A group of scientists from the USA and Japan assessed the ability of the ChatGPT chatbot to make medical diagnoses. The study was published in Journal of Medical Internet Research.
The specialists were faced with the task of checking whether modern chatbots can be used for medical purposes – in particular, when making a diagnosis. Scientists uploaded symptoms of five popular orthopedic diseases into the system and asked ChatGPT to make a diagnosis and make recommendations based on this data.
Over the course of five days, each of the scientists asked ChatGPT repeated questions. Experts noted that the chatbot was quite accurate in determining the diagnosis, but not for all diseases. Overall, the authors noted that artificial intelligence (AI) has proven to be a poor tool for these purposes.
Thus, ChatGPT diagnosed carpal tunnel syndrome with 100 percent accuracy. However, in the case of cervical myelopathy, the chatbot’s accuracy was only 4 percent. The AI was also inconsistent in its responses and made inaccurate recommendations regarding hospitalization of patients. Moreover, in almost 80 percent of cases, ChatGPT recommended that patients see a doctor.
“Given the risk of error and potential harm from misdiagnosis, it is important that any diagnostic tool clearly warns patients to seek expert medical opinion to confirm the disease,” said senior study author Koji Fujita.
In mid-September, scientists at the University of California stated that ChatGPT affects the increased water consumption intended for cooling data centers. Experts have found that the chatbot indirectly consumes about 500 milliliters of water while processing 20 to 50 requests.