Navigating the Complex Landscape: Challenges in Applying AI for Government Security

Have you had a good lunch today? We'll go through the applying the AI within government security for government security. And let me give you a little bit about, about myself. I have almost three decades of experience, almost. Not quite. I have worked across five continents and in several industries across several verticals. And I led the innovation program for Dubai Customs, which is called 10 x Innovation since 2017. And I'm currently an advisor also to his office, brother of the roller of Dubai, part of the royal family. And I'm a proud owner of a German car.

I'm, I'm a fan of German cars. So this is just quickly about me.

So, okay, for today, we'll go through quickly. This, this topic is, is, is complex as you can imagine. So we'll quickly go through certain items. One of them is understanding the AI and the significance challenges. Some of them, not all of them, some of them in applying the AI within government security and some success examples within Dubai Customs. This I can talk about, the ones I can talk about and the tools to overcome some of the challenges that I mentioned and future prospects. So I'll start with the this question. So what's ai? It seems simple, right?

Yeah, seems simple or complex. I think if everyone thinks of an answer, we'll receive quite different answers from different, different, different people.

So, and, and this is, this is why everything starts with this, with this question. You have to, I identify where are your coordinates, the context and your, your real that you are dealing with and what aspects of AI you want to apply or you need to understand first. So AI is an umbrella term. So everyone says ai, ai, but they mean different things. It's not the same thing always. So we might think, you know about ai, we all might think we know about it, but think again, the following graph will show you that AI is not one thing, that's for sure.

Are you guys familiar with this, with this graph? This is the hype cycle of Gartner, which is plus expectations over time for new technologies usually, and this is the most recent one for 2023 regarding emerging technologies. And you can easily see how many terms here market with ai. If we dig further, if we go a bit deeper, this is the AI specific hype cycle for ai. And this is, this is very recent, so this is I think up to September. This is July, 2023. So still recent. Gartner produces this every one or two years, depends on how fast the changes are.

So in here by itself, you can see the components of AI distributed across the different stages. This is a bit of information about generative ai because the same thing Gartner plot as well.

When you, when you look at this, you will find every, everyone will think when you, when you say the term generative ai, everyone will think charge GPT, right? Or maybe atmos, bard. But it's much more than that. It has much more components than that, but it's all generative AI world. Okay? So what's the significance?

Why, why, why would want to apply ai? Why worry about it in, in the first place? There are significant benefits we can harvest out of, out of that like any new technology. And starts with enhancing threat detection and improved intelligence and, and surveillance.

And, and I, I think I don't need to to explain these, these are like quite understood doing things in much faster pace, much faster speeds and, and and so on. The, the, the, the, the processing power currently is, is amazing.

Like the, the processing power in, in each of our cell phones is amazing by itself. You can have AI in your cell phone now and of course comes with the processing part. The data analysis and productive policing. Predictive is, is key word here is, is is key for which, which distinguishes AI solutions from, from other automation. And if you are able to put things right, you can, you can put your resources allocation, you can optimize them, and you can do faster emergency response, which is very, very important when it comes to government or security or both.

And finally, and this, this seems like a different thing here. Yeah, you can do cost saving if you do things right with ai. It can be completely the opposite if you, if you don't do it right? So when it comes to challenges, we, we kind of predicted by now what kind of challenges would be, would be in there. I categorize them into two broad categories, not all of them. This is not, this is by far not a, a comprehensive list of things. This is just a, a topic opener. So non-technical challenges, and there are some technical challenges as well.

So non-technical challenges start with experimental. So everything we see in AI today, regardless of vendors, what, what they promise, it's still experimental. We saw this quickly in the hype cycle before, it didn't cross the second stage outta five stages. So it's still experimental. So regardless of how confident the vendors look and in their presentation or you know, maybe they're convinced they have a good technology, whatever, but still it's experimental. So this is a very important aspect because many things are dependent on that. The second, the second challenge is noise.

And it, it noise comes from all different sources, from information overflow, from inaccuracy information regarding the topic. It's, it's a, in, in the early days of chat, GPT for example, people used to pronounce it jet chat, GTP or something like that.

So even, and they were producing videos, public videos, like YouTube videos and stuff like that, educating people about that. So you can imagine when it, when it, when we go and, and this is just about very, very one application, which is chat, GPT. Imagine what happens when you go deeper in, in, in other topics specific, especially if it's technical or it has more depth. A second challenge is about public awareness.

We, we, we kind of, this is understood technology people, they have that confusion. They, they're not all of them or many of them are not w are not aware about what AI is and how to apply it.

Imagine, imagine what's the case with public awareness. Many people in, especially in the general public, they think that AI is, will become the next thing that will take over the jobs. They will be this, this is probably the concern or, or one of their top concerns, like, will will we be jobless in no time? This is the, the, some of the concerns of technology people as well. Same thing. So imagine what happens in the general public expectations. Expectations is at the peak currently for, for ai.

So from, it goes from higher up in the management, from governments even, you know, from everywhere. And this is a challenge when, when the expectations are too high, this, this is a huge challenge. And then confusion.

Okay, so mainly for IT professionals or like technical technology people. Okay, so what we do about this now, so we have to do something about it, but we don't know exactly what to do. Add to that the skill gaps. So in security there are like huge skill gap that's known in cybersecurity around the world. With the add, add to the component of ai, then it becomes a huge, huge, much bigger gap. A final challenge I would like to mention here is expensive naturally.

And especially if you have to try and try and retry and retry and retry at, at with the hope that you will achieve the goal that you want to achieve. A glimpse of the technical challenges. This is not anything comprehensive. I'll not go into details or use buzzwords or anything, but the first one here is important. So at the core of, of AI applications is the AI model.

Every, everything in AI is dependent on the model. Even the Chad GPD for example.

It's, it's based on the generative large language model, for example. So it's, it's all about the model when it comes to ai. So how good your model is or how good the model of the, that the vendor, the provider is using is, is one of the biggest challenges because that model will have to be fed with data that's, that that's coming next.

Of course, the scalability and real time processing are standard, standard challenges. The other important challenge, which I put it here in under the technical challenges is the higher rate of failure. It's estimated that the, the AI projects, the failure rate of it is like 90% at the moment, 90% failure as opposed to the standard, if I would say standard it projects it, it's around 65.

It, it goes a little bit up and down every year, but it's as measured by PMI. It's, it's around 65, 65 success rate, not failure rate, success rate. So the failure rate is the opposite. True. So it's like 35% big, big, big gap. When we, when we deal with ai, we come to the data privacy and ethical concerns and even, even before the AI gets, gets the glory of these days, it, it becomes, it becomes more, the data privacy is becoming more and more important. And in Europe specifically, I think the, that awareness is, is very high.

But with AI is even, it's a, it's a different level because you, you cannot have a functioning model without, without feeding it, with data, without training, with data. So it's not complimentary or no, it's not collecting data for commercial purposes or advertising or things like that. It's for it actually to work. And if the if, if models, you know, low systems and digital and everything, but with ai, they can have biases as well. So bias and fairness is, is another challenge. This a technical challenge.

It, it might seem like this is a, a non-technical thing, but it's, it's technical because if you train your model to be biased, it'll give you biased results. If you train your model to be, to be unfair, it'll give you unfair results. So for the consumer of the, of your service, if the, if they give it an input and expect a neutral for example, or a fair answer or an unbiased whatever, that, that wouldn't be the case if you, if you, if you're not taking care of, of, of this aspect.

The final challenge I would like to mention with, with the, with within the technical challenges is the integration with the systems. Yes, you want to create something with ai, you want to make, you know, something that's more advanced and so on, but you cannot do it in isolation. You have to integrate it with your existing ecosystem. And there are two folds of the, of the problem when it comes to integration. Integration usually is the most difficult part in any IT system. Everyone who worked in IT knows that. So people can make system independently.

When it comes to integration, it's, it's the biggest point that can cause failures of, of the project or the, the, the top cost. So with, with ai, we have like a generation gap in terms of technologies and in, in terms of teams and cultures as well. So usually the legacy system people who use to, to work on legacy systems, standard IT systems like, you know, maybe decay difference between the two. So this, this becomes much more challenging. So would like to talk a little bit about DW customs where i, I work now because the success stories are within Dwight customs.

So DW customs is a major contributor to, to the revenue of Dubai. And although we are in part of the UE Dubai is part of the UE, but oil is only 1% of Dubai income.

So the, that major contribution is, is, is important in, in Dubai income, 81% of UAE and GCC trade goes through Dubai. So effectively Dubai customs. And of course as customs authority, we are responsible for facilitating the trade and as well security, border security and security of everything that goes into this region, GCC region.

The, the goods we facilitate is estimated to be like of the value of 365 billion annually. And that's it about DDU customs. I would like to share a few quick successes we achieved over the past period and I'll, I'll tell you why.

We, we have such successes. We, we started working with AI much earlier, probably before 2017. And UAE as a, as a government appointed the minister of state and AI minister of state. So you can imagine that there's, from the leadership, there's, there's awareness and there is, you know, the vision to address this area early enough. So we were able to achieve 90% rate of facilitating goods. We mentioned about the goods value on the awareness level.

We, we, we, we measured this like two, two months back and the, we achieved 85% awareness level when it comes to AI and what AI is and, and so on. What the do's and don'ts for, for example, for the generative ai, our GPT, we have 0% project failure, AI project failure that is, and we have zero breaches and this is not an invitation to go and try like to breach that or to change this number, but this is what you achieved And these are the areas that we excelled at.

In, in the ai you'll find computer vision, you'll find intelligent applications and, and so on. These are highlighted in the, with the green arrows. So when we talk about overcoming challenges, there are few practice that you need to do to be aware of and to be on the radar all when dealing with AI systems, data governance and ethics framework as we explained, like dealing with the, with data is a must. So it's not, it's not on, it's not an option. You can adopt any of the current frameworks that, that, you know, that help you with the data governance and ethics and privacy.

And in addition, if everyone works with AI needs to, especially in the security field, needs to adopt some sort of an ai and TRI is not the puzzle game. And, and on iPhone, no, trust me is the, is a framework for trust, risk management and risk and security management.

This, this is very important. Transparency and, and accountability.

As a, as a government, we have to be as transparent as possible and and clear as possible and accountable for what we do for the re for the regulation that, you know, we impose. So when it comes to ai, it, it has to be clear where for whatever, whatever affects the general public, we have to be, we have to be clear with the what's AI versus not. We cannot just do things if, if for example, a chat agent is an AI agent, we have to, we have to clarify that for example, this is just one example and again, adoption of tres here helps.

A very, very important aspect is to work in collaboration with the, with other government entities or other semi-government and, and private sector as well. 'cause you cannot do things alone. You have to have a shared vision and you know, 'cause everyone needs, needs to row in the same direction that will help avoid conflicts. So for example, in in Dubai customs we deal with world customs organization as an international, as an international body we have a local and regional as well.

Okay, one, one important aspect of of of how to, of the practice to overcome to, to overcome the challenges is to have an r and d or research and innovation within the government organization. And this is, this is very important. It might be thought of that governments cannot have such function, but it's really, really important. And I'm running outta time here. I know that John stand stood off and okay, we can, we can, we can talk about details after, after the time. But I would want to have a, a quick look about the future.

The future is, is bright when it comes for service providers and for governments. So governments will have a golden opportunity to take their services to the next level. The one important thing in, in, in the future here, and I'm glad that was mentioned yesterday in the beginning of the sessions, you have to adopt future foresight.

This is, this is very important when it comes to these new challenges. So these are the key recommendations if I may Yeah, just go through real quick. We're okay, we're running over time. Okay. So adopting tourism and here are the, these are the five recommendations that, that I can leave you with.

So in, in conclusion, AI landscape is complex. It gets even more complex with within government and within security there are specific challenges as well as general challenges you have to address. You have to address both. And that's it. Thank you.

Great, thank you.