Exploring the Role of Generative AI in Dental Education and Clinical Decision-Making: Endo-Periodontal Lesions as a...

This study aims to evaluate the accuracy and consistency of general-purpose language models' responses to endo-periodontal questions in the context of their potential use as an educational tool for dental students and as an accessible information source for the general public.

A total of 40 dichotomous questions were designed on the basis of the article "Acute periodontal lesions (periodontal abscesses and necrotizing periodontal diseases) and endo-periodontal lesions". The questions were posed to ChatGPT-4o, Scholar GPT, and Scholar AI by 3 independent individuals, who created new accounts for these services. They clicked on the "new chat" button each time a new question was posed to avoid influence from previously entered or generated data. The responses of the AI models were evaluated using Pearson's chi-square test.

Out of 7200 responses evaluated, 86.7% were correct, 8.2% were incorrect, and 5% included a recommendation to consult a physician. The accuracy rates of Scholar GPT, Scholar AI, and ChatGPT-4o were 90.8%, 78.1%, and 89.8%, respectively. The differences in accuracy were statistically significant ( P <0.05).

Scholar GPT performed most favorably, passing the accuracy threshold and providing reliable information. Although there is evidence supporting the increased efficiency of AI application in dentistry, the use of such technologies should be approached cautiously considering the possibility of erroneous responses. It may also be necessary to implement a control phase in a given project. Further studies should be conducted to develop GPT services specific to scientific database applications.