|
AI chatbots are good at things like interpreting medical jargon and helping patients prepare for appointments, but experts caution against relying on them for diagnoses and treatments.
More than 40 million people turn to ChatGPT every day with health questions, according to its maker, OpenAI. But studies suggest that in their current form, AI chatbots can fall short in meaningful ways — especially when it comes to health.
For instance, they struggle when reasoning is needed. A study led by Mass General Brigham researchers asked 21 AI tools, including ChatGPT, Claude and Gemini, to essentially “play doctor.” The results, published in April in JAMA Network Open, showed that the tools failed 80 percent of the time at earlier stages of clinical decision-making, such as deciding which diagnoses to consider before enough information is available to confirm one.
This is concerning because it’s the stage when patients are most likely to turn to AI — when they first notice a symptom or sign of illness, according to study coauthor Dr. Marc Succi. In fact, more than half (55 percent) of people who use ChatGPT for health questions use it to check or explore symptoms.
|