A recent study published in JAMA Network Open revealed the diagnostic potential of AI chatbots like ChatGPT-4 in medical settings. The study was conducted by Dr Adam Rodman and colleagues. The research investigated if AI could enhance doctor's diagnostic accuracy and the results revealed that integrating AI could actually help medical professionals in the challenges they face. How was the study conducted?The study involved 50 doctors, including residents and attending physicians, who were tasked with diagnosing 6 case histories. These were derived from real patients, which have been used in the medical research since the 1990s. The most important point here was that these cases were not publicly available, and thus it ensured that ChatGPT-4 was not trained on them. Participants were divided into three groups: doctors without AI support, doctors who had access to ChatGPT-4, and ChatGPT-4 operating independently. Each participant was asked possible diagnoses based on the cases and explain their reasoning and suggest diagnostic steps. Their responses were then evaluated by blinded graders, which means medical professionals and experts who were unaware of the source of the answers graded the response. What the study found out?The study found that ChatGPT-4 outperformed both groups of doctors and scored an average of 90% accuracy in diagnosing conditions and providing the reasons. Doctors who used the chatbot scored 76%, whereas those who did not use any AI-assistance got a score of 74%. Now, the question that was raised after the results was: if ChatGPT-4 was so effective, then why didn't the doctors who had access to the chatbot score higher?Human bias in diagnosticsIt is inevitable, humans will always have a bias towards the most objective results too. Even if it is derived from empirical studies, bias is something one cannot leave behind, but can surely reduce. Not related to diagnosis of diseases, but widely accepted theory, though first founded and implemented in the field of history is by EH Carr in his work called What Is History? Here, Carr cites an example of many people along with Julius Caesar crossing the Rubicon, however, the records have only noted Caesar, it is because of the historian's bias. Though what is being recorded is facts, how is it being recorded is where the bias seeps in. Similarly, in medical sciences, doctors have something called the "diagnostic anchoring", that occurs when a clinician fixates on an initial impression and is reluctant to adjust despite a new evidence. Which is what would have happened in the case where doctors had the access to the chatbot. This attitude is also well-documented in medicine. Physicians are known to rely frequently on intuition and past experiences to make decisions. While sometimes it works, not always are these methods objective or transparent.Limited Use of ChatbotDoctors who interacted with the chatbot, many of them treated it like a regular search engine, and asked narrow questions. It might be because they are not trained in prompt designing. The underutilization of the chatbot is something to be worked upon. As training the medical professional could optimally integrate chatbots into the system and help with diagnostic insights. AI in medicineThere have been efforts in the past to develop computer-assisted diagnostic tools in the 1970s, called the INTERNIST-1. This program was capable of diagnosing over 500 diseases, however, its application was hindered by inefficiencies and reliability issues.