ChatGPT won’t replace doctors: the chatbot often misdiagnoses complex diseases more than it successfully does
These conclusions were drawn by doctors testing ChatGPT 3.5. The chatbot was given 150 examples of complex diseases designed to evaluate the diagnostic abilities of medical professionals. The artificial intelligence provided the correct answer only 49% of the time. The results were published in the journal PLOS ONE.
In February 2023, ChatGPT barely passed the exam to obtain a medical license in the USA without additional interventions from human trainers. At that time, the researchers behind the experiment called the result a “notable milestone” for AI.
However, the scientists conducting the new study noted that while ChatGPT previously demonstrated an ability to answer brief medical questions, its capacity to handle complex cases is unclear, “despite the massive amount of information it was trained on.”
Nevertheless, researchers acknowledge that ChatGPT might have significant importance for training novice doctors due to its access to hundreds of terabytes of medical data.
