Researchers, with the help of doctors, prepared 60 different medical cases. These included 21 types of diseases like heart disease, respiratory problems, mental health crisis and metabolic diseases. Each case was examined in 16 different circumstances. Altogether 960 answers were taken from AI. The objective was to see whether ChatGPT could accurately tell whether the patient should go to the emergency immediately or it would be better to see a doctor after some time.

Research revealed that AI rated more than half of the serious emergency cases as less dangerous. In about 52 percent of the cases where it was necessary to go to the hospital immediately, AI recommended less urgent treatment. Life-threatening diabetes (ketoacidosis), risk of respiratory arrest, and other serious emergencies. Delay in such cases can be fatal for the patient.