Interview: Discover How TII's Falcon LLM And AI Safety Efforts Are Transforming Global Technology Landscape

As part of Abu Dhabi's Advanced Technology Research Council, Technology Innovation Institute (TII) is pushing the boundaries of innovation, with projects like the Falcon large language model (LLM) leading the charge. Hakim Hacid, Executive Director and Chief Researcher of the AI Research Unit at TII, discusses the institute's journey, its key partnerships, the future of AI safety, and how emerging technologies like quantum computing and robotics are set to transform industries worldwide.

On the sidelines of GITEX 2024 in Dubai, in an exclusive interview with OneArabia.me, Dr. Hakim Hacid sheds light on TII's groundbreaking work in the field of artificial intelligence.

Dr. Hacid, can you tell us a little about the Technology Innovation Institute (TII) and its role within the UAE's innovation landscape?

Dr. Hakim Hacid: TII is an applied research institute founded about four to five years ago. The core mission is to establish a world-class research center within the UAE that can compete with leading names in the global tech industry.

We have 10 research centers at TII, each focusing on distinct fields, ranging from robotics, cryptography, security, and biotech to secure systems, propulsion, and of course, artificial intelligence, which I lead. Our work is aligned with Abu Dhabi's priority sectors, including security, defense, transportation, and healthcare. While there's no strict one-to-one mapping between our research centers and these sectors, our approach is holistic, with each center contributing expertise to address different aspects of these industries.

We emphasize applied research, meaning that while we do fundamental research, our primary goal is to drive it towards practical, real-world applications. In this journey, we collaborate closely with our colleagues from Aspire, who manage our research programs, and our venture arm, which helps commercialize research outcomes.

Can you shed light on Falcon LLM, especially since we're a bilingual media house with a keen interest in both Arabic and English language models?

Dr. Hakim Hacid: Falcon is a large language model (LLM) that we started building about two years ago. At that time, LLMs were not widely recognized as they are today. It was a bold move by TII to invest in something so novel, but it's a testament to our commitment to innovation. We were building Falcon around the same time ChatGPT came onto the scene, so it was exciting to see the industry evolve rapidly alongside us.

Initially, our models, like Falcon 40B, were trained on relatively small datasets—1.5 teratokens. As we progressed, the Falcon 180 model was built on 3.5 teratokens, and Falcon 2, which is the most recent, used 6.5 teratokens. Our next generation will leverage an even larger dataset, around 15 teratokens. We're continuously improving, and Falcon models consistently top the leaderboards for performance.

One key aspect of Falcon is its open-source nature. We were among the first to release an open-source LLM, which sparked a broader movement towards transparency. Along with the model weights, we also shared the datasets used to build the model. This transparency, combined with the model's performance, has drawn a lot of attention to Falcon, making it a disruptive player despite coming from a smaller institution like TII.

In terms of capabilities, we're moving toward multimodal models. Falcon 2 can already handle text and images, and the upcoming versions will be able to process audio, video, and even time series data.

How does Falcon handle bilingualism and multilingualism?

Dr. Hakim Hacid: Multilinguality has been a major focus for us. Falcon 2 supports 12 languages, including Arabic, and we're expanding that number to 17 or 18 in the upcoming versions. The performance in each language has significantly improved as we refined the models. We're also pushing the boundaries in terms of efficiency across all these languages, making Falcon a truly global model that can serve diverse linguistic needs, including Arabic.

TII has several partnerships with the private and public sectors. Can you elaborate on some of the sectors and partners you're currently collaborating with?

Dr. Hakim Hacid: Collaboration is at the heart of what we do. We have partnerships with over 100 universities worldwide, focusing on working with teams and individuals who are experts in their respective fields. We've also partnered with major industry players like AWS and Intel, and we recently announced a new collaboration with Intel, to name a few.

At the same time, we work with startups that have innovative ideas, especially in areas where we may not have the time or resources to focus. These collaborations allow us to foster innovation and integrate promising solutions into our projects.

AI safety is a critical issue today. How is TII contributing to the development of AI safety standards in the UAE and beyond?

Dr. Hakim Hacid: AI safety is a global concern. The earlier generations of AI models, including LLMs, were not particularly safe. But as the technology evolved, so did our understanding of how to address these issues. At TII, we have developed the technical ability to control the outputs of our models to ensure safety, but the challenge lies in balancing safety with censorship. There's a fine line between ensuring that models behave ethically and crossing over into restricting free expression.

Moreover, safety is not universal—what is considered safe in one region may not be seen the same way in another. We need to take a localized approach to safety, one that respects the cultural and legal contexts of different countries. While the technical side of safety has progressed considerably, we still need to work on the philosophical aspects, as AI safety isn't just a technological issue—it's a societal one too.

What can we expect next from TII in terms of AI innovation?

Dr. Hakim Hacid: Without giving away too much, we are working on enhancing the performance of our models even further. One of the big upcoming advancements is in multimodality—combining text, images, audio, and more into a single model. We're also diving deep into reasoning. The goal is for AI to not just generate content but to think critically about what it creates. This reasoning capability is essential as we move closer to the idea of general intelligence. In essence, we want AI that can truly understand and engage with the world, rather than just respond to it.

Finally, the question on everyone's mind: Will AI take over human intelligence?

Dr. Hakim Hacid: (Laughs) No, I don't believe AI will surpass human intelligence anytime soon, at least not in the next 10 to 15 years. AI is a tool designed to support human efforts, particularly in handling repetitive tasks and making processes faster and more efficient. The intellectual capacity of humans, especially when it comes to creativity and complex problem-solving, is something that AI can't yet match. In fact, I often compare today's AI to the mind of a four- or five-year-old child—it has potential, but it still has a long way to go.

AI's role will evolve, just as technologies like cars replaced horses in the past. Rather than displacing humans, it will require us to adapt and find new ways to work alongside AI, just as society adapted to the advent of cars. In that sense, AI will complement human intelligence rather than replace it.

As a scientist, what disruptive technologies, beyond AI, do you see shaping the future?

Dr. Hakim Hacid: Robotics and quantum computing are two fields to watch. Robotics will revolutionize industries across the board, and quantum computing will disrupt everything we've built in the past, from cryptography to data processing. Another exciting frontier is the intersection of biotech and artificial intelligence—this convergence will unlock groundbreaking innovations, particularly in healthcare and life sciences.

As TII continues to grow and innovate under Dr. Hacid's leadership, it is clear that the institute is not only shaping the future of AI but also playing a key role in driving technological advancements globally.