💊 ChatGPT Diagnoses More Accurately Than Humans
Researchers at Beth Israel Deaconess Medical Center in Boston have found that ChatGPT defeats doctors at diagnosing illness.
The experiment involved 50 doctors from several major U.S. hospitals. Each doctor received six case histories and had to suggest three possible diagnoses, providing supporting evidence, and then make a final call. The illnesses were not the most common, but not extremely rare either. These cases were based on real patients and were never published, so ChatGPT could not have been trained on them.
➕ Results
ChatGPT performed best, correctly diagnosing 9 out of 10 cases. Doctors who turned to the chatbot for a second opinion were right 76% of the time, while those who acted by relying solely on their own knowledge and intuition diagnosed with only 74% accuracy.
"I was shocked. AI systems should be ‘doctor extenders,’" insists physician and study co-author Dr. Adam Rodman.
If ChatGPT pointed out an error to doctors, they were not persuaded by the chatbot and unwilling to change their minds. Also, many doctors had no experience with AI: they didn't even know they could copy a medical history into a chatbot and ask it for a comprehensive answer.
In the U.S. alone, more than 400,000 people are misdiagnosed in hospitals every year. Every second medical error costs patients their lives. The economic damage from the human factor exceeds $20 billion.
More on the topic: