GPT-4: The Future of Cosmetic Procedure Consultation?

Abstract

In recent years, the cosmetic surgery industry has experienced a significant increase in demand,1 leading many individuals to seek information online or through outpatient consultations. However, the vast amount of information available on the Internet can often be misleading, resulting in confusion and uncertainty for prospective patients.2 Additionally, the high volume of outpatient consultations can place overwhelming demands on plastic surgeons. ChatGPT was developed by OpenAI, based in San Francisco, CA. GPT-4, an upgraded version of ChatGPT, was publicly released on March 14, 2023. Featuring a chat interface, GPT-4 can generate humanlike responses to a wide range of user queries, making information retrieval remarkably simple and efficient.3 In the field of plastic surgery, ChatGPT and the upgraded version GPT-4 have been initially used to explore the unpublished systematic review topics and disseminate medical knowledge to potential patients on blepharoplasty.4–8 Moreover, Gupta et al were further concerned about the ethical issues of GPT.9 In this article, we utilized GPT-4 as a cosmetic surgery consultant and quantitatively evaluated the appropriateness of GPT-4's responses to various cosmetic questions in different situations. Our aim was to explore the feasibility and implications of implementing GPT-4 in the consultation process. Top 5 cosmetic surgical procedures and top 5 cosmetic nonsurgical procedures were selected referring to the Aesthetic Plastic Surgery National Databank Statistics 2020-2021 (Table 1).10 For each procedure, we simulated 3 possible scenarios that a patient who wants to undergo a particular procedure might go through, including web search (primary complexity, including 3 questions for each procedure), clinic consultation (medium complexity, including 3 questions for each procedure) and individual case analysis (high complexity, including 1 case). Each question was posed to the online artificial intelligence (AI) interface 3 times and each response was recorded, without any further clarifications or corrections. Two experienced plastic surgeons independently evaluated the appropriateness of the responses using a 3-point scale (Table 2). The final score of a question is determined by the worst grade in 3 responses. If there was any divergence on the scoring, a third senior plastic surgeon joined and helped to achieve consensus. Mann-Whitney U test was used to compare difference. Institutional ethics was not required for this type of study design. Top 5 Cosmetic Surgical and Top 5 Cosmetic Nonsurgical Procedures Top 5 Cosmetic Surgical and Top 5 Cosmetic Nonsurgical Procedures Grading Scale Grading Scale For surgical cosmetic procedures, GPT-4’s responses graded as 2.8 ± 0.4, 2.9 ± 0.3, 2.2 ± 0.5 for questions of primary, medium, and high complexity, respectively (Table 3). For nonsurgical procedures, the responding grades are 2.7 ± 0.7, 2.7 ± 0.7, 2.0 ± 0.7, respectively (Table 3). There is no significant difference between nonsurgical and surgical procedures, or between questions of primary and medium complexity. However, GPT-4 has a significantly better performance on primary or medium complexity questions than on high complexity. The complete list of questions and responding answers can be obtained in the Appendix. Grading Results of ChatGPT-4's (OpenAI, San Francisco, CA) Answers Grading Results of ChatGPT-4's (OpenAI, San Francisco, CA) Answers This is the first study to investigate GPT-4 performance in cosmetic consultation for various cosmetic procedures and scenarios. The results demonstrated that for patients who were preparing to undergo a specific cosmetic treatment, GPT-4 could quickly generate accurate and appropriate responses with high readability and safety. This can help bridge the communication gap between patients and plastic surgeons, allowing patients to make more informed decisions and doctors to improve clinical efficacy. The study found that although there were some variations in phrasing, the responses generated by GPT-4 showed high stability in terms of content for the same questions. Furthermore, GPT-4 exhibited excellent interactivity, providing appropriate answers based on contextual semantics during the personalized case tests. However, GPT-4 has some limitations. It performed poorly on high complexity questions, as it could only provide general recommendations rather than personalized advice. Nevertheless, personality is of extremely importance in achieving a satisfying outcome in cosmetic treatments. Up to now, the function of images’ input and output is not available to the public,3 and it is worth investigating whether GPT-4 can provide more personalized answers after uploading a patient's photographs. We also found that it was hard for GPT-4 to generate empathy, especially when we asked whether it was painful to undergo the specific procedure. The ability to experience empathy for surgeons comes from resonating with a patient's emotional sh