We aimed to test whether and how ChatGPT understood the epidemiological problems related to fluoride intake and whether ChatGPT could produce novel and feasible hypotheses to tackle the challenges in the research for the disorders caused by a deficient or excessive fluoride intake. We designed a set of questions to evaluate the knowledge of ChatGPT version 4o on the epidemiological problems related to fluoride intake. Three evaluators then reviewed these answers. We then requested ChatGPT4o to produce hypotheses for the eight disorders related to insufficient or excessive fluoride intake. These hypotheses were then evaluated independently by three evaluators. Finally, summaries were made through group discussions among all the authors. For the three questions on basic knowledge about the effect of fluoride on public health, the answers from ChatGPT were rated as excellent or good. For the 12 answers from ChatGPT to the epidemiological questions, 8 out of 12 answers were graded A, as excellent. Four answers were rated as B for good. The descriptions provided by ChatGPT on the effects of fluoride intake were comprehensive and well-structured. Six out of 8 answers were graded as excellent and the other 2 as good. ChatGPT proposed a hypothesis for each of the 8 disorders that are caused by either a deficiency or excess level of fluoride. Four hypotheses were rated as novel and feasible. Three hypotheses were considered relatively new and feasible. Only one hypothesis was regarded as an established hypothesis. As AI technology develops, it can assist health professionals in understanding the disorders and researchers in their work on the mechanisms behind the disorders caused by insufficient or excessive fluoride intake.
Read full abstract