AI at the Edge: 6G, Arabic LLMs & the Middle East’s AI Leap with Mérouane Debbah

Viniit welcomes Mérouane Debbah, professor at Khalifa University and founding director of the 6G Research Center, highlighting his work on LLMs, distributed AI systems, and his role in launching the Arab world's first LLMs, Noor and Falcon.

Mérouane shares his journey to the Middle East as chief researcher at Technology Innovation Institute, the creation of Noor (10 billion parameter Arabic LLM) in 2021, and the development of Falcon, which became the top-ranked open-source model globally in 2023.

Discussion on how AI value lies not in foundational models but in verticals and production use. Mérouane explains the shift toward building infrastructure that connects AI models and enables agent-to-agent communication, leading to his work on 6G.

Comprehensive explanation of telecommunications generations from 2G (voice) through 5G (IoT), with 6G being designed for mobile agents and intelligence. Discussion of how future internet will be assistant-to-assistant communication rather than human-to-human.

Mérouane explains semantic communication using the example of image-to-text-to-image transfer, where only instructions (prompts) are transmitted instead of raw data, requiring new infrastructure for fast, reliable AI-to-AI communication.

Overview of the 10-year development cycle for telecommunications generations, with 6G targeted for 2030. Discussion of standardization requirements to ensure universal compatibility between different AI models and platforms across manufacturers.

Vision of future phones without apps, where large language models act as operating systems. Users will interact through voice commands converted to actions via large action models, fundamentally changing the mobile ecosystem.

Discussion of AI moving to the edge (devices, base stations, routers) requiring a shift from large to small language models. Challenges of embedding intelligence on mobile devices while managing energy consumption and efficiency.

How edge computing creates opportunities for telecom operators to serve AI as a new service category alongside voice, data, and internet. Vision of AI inference subscriptions bundled with traditional telecom packages.

Explanation of why regional LLMs are necessary beyond translation, using the example of how an English-trained model would give culturally inappropriate responses when translated to Arabic, emphasizing the need for culturally-aware AI.

Discussion of data acquisition challenges for regional LLMs, including the English bias in internet content, cost barriers for data upload in developing countries, and the complexity of capturing spoken language and dialects.

Challenges in voice-based conversational AI including dialect diversity, intonation differences, and varying speech speeds across cultures, requiring sophisticated real-time processing capabilities.

Critique of current AI model evaluation methods that focus on academic benchmarks rather than real-world usage. Discussion of the gap between how models are ranked versus how they're actually used in production.

Mérouane's work with GSMA to develop telecom-specific benchmarks and evaluation criteria, announced at Mobile World Congress, highlighting the importance of industry-specific model assessment.

Personal insights on managing the intense pace of AI development, including strategies for following the field through newsletters, arXiv papers, and community connections to separate signal from noise.

Emphasis on collaborating with researchers to understand what's coming 1-2 years ahead rather than reacting to what's already been demonstrated, providing competitive advantage through early insight into emerging technologies.

Reflection on the UAE's visionary approach since appointing the first Minister of AI in 2017, discussing the massive changes in talent attraction, infrastructure development, and the results now becoming visible.

Final thoughts on the exceptional moment for AI development in the Middle East region, invitation to visit Khalifa University in Abu Dhabi, and appreciation for the conversation.