Hello, everybody, and welcome to another edition of C&F Talks, where I get to speak to one of the speakers at one of our forthcoming events. Today, it's my pleasure to have with me David Sully, who's the CEO and Co-Founder of Advai. And Advai is a world leader in testing, evaluation, and assurance of AI systems. David's going to be speaking at our UK National AI Policy, Infrastructure, and Skills Summit, which is being held in London on the 12th of November. David, welcome.

Hi, good morning, really nice to see you.

Very good to have you with us, David.

Could you perhaps share with us first your thoughts on the AI action plan and how it's been implemented so far? What are the actual opportunities and perhaps the hurdles as well that need to be overcome? And how well positioned is the UK in comparison to other countries, I suppose, particularly in terms of getting the necessary investment for AI infrastructure?

Okay, so yeah, multi-layered question there.

Multi-layered.

Multi-layer, absolutely. So, I'll approach this through the prism of Advai, where obviously we're looking at the safety and security of the system. So, the action plan, there's certain parts of it that are very, very specific to us in terms of the insurance industry, for example. Working from a top-down level, though, the plan itself is very welcome. It's been very welcome across the industry.

I think what Matt and Number 10 sort of laid down is really clear. It's exactly what the UK is looking for. The question is always sort of in the action and the implementation phase, which is the really, really hard bit always.

And on that, I think there's some really positive moves. You saw all of the big US tech companies coming over, offering up investment, putting in data centres, so on. That backbone is really, really important. So that is very, very welcome. And I think you won't really see the effects of that really until for another year or so, but hopefully that's going to make a tangible impact.

I think where there's a bit more of a question mark around the action plan is more around sort of the, at a business level, the UK still feels quite uncertain about what it's implementing in terms of AI, its confidence levels around that, and sort of what are the rules of the road for it.

So, I think there's a lot of excitement, but you're still seeing, particularly within listed companies, for example, regulated areas, the really sort of the valuable parts of the UK economy. I'm not sure that you're seeing sort of a go-for-it attitude yet in terms of AI adoption, and I think there's a real confidence and trust sort of area around that that needs addressing.

I mean, in terms of just picking up on that last point, do you think that large companies or all companies are slow off the mark with AI in terms of their investment and use of it?

We're seeing a lot of interest and a lot of sort of effort, but what I term a lot of it as is it's at proof of concept stage. So, there's a lot of playing around, a huge amount of adoption of, say, Microsoft Copilot and, say, chat agents such as ChatGPT, but actually getting it into automated functions, and this goes particularly if you're a large company with a risk function or anything like that. That's where I'm seeing sort of a blocker. So, I'm not really seeing a huge number of deployments into fully automated functions.

So, more toe-in-the-water testing on the fringes, perhaps.

I mean, turning to what Advai itself does, how do you, I had a look at your website, how do you actually determine, from a Layman's point of view, please, how do you determine the vulnerabilities or weaknesses of AI and machine learning systems, and what are the sort of broad lessons that you and your clients have learned from that experience?

So the way to think about what Advai does is within the AI industry traditionally, so when you look at open AI or most of the AI ecosystem in the UK, you have what we term as the builders of the AI systems, and the way to think about it is the systems that they're building are incredibly complex and they're sort of like an infinitely complex building. The thing that Advai looks at is the other side of that, which is how do you explore the building? And that takes a completely different set of technology and algorithms.

So all of our expertise, all of our technology focuses on that building that, say, open AI is built, and you're trying to integrate into your company, which parts of it are safe, which parts of it are secure, and which parts need to be hived off, for example, and ultimately what's useful for you. And that technology set is very, very limited worldwide because it takes an entirely different way of thinking. It's an entirely different set of algorithms that need to be built from base principles, and that's what Advai does.

But ultimately for the end user, what they see is simply, you can try this on a really low-level scale. You can do it yourself. If you ask the same question of ChatGPT, Gemini, and Claude, you will get different responses back.

If you're a company, which of those is best aligned to your company and the way that you operate and the way that you need to adopt those AI systems? And that's what we determine.

Interesting.

Just revisiting slightly corporate strategies with AI, I mean, from your vantage point, do you think companies are missing out? And if so, which are the most important applications of AI that corporates should be considering rather than dipping their toe in the water?

That is a really, really good question. So, the first thing is that there's huge potential when the systems can be made reliable enough for automation. And that's the bit that everybody's tripping up on at the moment is how to make them reliable enough.

So, in answer to the question, it's sort of the systems that we've got at the moment that are widely available, those aren't easily automated in a secure and reliable way. So, you've got two parts to this question or two answers. The first is with the current technology, the advantage comes from the way that a lot of people are using it.

It's an individual using, say, ChatGPT or Claude to start accelerating and improving the content of like their code, for example, or helping make sure that marketing material is like finessed. But where I think the really, really valuable stage is, is when those systems are reliable enough that you can put them into automated workflows. And ultimately, what everybody really wants is agentic systems.

But so that's where the real opportunity, the real value, the ability to earn more money off sort of this product to find efficiencies, all of that side of things becomes possible. But it needs to be reliable enough to do that. And so, I don't think we're at a stage there yet, or we're in the early days of that.

Yeah. So really, the key is at that point where reliability has been established, where everybody has confidence in that, and then perhaps they'll go further down the path. David, very interesting speaking to you.

For our viewers, I very much hope you'll be able to join us in person at the conference, the UK National AI Policy, Infrastructure and Skills Summit, which is being held in London on the 12th of November. Further information available on our website, www.cityandfinancial.com.