Scientists have found that modern language models of artificial intelligence, such as ChatGPT, are able to effectively diagnose diseases if they analyze descriptions similar to the fact that they are found in medical textbooks. However, when it comes to the analysis of the symptoms described by patients themselves, these models are often mistaken.
It is interesting that despite the lack of initial specialization in medicine, many people are increasingly turning to artificial intelligence for self -diagnosis. according to the survey conducted in 2023 by Virginia University, 78.4% of respondents are ready to use ChatGPT for Assessments of your health.
II technologies have already firmly entered medicine, where various tasks are successfully performed, including analysis of medical images and automation of routine processes. Numerous studies confirm the ChatGPT potential in healthcare. For example, this AI copes well with the provision of consultations on mental health, management of drug treatment and teaching patients. However, as the data is shown, the accuracy of ChatGPT in the diagnosis of diseases in children is only 17%.
Particularly complicated for AI are tasks associated with the diagnosis of genetic diseases. .
The results showed that the accuracy of the diagnosis was significantly varied depending on the model, starting with 21% and reaching 90%. The GPT-4 model turned out to be the most successful. However, when scientists tried to simplify the language of questions, replacing medical terms with simpler, the accuracy of models decreased.