Taibah University in Medina has concluded the 6th International Conference on Artificial Intelligence and its Technologies (MCAIT 2026). The three-day event featured scientific lectures and specialized research sessions, showcasing the latest advancements and applications in the field of artificial intelligence.
The third day of the conference included sessions focused on the development of Arabic language models, as well as AI applications in healthcare, education, cybersecurity, and smart cities. It also featured a specialized workshop and parallel research sessions—held both in-person and remotely—where various research models and findings were discussed.
During the third day, researchers presented studies on the reliability of Arabic language models. Researcher Reem Al-Tamimi analyzed 17 studies evaluating the ability of AI models to process Modern Standard Arabic and its various dialects. The findings revealed significant variations in the models' capacity to understand dialects while preserving meaning and cultural context. The study proposed a research framework for developing more comprehensive criteria to assess the efficiency and reliability of these models.
In the field of silent speech recognition, researchers from the university presented a model capable of identifying four Arabic words by analyzing lip movements. Utilizing deep learning techniques, the model achieved an accuracy rate of 87.02%. When an additional category was included to identify unknown words, the accuracy reached 78.01%. The experiment focused on the words: "yes," "no," "sorry," and "thank you." In the field of medical applications, a study by researchers Inas Al-Ruhaili and Areej Al-Huthali addressed the classification of approximately 170,000 medical questions in Arabic, categorized into three levels of severity. The AraBERT model achieved an F1-score of 98.29%, compared to 95.59% for a CNN-based model—results that support the development of intelligent tools for triaging health-related inquiries and prioritizing their handling.
Conference sessions also featured a study by researcher Bashayer Al-Saidlani, which compared automated text recognition techniques for Arabic by analyzing 500 lines of historical manuscripts. The findings revealed varying error rates among the systems tested; one system recorded a character error rate of 27.92%, while another reached 43.66%. Meanwhile, a model specialized in diacritic restoration (reconstructing Arabic vowel markers) achieved an error rate of 4.06% in an independent experiment—a contribution that helps identify suitable technologies for digitizing and preserving the content of historical manuscripts.
Regarding the reliability of artificial intelligence in legal applications, a study by Noura Al-Rajhi and Mayada Al-Masri examined 90 legal questions—posed in both Arabic and English—concerning the Saudi Personal Data Protection Law and the European General Data Protection Regulation (GDPR). The study aimed to measure the consistency of responses generated by three language models when presented with the same questions in different languages. The results showed an 87% consistency rate across 254 pairs of responses, highlighting the importance of developing standards to assess the stability of legal answers and the reliability of multilingual models. At the conclusion of the conference, the university expressed its gratitude to the National University of Malaysia for its academic partnership in organizing the event, and honored the teams that contributed to its organization and the success of its sessions and activities.

Authored on
28-Jumada Al-Ula-1448-08-November-2026
Target Audience
Employee
Faculty Members
Students
Visitors