ChatGPT-4 diagnoses more accurately than doctors, according to study

The artificial intelligence It is increasingly capable of solving problems, even in the medical area. At the beginning of the year, for example, a study showed ChatGPT obtained a better score than Peruvian students on the National Medicine Exam. Despite this, health professionals still do not feel confident enough to use this type of tools.

Other evidence of the usefulness of technology in the health field is a recent study carried out in the United States, which has shown that ChatGPT-4, OpenAI’s artificial intelligence, can diagnose diseases more accurately than doctors. The analysis compared the ability of AI to resolve clinical cases against doctors who used traditional tools and those who also had access to the chatbot, evidencing the potential and limitations of the use of AI in medicine.

In the study, which involved 50 doctors from different specialties, ChatGPT-4 achieved 92% accuracy when working autonomously, surpassing the doctors so much who relied on the tool (76%) and those who depended solely on conventional resources (74%).

An ally wasted by human bias?

The researchers highlighted, however, that many doctors did not take full advantage of the chatbot’s capabilities.

“They weren’t listening to the AI ​​when it told them things that contradicted their own ideas,” said study co-author Dr. Adam Rodman.

This resistance, known as confidence biaslimited the usefulness of the chatbot as a complement to clinical decision making.

Additionally, doctors treated ChatGPT-4 as a simple search engine, asking specific questions instead of using it to analyze complete clinical cases. This not only reduced the effectiveness of the tool, but also reflected the doctor’s lack of training in its use, a barrier that the study identifies as key to maximizing its potential.

Large language models have changed the way medical diagnosis is approached. Unlike previous computational systems like INTERNIST-1, which sought to replicate human reasoning, tools like ChatGPT generate diagnoses based on previous data and linguistic patterns, facilitating their integration into modern medical practice.

Despite their limitations, researchers believe that tools like ChatGPT could become valuable assistants for doctors, providing second opinions and improving diagnostic accuracy. However, to achieve this potential, it will be crucial to overcome skepticism and train professionals in its use.