AWS Summit Dubai 2025 - AI Innovation & Responsibility with Dr. Werner Vogels & Dr. Najwa Aaraj

Yes. Yeah. Naja Arya, the CEO of TII, but maybe you can better explain what TII and your role is there.

Yes. So, first thank you Werner and thank you for AWS Summit for having me here. Thank you for your patience as well. Thank you for Tanuja's patience because I've changed the schedule quite a few times but finally I made it here which is good given that you're on silent disco all of you but I'm not. I think people you can hear me right? Okay, awesome.

Great, great, now you can hear me as well. Fantastic, very good. So, the Technology Innovation Institute is part of the Advanced Technology Research Council, which was formed in May 2020 with the main mandate of driving R&D and establishing a whole R&D ecosystem in Abu Dhabi. TII is the applied R&D arm of ATRC. In addition to TII, we have two more entities. One is Aspire which does the business development and program management and looking into what are the use cases which can be benefiting from the technology that we develop, as well as Venture One which is the commercial arm. In TII today we build the technology, we reach prototypes in the lab. One very important aspect of what we do is we deploy in operational environments and when the technology is ready to be commercialized it's pushed through Venture One either through licensing or through startups. Today through Venture One we have established four startups. One of them is AI71 and I do believe Tanuja earlier announced the collaboration or strategic partnership between AWS, TII, and AI71, and we have Kiara here who's the chief AI and product officer in AI71. In addition to AI71 we have three more startups that were established a few months ago. One of them is called Quantum Gate which we're driving all the post-quantum crypto and identity and access management offering through that. We have another startup called STAI which is a company or a startup focused on autonomy for ground vehicles. And the last one is robotics for climate tech. It's called Nabati which uses swarm robotics in order to do ecosystem restoration and in particular mangrove seeding, mapping, and monitoring in the country and at a global level. So that's in a nutshell who TII and the whole ATRC ecosystem is. And I know not everybody here wants us to talk about AI but I assume you've heard about AI all day.

Yes. So maybe allow me to speak first about the impact of quantum and security. Right? Today there's a race globally to develop the quantum computing hardware and the number of algorithms that are known to be running efficiently on potentially large quantum computers is not that high. But there is a quantum algorithm which is called Shor's which basically threatens the security of all the public key encryption that we have today. So when we go to a website and we browse and there is HTTPS, anything that's using certificates, all our VPN boxes, all our communication applications where you have to do certain key exchange, all of these algorithms will be broken. And these algorithms are the public key algorithms RSA and ECC. Today we rely on the hardness of two fundamental problems which is integer factorization and discrete log. This complexity class will be broken by the quantum computer. So meaning that 70 to 80% of our crypto today will be broken. And this is where the whole concept of post-quantum crypto comes in, which basically is trying to find algorithms for key exchange and digital signatures where the foundational math is not broken by quantum computers. So these are initiatives that have started a long time back with Kyber which is code based, and then you see NIST has started from 2016 with the competitions to find standards for KEM and digital signatures. Last year or the year before they put another call for signatures as well in order to come with the standards that should be adopted for quantum-safe security. Now, the reason why we are looking into them today is we're worried about passive quantum attacks. The quantum computer that would break crypto does not exist today. It has to be around 1 million qubits or 1,000 logical qubits. And this is expected between 8 to 10 years or even more. However, if you think about government entities, critical infrastructure, they all store information. And this information once the quantum computer is ready you can go back and attack it and reveal all state secrets, infrastructure secrets, etc. So this is where the whole store now decrypt later threat comes from. This is why there is a very big advocacy to start into this post-quantum migration. And we are in a country where they are very well aware of this problem. There was a law last year and an executive regulations that came up in order to start with those post-quantum migration, and the Cybersecurity Council today is responsible for that. So this is the threat on encryption. If we talk about...

No, it's actually a great advice. And another thing is the migration will need time, right? And this is where the Mosca's theorem comes into place as well. Now as an impact on the overall cybersecurity theme, we can think about pros and cons, right? In terms of cons, things like malware generation, obfuscation of malware, and all of these will become much more, it will become exponentially sped up with quantum computers, right? At the same time, incident response will be more manageable. Feeds and the analysis of the feeds and then coming up with patches for systems for network etc. will be sped up given the computational acceleration that quantum computers will bring. There is a very big, going back to AI, I don't think we can speak without putting AI in the picture right. Today you see a lot of vulnerability management systems, exploit detection, binary analysis that are relying on LLMs or AI etc. So this also will have a big potential when quantum computers will be into play.

Absolutely.

Exactly. Yeah. They have all the advantages.

Yes.

So before I come, just a parenthesis. Today we have announced and released a new model that was also trained on AWS which is Falcon 34B H1. It's unique in a sense that it is a hybrid model. It is a Mamba plus transformer-based model. So it's a hybrid architecture. It is of course topping the leaderboard as we are used in all of the Falcon series that was released, and with a performance that is even better than models that are twice the size, right? So this was released this morning. His Excellency Faisal Al Bannai, who's the Secretary General of ATRC, announced it in his talk and made it in the Emirates. So this was very special because Falcon is made in the Emirates. And the other announcement was we have enabled API access to Falcon Arabic, right? And Falcon Arabic is a pretty, it's a Falcon 3 that was extended with Arabic capabilities. It is pretty special. It has been trained with native Arabic and it is performing as the best Arabic model that exists, right? So to go back to your question, the decision to deliver Falcon of course has multiple aspects, right? So since April 2023 we have started releasing the Falcon series of large language models, multimodal models, transformer architecture, Mamba or SSM architecture, or a hybrid architecture that we released today. Initially the models were done as a research program and initially also there were not that many models that were open source, right? So it has been released as large language models, not really production ready, but we released them in order to have sovereign models and the second one in order to start bridging the gap between closed models and the majority were closed at that point, in order to enable access for the entire community to start exploring with LLMs, looking into how they function, into architecture, in order to start using them in actual use cases. And as we progressed and we matured, today the Falcon models are production ready models. These are models that can be deployed in solutions to solve specific use cases. The more we are releasing, and if you look into the history, the models now they are more performant even though they have smaller sizes. So we're focusing a lot on data flow quality, on data processing, etc. There is a lot of optimization in the training pipeline that is happening as well. And the more we will be growing, we will also be specializing the models more. And I mean we all are very familiar with this curve of AI, right? Which I really love is moving from these LLMs to more focused and smaller sized large language models and VLMs, moving into the physical AI. And hopefully by the second part of the year you will hear a lot of announcements of those specialized models from TII especially in the robotics field.

Absolutely.