Generating Fast and Slow for Entree Level Medical Knowledge

Overview

The generative artificial intelligence tool, ChatGPT, has garnered widespread interest since its launch. This innovative multimedia platform has the potential to enhance medical communication and health education, thereby improving medical accessibility and reducing the burden on healthcare professionals. Some studies have indicated that ChatGPT achieves higher levels of satisfaction in counseling compared to human healthcare professionals. Additionally, research has shown that ChatGPT's performance in answering objective structured clinical examination questions is comparable to that of typical medical students. However, in both scenarios, it still requires editing by professionals before being used. Moreover, a recent meta-analysis evaluating ChatGPT's ability in various types of medical exams revealed inconsistent results. Before its application and actual integration into clinical practice, the investigators need to comprehend the advantages, disadvantages, and relevant limitations of ChatGPT in the field of medical communication. This study aims to simulate virtual consultations between ChatGPT, acting as a health professional, and study participants, serving as patients. It will evaluate the participants' satisfaction with the virtual consultation questions, categorized by different levels of cognition through Bloom's Taxonomy. This study plans to recruit medical professionals, healthcare-related professionals, medical students from National Taiwan University Hospital, and the general public. Two researchers will select 20 questions from the USMLE step 3 practice tests and the second stage of the Taiwan Medical Licensing Examination. These questions will be categorized into different levels of cognition based on knowledge, comprehension, application, analysis, synthesis, or evaluation. ChatGPT 3.5 and 4.0 will answer these 20 questions without specifying the question's level of cognition. Each participant will review the answers to these 20 questions and assign a satisfaction score based on the appropriateness of each answer.

Conditions