Performance of ChatGPT on Specialty Certificate Examination in Dermatology multiple-choice questions
Lauren Passby, Nathan Jenko, Aaron Wernham
Abstract
ChatGPT is a large language model trained on increasingly large datasets by OpenAI to perform language-based tasks. It is capable of answering multiple-choice questions, such as those posed by the Specialty Certificate Examination (SCE) in Dermatology. We asked two iterations of ChatGPT: ChatGPT-3.5 and ChatGPT-4 84 multiple-choice sample questions from the sample SCE in Dermatology question bank. ChatGPT-3.5 achieved an overall score of 63%, and ChatGPT-4 scored 90% (a significant improvement in performance; P < 0.001). The typical pass mark for the SCE in Dermatology is 70-72%. ChatGPT-4 is therefore capable of answering clinical questions and achieving a passing grade in these sample questions. There are many possible educational and clinical implications for increasingly advanced artificial intelligence (AI) and its use in medicine, including in the diagnosis of dermatological conditions. Such advances should be embraced provided that patient safety is a core tenet, and the limitations of AI in the nuances of complex clinical cases are recognized.