Generative AI as a Tool for Environmental Health Research Translation

Abstract

AbstractGenerative artificial intelligence, popularized by services like ChatGPT, has been the source of much recent popular attention for publishing health research. Another valuable application is in translating published research studies to readers in non-academic settings. These might include environmental justice communities, mainstream media outlets, and community science groups. Five recently published (2021-2022) open-access, peer-reviewed papers, authored by University of Louisville environmental health investigators and collaborators, were submitted to ChatGPT. The average rating of all summaries of all types across the five different studies ranged between 3 and 5, indicating good overall content quality. ChatGPT’s general summary request was consistently rated lower than all other summary types. Whereas higher ratings of 4 and 5 were assigned to the more synthetic, insight-oriented activities, such as the production of a plain language summaries suitable for an 8thgrade reading level and identifying the most important finding and real-world research applications. This is a case where artificial intelligence might help level the playing field, for example by creating accessible insights and enabling the large-scale production of high-quality plain language summaries which would truly bring open access to this scientific information. This possibility, combined with the increasing public policy trends encouraging and demanding free access for research supported with public funds, may alter the role journal publications play in communicating science in society. For the field of environmental health science, no-cost AI technology such as ChatGPT holds the promise to improve research translation, but it must continue to be improved (or improve itself) from its current capability.