AI Summit Special: The Return of Matt Armstrong-Barnes, Group AI Practice Director at Servita

26:52They run the most algorithmic techniques that they can associate with car performance. I think we could see probably as we start to go out into the future, we start thinking about simulations. So how can we bring players into a digital simulation world? Oh, wow, okay. So you can test tactics and ways of playing through digital twins, essentially, in a virtual game. Wow, okay. I think if we go back to EA Sports, I think EA Sports, back in 2018, they predicted the winners of the World Cup. In 2018, I'm not a football guy. We've seen this ability to run simulations in virtual worlds that have got pretty big mechanisms of predicting outcomes, especially AI is based on probabilistic mathematics.

27:39So the sporting world is all about probability. Yeah, of course. So being able to model that at a pretty macro level really is going to be game-changing. Yeah. And I think it's going to have a pretty significant impact on all sports at a professional level. It's going to be, I think, a pretty big game changer. Yeah, well, hopefully the Welsh rugby team will get a good AI. Maybe you can go and help them.

28:06My stepdad's Welsh. I think you mentioned that. We do have a good degree of banter about the rugby. So it's good seeing you on the way up. Well, as a Welsh, but I don't want to talk about rugby at the moment. It's been a rough couple of years, but I'm sure we'll be back. But maybe this is the stimulus to get them back. Let's talk about safety, Matt, okay? This is a really important topic right now. I'm certainly not alone in being a little bit nervous about the next couple of years, given the speed and just the sheer pace of innovation. And let me just give you an idea of what I'm most concerned about is capitalism is playing out, okay?

28:42And that's fine. I'm a capitalist. You know, we've got Anthropic, OpenAI, Meta, you know, Google, all the giants are all competing for this sort of AI supremacy and having the best model, having the best agentic AI platform or whatever it is. But capitalism is competitive inherently, and safety may be taking a bit of a backseat, potentially, if we're not careful. And there's a lot of things that can go wrong with this, isn't there? What are your thoughts on that? So I think, firstly, the thing is, AI can make a lot of bad decisions really quickly. That's one of the fundamental challenges we've got.

29:19At scale. Yes. However, I think we are seeing legislation. There's been some proposed legislation here in the UK about the formation of an AI authority, creating some sandbox environments. We've got the EU AI Act, which I think has got some challenges going forward. You know, it's a very complicated piece of legislation to implement. The problem is technology moves quickly, AI especially, and legislation and government move slowly. So it's going to be out of date in a year. We are seeing governments now gearing up towards this. And I think in terms of the pace of innovation, if we look at the TikTok between language models, you know, 165 days or something, and that assumes you're looking at one language model.

30:02If you start looking across all of them, the pace of change is absolutely phenomenal. Ultimately, are we going to be able to change that dynamic? I don't know whether or not we are. The innovator in me says, well surely we should be continuing to drive the bounds of what we can and to make sure that we have the greatest capabilities available. Of course. We then need to make sure that we have the relevant ethical frameworks and this is a bit I think that we're possibly missing in some of the really innovative organizations, the ethical frameworks because the innovator in me says, should we innovate?

30:37Yes. The pragmatist in me says, just because we can, doesn't mean we should. Should we? Yeah. And I don't think we have enough people who have a pragmatic approach to say, just because we can, should we? And I think we should also look about where this type of technology is being implemented in organizations and at what kind of scale. And whether or not they're actually getting the right kind of benefit from it, ultimately out the back, based around what they're implementing. So come back to start with a real business problem and work back from there. Yeah, that's very well said. So we did loosely talk about this earlier on, about this story which emerged a week or two ago of the anthropic engineer getting blackmailed by a large language model who read his emails for the listeners who didn't catch it.

31:31I think the model read said engineer's emails and then blackmailed him to out his extramarital affair if it shut that model off. The story appeared on a number of reputable sites. I know maybe it was probably sensationalized, but it does make you wonder if models are blackmailing humans now with things like that. That's pretty scary, isn't it? I think we've seen a number of sensationalist stories coming out quite recently. We've seen other ones, you know, engineers, absolutely adamant that models have become sentient. So I think one of the things that we're seeing in the press is there's sort of a lot of jumping on sensationalist stories.

32:11Yeah. Because fundamentally, AI is really complicated predictive mathematics. Of course. Language models are very complicated next word predictors. Yeah. So applying that kind of humanistic behavior to them, potentially we need the press and commenters to have some practical reality about what is or isn't happening in the space. and I do work with PhD researchers and data scientists and they're incredibly passionate people and I have seen them get very close to specific implementations and they do start to see some things that aren't necessarily there which is why in a business context you need to have that step back to say actually is this going to achieve the right kind of business benefit or actually is this model trying to blackmail you or did you subconsciously create that by the way that you've asked it questions?

33:06So you've zeroed in to a statistical distribution. Because you're so immersed in this stuff, aren't you? That you are going to now and again have a bit of a, yeah. Yeah, no, like a mirage. Do you know what I mean? You're going to see things that are not there sort of thing. Absolutely. So having that practical background, and this is why I think we need sort of educated business leaders to understand, right, okay, it's great that you've got a model that is going to perform like this. However, when I implement it into my business setting, is it going to drive me the right kind of value? And we see this with an uptick in performance metrics.

33:45If I look in my laboratory, if I've created in a lab, and it's performing with a specific level of performance, it's F1 harmonic, then if I look at that, is that good enough? Am I willing to take the level of risk associated with it not performing at a certain level? Or do I need to take uptick it? Do I need to bring in more data? Can I look at the statistical distribution of that data? Give you a different example. If we see everyone's talking about using AIs to code. Yeah. Most algorithms are looking at existing places on the internet where people ask coding questions. If you look at those places, Stack Overflow, for example, most of the code on Stack Overflow is code that doesn't work.

34:31So that means we're... Because I'm asking a question. I'm putting some code onto Stack Overflow and saying, this doesn't work. Lots of people are giving examples which aren't necessarily right. So if I have 10 examples, I've got one that's right. Statistically, that is at the wrong end of the curve. So if I'm training my algorithm here, I'm training it on poor quality code, not high quality code. Of course. And we're seeing this manifest in other areas as well. if you don't understand the significance of a statistical distribution, what can appear to be high-quality data, because of the curse of dimensionality, can actually end up being a poor distribution, a poor representation of your data.

35:09And the algorithm, because it's based on probabilistic mathematics, will zero in to the peak of the curve, which might not necessarily be where you want it to derive information from. Yeah. Well, that's an interesting take. I heard that Stack Overflow or GitHub, something like 85 % of the code on there is AI generated already. But then again, I suppose it's easy to create AI generated code at scale, isn't it? Yes, very much so. But yeah, so just going back to the safety thing, you mentioned when we last spoke, I think, about trust and transparency. So I'm just wondering, in terms of your new role, what are you saying to your customers in order to instill a culture of transparency and safety first in terms of adopting AI technology?

35:56So I talk to organizations. So I think there are lots and lots of routes into the AI journey. That's probably the first thing. What's your strategy? Have you got a strategy? And your strategy breaks down into a number of major streams that you need to kick off. You need streams that cover trustworthy AI. Sure. What's your operating model? How are you going to organizationally structure yourself? How are you going to understand which AI initiatives that you want to go forward with? Do you want to do AI or Gen AI because they tackle different problem spaces, different use cases, and have different paths to production?

36:30Yeah. How are you tackling your data? What tech stack are you going to put in place? How do you engage with the people in your organization? So you start right from the beginning, basically, then, yeah. Some organizations already have some frameworks in place. And it might well be that a lot of organizations already have governance. Organizations already have views around how they're implementing trustworthiness across other things they're doing. Lots of organizations have AI projects or initiatives already. AI is a very, very broad term. So it might well be that they're using machine learning already.

37:04So they have some of these things in place. so it is about working out benchmarking where you are from a maturity perspective and whether or not these things need a little uplift or a lot of uplift or whether or not they don't exist and need creating and I think that's when you talk about this sort of AI piece it's not about building an algorithm and deploying an algorithm if you're going to do it successfully there's a lot of things that you need to wrap around that to make it accepted in your organisation make sure that you've got the right class balances in your data you're mitigating bias you're leveraging your tech stack in the right way you've got the right partner ecosystem of people who understand how to do this stuff you've got governance in place you've got an operating model you've built business cases all of these things if you start off on the right footing you'll absolutely be successful at creating deployable AI systems yeah I think what I've noticed is is that a lot of organizations may not have a conscious directive or strategy to implement AI capability.

38:08But in terms of their IT estate, obviously AI capability is emerging within that estate. You know what I mean? It's not unbeknown to them, but nonetheless it's just natural progression of these products. I'm just wondering, from an asset management standpoint, how would you suggest CTOs keep track of this stuff to ensure that their data isn't training other companies' models or whatever else you know what i mean so how do you keep hold of your assets and manage them from an ai standpoint absolutely i think having proper data controls in place data loss prevention making sure that you're using it in the right kind of way if you are engaging with organizations to collaborate on the construction of an ai you need teaming agreements you need to understand who owns intellectual property all of these kind of challenges you need to factor in in the commercial engagement models that you're going to run yeah because ai talked about this all the time ai is a It's about having the right collection of people collaborating together to ultimately drive business benefit.

39:08One of the great things about Servitude is we're an absolute catalyst. We've got a great ecosystem. We can bring in the relevant stakeholders to make these initiatives successful. So, yes, having a good solid grip over the assets that you've got deployed, including those data assets, is going to be of paramount importance. Yeah, no, absolutely. Skills. I wanted to talk to you about that. So you mentioned, you brought this up earlier on. There's obviously a skill. This stuff is moving so quickly. There's obviously a dearth of talent around to support this type of, you know, various parts of the AI transition journey.

39:48How do you think organizations are going to be able to keep up with the technology that they're adopting and implementing, especially from a governance and safety standpoint? What do you say into organizations to ensure their workforce are suitably skilled to support the new technology they're implementing? I think that comes across multiple dimensions. So firstly, there's kind of like a base level of understanding. So if you're going to drive a car, you don't necessarily know how the engine works. But you do need either partners that you can work with or skills in your existing organization who do.

40:22You know, you need to maintain it. What happens if it breaks? It needs fixing. All of these kind of things. So most people in an organization are going to be AI users. So they do need some foundational knowledge, especially if they're going to start to use generative AI, because they need to ask the right questions. Of course, asking the right questions gets better answers, but you don't need to know ultimately how retrieval augmented generation works, how language models work, how tokenization works, how all of the different AI algorithms operate. But you do need some capability in that space, because if you've got a problem, you need to choose the right technique to tackle that problem.

41:00Sure. And that's where I definitely see that there is a gap from a skills perspective. And it goes all the way through, right the way from this sort of broad education across the ecosystem. So I think there's a couple of ways of doing that. One is from board level down, get leaders in organizations to understand where AI can play a role, what type of problems that you can address with AI so that when these leaders find these problems, they know the right kind of tools to bring in. Yeah, of course, yeah. I think there's an augmentation play as well, which is you can go on the AI journey all on your own.

41:38It's tough. There's loads of lessons that you'll learn and you will arrive with an AI implementation, possibly a bit battered, falling on some banana skins, swum through some moats, all of that kind of good stuff. Or you engage with organization. by Servitor, where we absolutely know how to avoid all of those pitfalls and get successful deployments in that map against business problems. It is all about driving all of that through the organization. And when you start thinking about AI initiatives, it's a team sport. Make sure that you've got some diversity of thought. Make sure that you've got some people with the right kind of skill and those who understand your business and your business domain data, as well as those people who can bring in the technology-based skills.

42:21Because it's one of the major areas where it's such an overlap between some of the disciplines that we've seen from IT, some of the disciplines we've seen from computer science, because they're slightly different, combined with some of the business knowledge with also some of the data science-style capabilities. Yeah, very well said. I think there's loads of bits to tackle there. But I wanted to come back to the skills part. I'm doing a talk tomorrow at London South Bank University to a load of students. And I know, Matt, you've got children who are potentially entering the jobs market soon. So the question I want to ask you is, if there's any young people listening to this, or actually people who want to change careers of any age, if you are looking to work or maybe capitalize on this explosion of interest and the proliferation of AI capability within business and government and so on and so forth, what sort of emerging careers do you think, emerging vocations, will likely pick up speed over the next couple of years.

43:22We talked about AI ethics, I think is probably one. Yes. Obviously, there's a lot of prompt engineering courses and around things at the moment. What ones really stand out for you as a, say, to a young person? If you get into this line of work, if you're interested in that type of thing, I think there's going to be jobs in the future with that, for those type of skills. I think we're going to see an emerging need for somebody who understands business problems and how to translate them into ways that algorithms can implement them. So this is the formation between... That's like a business analyst type skill set then, is it?

43:54With prompt engineering. Yeah, sure. So we've heard of vibe coding. Yeah, of course. So vibe coding, marketing, vibe is... It's the latest one, yeah. The latest trend. Yeah. However, you still need to ask and formulate your questions in the way that you get the right outcome. I'll give you an example. if you wanted to build a CRM system, you can't just turn around to an AI and say, write me a CRM system. You need to understand at a granular level of detail what you want it to do, what you need it to build, how it's going to build it. Because it may not need to be a CRM. It may need to be something bigger than that, like an ERP plus CRM.

44:32Yes. So some of that beyond, so this is sort of the creative side meets the business analyst, meets the prompt engineer. And I think these are roles that are going to start emerging. where you can have somebody who can speak to a business person or understand business problems, analyze them to a degree that they understand how to turn that business problem into the logic that you need and then turn that into an effective prompt to go through a Vibe coding style process where the algorithm is creating the code that you need. I still think we've got a way to go with Vibe coding, by the way, because most of the stuff that I get out of algorithms, old-school mainframe programmer, it's not quite what I want.

45:17It doesn't quite work. It makes use of some libraries that I hadn't heard of before and whether or not I'd want to bring those into my enterprise organization might not be the way that I wanted to tackle a problem. But absolutely, I can see AI is playing an increased productivity role with developers. And I can see this. If anything, I almost see the role of the developer, which is somebody who picks up lots of bits and stitches them together, probably translating into this business analyst meets prompt engineering. The programmer is the person who's going to be writing the underlying algorithms into code.

45:54So much more technically focused. So I actually see it's the developer role that needs to move up and probably the programmer role that needs to move closer to the technology with AI filling the gap in between. I mean, with most, we talked about this in our last conversation, I'm sure, but with every technology innovation through history, going back to the Luddites or going even back a lot further than that, there's generally been a net gain in jobs, isn't there? Do you think that's the case? It does feel different this time, I'll be honest with you. And funnily enough, it's coming for the white collar, not the blue collar workers, which is different to normal.

46:28And there are lots of differences, obviously, that always are. But what's your general forecast? Are we looking at a net gain in jobs for humans? Or are we all going to be on universal credit in two or three years? I think the analysis shows that it will fundamentally shift the job landscape. Yeah. The analysis says that overall, there will be a net increase in jobs. Okay. Optimistic. So yes, I like to be optimistic about these things. But you're talking, I think it will get rid of a significant volume of jobs and then it will create a small 10 % uptick. What that does mean is that, and I think fundamentally this is right, if you don't have any AI skill today, the chances of you going into this new job paradigm are very low.

47:11So you need to be AI native. You need to have an understanding. You don't need to know how the algorithms work or anything along those lines, but you need to understand, have a broad understanding of what the technology does. Basically, it's like how to write a good prompt. These kind of skills, because it's not just a case of writing wood line. It blows my mind. So many people are still not using AI for more things because they don't really get how to properly. It's just crazy to me. But there's still a lot of people like that. Most people are probably like that, aren't they? I think over the next few years, we'll see people having a greater interaction with AI or generative AI, specifically on a more regular basis.

47:55And I think the people who are in the AI world will still use AI techniques to find hidden patterns in data and then expose that through generative AI techniques which users will consume. And I also think as we start to see agentic workflow, increase in the probabilistic nature of it because at the minute, you're timesing probabilities together. So it's not a deterministic outcome, which is what people are expecting. As we start to see these workflows play out more generally, they will start to be able to achieve some fairly straightforward workflows that historically have been mandrolic, you know, people driving them.

48:33I'm just talking about this event then today. I can't believe we've gone through 45 minutes already. What are you most excited about? The day and a half in front of us, or maybe have you enjoyed anything this morning? What are you looking to get out of this event? Oh, I think it's a fantastic event. I've been coming here since 2018. I think this is the 10th. Every year since then? Yes. Wow. Every year it's run. It's only my second, oh, third time. It's my third time. Every year that it's run. I think they did one. Oh, yeah, they stopped it for one or two years. Only one, I think. I did come. But yes, every year the buzz gets better.

49:02Oh, it's crazy. year isn't it people coming along more people every year i'm part of the ai ambassadors chapter for 2025 yeah and i've really seen the conference grow it's become very how we share how we collaborate how we grow this discipline when i was asked why i wanted to become a member of the chapter six nine months ago i said there's a small quorum of people who've been doing this for like 10 years and the discipline itself has exploded massively. We have a fiduciary responsibility to help those people who are joining the industry more recently to understand how to do it in a sustainable, responsible, maintainable, supportable way that ultimately achieves the benefit of AI globally.

49:51And this conference is an absolutely fantastic way of achieving that. Yeah, absolutely. Well said. But I mean, it's an incredible atmosphere here today. And I'm sure it'll be the same tomorrow. I was a bit disappointed Jensen wasn't here. But I listened to a bit of his speech on the live feed. And it was fantastic. But I have to say, there's some incredible speakers across the two days. And it's a brilliantly run event. And yeah, it's fantastic. Anything in particular you're excited about? You signed up for any talks or anything that you're looking forward to? If anything, I've only been to a couple because I've been so busy sort of talking to people.

50:27Chatting to me and getting tech set up. Literally, I'm walking around upstairs and people are stopping me and saying hello. So it's been fantastic from that perspective, bumping into people. I have a few sessions this afternoon. It's great to see some of the tech giants here. Yeah, all of them are here. I'm there coming to talk about some of the initiatives that they're doing. And also the one that I really like is we've got customers who are on stage saying, this is what we did. This is how we did it. These were the successes. These were the challenge points, the sessions on, this is how we built governance.

50:59So we've moved into people sharing their knowledge about what they're doing to absolutely make the next wave of AI be hugely successful. Yeah, fantastic. So we finished last time. I asked you, what are you most excited about in relation to AI? You obviously answered that question. Has anything since then, anything that's popped up on your radar that you're really excited about? Any startups or tools or anything like that? What is occupying your mind in terms of AI, in terms of new technology and new stuff you've seen in the last year or so? So I think probably last time you asked me this, I would talk about microservices architectures, which I thought were fantastic.

51:38I thought it was the direction of travel. It was great to see Anthropic come out with MCP, the Model Contacts Protocol. Yeah, we were talking about that earlier when we were... Absolutely. ...a game changer, isn't it? Yeah, it went from an idea in November, December last year through to being talked about Google I.O. Yeah. I've seen IBM mention it today in terms of them building it into their tooling. So this is about how you keep context within multiple agents collaborating. There are others, there's agent to agent, there's a number of other things which are all about the communication side. This is busting the field wide open.

52:15It really is going to be the next evolutionary step as to how we get proper agentic AI workflows beyond the point where they are primitive, much more into the sort of next evolution where they can collaborate, they can achieve some absolutely incredible things. And that, I think, is going to be the big game changer. Everyone's talked about it. 2025 is the year of the AI agent. Yeah, it is indeed. 2025 is the year that we'll talk about the AI agent. 26, when we're left to deploy it. Yes, will be when we start to deploy some, in anger, that absolutely achieve the right kind of business benefit. for the average organization that isn't Google DeepMind.

52:57Fantastic. Well, on that bombshell, I think it's been lovely talking to you, Matt. And I'll let you go and enjoy some of the talks and things. And obviously, you've got loads of people to catch up with. But thank you so much for talking to us. It's been great to catch up. Where can people find you? Keep tabs on what you're doing. LinkedIn. LinkedIn. LinkedIn is a great place. Come and follow me on LinkedIn. I have posted a couple of videos. I will definitely post some more. I think there's a need for some primer stuff. So I'm going to go back to first principles and drop out some Primer Explainer videos as well as talking about explaining what the model context protocol is and how it works and why it's right.

53:29That would be fab. Brilliant. Well, thank you so much.

53:38This episode was brought to you by Be Digital. Be Digital support leadership teams to optimize cost and get more out of technology investments. Be Digital and the team have unrivaled expertise with technology license management and data remediation and are therefore perfectly positioned to help prepare organizations for AI technology capability. And on the last point, Be Digital have developed a cutting edge AI readiness assessment, which provides tech leaders with a platform they need to make well-informed decisions about AI adoption strategy in 2024 and beyond. Go to Be Digital UK to find out more and get in touch.

54:22Thank you.