Related Experiment Video
Updated: Jul 9, 2026
Digital Hybrid Model Preparation for Virtual Planning of Reconstructive Dentoalveolar Surgical Procedures
Published on: August 5, 2021
Comparative Efficacy of Trained Versus Untrained Generative Artificial Intelligence Platforms in Providing Case-Based
Chaitanya P Puranik1, Jakob Turley2, Kaci Pickett-Nairne3
1Children's Hospital Colorado and School of Dental Medicine, University of Colorado Anschutz Medical Campus, Aurora, Colorado, USA.
Background/Aim:
To determine the comparative efficacy of trained versus untrained generative artificial intelligence platforms in providing multiple-choice questions on traumatic dental injuries in a pediatric dentistry curriculum.
Material And Methods:
In this cross-sectional study, a standardized prompt was used on three generative artificial intelligence platforms, accessed via web interfaces in 2025 (United States), to generate case-based multiple-choice questions on four domains: avulsion, crown-root fractures, primary teeth, and permanent teeth injuries. The three generative artificial intelligence platforms used were: ScholarGPT, untrained, and trained ChatGPT4o (OpenAI). Evidence-based guidelines from the International Association for Dental Traumatology were used to train the platform. The generative artificial intelligence platforms were asked to select one correct answer from four choices and provide a rationale for the selection. Two calibrated, masked, board-certified pediatric dentists scored the case-based multiple-choice questions using a validated Artificial Intelligence Study Material Assessment and Reliability tool and noted subjective responses for the questions. Statistical analyses were performed with an alpha value of 0.05.
Results:
All three generative artificial intelligence platforms demonstrated no statistically significant difference (p > 0.05) in terms of their Artificial Intelligence Study Material Assessment and Reliability scores. Some questions had incomplete clinical information (37%), while the options were simplistic (56%) with incorrect rationale (31%). Newer or trained generative artificial intelligence platforms have scores similar to those of untrained platforms, suggesting that publicly available evidence-based information from the International Association for Dental Traumatology enabled the platforms to access these resources for enhanced accuracy.
Conclusion:
The newer or trained generative artificial intelligence platforms show potential to augment dental trauma education within pediatric dentistry through the development of case-based multiple-choice questions; however, limitations in rationale accuracy and answer quality highlight the need for expert oversight.
