An artificial intelligence-based dental semantic search engine as a reliable tool for dental students and educators

Area of Science:

• Artificial Intelligence in Education
• Natural Language Processing
• Dental Education Technology

Background:

• AI tools in education face accuracy, bias, and hallucination challenges.
• Efficient information retrieval is crucial for dental students accessing study materials.

Purpose of the Study:

• To develop and evaluate DentQA, a specialized question-answering tool for dental education.
• To address concerns regarding accuracy, bias, and hallucination in AI-generated content.
• To facilitate student access to institutional dental study materials.

Main Methods:

• DentQA is built on the GPT3.5 language model, employing prompt engineering.
• Information is extracted from expert-verified external dental documents.
• Evaluation uses both non-human (BLEU scores) and human metrics for performance, relevance, and accuracy.

Main Results:

• DentQA demonstrates strong linguistic proficiency with a Unigram BLEU score of 0.85.
• Human evaluation shows DentQA is more accurate (p=0.00004) and hallucinates less (p=0.026) than standard GPT3.5.
• Consistent performance across question types (p=0.012) and high user satisfaction (>70%) confirm usability.

Conclusions:

• Semantic search engines like DentQA can mitigate misinformation in dental education.
• The study outlines a workflow for developing customized AI tools for dentistry.
• Further optimization and integration of AI advancements can significantly benefit dental education.