Home / Technology / AI Health Advice: Accurate or Dangerous?
AI Health Advice: Accurate or Dangerous?
20 Apr
Summary
- Chatbots provide problematic medical answers nearly 50% of the time.
- AI references are often incomplete or entirely fabricated.
- Users struggle to correctly interpret AI's medical information.
Recent research indicates that widely used AI chatbots, including ChatGPT and Gemini, exhibit significant inaccuracies when answering health-related questions. A study published in BMJ Open found that almost 20% of chatbot responses were highly problematic, and half were problematic overall, with errors in reference citations being a common issue. Chatbots struggle most with nuanced topics like nutrition and athletic performance, often providing fluent yet potentially harmful advice for open-ended queries.
Further studies highlight user comprehension challenges, with real people obtaining correct answers from chatbots less than 35% of the time. AI models also faltered in providing accurate medical diagnoses based on limited patient information, though accuracy improved with more data. These findings underscore that while chatbots can be useful for research starting points or question preparation, they are not reliable medical authorities and users must verify all health claims.
