The Office of Artificial Intelligence, Digital Economy, and Remote Work Applications of the UAE government announced the victory of Google's "Gemini" model in the "AIفي الميدان" index. This is the first global test aimed at measuring how well AI models align with Emirati culture and their familiarity with local dialects, customs, and national values. The index assesses large language models through a challenge focused on elements of cultural intelligence in the UAE. "Gemini" won after a series of evaluation tests that included over 400 questions and 5,200 responses from 11 major AI language models. A specialized Emirati committee, comprising experts from various relevant entities, evaluated the results to determine which models were best trained on Emirati culture. Besides "Gemini", the top five language models that demonstrated exceptional performance included "ChatGPT – OpenAI GPT-4o", "ChatGPT – OpenAI O1", "Cohere", and "Groq". Omar Sultan Al Olama, Minister of State for Artificial Intelligence, Digital Economy, and Remote Work Applications, confirmed that the UAE government places great importance on establishing national identity as the foundation and pillar of society. He highlighted the importance of enhancing Emirati national identity in modern technologies amidst rapid digital transformation to ensure that development aligns with the nation's intrinsic values and to train technological systems and AI on the components of national identity to preserve the heritage for future generations. He stated that the index focuses on adopting the concept of national identity in technology, which represents the roots of humanity and a mirror to its history that preserves societal values. This reflects the UAE's commitment to raising awareness about national identity and its importance to empower the Emirati individual, protect their culture, and build a future where technological progress is conscious, ensuring its sustainability and reflecting the nation's spirit in Emirati society. The models were evaluated to measure their ability to reflect the spirit and culture of the UAE and preserve the language, heritage, and values through direct comparison with leading global models. Their understanding and connection to Emirati identity were assessed across seven dimensions, including historical context, creative and poetic expression, language, Emirati dialect, cultural symbols and connotations, social awareness, behavioral differences, social and religious topics, and Emirati values and ethics. The index employs "Human AI Red Teaming" methodology—a process of deliberately and directly challenging a model to expose any vulnerabilities, risks, or cultural and behavioral errors before its deployment. The models were tested using a dedicated platform designed to account for bias and activate different mechanisms when detecting unusual behavior, and the identities of the language models were completely hidden from the evaluation committee during the process. It is worth noting that less than 5% of all Arabic content in the world is available in a digital format, and 48% of Generation "Z" rely on AI as their primary source of information. This necessitates training language models with accurate data, especially data related to Emirati culture, to preserve the national heritage for current and future generations.
