Source-grounded Artificial Intelligence–Driven Transfer of Plastic Surgery Textbooks to Podcasts: Creation of Content and Trainee Satisfaction

Abstract

Source-grounded artificial intelligence (AI) models have emerged in the public domain and represent a powerful tool for content creation. These models allow users to "ground" the language model in user-input sources and convert text to audio. Although plastic and reconstructive surgery has traditionally relied on textbook sources for trainee education, we demonstrate the potential to migrate information to audio sources. We created podcasts using source-grounded AI technology based on selected chapters of <i>Grabb and Smith's Plastic Surgery</i>. Generated audio was created using NotebookLM and featured a conversation between 2 AI "broadcasters." Surveys were distributed to assess trainee feedback on AI-generated podcasts compared with the source material. Eight source-grounded AI podcasts were created on topics including wound healing, breast reconstruction, lower extremity reconstruction, and aesthetic principles. Generated audio had a mean length of 14.3 ± 0.2 minutes. Generated podcasts had excellent accuracy with no extraneous insertion of information. Weaknesses of AI-generated podcasts included under- or overemphasis on topics that were emphasized in the source. Survey participants (n = 10) on average spent 14.7 ± 4.7 minutes with the podcast and 34.0 ± 13.9 minutes with the book chapter. Participants preferred a mixed-media approach (90%) over text (0%) or audio alone (10%) for learning the specified material. We demonstrate the potential for source-grounded AI models to aid in the creation of audio-based plastic surgery educational content with high accuracy. This technology can translate plastic surgery literature into audible content to diversify the media available to trainees and practitioners.

Ähnliche Arbeiten