Like any neural network, ChatGPT has no medical background. It deftly manipulates only textual information to generate relevant responses. Such algorithms are trained on huge arrays of data that are “fed” to them by the developers. Using various methods, this information is compiled into something new.

It is very complicated

A group of enthusiasts from the Ansible Health initiative tested the ChatGPT neural network for its ability to pass one of the toughest exams in the US, the USMLE (Medical Licensing Exam). This exam consists of three parts and the preparation takes 300-400 hours. The list of questions ranges from basic medical knowledge to the paradoxes of bioethics.

The result surprised all participants – text The AI ​​nearly passed this test, with success rates ranging from 52% to 75% with a 60% standard..

But the researchers say that during the exam, the AI ​​demonstrated an excellent ability to create logical chains to justify answers. He managed to find the necessary information in the database by indirect signs to give a detailed answer.

Unfortunately, in some cases ChatGPT’s decisions were not only notoriously wrong, but they could also have tragic consequences. This means that in the near future artificial intelligence will not replace medical professionals, although it may become an assistant for them as an additional source of data.