Using generative artificial intelligence in the production and dissemination of innovation in otolaryngology—ethical considerations

Abstract

A team of otolaryngologists is studying the efficacy of early detection strategies in patients affected by idiopathic subglottic stenosis. They use an interactive artificial intelligence (AI) chatbot to prompt patients to report symptoms from home to remotely monitor the progression of stenosis. The chatbot prompts patients to provide feedback on their satisfaction with remote monitoring. Primary endpoints include extent of stenosis at evaluation (noted by flexible laryngoscopy in the office), number of unnecessary visits, and frequency of surgical interventions to relieve stenosis, such as CO2‐laser use, balloon dilation, or laryngotracheal resection. Secondary endpoints include oxygen saturation, spirometry score, pulse, and patient‐reported outcomes. Because idiopathic subglottic stenosis is rare, the study did not have as large of a data set as they had hoped. Therefore, the team augments their existing data set via a generative adversarial network (GAN) trained on their existing data. The team reports significant improvement in the measured endpoints, with marked decrease in unnecessary visits and surgeries. The team has become occupied with clinical duties and does not have time to compose the manuscript. Moreover, they have a history of submitting poorly written manuscripts. Therefore, they decide to use generative AI to aid in manuscript production.