Chat Generated Pre-Trained Transformer를 통해 얻은 돌발성 감각신경성 난청 정보의 정확성

Abstract

Background and Objectives Chat generated pre-trained transformer (ChatGPT) is a con-versational artificial intelligence model, which has recently attracted worldwide attention byenabling natural conversations based on huge information from deep learing in various fields. Several studies have reported usefullness and reliability of medical information obtained byChatGPT, but there are no studies explaining reliability and accuracy in the field of otorhino-larynoglogy. On this regard, we investigated the accuracy of information on sudden sensori-neural hearing loss obtained by ChatGPT. Materials and Method Twenty-five questions and answeres related to sudden sensorineuralhearing loss were recorded from ChatGPT based on textbook from Korean Society of Otorhi-nolaryngology-Head and Neck Surgery and Clinical Guidelines of American Academy ofOtolaryngology-Head and Neck Surgery. Answers were shown to one specialist in otorhino-laryngoly in a blind test and asked to assess their accuracy. Each question was rated as ‘accu-rate’ or ‘inaccurate.’ If the contents were not found in the textbook or the guidelines, it wasrated as ‘unreliable.’Results Of the 25 questions, 19 (76%) were identified as ‘accurate,’ 6 (24%) were ‘inaccu-rate,’ and 0 (0%) were ‘unreliable.’ Questions about definition, prevalence, hearing rehabilita-tion, diagnosis and treatment were found to be more accurate than the average, while causesand prognosis were less accurate than the average. Conclusion The information on sudden sensorineurla hearing loss obatined from ChatGPTwas quite accurate. It is expected to provide substantial help to patients and doctors. As medi-cine and medical artificial intelligence develop together, further research is needed on reliabil-ity and accuracy in vaious diseases and fields of otorhinolaryngology-head and neck surgery.