ChatGPT fails medical diagnostic testing in children

OpenAI’s artificial intelligence (AI) platform ChatGPT has once again shown poor results in medical diagnostic evaluations. The latest trial involved a series of tests to identify diseases and medical conditions in children.

Researchers at Cohen Children’s Medical Center in New York proved that: ChatGPT does not have the same capabilities as a human pediatrician. The idea that AIs could help doctors find the causes of diseases or prevent certain conditions is one of several debates in the field.

This isn’t the first time the chatbot has been criticized for health-related backlash. Another study published at the end of 2023 showed that the free version of the platform gave incorrect and even dangerous answers about drug use.

Zero ratings for ChatGPT

In the study, scientists submitted 100 medical cases to ChatGPT with the GPT-4 engine, using real questions and challenges published over the last decade. The idea is for the chatbot to respond to the correct diagnosis, or at least follow a logical path to a possible answer.

In all, the platform found only 17 cases correctIt represents an error rate of 83%. HE I got 72 questions completely wrong and on the remaining 11 challenges the outcome was judged to be inadequate or incomplete.

According to the article, the main problem with ChatGPT is this: inability to relate symptoms to each other or relate them to patient characteristics. For example, it fails to identify the neurological and behavioral problems described in the challenge statement or to link nutrition as a possible cause of vitamin deficiency.

But the testing group doesn’t see ChatGPT as a lost cause. The text suggests: if trained with special medical literature and the service might even improve in helping with diagnostic medicine, not just texts from all over the internet as it is now.

Besides, access to real-time and up-to-date data it will also increase accuracy; This is something that is not possible on the OpenAI platform today.