Should we see the glass half full or half empty? A new study by Dr. Adam Rodman, an internal medicine expert at Beth Israel Deaconess Medical Center in Boston, looked at how chatbots like ChatGPT could help doctors diagnose illnesses. -even excellent results, it seems that practitioners are not properly taking advantage of its potential. Let’s take stock.
In detail, the researchers of this study published in the journal JAMA Network Openbrought together 50 doctors operating in American hospitals. They looked at 105 clinical cases based on real patients used by scientists since the 1990s.
Disturbing results
The conclusions of this experiment are surprising to say the least. Indeed, they even shocked Doctor Rodman. Not surprisingly, ChatGPT obtained a satisfactory average score of 90% in establishing the diagnosis. Doctors who did not use this tool obtained an evaluation of 74%, while those who used this AI did only slightly better: 76%.
So the main question was why were these practitioners not taking advantage of ChatGPT’s capabilities? The answer is simple: doctors did not trust the AI when it made a diagnosis different from theirs. They clung firmly to their idea without trying to question it or confront it with the analysis of the AI.
Likewise, doctors used OpenAI’s language model very poorly. Thus, the latter tended to use it as a search engine by asking it certain specific questions about a case. Only a small part of them understood that they could ask him for an overall point of view on a patient file.
Learning therefore seems necessary if we want health professionals to benefit from the capabilities of this technology, even if they must always maintain their critical thinking.
Do Internet users trust ChatGPT (too much?)?
As a reminder, user confidence in ChatGPT’s skills is on the contrary surprisingly high. According to a recent study by the Life Span Institute at the University of Kansas, parents are unable to distinguish responses produced by AIs from those produced by human doctors.
Likewise, the majority of them judged that the texts produced by chatbots were the most reliable within the framework of this experiment. More information in our previous article here.
???? To not miss any news from Presse-citron, follow us on Google News and WhatsApp.
Health
