the results were disappointing. AI users could only correctly identify their condition about one-third of the time, and got treatment recommendations right less than half the time. This makes AI even less reliable than a basic internet search. Why do AI tools perform poorly? Researchers attribute it to "human-machine communication gaps." Real patients often fail to provide all necessary information, misunderstand the AI's instructions, or ignore the advice given. This is why researchers warn that using AI for urgent health needs can delay treatment and potentially be life-threatening. In the US, 1 in 6 adults now consult AI for health advice at least once a month — a genuinely dangerous trend. Experts strongly urge people to verify any health information with reliable medical sources rather than blindly trusting AI. Ref: Bangkok Post