Awareness and use of generative AI-powered tools: results of one-year follow-up prospective cross-sectional global survey.

Abstract

Background: Generative AI (GenAI) frameworks, such as generative pre-trained transformer (GPTs) and large language models (LLMs), promise to transform clinical and research practices. Informed human opinion is key to guiding appropriate technological development and task refinement. Detailed data on how GPTs/LLMs powered-Chatbots usage, perceived risks and benefits among physicians has evolved over time and their impact on clinical and academic activities remain unclear. The aim of this study is to assess how the use of GPTs/LLMs chatbots by professionals working in urology has changed over time in the setting of academic and clinical activities.

Methods: A one-year follow-up prospective cross-sectional survey was conducted among urology professionals. Results were reported highlighting the differences in baseline characteristics of participants and their responses between the initial release (April 20^th, 2023 and May 5^th, 2023) and re-deployment of the survey 12 months after chi square and t-test were used to compare categorical and continuous variables.

Results: A total of 129 participants completed the second survey. Eighty-six percent of participants reported having used any GPTs/LLMs chatbot for academic tasks, a significant increase from the previous survey (52.4%; P<0.001). When asked if they were using GPTs/LLMs chatbots more in academic settings compared to one year prior, 70.1% of participants answered affirmatively. Participants, when asked about the use of GPT/LLMs in particular clinical tasks after one year, reported less frequent use for deciding treatment options (18.6% vs. 31.0%; P=0.03) and patient follow-up care (10.1% vs. 21.4%; P=0.02). When participants were asked if they were using LLM chatbots more in clinical settings compared to one year before, 35.6% answered affirmatively.

Conclusions: GPTs/LLMs have a consolidated role in academic tasks, with increasing usage, while some resistance to their use in clinical practice remains. These results are relevant for driving the human-centered development of GenAI technology.