Associate Professor of Emergency Medicine and Global Health University of Washington Mercer Island, Washington, United States
Objectives: This work provides an overview of AI use in the academic emergency medicine with a focus on content creation as demonstrated by the use of chatbot technology.
Background: Artificial intelligence (AI) refers to the use of computers to preform intellectual tasks such as interpretation of data to solve problems. AI use in healthcare includes interpretation of radiology studies, predictive analytics, personalized healthcare using genetics and medical data analysis. Use of AI in education already exists in the form of adaptive learning platforms, automated assessments, learning analytics and more recently content creation. Chatbots are a form of AI that uses conversational language, search engines and analytics to produce content simulating that produced by humans. Content creation is a major role of academic emergency medicine clinicians.
Methods: Using a publicly available chatbot https://chat.openai.com/ (March 23, 2023 version) the authors built content simulating an academic work surrounding the interplay of AI’s role in emergency medicine and possible use of AI in the future of emergency medicine. Using plain language, the chatbot was tasked with finding supporting literature, writing a conference abstract, a research paper and developing a lesson plans, case studies and a curriculum. The chatbot was then asked to critique its own work, analyzing the weaknesses of the content it produced.
Results: The chatbot created content consistent with the tasks asked of it, written at a level appropriate for the audience (physicians). The created content contained detailed information, but it is unclear the pathway of how it obtained that information, thus not allowing judgment (or bias) regarding the veracity of that information. The content also contained fabrications such as reporting a literature search when none was conducted nor is the chatbot capable of conducting one. The chatbot created content with ingenuity, such as a case studies of fictitious emergency departments, incorporating an understanding of emergency department processes (nursing triage and patient flow), current challenges (delayed care and ED crowding), and other nuanced and detailed narrative prose. The chatbot was ‘aware’ of its own weaknesses stating ‘I don't have the capability to perform an actual search on databases’ and ‘implementation of AI must be done carefully, taking into consideration the benefits and limitations’. The overall results would be indistinguishable from content created by academic emergency clinicians.
Conclusions: AI in emergency medicine is dynamic. Content created by chatbot technology creates products that appear to satisfy the academic needs requested of it. However, not unlike a math problem in which the answer is provided without demonstrating how that answer was obtained, the created content lacks veracity and depth. To quote the chatbot “Overall, AI has the potential to transform emergency medicine education by providing new and innovative ways to teach and learn. However, it's important to ensure that AI is integrated into emergency medicine education in a way that complements traditional teaching methods and does not replace them entirely.”
