AI Changes Everything

About AI Changes Everything

AI Changes Everything is an Oracle podcast that explores how business leaders, innovators, and practitioners are putting artificial intelligence to work across their organizations. Through conversations with experts shaping the future of business and technology, the series examines real-world AI strategies, emerging trends, and the opportunities and challenges organizations face as AI adoption accelerates. Each episode offers practical insights into how AI is changing the way people work, make decisions, serve customers, and drive innovation. Whether you're a business leader, technology professional, or AI enthusiast, you'll hear perspectives that help connect today's advancements to what's next. Explore the series at http://www.oracle.com/aichangeseverything

Publisher

Oracle Corporation

Category

technology

Language

en

Explicit

No

First episode

24 Jun 2026

Latest episode

13 Aug 2026

Latest episodes

5 episodes in the feed.

1. 

   13 Aug 2026

   AI Changes Everything: Protecting IP with Mel Morris

   Explore how AI can accelerate research, uncover hidden connections across massive data sources, and help organizations protect intellectual property earlier in the innovation cycle. In this episode of AI Changes Everything see how increasing innovation velocity can help leaders move from ideas to commercial value faster at https://social.ora.cl/6008BDMPGC In this episode of AI Changes Everything, Mel Morris CBE, serial entrepreneur, investor, and CEO at Corpora.ai, discusses how AI is changing the way organizations approach research, innovation, and intellectual property protection. Drawing on more than a decade of work building technology designed to analyze vast amounts of unstructured information, Morris explains how researchers can move beyond traditional search methods to discover connections across disciplines that might otherwise remain hidden. The conversation explores how AI can help teams define problems more effectively, identify unexpected sources of insight, and accelerate the path from research to innovation. Morris shares how Corpora.ai enables researchers to search across a broader universe of knowledge, uncovering relevant information from domains they may never think to explore. He explains why innovation velocity is becoming a critical competitive advantage and how earlier problem definition and solution generation can help organizations protect intellectual property before competitors reach the same conclusions. The discussion also examines sovereign AI, the United Kingdom's opportunity to strengthen innovation through AI-enabled research, and the challenges organizations face when turning ideas into commercial outcomes. Morris offers lessons from building and scaling successful businesses, including his experience helping grow King during the Candy Crush era, and shares why leaders often underestimate the importance of scale when pursuing growth. The episode concludes with a practical prompting technique, perspectives on responsible AI adoption, and a thoughtful discussion about areas where AI requires careful oversight, including cybersecurity, healthcare, medical innovation, and warfare. For business leaders, researchers, innovators, and technology decision-makers, this conversation provides valuable insight into how AI can help accelerate discovery, protect innovation, and create new opportunities for growth. ------------------------------------------- Episode Transcript: 00:00:00:00 - 00:00:15:04 Unknown The views expressed in this podcast are those of the individual speakers, and do not necessarily reflect the views or policies of their organizations or or its affiliates. 00:00:15:06 - 00:00:38:23 Unknown What if I could surface critical insights from sources you'd never think to explore? What if you could compress months of research into days? What if it could protect your IP from the outset? And what if it could unlock a diverse universe of knowledge for researchers? Imagine the impact that could have. 00:00:39:01 - 00:01:01:01 Unknown Hello everyone, and welcome to the AI Changes Everything video series. I'm joined today by Mel Morris, serial entrepreneur. Hope you don't mind me calling you back. Mel, investor and now CEO of corporate AI. Okay. Mel, if you don't mind, just to kick off. Just describe the journey that's led you to computer AI and what it is and what it stands for. 00:01:01:01 - 00:01:27:08 Unknown If you could please the journey. Well, it's been a long journey. Go commit to a decade. In September. So we began looking at how you can manage a massive corpus of information. At the same time, people are looking at developing AI, so it's quite an interesting journey. But our view was we wanted to be able to look at really huge sums of unstructured data. 00:01:27:10 - 00:01:50:21 Unknown Of course, the starting point. What is huge? We had designs on hundreds of petabytes and looking around to see what technologies existed, particularly in terms of database technologies, thinking about graphs and vector. No one had really ventured into anything like that scale. So, Neil, I think the key thing with this is that essentially it was all about scale. 00:01:50:23 - 00:02:12:15 Unknown So you spoken about the scale. This is someone out there who doesn't operate in the field of research. What's the problem you're solving and who are you solving it for typically. Research is about finding information, and it's also about connecting the dots. So the number of dots that you can connect is entirely depend on the size of the corpus you're looking at. 00:02:12:17 - 00:02:34:22 Unknown So we allow researchers to connect more dots across for more information than probably any other tool. Right. So today these researchers we have access to limited amounts of information, shall we say, in research. And suddenly you just open that up for them. What are the other sort of benefits that it has for them? If you think about typical research, think about universities higher education. 00:02:35:00 - 00:02:58:00 Unknown As people develop their skill set, they tend to narrow their focus. As you get further and further up the scale, that focus reduces, the depth, increases, but the focus reduces. So university projects typically have, if you want to draw a sort of a picture of a Venn diagram or vice of people there, and you can say, well, the this group of people may have different skill sets. 00:02:58:01 - 00:03:19:02 Unknown There's some overlap. There's lots of diversification. The same is true with data. So if you're looking at a corpus of information that's quite narrow, then you're going to find certain things out. But there might be things in relate to fields of science or even unrelated fields of science that could answer things that you're looking at. You could be looking at electrical process. 00:03:19:04 - 00:03:40:14 Unknown There might be something inside of fish that actually might be relevant to what you're doing, but you might not think to look there. With corpora, we connecting more of the dots across more of the domains easily. And will corpora understand the context of what is being asked and what's being researched. And go and find almost those adjacent pieces of information automatically on behalf of the researcher? 00:03:40:17 - 00:04:06:00 Unknown Yes. And he points to the interesting point, because, all the new tools, particularly if you look at, say, the transformer technologies that became ChatGPT, these sorts of toolsets lend themselves now to the fact we become a bit lazy. We say we're looking for this, and then we go a step further. We talk about where we think they're likely to find information that narrows down the scope. 00:04:06:02 - 00:04:23:18 Unknown So we have a saying that when using corpora. Brevity is king, right? Do not try and tell us where to look. Give us the challenge. Tell us the problem. Let us decide where all the information that might be relevant that exists. And then you can choose to decide which elements of that you want to follow. You want to use. 00:04:23:18 - 00:04:42:22 Unknown You want to ignore. This is a fundamental shift, right. Because I think what you're saying is where I was in the past, I had to understand my problem and know where to look and then go look for it. But now you're just saying, tell me the problem. Copy will figure out where to look. And there may be places where you haven't even considered and bring that information back to you. 00:04:42:22 - 00:05:00:10 Unknown Is that. That's exactly right. And I'll give you a simple example. So let's say I'm looking at a medical problem. So I might want to look at a new method of treating this form of cancer. Now I probably would look in PubMed. I would look at most of the medical journals. I probably wouldn't look inside SEC filings. 00:05:00:13 - 00:05:26:06 Unknown No. But the SEC filings might include really important information on an 8-K or some other document in a filing that relate to exactly the problem you're looking at. But if you didn't think to look there or you didn't know to look at that, you'd never see it. So you might waste time going to file a patent for your new medical invention, and then realize that the 8-K, the SEC filing, became a piece of prior art that would invalidate your patent. 00:05:26:07 - 00:05:54:08 Unknown Interesting. Wow. Okay. And what what is the process look like? You know, there is sort of an understanding the problem, but how does it talk me through? Sort of like the process of how this works. Well, we decided that part of the challenge was that people particular with tools like ChatGPT, the transformer technologies, tend to want to provide this almost so copious definition of what the problem is. 00:05:54:10 - 00:06:20:08 Unknown And in so doing, they're actually proffering part of the solution. At the same time. And that effectively now is going to narrow down again what you're trying to do. So with corpora we would advise starting with a simple statement. All right. The example the same would be something like the challenge the let's call it as well let's say the the harmful biological, physiological and systemic impacts of human dialysis. 00:06:20:10 - 00:06:40:01 Unknown That would be a perfect statement to start with corpora. You would then ask it to produce a problem definition, which would actually probably probably be with 20, 30 pages of all of the different things that might constitute those sorts of challenges. You would then get that problem definition. You can review that, you can add to it, you can change it. 00:06:40:03 - 00:06:57:01 Unknown Then you can take the problem definition and you can then say, okay, so that's that's a problem. But what's the solution? And you can literally say now produce a solution. And it will idea to solution that goes against that problem definition and gives you a chapter and verse about how it can be solved by looking at what the problems are individually. 00:06:57:02 - 00:07:18:14 Unknown Maybe there's 10 or 20 then taking each one. Understanding the state of current convention. What's the current methodology used? And then you can then see the idea, the solution to overcome the challenges with that. That must be invaluable to that. To the researchers. Right. Not only having access to the information, but having an indication as to how valuable this information is to them. 00:07:18:18 - 00:07:51:06 Unknown Yes, it is, and in several ways. So it almost inverts the model. And this is really a key to how the value of corpora really comes to the fore. And what I mean by that is that in most research projects, you start out with an idea or a notion of what you want to do. You invest a lot of time doing research behind that lab, testing whatever it happens to be, mathematic modeling, etc. but it's going to be a fair time before you get to a view of what your solution is going to look like. 00:07:51:12 - 00:08:22:22 Unknown It might be one year, two years, three years, or even longer. And the cost of that increases with time. But but other things also happening at that time. It's no longer just a static scenario. Other people will probably be looking at holes or parts of that particular solution, and therefore you're in a race against time. So if it's now taking you a couple of years to get to a point where you can define your invention before you can think about finding a patent or protecting it, there's a significant exposure with corpora. 00:08:22:23 - 00:08:57:11 Unknown We invert the problem. We start with a simple statement of the challenge. We come up with an idea, the solution, which a scientist can easily look at. Review and endorse and say that really works, or we need to do this. That happens literally within days of starting a project, actually within minutes of running the product. So in this scenario, you can go and you can file patent applications, provisional personal applications right up front, protecting your invention, allowing you then to talk freely about what it is that you're working on to pre-market what you're doing. 00:08:57:13 - 00:09:24:01 Unknown So in that scenario, you can publish on arXiv and other publication tools, and that gives you the chance to promote what you're doing while you've actually now preserved the intellectual property. So it's inverting the model. So the value of this thing is in more clear that problem definitions fully worked out idea to solution, fully validated ideate solution. And then there's a range of other tools in that that you can work on from there. 00:09:24:03 - 00:09:42:19 Unknown So one of the things I feel I have to ask you about is the whole area of sovereign AI. Okay. That's the word that's on everyone's lips at the moment. Do you have a view on that when it comes to corporate AI? I do. Firstly, we need it as a country. We need this. We we've got a strong history of being innovative. 00:09:42:21 - 00:10:07:22 Unknown The challenge is how we monetize that, how we actually generate real value and real growth. On the back of that, the new opportunities that are coming from the government now, the DC sovereign AI program is one of those 500 million going in to help companies utilize AI to further science and innovation. It's a great initiative, and those things will help for projects that are already defined. 00:10:08:00 - 00:10:39:09 Unknown But I think we need to go further. We need to be able to help inventors create ideas faster. I call it sort of innovation velocity. We need to increase that. And in order to do that, we need tools like Cooper and probably others as well that can effectively give us this asymmetric advantage. If we can innovate faster, we can protect it, protect it faster, and we've got a government back into further those programs, then that really does go towards making Britain probably a superpower in terms of this use of AI in science and innovation. 00:10:39:10 - 00:11:05:03 Unknown Absolutely. And Kapoor. So AI is of course a UK company, right. So the IP is pretty much all been generated here. Yes, in the UK. Absolutely. And I think it's also further than that because the way we design the application, let's say universities, commercial organizations of scale, if they want to run the entire system in-house, we can let them do that in that model. 00:11:05:03 - 00:11:26:20 Unknown Then they can use open white models. They can use a copy of corporate internally and their entire ideation, all of it. Then it's within their own boundaries. It's a real definition of sovereign and this is actually quite important because, you know, we all know the statistics of how many AI pilots fail. It's more than 80%. And universities have a strong track record for innovation. 00:11:26:22 - 00:11:50:07 Unknown But with respect, they don't always have a strong track record of commercializing or protecting their IP and innovation. I mean, do you have a view on that or just this two things that are the basically sort of, if you like, are are the impediments to us being more successful? The first one is it takes too long into a cycle before we can actually crystallize the IP, before we can actually get to a point where we have patents that can be filed. 00:11:50:09 - 00:12:09:18 Unknown That's a big cost. Also, by the time we get to that point, other people may have beaten us to it, in which case most of the R&D work we've done is sunk cost. So the failure rate is what kills us. We have to find a way to be able to innovate faster, protect our innovations much faster, and then actually bring that to market much more quickly. 00:12:09:20 - 00:12:27:10 Unknown We're talking about an exponential change in the cycle of this thing here. And I think that's something that I believe we are a part of that. I think that these these funding probes, that the government are a part of that will be all hard to make these things work together. Now, I mean, we've managed to get through this interview so far without mentioning AI once. 00:12:27:12 - 00:12:50:01 Unknown Okay, I think I said it at the very, very start. What has AI changed? Because, you know, you've been Cooper as an idea, as a company has been sort of in your mind for ten years now, right? If you were doing this five years ago, that that would be an incredibly complex thing to do. So what has AI changed in terms of how you are able to deliver this service to your end clients? 00:12:50:03 - 00:13:14:06 Unknown Well, I recall a lot of the work going in Google, the time when Transformers were being brought into into play. And, I think the paper was entitled Attention is Everything. That's a paper. It's a seminal paper. Yeah, we saw that coming out. And Joe, my co-founder myself, looked at each other and said, this is significant to change the game. 00:13:14:08 - 00:13:37:15 Unknown And I said, you know, we have to think very carefully about how do we play in this space, but more importantly, how do we leverage the wonderful things that this transformer technology is going to do? And so now Cooper is the fusion of, wonderful database technology for unstructured data, together with be able to harness literally any piece of AI, literally any model you want to use it. 00:13:37:15 - 00:13:58:03 Unknown I say, okay, and what about acceleration? You know, a couple of years ago, for the data sets you're working with and what you're generating in real time? I mean, you know, today we work on projects where we run simulations and we go, come back, go, go home, come back the next day. And if we're lucky, the simulations finished, right? 00:13:58:05 - 00:14:15:16 Unknown And that's using GPUs. I mean, what sort of speeds are we talking about for the return of information? It's orders of magnitude, because a large part of and separate out two pieces of AI are here, what I call computational. I was a very different feel to what we operate in. We're operating in the field generative AI. 00:14:15:16 - 00:14:39:00 Unknown Yeah. Okay. Now, in generative AI, things happen quite slowly. People think it's fast. That's the Gemini or fast open AI, GPT or I've gone too close to anthropic to say or matter to say. Produce me report. Doing this, they might take 20, 30, 40, 50 minutes. And they look at it go, wow, that's fantastic. Yeah. But we often find those reports are quite surface level. 00:14:39:06 - 00:15:00:01 Unknown They're not deep in the way that we would CD because they haven't had time to access. Going back the same challenge, how do I find all the information? How do I then condense that into a form that I can process efficiently? And that's still hurting the whole deep research products that are out there today? Because they have to be commercial. 00:15:00:03 - 00:15:22:00 Unknown People won't wait five days for the report, so they have to do a certain amount. So they compromise on the amount of searching they can do within the time available. We've inverted that problem so we can search literally 100 times much information in a 20th of the time it would take for them to do their job. So the answer is with corpora we're talking about answers that take seconds to minutes. 00:15:22:02 - 00:15:52:10 Unknown Wow. Okay. So would it be fair to say that the the frontier models are almost horizontal? General purpose models can be applied to almost any situation. Cooper it's almost more of like a vertically specific, very, very deep, highly optimized, model. I think that that probably is underselling Cooper because I think so broad and deep. Okay, okay. So we have far more data than any model could say. 00:15:52:13 - 00:16:11:09 Unknown Right. Yeah. Right. So the breadth is there. We're also able to find more information and get it down so we can go deeper as well. So we go both broad and deep compare with the life sciences models. And we have full attribution to where the information came from. Can I just switch the conversation more to talk about the human element? 00:16:11:09 - 00:16:37:04 Unknown Okay. I mean, where's the expertise coming from? I mean, who are you typically hiring? What does the computer team look like? Is it full of data scientists? MLOps engineers? Yeah, we call it our ambassador program. So we have some really world class experts on board. People like Doctor Alexander Rape, who's a material scientist, Professor Paul Stuart, who's very much into the aerospace and engineering control side of things. 00:16:37:06 - 00:16:59:06 Unknown Doctor Andy Pardo, who is a, a professor in, I at, Port University. Aki Batty, who's, the medical director for a big medical operation in the UK. So we have a lot of domain expertise. This is the bit that I find quite fascinating, this combination of, you know, computational science with, research or domain experts. 00:16:59:06 - 00:17:21:02 Unknown Do you mind if I ask you about the business model? What is the business model for corporate AI? Is it subscription based? How does that work? So we wanted to go with a model that was more exclusive. And we started working with start up companies, working on really extreme frontier science projects. And the results are outstanding. So our model is not based on revenue. 00:17:21:04 - 00:17:49:03 Unknown It's based on building up, stakes in companies leveraging our technology to help them build IP and then leveraging the value of that. So if someone those our revenues that that that almost zero. But the value we're building is in building IP that we're sharing the value of with the companies we're working with. So you're building this this network of IP and actually some stakes in these startups, as they grow, you grow as you grow they grow. 00:17:49:04 - 00:18:10:04 Unknown Yep. So you've almost got this symbiotic effect, right. It's a virtuous circle. Yeah. Building is key. And we've got plans now to to tie that now up to the universities to be able to show how we can actually let their office share in this same type of capability, which would transform the capital model for university operations. Right. Okay. 00:18:10:05 - 00:18:26:03 Unknown So if it's okay just switching over, I'm going to do a quick fire round. Okay. And then after we've done that, maybe I'm going to come back and ask you for a little take off, a little bit of advice, to anyone starting a business. I regret this then, but go for it. I think you my definition, dash. 00:18:26:08 - 00:18:55:17 Unknown Okay, so I think the plan is there's, 5 or 6 questions. You got to answer as many as you can in 30s get. Okay. Now thinking time. So here we go. One word for AI today transformational. Brilliant. One word for AI in five years. Metaphoric beautiful AI agents hype or real shift. In between the biggest risk in AI right now, people fail to understand how good it is. 00:18:55:20 - 00:19:22:17 Unknown Very good. Biggest opportunity is people actually leveraging it. Excellent. And one thing leaders consistently get wrong. Scale, scale. Okay. You did very well. There was one last question. I'm still going to ask it a bit like, you know, a mastermind I've started. So I'll finish. One AI term you wish people would stop using. 00:19:22:19 - 00:19:47:13 Unknown Panacea. Panacea. Brilliant. Okay. Thank you very, very much. Okay. I prompt of the week. Mel, do you have a favorite prompt? You like a prompt for the week that you would encourage, viewers to use? I do, it's a bit of a double edged sword, this one. So instead of putting your prompt directly into the model, ChatGPT wherever, try the following. 00:19:47:14 - 00:20:17:03 Unknown Change this into a concise prompt for feeding into ChatGPT, and then put your prompt after it. It's masterful, very, very cute. Right. So I think we've covered more or less everything. Well, that's not true. We could we could go on talking for hours, I'm sure. Now, but one of the things I didn't reveal about to you, I said, you're a serial entrepreneur, is that you are actually the chairman of King Games, which, of course, is responsible for Candy crush. 00:20:17:05 - 00:20:46:15 Unknown Probably the most successful mobile application ever to have existed. If you were speaking to an anyone that was starting a business or trying to scale an idea or business in a corporation, is there any advice you would give them, having learned the hard way? I think King's a great example. And, I would say that the perseverance, the tenacity, the stamina, because King was an overnight success. 00:20:46:17 - 00:21:06:04 Unknown The early years were brutal and, no one wanted to back the company. None of the VCs at all. We we scoured Europe talking to no end of X. No one wanted to know. All right? We'd probably got to $55 million a year in revenue. I think it was still with any external. It's it came in all right. 00:21:06:04 - 00:21:24:14 Unknown And of course, can of course put it on the, on the map. But don't forget, Candy crush was probably ten years into the company's, existence. So, so there were the key things, but and then the other one is people. People who in a startup environment can get on with one another but have the complementary skills to make it work. 00:21:24:15 - 00:21:55:00 Unknown And the team at King was exceptional, right? Excellent. And the one thing people always ask about is scale. Okay. Looking back now, across many of the businesses, was there anything that you've learned now that you realized was an accelerator to helping scale? Yeah, I think first, and Cooper is a great example here. I mean, if you look to some of the things we thought was, I, I go to market plan three years ago, it was rubbish. 00:21:55:02 - 00:22:12:00 Unknown It was rubbish. It simply was never going to scale, you know, not without putting in millions into marketing or whatever. It was never going to work. And we had several attempts at that. And I think, you know, quite rightly, the people we looked at that we thought we might as it's still backing who passed on, they were right to pass on it. 00:22:12:02 - 00:22:33:02 Unknown Okay. But that is a favor, because how we have gone down those paths with the backing we do for become a mediocre business in each of those different markets. And so I think sometimes to spend the real time to figure out not how you go to market, plan, but one that can actually work with the so capital that you have available to support it is critical. 00:22:33:05 - 00:22:57:11 Unknown Yeah. Very good. Okay. So we've come to the end of our session. You know this series is entitled I Changes Everything. And my last question to you is is there anything that I shouldn't be being useful or it's not fit for purpose in your in your view? I have concerns along with the rest of the AI brigade in terms of its malign use and areas like cyber. 00:22:57:13 - 00:23:23:23 Unknown I think there's concerns in terms of the mental health implications, particularly of sort of, you know, the so GPT style sort of offerings. So I think we have to be very careful with that. I think there's going to be an acceleration of innovation, and that's going to bring along a whole raft of challenges. We'll have medical solutions that probably are going to be deemed to not necessarily have to go through the same hurdles to get accreditation. 00:23:24:01 - 00:23:46:03 Unknown We have to be careful with those. We shouldn't stop doing those things. We just have to be mindful that it will make mistakes from time to time. And yeah, we have to be careful. Those mistakes aren't to too huge in terms of the cost of lives or whatever. But I think also on the other side, there's people with diseases today that have no options, and they are potentially offers great solutions to those people to find cures. 00:23:46:05 - 00:24:03:10 Unknown So we've got to think about those sides to it. But I think that uses of the technology in a malign form is very dangerous. You can think of warfare. Every nation is going to have to use it in warfare. All right. But we also have to think about how can we be innovative about the defensive side of that. 00:24:03:12 - 00:24:21:23 Unknown I have lots of concerns. Yes, sir. Well, we've come to the end of our discussion. Thanks very much for the session today. I've really enjoyed it. I want to wish you and the AI team the very, very best of luck. Thank you. Really enjoyed it. Thank you. So if you're interested in watching more go to our Qualcomm forward slash I changes everything to see more episodes. 00:24:22:05 - 00:24:46:12 Unknown Thank you for watching.

2. 

   30 Jul 2026

   AI Changes Everything: Inside Oracle Red Bull Racing's AI Edge

   See how Oracle Red Bull Racing uses AI, simulation, software engineering, and data science to improve performance in one of the world's most competitive environments. In this episode of AI Changes Everything, see how technology, race strategy, and engineering innovation help teams make faster decisions and build competitive advantage at http://www.oracle.com/aichangeseverything In this episode of AI Changes Everything, Martin Galpin, head of technology at Oracle Red Bull Racing, explores how AI, simulation, software engineering, and data science contribute to performance in Formula One. Galpin explains how Oracle Red Bull Racing develops the specialized software it cannot buy elsewhere, creating tools that help engineers, strategists, and performance teams gain an advantage in a sport where every fraction of a second matters. The discussion examines how simulation and modeling have become essential to modern car development, especially as Formula One teams operate within strict constraints on testing, wind tunnel usage, and computational resources. The conversation explores how AI and machine learning support a wide range of activities across the organization, from information retrieval and engineering productivity to sensor modeling and vehicle performance analysis. Galpin shares how generative AI is changing software engineering, why the cost of creating code is rapidly declining, and why understanding, maintaining, and governing software remains a critical responsibility for engineering teams. The episode also looks at the role of AI in race strategy, where explainability, trust, and human judgment remain essential. Galpin discusses why high-pressure decision environments require engineers and strategists to understand the recommendations AI systems provide rather than simply accepting them at face value. He explains how Oracle Red Bull Racing evaluates new technologies, balances physical and data-driven modeling approaches, and focuses on measurable outcomes when adopting AI. Beyond technology, Galpin shares his perspective on the future of software engineering, the importance of domain expertise, and how AI is reshaping the skills organizations need to succeed. The discussion concludes with practical advice on working with AI, thoughts on the future of Formula One technology, and a reminder that AI is most effective when it acts as a force multiplier for talented people rather than a replacement for them. ----------------------------------------- Episode Transcript: 00:00:00:00 - 00:00:15:06 Unknown The views expressed in this podcast are those of the individual speakers, and do not necessarily reflect the views or policies of their organizations or iCal or its affiliates. 00:00:15:08 - 00:00:33:15 Unknown In Formula One, every decision and every second can define a championship, but when the stakes are at their highest and there's no margin for error, is technology alone enough? Or does human judgment remain the ultimate competitive advantage? 00:00:33:17 - 00:00:53:17 Unknown Hello, everyone. I'm here today with Martin Gallatin, head of technology at, Oracle Red Bull Racing. And we're gonna have a fantastic session, I'm sure. Probably one of the most exciting video series we've done. Let's get right into it. So, Martin, can you just talk about your role at Red bull? What is it that you and your team do? 00:00:53:19 - 00:01:27:11 Unknown Thanks for having me. So, yeah, I'm Martin, head of technology. Oracle Red Bull racing. We, a group of software engineers and data science, that work on, projects for, the entire campus. We work on aerodynamics, vehicle performance, race strategy, the, the simulators, it in the mix. And then we have, some machine learning data science projects that that look across all of those different functions. 00:01:27:13 - 00:01:50:19 Unknown The, the overall goal of what we do in the department is to build the software we can't buy from, from others. So software the differentiates ourselves from our competitors. So I mentioned and I realize there's a lot of things you can't go into because of, you know, company confidentiality. But I mentioned earlier what your team does is that sort of that secret sauce, that extra thing that allows you to go a little bit further. 00:01:51:10 - 00:02:15:19 Unknown Can you talk a little bit about what the role of technology is in things like, developing brake strategy or modeling the car or just getting that little extra oomph? I think formula one is a a unique environment compared to most in as far as I think, simulation modeling data has been part of how we operate for decades. 00:02:16:13 - 00:02:49:10 Unknown I think Oracle Red Bull Racing were a lot quicker to recognize the value of software and modeling on driving confidence and I think the rest of the grid is now catching up to that. I think I see driving a lot of that, but I think fundamentally in Formula One with the restriction that we have now for testing at the track for the wind tunnel, and the CFD, fundamentally, most of your car development relies on the tools that you have available to you. 00:02:50:00 - 00:03:11:02 Unknown So it is pretty fundamental in designing, a winning racecar these days. And what was it about the Oracle Red Bull Racing that kind of made you have that focus on software engineering, where perhaps some of your competition? Well, was that a design ethos from the start when the team was set up or was that all down to you and your team? 00:03:11:04 - 00:03:42:16 Unknown No, I mean, I've been, Oracle Red Bull Racing for over ten years now. And so, I have been a long time, but largely I think the, the cultural, the philosophy and the investment in and simulation modeling, was a lot earlier than that. I think the, the, I guess the easiest way to it to think about it is the, the leadership at Red bull from a, from a, from the technical side comes from a very, modeling, heavy background. 00:03:42:19 - 00:04:16:15 Unknown And I think that as a culture as approaching problems in a, in a computational way, the when you go back 15, 20 years was actually, was actually quite novel. Whereas now we've accumulated 20 years of experience in the problems where maybe others have five, ten, and the tools that we have in the foundations that we, we've laid over the years, I think, a lot of the continuity of success we've had, over the last decade and it's clearly showing in the results. 00:04:16:17 - 00:04:34:09 Unknown Now, Martin, you spoke just a few moments ago about some of the limitations when it comes to things like testing and simulation, which was a surprise to me because, you know, the impression I have is that you're pretty much unconstrained. But that's not the case, is it? Can you just give some color around what these limitations are? Yeah. 00:04:34:09 - 00:05:03:11 Unknown I mean, there was a time when we were unconstrained. The that was 20 years ago now we have a project to, to start with. So we have an overall capital, minimal, but we can spend in a given year. We have restrictions on track testing, so we are only allowed to run this year's car. At a certain number of events a year, most of that pre-season. 00:05:03:23 - 00:05:29:09 Unknown So there's no opportunity to run the race car outside of race events during the season other than some marketing events. And then we have restrictions on the wind tunnel. So how much time? We have to test the model of the car. Before we build components and, to develop their dynamics, as well as some regulation around the computational time to do safety. 00:05:29:09 - 00:06:00:01 Unknown So fluid dynamics, so that the only real areas at the moment that are not constrained is, is the simulation tools, with the exception of the wind tunnel and CFD, obviously. So it was a higher importance on those those tools being being good. Yeah, absolutely. So this is where I sort of tell it really comes in having people in the team that can just sort of be highly creative or squeeze the absolute and degree out of a piece of technology, either of any limitations on the technologies that you can use. 00:06:00:01 - 00:06:24:16 Unknown It's all within, formula one. There's not really limitations on the technology itself. There's just limitations on the constraints of our users. So as I say that the cost is is one aspect, but if we take, say, the CFD regulations, where obviously the, the fluid dynamics is more cost effective than building a model and running in a wooden tunnel. 00:06:24:18 - 00:06:50:18 Unknown But the the constraints of such the, the there's not an unlimited amount of compute capacity that we can use to do CFD in an attempt to bias the advantage away from teams. Who can you can find more, more compute capacity than others. So yeah, that's not really a constraint on the technology as such. But there is constraints that change which technologies you pick and how you approach problems. 00:06:50:18 - 00:07:13:23 Unknown And just talk a little bit more specifically about AI. Where are you using AI today? You know, can you give an insight into what sort of areas of the business where AI is playing a pivotal role? So I think when we talk about AI, obviously AI is lots of different modeling techniques. Lots of different applications of technology again, available. 00:07:13:23 - 00:07:49:11 Unknown I think we were quite early in that journey in this as far as we've had a machine learning group, in the department since, I think 2016. So almost ten years now. And obviously the landscape for AI is changed a lot in that time. But it really depends where, where in the spectrum of car development you look, there's aspects where like what organizations we can use, recent AI to drive efficiency and operational, efficiency. 00:07:50:03 - 00:08:20:16 Unknown In terms of information retrieval and building software systems that are, you know, better, integrated and smarter. But then there's also aspects closest to the car where perhaps we can use, machine learning techniques to, either that, say, model sensors on the car to reduce mass or, or improve the physical models that we, we would normally have, with, you know, surrogate model and a lot of the techniques. 00:08:20:16 - 00:08:42:10 Unknown So, yeah, it depends where in the organization you look. But I think the overall impact, I think is everywhere now. Yeah, it's become infused into what you do. And, you know, thinking about generative AI, is that something that you and your team are starting to use and might even be in back office capability as opposed to on the car itself, which is gentrified? 00:08:42:15 - 00:09:12:04 Unknown You're making use of the large language models at all? Yeah, absolutely. I think in the last few years, obviously the viability of a lens for, so for making tools better has become, genuine. And I think, again, there's two aspects is using generally as software engineers to build better tools, faster. And then there's also using a lens to help us build better tools that engineers use. 00:09:12:04 - 00:09:34:11 Unknown And I think you're starting now in the last couple of years to see that becoming a reality. In addition to the the gains, I think that everyone is experiencing with software engineering, at the moment. Yeah. And have you seen that impact? You know, because your background is software engineering, right? Where's the most profound impact been in that software engineering sort of continuum life cycle? 00:09:34:14 - 00:10:06:20 Unknown Where have you seen it? I think the the biggest thing is the pace at which has changed in the last 18 months. And I think we're in a position now in 2026 where the cost of creating code or the cost of creating software is almost fallen to zero, right? So the the the problem is not how you create software, but the cost of owning that software or the cost of building the right software and maintaining it over a long period of time is not zero. 00:10:06:22 - 00:10:30:04 Unknown So it's kind of just changed the the way you approach problems. And I think we are on that journey like I think everyone in the world is at the moment, but I don't think anyone yet has either realize the full potential of what's happening or have we reached the end game for what I think will happen in the next couple of years? 00:10:31:01 - 00:10:53:23 Unknown The the way the AI is built software, and most people have written in software over the last decade is not the way you would write software today. So it's already changed in a fairly fundamental way. Yeah, absolutely. It's been quite stark to actually watch that change over the loss. And it's been fairly recently, last 1 to 2 years. 00:10:54:01 - 00:11:19:01 Unknown Okay. So look, we've been talking about the use of AI. Are there any areas where you would prefer not to use AI or our code Red Bull Racing of taking a a decision? Not actually, no. We're not going to use AI. Perhaps that's where we need human because at the end of the day, the times when you want humans to exercise judgment control, is there any areas that you perhaps would prefer not to use AI? 00:11:19:03 - 00:11:55:19 Unknown Many, I guess to give two examples. If we take how we approach race strategy, which I know we've, we've talked about in the past and the projects we have with Oracle, over the years are pretty well documented. You would traditionally approach, race strategy projects using numerical methods, Monte Carlo simulation. And over the years for the project we've we've had with Oracle has been to to bring the, the, the barrier of compute capacity down so that we can run more simulation during the race. 00:11:55:21 - 00:12:23:03 Unknown And I think the natural evolution of that tooling is towards AI and towards rather than taking advantage of Monte Carlo methods and law of large numbers, you try and take a more data driven approach where you learn behavior, at the track instead. But I think the key, challenge with methodologies that are learned is understanding their behavior. 00:12:23:05 - 00:12:50:17 Unknown And I think the, the, the challenge that we have in adopting that technology in race scenarios where it's heated, the moment high stress environment, the stakes are very high. Is the explainability of the technology that you use. And obviously there's lots of academic research. There's lots of progress in explainable AI, over the last year. And that's it's definitely improving. 00:12:50:17 - 00:13:18:04 Unknown But I think the there's a, I guess, intellectual skepticism that you have to approach it with the means, the where we are at the moment, the human or the the strategies on the pit wall still has to take a decision and has to understand the options he was presented with. And then if we live a in another area, I think, 00:13:18:06 - 00:13:43:11 Unknown If we can model a problem physically, he would model it physically. Right. So I think where machine learning particularly maybe slightly different to AI, helps is the ability to learn models from data that they are very difficult to model physically. And I think that balance of when it's right to use a physical model, when it's right to use a data driven model, is is always a challenge in formula One. 00:13:44:06 - 00:14:11:04 Unknown Sometimes you want to use a data driven model that maybe the data is too sparse, so you can't. But the trade off between physical data driven modeling, I think is is probably the biggest challenge that that we have. Yeah. And I think anyone watching this, this interview would be quite reassured to know that there's still a human exercising decision in the pits, because otherwise potentially a device could get a little bit too boring, right? 00:14:11:04 - 00:14:37:01 Unknown If we knew that, if I were controlling it, I can I just ask a question based on sort of where you see things going. So first, but I wanted to ask, is your team in software engineering? Are you seeing a change in the profile of the software engineer and people that are being brought in and recruited into Red Bull Racing, or not, you know, has that changed over the last couple of years? 00:14:37:01 - 00:15:03:19 Unknown Do you see it changing in the future? Yeah, absolutely. At the moment, I'd say we haven't consciously changed our hiring practices as an I think it's it's premature to trade people for technology at this point in time. And obviously, this question marks about whether we should do that as a society anyway. But I think clearly the work that a software engineer does has changed, and that that is genuine today. 00:15:03:21 - 00:15:40:23 Unknown And I think that the way you will approach, entry level software engineers or graduates now on over the next few years will also change in terms of understanding the right way to use this technology to accelerate things that you build, but also creating an environment where junior engineers continue to learn. Right? Because cognitive atrophy is real, when the more you use AI to write software, particularly in your early stages of your career, the less perhaps you learn about the systems you will you're building. 00:15:40:23 - 00:16:03:12 Unknown So I think, yes, today the impact is profound. We're still on that journey of figuring out exactly what it is that we need to change. But, for sure, over the next three or 4 or 5 years, there will be, I think, a revolution in the way that you approach building software, which we're now starting to understand again. 00:16:03:12 - 00:16:41:12 Unknown That's really encouraging to hear because I think there's still this, this need to hire problem solvers. Right. And that's a very human quality. Yes. I can do a large amount of that. But the creativity in that innate problem solving, like curiosity, you know, there's no substitution for that, right? Yeah. I think I think it's interesting the it depends what you think the value in a software engineer is, is the value in the software engineer in the code that we write, or is it in the the tools that we build, understanding of how you marry requirements to, to tools or to output outcomes. 00:16:41:12 - 00:17:08:00 Unknown And I think we'll see it bias more towards that domain understanding more than it is the practice of of coding I completely agree. Yeah, but domain expertise will always jump up since we're talking about software engineering, there'll be a number of people watching this kind of thinking, you've got a dream job, Martin writes, myself included. Do you mind just giving us a little bit of insight into your own personal journey? 00:17:08:00 - 00:17:33:21 Unknown I mean, how how did you arrive at this position? What was your background? The journey? I guess the first thing to say is, yeah, obviously I'm very lucky to do what I do. I get to go to work every day and enjoy the problems that we work on and the people I work with. You know, it really is a luxury to work with people that are smarter than you, that you can learn off all day. 00:17:33:23 - 00:17:54:15 Unknown You know, whether or not it's the software engineers we have or, people in in the other engineering functions that are truly world class in aerodynamics and different topics. So that's the first thing is to recognize that, yeah, I'm very lucky in terms of how I, how I got there. I've been involved in this for a very long time. 00:17:55:02 - 00:18:19:14 Unknown I used to race cars as a kid. I, I won races and championships, but eventually I had to, get a real job. I also studied, computer science, so, naturally, computer science and racing cars. Fancy formula is the kind of nexus of those two, desires, I guess. And then I started off working at a former team. 00:18:20:09 - 00:18:38:04 Unknown I was there for five years. And then I moved to rebel, over ten years ago now. And I've really enjoyed pretty much every, every year that I've been on the team. Excellent definition, dash bottom, but you can have a little bit of fun if that's okay. And then we'll come back to a few more serious questions. 00:18:38:04 - 00:18:54:11 Unknown I always do this sort of quick fire round. So the idea behind this is you got about 30s whatever comes to your mind first, one sentence or one word answer. There is no right or wrong. Okay, here we go. One word for AI today. 00:18:54:13 - 00:19:36:19 Unknown Potential nice. One word for AI in five years. Revolutionary. Nice. AI agents hype or real shift? Real shift. Cricket. The biggest risk in AI right now. Not understanding the technology. Biggest opportunity. Real productivity gains. Very good. One thing leaders consistently get wrong misunderstanding the technology. Okay, Bell's gone, but I'm still gonna ask you one more question. If it were one AI term you wish people would stop using, what would it be? 00:19:36:21 - 00:20:00:16 Unknown People's preoccupation with Rad at the moment is something I hear a lot, and I think that comes back to to misunderstanding that the technology, and thinking that we can apply some of these techniques to every problem and and it leads to good outcomes. Yeah. That's interesting. Rock. Yep. I thought we'd got past the rank discussion, but it seems to be hanging around, isn't it? 00:20:00:16 - 00:20:27:08 Unknown Yeah, yeah, yeah. I'm hoping it's not that long. Right. So I wanted to sort of look a little bit further out. You know, two years is a long time in Formula one, so I won't ask you to plot forward more than two years, but where do you where do you see the use of technology being applied to formula one evolving over maybe the next 12 to 18 months, if you can sort of look out, please? 00:20:27:10 - 00:20:59:17 Unknown I think the role of technologies will just continue to grow in terms of its impact. I think as I say within data driven and simulation modeling led for for decades. But I think that if the trend of development or the rate of development in modeling techniques increases in a way anywhere similar set of elements for foundational models and physical modeling, I think you could start to see real profound impacts across actually the way that you design the car. 00:21:00:15 - 00:21:29:01 Unknown Obviously lens helps you with, with the building of better tooling for engineers these. But they, they are unlikely to help an engineer come up with a novel solution or for you to model a problem in a different way. And I think you're starting to see signs that similar trends are happening in, in, you know, foundational modeling for physics, for fluid dynamics. 00:21:30:09 - 00:21:55:22 Unknown And I think the, you know, the impacts of that change will potentially be greater than, than what we've seen at a lens in terms of the actual direct impact on commercial performance. Yes, absolutely. And I'm one one more question. We've the whole this whole series is called, I Changes everything. Is there one particular thing that you think I should not change? 00:21:55:22 - 00:22:17:22 Unknown Or even if we can change it, we should leave it alone. Where should not be used? I think we should. We shouldn't be approaching problems thinking that AI replaces what a person does with a. Let's say we said earlier about the strategies on the people, but I think it also applies in software engineering and, and other, other areas of the team. 00:22:18:03 - 00:22:49:23 Unknown I think formula one is fundamentally, a war of brains. And I think that you still want to have the best people. And I think the AI acts as a superpower for them in order to be more effective, but it doesn't replace them. And I think, yeah, when we approach, the adoption of AI and the introduction of it into the tools that we build or the organization, we we still have to, I think, value the experience of, of humans and engineers. 00:22:50:04 - 00:23:10:11 Unknown Absolutely. Well said. I product of the week. Martin. What are the questions I always ask is, do you have a favorite prompt or a way that you, you tend to interact with the LMS that you'd perhaps like to share? You know, I think it's quite easy to take for granted just how much information and knowledge you have available to you any moment now. 00:23:10:11 - 00:23:35:12 Unknown And one thing I find myself doing more by default is talking through technical problems, design algorithms, architecture with an alarm or a chat bot as my default for how I solve problems. And I guess more specifically in terms of prompts. When I'm doing that, I find that sometimes you actually really have to ask the alarm. Are you sure? 00:23:35:14 - 00:23:57:22 Unknown Are you sure? Can you check that before you really arrive at the right answer? Very good. That's great advice for all of us. Thank you. Well, Martin, unfortunately, we've come to the end of our discussion. I think we could go on talking for ages. And I have to say, I think you've given us a rare insight into what, you know, what's going on behind the scenes of formula One and also Red Bull Racing. 00:23:57:22 - 00:24:40:12 Unknown So thank you very, very much. I really appreciate your time. And to everyone else watching, I hope you've enjoyed this as much as I certainly have. And I'd encourage you if you want to watch more, interviews and discussions in this series, go to our Qualcomm forward slash I changes everything button. Thanks again.

3. 

   16 Jul 2026

   AI Changes Everything: Using AI to Help Prevent Human Trafficking

   Discover how AI, data sharing, and cross-sector collaboration help organizations identify trafficking patterns, improve prevention efforts, and disrupt criminal networks. See how technology can support measurable action against human trafficking by exploring more at www.oracle.com/aichangeseverything/video-series How can artificial intelligence help address one of the world's most complex humanitarian challenges? In this episode of AI Changes Everything, Neil Sholay, senior vice president of AI and Data at Oracle, speaks with Ruth Dearnley OBE, president and founder at Stop the Traffik, about how AI, trusted data sharing, and collaboration across industries can help prevent human trafficking. Rather than focusing only on technology, the conversation explores how AI becomes more valuable when applied to real-world outcomes. Ruth explains how Stop the Traffik developed the Traffik Analysis Hub to combine data from financial institutions, businesses, law enforcement, governments, technology platforms, and communities to reveal patterns that would otherwise remain hidden. Together, they discuss how shared intelligence can help expose trafficking networks, improve prevention, and support faster, more informed decision-making. The discussion also explores the role of predictive analytics, AI agents, data quality, responsible innovation, and collaboration across organizations. Ruth shares examples of how insights are being applied to major global events, including preparations surrounding the FIFA World Cup, and explains how measuring prevention has become a meaningful way to demonstrate impact. Throughout the conversation, both speakers emphasize that technology is most effective when it complements human expertise, helping people make better decisions while preserving the human connection at the center of every solution. STOP THE TRAFFIK is a global organisation dedicated to preventing human trafficking. We combine data, technology, and lived experience to identify patterns of exploitation, empower communities to protect themselves, and work with businesses, financial institutions and governments to drive systemic change. Our vision is to create a world where no one is ever bought and sold. If you would like to help, or donate to Stop The Traffik, visit: https://stopthetraffik.org/donate/ 00:00:00 Disclaimer 00:00:15 AI for Fighting Human Trafficking 00:01:00 Meet Ruth Stanley OBE 00:01:23 Stop the Traffic Origins 00:05:07 Building a Global Movement 00:08:35 Human Trafficking as a Business 00:12:37 Creating the Traffic Analysis Hub 00:17:03 From Insight to Prediction 00:20:09 World Cup Trafficking Prevention 00:24:04 Ruth's Personal Journey 00:29:21 Technology, Scale, and Vision 00:33:17 Building a Network of Networks 00:36:14 Rapid-Fire AI Questions 00:39:23 AI in Everyday Work 00:40:17 Closing Thoughts

4. 

   24 Jun 2026

   AI Changes Everything: What Leaders Must Get Right About AI

   Discover how leaders can move beyond AI experimentation to build trust, scale adoption, and create meaningful outcomes across defense, government, and enterprise organizations. In this episode of AI Changes Everything, see why leadership, mindset, and responsible AI adoption matter more than any individual technology. Explore more AI topics at http://www.oracle.com/aichangeseverything In this episode of AI Changes Everything, Tony Reeves, partner for AI in Defense at Deloitte, shares his perspective on one of the most important challenges facing organizations today: how leaders can successfully adopt AI in complex, high-stakes environments. Drawing on experience across defense, government, technology, and consulting, Reeves explores how AI is being applied across command, operational, and administrative functions. While public attention often focuses on autonomous systems and battlefield applications, he explains that many of the most immediate benefits are emerging through decision support, productivity, logistics, planning, and knowledge work. The conversation highlights how organizations can create value by focusing on real user problems rather than becoming distracted by technology for its own sake. A central theme throughout the discussion is leadership. Reeves argues that mindset often matters more than toolset and explains why organizations must move beyond traditional approaches to innovation if they want AI initiatives to succeed. He discusses the importance of multidisciplinary teams, user-centered design, trust, assurance, and creating environments where experimentation and learning can happen safely. The conversation also explores why many AI pilots fail to scale and what leaders can do differently to improve adoption and long-term impact. The episode examines some of the most important topics shaping the future of AI, including sovereign AI, resilient infrastructure, data strategy, energy requirements, assurance, governance, and the balance between human oversight and automation. Reeves shares his views on the future of competition, the changing nature of national security, and why organizations must prepare for a world where AI becomes deeply embedded in decision-making and operations. The discussion concludes with practical leadership advice, insights into preparing the next generation of leaders, and a thoughtful exploration of why human ingenuity, curiosity, and judgment will remain essential as AI capabilities continue to advance. For leaders navigating AI adoption, this episode offers valuable guidance on how to build trust, create value, and prepare for a rapidly changing future. Tony's Prompt of the Week: Please search my emails this week for any outstanding action items that need my attention, do not just include emails where I am directly @ mentioned but include emails that ask or imply that I need to take action. For instance, "What do you think?" or "Can you..." Estimate how much time it would take to complete the task based on these criteria: ○ Do now – 2 mins ○ Focus – 5 mins to 30 mins Do last - more than 30 mins Next, organise them into a table labelled as 1. 'Do Now,' 2. 'Focus,' or 3. 'Do last,' the date requested, a description of the task, a link to the message, and your estimation of the time required to complete each task based on the time estimation criteria provided, and present the table sorted by the label category. ------------------------------------------------ Episode Transcript: 00:00:00:00 - 00:00:15:06 Unknown The views expressed in this podcast are those of the individual speakers, and do not necessarily reflect the views or policies of their organizations or call or its affiliates. 00:00:15:08 - 00:00:32:13 Unknown What role does I play in national defense? How does leadership need to evolve with the use of AI in that scenario? Where should I be applied and not in the context of command and control? 00:00:32:15 - 00:00:51:19 Unknown Hi everyone. I'm here with Tony Reeves, partner for AI in Defense at Deloitte. We're going to be talking today about the use of AI in a defense context. Okay, Tony, great to have you on here. I don't even know this. We obviously met previously. You are the first person I had on the list to interview for this. AI Changes Everything video series. 00:00:51:22 - 00:01:09:05 Unknown And I'm not just saying that. And it's because I had. I thought you had some really unique views about how I could be used. So no pressure then at that point. No, no, we're going to keep this very, very light. So tell us a little bit about your role at Deloitte. What you do, if you could please give us some, some background. 00:01:09:07 - 00:01:34:20 Unknown The journey. Sure. So I've been doing it six years. We are traditionally seen as a tax and audit company, a professional services firm. And six years ago, we deliberately started thinking, what's the future going to look like, especially in defense around technology and this idea that we have to be continuing what we've always done, helping clients fix difficult problems and challenges, but also be relevant in that future space. 00:01:34:20 - 00:01:56:07 Unknown And that means actually understanding the technology, how it's applied, what it does. And so I spent last six years in Deloitte trying to help us work out who is defense, security and justice. What does I mean for that sector? How do we start to adopt AI responsibly, safely and effectively? We need to be making sure that it does and lives up to its promises. 00:01:56:09 - 00:02:16:00 Unknown Prior to that, I've spent time in Microsoft, as a digital advisor. I've worked in the government. I've led major programs. And I spent 13 years in the Army. And so I have this weird, background of being an infantry officer, some who used to hide in hedges and carry heavy kit, as well as a techie geek. 00:02:16:00 - 00:02:32:06 Unknown And I was always a geek at heart. Coding in the early 1980s when I was a kid, and continue to do so. Excellent. But I think that that's what makes you quite unique. You know, you've sort of you've been on both sides of the fence, as it were. Okay. So why don't we start with just talking about where does it make sense? 00:02:32:06 - 00:02:55:05 Unknown Where is it appropriate to use AI in a command and control context within defense? And then perhaps where does it not? Yeah. I think that's, you know, that's a heart of I think so to the audience, I mean, most people saying what's defense doing with AI? And naturally there's a concern is a worry that like, you know, whenever something is not visible or not clear, people fear the worst. 00:02:55:07 - 00:03:17:01 Unknown And I think that, you know, often and sadly on regular as you know as well, panels will always be what about the Terminator? Robots are going to come and kill us, of course. And it's like, well, yes, there is a risk. But yet actually defense is taking that extremely seriously as an organization. And so where we're increasing is seeing the use cases really falling into two three parts. 00:03:17:01 - 00:03:38:00 Unknown Is this the yes the warfighting the battle space piece where of course in you know, Ukraine we see this all the time where where drones where they're controlled or automated. On both sides are causing damage and destruction. But that's a fairly small use case, although it's of course, going to grab the headlines, in the command space. 00:03:38:02 - 00:03:59:03 Unknown There's many activities which we as a civilians will just take for granted the idea that plotting a route from A to B, going from London to Birmingham, we'd immediately go to an AI and and a map system, say, what's the quickest route? And yet for defense, this is still relatively new in many cases because partly attention's on the battlespace as opposed to that command. 00:03:59:03 - 00:04:17:09 Unknown And control. And how do we make things more efficient? But then, of course, the biggest area, like every organization is in the office space, and that's where I think everybody is looking and saying, here's how AI today is making a difference to how we work, what we do and how we deliver the the outcomes that we have to achieve. 00:04:17:11 - 00:04:40:03 Unknown So there's sort of battlespace office space, command space. And right now you're saying fair amounts of it's happening in the office space and the command space okay. It's interesting because as you rightly say, all the headlines immediately go to the battlespace, but all the values and the impact is happening in the other, too, right? Yeah. And I think that, yes, there is a lot of work in battlespace. 00:04:40:03 - 00:05:00:23 Unknown And yeah, we see, many things in the headlines because it's always going to grab the attention. Last week, Ukraine announced that they had seized ground in Ukraine. Back off Russians, using purely autonomous systems. No humans involved at all. Apart from on the Russian side and Russian surrender to a small group of, remote controlled drones. 00:05:01:01 - 00:05:26:17 Unknown Yeah. So they is always going to be a headline grabbing story because nobody's interested in. I've used this AI to help me write this email. It's not going to make the 9:00 news isn't understood. Yeah. I like you. I spent a lot of time talking to mid and central government. And the impression I get is that there's a, there's a real willingness now to adopt AI and apply it to things like logistics and stuff like that. 00:05:26:19 - 00:05:57:12 Unknown You know, who are you typically working with? You know, is it the generals? Yes, I rose, and, you know, what are you hearing from them in terms of how they want to use AI? I think that's a that's a really big question is going to be lots of different ways to answer that. Normally my clients range from, government ministers and departments, who are probably trying to either implement a strategy or been told by, you know, that, head of state that we must do more in AI, either to create jobs or to move things or just to become more efficient. 00:05:57:14 - 00:06:18:11 Unknown The generals, of course, are increasingly starting to look at what does it mean? What how should I go? And, you know, and so in the UK, we've had some very forward thinking leaders. I mean, Tony Radican kicked off the Nelson project a decade ago, which was like, I don't know how much I is going to impact the military in the Royal Navy, but if we don't start thinking about it today, we'll never know the answer to that question. 00:06:18:11 - 00:06:37:14 Unknown So let's start exploring. It might be we spend a pound. It might be we spend more than a defense budget. And I bet if we don't explore, we're not going to know. And so I think that that sort of, you know, most generals now today are starting to feel we need to be moving faster out of the explore space and much more into the implement and adopt in the use space. 00:06:37:16 - 00:06:54:15 Unknown And I think the other bit, you know, the other group that we spend most of our time working with are the users, you know, in a way that, you know, AI is changing what we do. It's also changing how we do it. And traditional defense procurement of in 20 years time, I need to replace this battleship with aircraft Kiowas tank. 00:06:54:15 - 00:07:16:06 Unknown And so I need to define the requirements. Today is changing because the user is saying, well, I need it faster, quicker to day in in the battlespace. The technology is changing much quicker. And as a result, we have to spend more time listening to the people using the capability and the problems and the use cases they have, than perhaps the people more removed from that immediate problem. 00:07:16:08 - 00:07:33:10 Unknown And again, this is very similar to most civilian organizations adopting AI. We have to have the problem to fix, and that needs to be the person who has the problem helping us fix it. Understand that. Yeah, you're absolutely right. This is something we're seeing in banks and retailers, right? Start with the outcomes. Start with the user experience and work backwards from there. 00:07:33:12 - 00:07:57:00 Unknown Just can I talk a little bit about, you know, the mindsets required for AI. Because you're right. We see you know, generals and ministers say we need to adopt this. We need to move faster. But that's the mindset change, right? So a general that's been in position in well, for 30 years, compared to one that's maybe 4 or 5 years into their tenure, is in a different mindset that's required today. 00:07:57:02 - 00:08:16:10 Unknown I think that mindset is far more important than toolset, and it is much easier to talk about the toolset. It's much easier to to understand what a, you know, what a tank does. And, you know, as a young officer, you're trained. This is what a tank this is how it works, is how fast you can go. This is the type of hill it can go up. 00:08:16:10 - 00:08:40:01 Unknown And so all of that is defined. It's known. It's it's it's like you it's certain. And so the toolset, whether that software toolset and you know, products that are helping make decisions or it's the remains of the physical hardware toolset of a drone or an unmanned vehicle, you know, is also relatively easy to know. It's a comfort zone mindset and changing mindset. 00:08:40:01 - 00:09:09:23 Unknown I think that, you know, military have always been quite comfortable fighting the last war. And, you know, the cliche of, you know, the hardest thing of it, harder than putting a new idea into a general's mind is getting the old idea out of it. And I think that ability to put stuff into people's minds and say, you need to think about the problem to resolve it and then think about it all said people instinctively get when you explain it, when you talk, when we have conversations like this, people go, oh, so obvious. 00:09:09:23 - 00:09:29:14 Unknown Yes. And yet when you speak to somebody who has spent their entire career 30 years doing the thing that isn't so obvious, it sometimes needs people like us to say, that's how I think. Let's talk about mindset. The other thing I'd add about mindset as well is that often if we start with the toolset, we look at what do we have today? 00:09:29:16 - 00:09:48:23 Unknown And so look at I have a tank today and tomorrow I'm probably going on another tank because I understand tanks. So the toolset thinking that I so like you that known that's certain approach. You know suddenly encourages people to go I know this. That's what makes me valuable. That's what makes me important. That's what makes me a general. 00:09:49:01 - 00:10:04:04 Unknown And so I'm very comfortable just replacing like the like the leap of mindset is hard. And this is, you know, where we talk about use cases and the problem set. Once you have a tool set and you apply it to a problem set and you speak to the user and say, what are you facing? Is the problem the issue? 00:10:04:04 - 00:10:24:05 Unknown What's the thing that challenge you the most to achieve that overlap? So like I said, actually this tool sets useful, this tool sets obsolete. And that starts shifting people into the mindset of how do I learn more rather than being a know it all how to become a learn it all? And this is something I think Deloitte in particular have done very, very well. 00:10:24:05 - 00:10:49:15 Unknown Right. This whole thing of you know, the theory of AI, it's very, very hard for someone to get their mind around it. But the minute you identify a problem and you develop into a use case, and then you start building something, using the tool in a meaningful way, it sort of comes to life, right? Are there any techniques that you're using that are working quite well where you're you're helping people move along? 00:10:49:15 - 00:11:12:19 Unknown That sort of mindset journey towards going from theory to practical application of AI. Any best practices or techniques that you've seen that have worked well? I think that one of the advantages that organizations like Deloitte have, compared to, I'll say, pure technology companies, is that we have a long history and, you know, business in helping people change how they think that transformation of business change that. 00:11:12:19 - 00:11:34:16 Unknown So like you, here's a problem. Let's think about differently. What ideas have you got. And sometimes, you know, being a consultant is a is a blessing and a curse, you know, because the first bit of it is like you're going to tell me what time it is by looking at my watch. And that sort of, you know, consulting approach, has a certain oh, it's trickery, it's games, and yet it's really effective. 00:11:34:16 - 00:11:58:19 Unknown It's a, it's an effective way of getting people saying, let's think this differently. Let's think about this in a different way. Let's let's start putting together all these ideas and prioritizing them then. And this is what the change has been in the last six years is how do we start applying that technology mindset, that technology delivery, that way of working in a more agile, flexible, lean startup manner to that consulting business and also with our clients to fix our problems. 00:11:58:19 - 00:12:28:08 Unknown And so I think that we had a a natural advantage. And we can merge those two. We've had to learn far more about delivering technology with hands on to bring that capability in. And that's giving away our secrets here because people said, well, perhaps I can start learning some consultancy stuff, but I think know very much, you know, if you're talking to users and especially users who aren't used to talking about their problems, you know, more used to a system where I've given to the requirements manager, they're going to fix this for me. 00:12:28:09 - 00:12:48:01 Unknown Yeah. If you're talking to users, you need those techniques. You need those abilities. Make them feel comfortable to share ideas, to have open discussions, to realize that failure is going to happen. But that's okay. Yeah, that reassurance that we're going to help them through the failure to learn, we're going to do that quite quick. So all those techniques, which we'd always have as an agile delivery team, that's what we've been trying to bring together. 00:12:48:03 - 00:13:12:12 Unknown So it's understood. Yeah. We spoke a little bit about the human element, the mindset element. And one of the phrases we always use in sort of corporate AI is human in the loop, you know, in the flow of work. Do you have a view onto when there should be a human in the loop for use of AI, or if there are situations where you absolutely want completely autonomous control when it comes to defense? 00:13:12:14 - 00:13:34:20 Unknown A couple of things on this, I think that the average general thinks there is strategic as Eisenhower, as good looking as George Clooney. You know, and as competent to you as a far better than anybody else. It's in their command. And yet the average general is kind of average by definition. And so we end up trying to get the average. 00:13:34:20 - 00:13:51:19 Unknown Don't think, well, you know, there's things I can do to enhance you, to make you better, can just support that decision. And we've had histories of this, you know, not just, you know, in the last ten, 20 years, but if you go back to Second World War, people using science and data to make decisions like, do we invade Normandy this day or do we delay it? 00:13:51:19 - 00:14:11:06 Unknown You know, those difficult, hard decisions scientifically based, all the AI is doing is presenting that science in a faster, quicker way. In many cases. And so I think part of the challenge with people is to try and get them into a point of saying, I can use AI to enhance what I do. Once we get into that space, then it's a question of what do I not want to use AI for? 00:14:11:08 - 00:14:35:16 Unknown Yeah. When we talk about the battle space area, people quite rightly get worried that without human control, we're going to have a similar problem to the mine issue that we've had in places in Croatia and Bosnia. You know, Croatia last month declared that finally, you know what, 25 years after the conflict was resolved, so I 30 years after conflict resolved, then I mind free as a nation. 00:14:35:16 - 00:15:00:05 Unknown You think in the 30 years they've been living with the consequences of of uncontrolled munitions? So people are quite rightly worried about AI without humans. Yes. Causing unknown issues and problems. And so we certainly need to have humans with control over AI and consciously thinking, how do I use it? But the debate we often get into then, do we have humans in the loop? 00:15:00:07 - 00:15:36:03 Unknown You know, the average normal military officer will say, of course we do. And the ethical AI technologies will say, of course we do. We must always have humans in the loop. And yet the concern is this if an adversary is using AI to move faster, to think quicker, to create damage, to our soldiers, to kill them more efficiently, effectively and more rapidly, then will we as a society be comfortable that we are maintaining humans in the loop at the cost of human life? 00:15:36:05 - 00:16:09:15 Unknown Now, we could argue that this is one reason why we need international regulation around the use of automated and autonomous systems in the battlespace, and we certainly should discuss that. But the same time, we need to recognize that we need human control. But it does not necessarily mean that humans need to be in the loop. I'd say that in the next couple of years, we increasingly going to see autonomous systems acting without humans, but controlled by humans on what their targeting is, on what they do, and with controls to deactivate those devices once their mission has either succeeded or failed. 00:16:09:17 - 00:16:28:05 Unknown That's where I think we're going. And it's exactly the same as you say in the office space. Right there are in the same over. I have conversations with with banks and governments and retailers every day as to where you want to give agency to the agents and allow them to have full autonomy. You know, you will have those exact same scenarios in defense. 00:16:28:08 - 00:16:54:01 Unknown Right? I think just don't continue that. You know, back in 2017, I was one of the people who signed the Future of Life paper to say that we need to have better control over lethal autonomous systems. And I maintain that view. We need to have better control over lethal autonomous systems. And I often say, I don't think that that is putting humans in the loop to do that control, because we will end up with a worse problem than what I'm trying to solve. 00:16:54:03 - 00:17:17:23 Unknown But things I think we do need to discuss is, you know, when, in 2007, Estonia was attacked and hacked. We had the Tallinn Manual that got published afterwards, and the Tallinn Manual is guidance for military commanders by lawyers as to what is, within the law of armed conflict for the use of cyber warfare. 00:17:18:01 - 00:17:45:09 Unknown That manual gives very clear direction which military commanders can understand, which says if I use cyber warfare to do this, if I destroy hospital systems, that is probably going to be a war crime with serious consequences as a result, that guidance is essential for the conduct of any type of cyber attack or defense. Yet we don't have that similar option, that similar manual, that similar guidance for the use of AI in a battlespace. 00:17:45:09 - 00:18:06:09 Unknown And I think we're increasingly seeing AI that can do significant damage and significant harm. And we are leaving it to the commanders in the field to make those decisions about law of armed conflict, which they are trained in, and they get regular updates on as to what is legal and not. Yeah, but the application is hard. So that's why I say we don't need a human in the loop. 00:18:06:11 - 00:18:33:22 Unknown It's why we need to be debating what are we comfortable? It's important to balance, and we must always consider the human element, particularly for the commanders in the field, to shift in a little bit. You spoken, in the past about, you know, that mix of algorithms, compute data, energy. What are you seeing? Are we getting better at using compute using the data energy is obviously something that's being discussed. 00:18:33:22 - 00:18:44:08 Unknown Now, you know that mix of those four elements? Are there any that, we're, we're doing better at or worse at some. What's your view? Yeah. 00:18:44:10 - 00:19:01:07 Unknown My first defense AI job was 1999 during the first bubble. And it was like, is like discovering a new world. It was just like, we can do all this things. It's amazing. It's going to be awesome. It's going to be incredible. It wasn't and is really dangerous and is really bad. And it was just like, yeah, oh, really doesn't work. 00:19:01:07 - 00:19:27:16 Unknown Oh, no. But where that drive came from was compute an algorithm, know the algorithms which had been built and developed over many years. Suddenly, with the arrival of the internet, we saw anything at all. We've got. We've got compute now. We can now start doing it. We got really high. Great compute capability. Surely if you throw enough compute at the problem with a really clever algorithm, we can infer confident and right decisions. 00:19:27:18 - 00:19:46:18 Unknown We couldn't, we didn't. It wasn't a good answer today. I think that, you know, with the internet embedded into our lives, that gives the data aspect, that gives us the access to understanding, which is obviously created large language models. And, you know, GPT and I mean, jump in this whole world today where people have AI, multiple eyes on their phone. 00:19:46:18 - 00:20:11:05 Unknown That's all because the data has been understood, exploited and used to help create, artificial intelligence, data is the real shift, isn't it? The data is actually being able to understand, to comprehend, to place your more sensitive data in a field of wider understood data and, and say, what are the differences, what is happening and whether that's health care and saying, here are my patients data. 00:20:11:05 - 00:20:29:06 Unknown How do I share that in a trust environment to see, do I have patients with a case that I can help cure quicker, or whether that's defense data is at the heart of where AI is? The last one, as you mentioned, is Sunny Energy. Yeah. And I think that the military adoption, we're still starting to work out that. 00:20:29:08 - 00:20:51:21 Unknown Yeah, all this compute, all this data, all this algorithmic driven decision making costs, energy. And the one thing that defense is good and bad at is energy. I mean, on the bad side, we have fast jets that burn thousands of gallons of fuel in a minute. So we have high capacity energy and it really bad at making it efficient and effective. 00:20:51:21 - 00:21:11:01 Unknown Nobody's like talking much about the energy management for algorithmic use in defense. They're talking about do I have access to some energy? And energy will always be there. So I think data and energy of the two real big topics, which will be it's interesting. I hadn't considered the energy part. Yeah. Obviously there's a race energy when it comes to AI. 00:21:11:01 - 00:21:36:21 Unknown We're looking at alternative approaches, but this whole drive of efficiency, it takes on a whole new meaning when you're, you know, away from central HQ over the centralized cloud data center. So just using that same amount of energy much more efficiently, something I hadn't considered. One of the things that one of my old programs was the fitting control systems onto military vehicles and and replacing tactical command and control. 00:21:36:22 - 00:22:01:18 Unknown And, you know, one of the debates we saw get into was on one side, every vehicle becomes a node for artificial intelligence was back in 2015. So just like every vehicle is a node for artificial intelligence and, you know, decade ago, ever seen what nonsense is this? Yeah. Well that's decades away. Artificial intelligence. But the biggest hurdle was the power supply within the turret and trying to say, well, actually, if I want to have a faster compute capability, I need more power. 00:22:01:20 - 00:22:21:16 Unknown That power is coming off of the engine running, which creates thermal signature, or the batteries running, which drains the batteries. And I probably need that for other things like life support and surveillance and total control and weapons systems and stuff. And so in parts we're quite good at energy management, but scaling and consciously thinking about saying, I need to upgrade this capability with an edge, compute that. 00:22:21:17 - 00:22:40:01 Unknown Yeah, it will suddenly provide the comms network for a bigger compute back there. Yeah, higher bandwidth going backwards and forwards plus the energy management. That's a debate still behind our defense. I was I'm actually seeing the same thing playing out actually in police forces. Right. Yes. How do you create almost this private mesh network where it's completely distributed and highly resilient? 00:22:40:03 - 00:23:00:16 Unknown Just going back to data, one of the things I think that has changed is if we were having this conversation maybe five years ago, it would all be about big data. Yeah, take all your data, put it in one place, prepare it, unify it. Off you go. Now, I think we're becoming much more selective around which data we use. 00:23:00:18 - 00:23:21:15 Unknown But also we've learned this new trick of bringing the algorithm to the data and not the data to the algorithm, which which I think has just opened up so many more scenarios for how we how we use AI. Are you seeing something similar as well? Certainly not that big data. We need a data lake. We need to publish publicly, you know, so put everything into one big puddle when there's often a puddle as puddle like that. 00:23:21:17 - 00:23:41:10 Unknown If one big problem is like, that's all my data. And you're entirely right. And that's like you ability to. Can we put the algorithm and the compute where the data is and then save bandwidth, especially for military purposes, because bandwidth is the hardest thing to command on the battlespace, where it's all being denied by the enemy or you don't have enough of it. 00:23:41:10 - 00:24:02:06 Unknown Yeah, there's no organization apart from the US military that has relatively unlimited data bandwidth and even they have a limited relativity. So I think that sort of I get that that requirement to start putting capability forward and whether that's, NPU chips or whether that's going to be GPU chips push forward. Yeah, that's something that's a really exciting possibility. 00:24:02:08 - 00:24:25:12 Unknown But increasingly we're not seeing that sort of necessarily that thinking in, in many parts of defense. The other thing as well, of course, is creating small and medium scale language models, you know, highly bespoke, trusted, safe capability that does limited create limited solutions or use cases, but therefore run on lighter compute, lower energy, smaller amounts of data as a result. 00:24:25:14 - 00:24:47:23 Unknown I think we're starting to see a shift in the the large AI providers and more bespoke solutions. You could argue that they're starting to differentiate about we do this, we do coding better, we do imagery better. We do understanding or language better. But nobody's really shifting. Perhaps Mistral, maybe a sign shift into that deployed edge. Bespoke compute capability. 00:24:47:23 - 00:25:06:14 Unknown Yeah I agree I think definitely starting to see the early signs of, you know, highly optimized specific small medium language models that solve 1 or 2 tasks incredibly well, and the footprint is much, much lower, right. The ability that you could deploy potentially on a small device, I don't know as well. You have that high degree of assurance as well. 00:25:06:18 - 00:25:24:01 Unknown Yes, because now you have an algorithm which is doing 1 or 2 things, and therefore all the training and the understanding and the knowledge there to use that safely is easier and simpler. Yeah. It's going to identify this. It's going to say yes or no to that. It's going to tell you this or information if it's telling you anything else, it's not working. 00:25:24:01 - 00:25:41:22 Unknown And in terms of cost efficiency but also security, it's a lot easier to lock these things down. And yeah, and we're working with Mid-Year assurance partner for AI and that's very much what you want to get into is to saying, okay, if you work out what that, you know, same as the secure by design or to about trustworthy by design. 00:25:42:04 - 00:25:58:22 Unknown Start early, start thinking about what to use cases. Start thinking about what a data is, stopping what the user is going to do. That's a much easier thing to accredit and assure and to trust than if we're trying to say, well, it could do this and it could do that, and I might get this bit of information, you know, simplify it. 00:25:58:22 - 00:26:17:13 Unknown Problem sets allow us to start saying, okay, that is a simplified tool set to govern and control. That's interesting because that's another thing that I'm a big believer. And, you know, you know, as I said five years ago, we be talking about big data, but also five years ago, we'd be talking about solving the entire problem before we get moving. 00:26:17:15 - 00:26:43:15 Unknown But now this ability to break it down and simplify it, get going and then iteratively go build as we go, I'm seeing that a lot in sort of the enterprise space. Is that something that you're seeing in defense, where we're breaking the problem down into smaller chunks and simplifying 100%? But you are. But I imagine you also seeing that, you know, chief finance officers are less comfortable with that because from an enterprise scheme, they want to say I want X amount return on investment. 00:26:43:15 - 00:27:01:01 Unknown And I'm going to introduce the AI to do that, return on investment in the same way as, you know, back in the 90s, probably the migration to yeah, we're both old enough to remember Microsoft Office 95 coming out and stuff. Yeah. Office was seen as this is going to fix this problem. This is going to and we all did the same thing in the same way. 00:27:01:01 - 00:27:25:03 Unknown If we send an email subject to type two email press send. It goes it's it. And the training is very simple. The return on investment is I now don't have to post stuff using stamps saving I can I send millions of emails every day less we're saving I mean but that's why I go. That shift is quite easy for economic chief finance officers to calculate once we start decomposing almost individual use cases. 00:27:25:03 - 00:27:42:19 Unknown This the soldiers going to use AI for that. This office is going to use it for that. This person is going to use AI for that. So the return on investment becomes very individualistic. Yes, I think all of that ends up and you can quantify it much easier. Well, you can, but yet when you aggregate that up it's much harder because it's not a it's not a simple calculation. 00:27:42:19 - 00:28:02:05 Unknown So we did some work with the minister defense where we're looking at the adoption of AI. One of the feedbacks we had was that for people who are neurodiverse, it was a far higher net satisfaction score that came out because they suddenly found they didn't have to get stuck in the process, which they struggled with. They could use AI to guide them through an answer, the process. 00:28:02:05 - 00:28:25:21 Unknown So that return was significant for that percentage of the workforce was quite large. Who struggled with process. I'm one of those people who really struggled to process. And so, you know, having I do it for me it but what we end up with is when we try to aggregate that up, the question is, well, how many minutes did we save per person or have you that stuff that's like enterprise way of thinking about return on investment. 00:28:25:23 - 00:28:52:10 Unknown Is being challenged by AI very, very hard because we are having to almost create an individual return on investment and somehow aggregate those up. I'd save many hours by not attending meetings when I'm not needed, but seeing the summary of it and then getting that understood to me does everybody have that use case? No. And so we can't just blank it out so much as you do to the email to every we use email on average X people send X amount of messages per day the return and say that's interesting. 00:28:52:10 - 00:29:25:20 Unknown Yeah. Because I'm it's almost like you Evans paradox. Right. If you free up 50% of someone's time, doesn't mean you all get to go home 50% earlier, right? Yeah. Actually, the research shows they actually end up working more because they have, you know, more time to invest. Are you seeing a level of scrutiny from the finance teams around the, the, in the initial investment I pilots and then the return on investment in the same way that we're starting to see now, because we all know the figures, like 80% of AI pilots failed. 00:29:25:22 - 00:29:44:02 Unknown I think that's changing now. But are you seeing similar levels of scrutiny, I think is the scrutiny, commercial engagement. And one of the big things I've seen, certainly in the United Kingdom, starting to shift in some parts of NATO, is that the commercial teams are changing how they contract for digital capabilities, not just AI, but digital capabilities. 00:29:44:04 - 00:30:00:16 Unknown And so they're increasingly starting to, you know, to be interested at the start of the idea. And when, you know, traditionally, you know, security creditors had this the same issue is that they get called in at the end. We've built this. So we're going to do this now. Come in. Accreditation and commercial teams often were doing exactly the same. 00:30:00:22 - 00:30:17:13 Unknown We decide are we going to buy this go and buy it for us. And so the commercial teams really struggle with that. So what we ended up with is commercial teams now again, very engaged early on to say, what are you trying to do? How are you trying to do it. What timescales you trying to do it. And so they are able to say, oh, here's the best way to build that capability. 00:30:17:13 - 00:30:33:04 Unknown And we've seen some really significant changes over the last 4 or 5 years with the mindset of the commercial team who are curious and want to know what's going on. You know, first time I had a commercial officer turn up in a meeting with a client just because they want to know what's happening as opposed to something gone wrong. 00:30:33:07 - 00:30:50:17 Unknown Yeah, that was a shock. But actually, that's exactly what I'm seeing regularly, regularly, regularly now with commercial team. And the same with that accreditation team. You know, that the security, the assurance piece. Yeah. So much easier. You start with that. You work through the process, you provide evidence as you go that it's working. That was the use case different. 00:30:50:19 - 00:31:09:05 Unknown Well this is what the consequences. And Sean's team couldn't say yeah okay. We understood that we anticipated that or we didn't anticipate that. But we understand where it's going. So I think is a real shift. Yeah. And I'm seeing the same thing, you know, commercial teams, finance teams, but also HR are suddenly turning up, sitting in the back of these meetings and starting to participate. 00:31:09:06 - 00:31:38:17 Unknown I think as a sign of maturity. I welcome it actually, because I think it's really good because just keeping this in the labs or in the tech teams, it isn't going to scale, right? We have to get this out into the other of business. Oh yeah, I think I 100% agree. And I think that if any organization that's deploying AI through their CIO is probably seeing that 80% for exactly where we have diverse multi, skilled teams coming together to say, I've got this problem, I think I can help us. 00:31:38:17 - 00:31:57:11 Unknown And then the tech team are helping them deliver it. The HR team is saying, okay, how do we deal with this? Where does it go? You know, the commercial team saying, what's the best way of not just contracting it for day one, but supporting it and refreshing it? And how often do we need to change the model, the support, the package, perhaps need to a whole new project in a year's time? 00:31:57:13 - 00:32:21:00 Unknown You need to mingle those two people together. Those intelligence organizations, to me, is is the most fundamental shift that we're saying, right. That's the key to success. And it's ironic. I can help them do that. Yes. But this reality becomes self-reinforcing. So one of the things I have to ask you, you know, the phrase that's on everyone's lips, we've kind of touched on a little bit is something I, okay, trust. 00:32:21:02 - 00:32:48:16 Unknown I'm certain that this must be something that comes up almost every day when you work with your clients. It certainly is for me. What's your view on solving AI? What are you hearing from defense? I think the, the current situation globally where we have an increase in conflict, you know, we've had a long period, especially in the west of a decrease in conflict, and a more stable and peaceful society. 00:32:48:16 - 00:33:12:04 Unknown The whole peace dividend, post-Cold War, has led society to think, well, we're in a period of peace and will always be in a period of peace. And that, I think, is a flawed assumption. It assumes that everybody is interested in maintaining the current status quo. And we're seeing many world leaders who are either deliberately or accidentally disturbing that status quo. 00:33:12:06 - 00:33:37:05 Unknown And so we end up with, first of all, this increased conflict, uncertainty. We've also seen a rise where national interest has since become more important, not the most important. And we still see a huge amount of, a global cooperation and collaboration. But national interests are more important going the national interest is this where do I go with my data? 00:33:37:05 - 00:34:00:16 Unknown My. I may take technology. And if you start looking into, you know, I as a fundamental driver of growth in a society, then owning, being responsible, being it trust it, be able to just declare it as I have confidence that won't be switched off by an adversary or by competitor, or somebody else will outbid me. And then suddenly is a national organization, national state. 00:34:00:17 - 00:34:41:00 Unknown I'm not providing that critical capability. That's obviously on every political leader and senior military officers mind. It's like, can I trust the access and availability of that system, not just the nationality and the origination of that system. And so we I think seeing every NATO country starting to come together and saying, at what level do I need to have purely sovereign capability, where everything from the energy supply to the data to the compute, to the facilities, to the failover capability, to the management, to the build, all of that is entirely done by my single nation. 00:34:41:02 - 00:35:01:23 Unknown How much of that can I collaborate and share with and trust allies to build, and how much of that can I genuinely say? It gives me higher resilience, higher fail over higher assurance, higher trust by being able to suddenly say, well, actually, this energy system has been destroyed in my country and therefore I need to switch over to this instead. 00:35:02:01 - 00:35:31:14 Unknown I think it's a difficult debate. If you look at the start of Ukraine, you know, in day one, their largest data center was destroyed. You know, it has become critical. National data is a critical national infrastructure. On day two, they started using data centers outside Ukraine to maintain capability and decision making. So I think that the idea that sovereignty is really important, sovereignty does not mean nationally everything is stored in a bit of tin next door in the office, and I can go and kick it occasionally and switch it back on and back off. 00:35:31:16 - 00:35:50:08 Unknown It's a very difficult debate. Something that's interesting. You say that because you ago we had this conversation around deploying Oracle, deploying data centers physically in the nation. But if you look at what just happened in the Middle East, we did have a lot of clients saying we want to take our applications and our data and physically to put them outside. 00:35:50:08 - 00:36:13:09 Unknown So it turns out actually to have sovereign capability. Requires you to work with a global provider. And one of the things I had and if you have a view on this, do you think it is even possible when you look at the supply chains or the technology that is provided to be 100% sovereign in building a cloud data center or providing AI? 00:36:13:11 - 00:36:34:06 Unknown I believe that some nations in the world are achieving that goal in China, for instance, where they're going to build, not just that, their own operating system and internet, but also their own GPUs and CPU to do it. And they are struggling. Yeah. And if China is struggling with that capability, the many other nations will probably struggle as well. 00:36:34:08 - 00:36:56:20 Unknown Going back to that use case example, I think that all use cases where there are things which a nation would probably want to say, I want capability to be guaranteed to provide this for my national survival, infrastructure, credibility and AI stuff. You could probably run on your own compute provider, energy supply to do it, but it's going to cost. 00:36:56:22 - 00:37:19:22 Unknown And I think that's the big challenge we've seen. If we talk about compute algorithm, data energy, the one that's hidden and seldom talked about is cost. Nations like that ability to be able to go and kick a bit of tin next door and make sure all their things are still green and it's all very good. They don't like the bill that comes with that, which says that it's going to cost you X amount per year to maintain it. 00:37:19:22 - 00:37:41:03 Unknown And one of the big decisions we're seeing at the moment around energy is energy cost. As a driver for data location, whether that's European nations suddenly realizing that it's much cheaper to use another nation's energy than their own energy, and therefore to say sovereign is important. But cost is also a factor. And therefore I'm going to not necessarily rely on purely. 00:37:41:05 - 00:38:08:21 Unknown But I think there are some use cases where you can put very bespoke critical systems into national and national stacks. But when 100%. No, I don't think you're not today, you certainly wouldn't want to be there. And you touched on sort of the competitive nature potentially even at a national level. How are you seeing AI changing this nature of competition between nations or even between organizations? 00:38:08:23 - 00:38:33:14 Unknown You know, how long have you got? So it's, this is one of those things that I spend all my time talking about. I think that today we are seeing AI warfare. There's loads of books written on AI warfare. You know, that. So I advise you on myself. And I'm lazy and therefore stopped after a while. But you know, but the the the idea that we seeing AI warfare today, we're seeing this concept that Ukraine is the future of warfare. 00:38:33:16 - 00:39:04:00 Unknown I think it's flawed. Yeah. I think it is a nature of warfare. It's evolved in the same way. As, you know, the tank evolved from 1916 to 1939. Yeah. We see this evolution of military capability to do what we do today. We're a little bit faster, better, smarter. The true difference that's happening is when we see AI, a genetic warfare, which is starting to be built where we have AI as a capability to fight different wars in different ways, at different paces. 00:39:04:02 - 00:39:29:16 Unknown So we don't have to fight a kinetic, physical battle space war. We are fighting a war against a nation and defeating that nation before the war even starts. This, you know, this concept of the seven minute walk, idea. You know, I generally, I generally believe that that shift, many people don't realize is happening, but it places AI not as a tool that enhances current warfare. 00:39:29:18 - 00:39:54:14 Unknown But I as the tool of warfare. That's a second shift. A second shift is this long and I suddenly critical for my national survival. And you could possibly look at that in recent conflict, where the initial phase of conflict hasn't just been the destruction of airborne, systems and early warning systems and air defense, it's also been cyber attack using AI capabilities. 00:39:54:15 - 00:40:29:07 Unknown And yet the third phase, which I'm more worried about, is that if you're a nation where you are competing with a country where AI is the distinct advantage they have over you as a military nation, then you need to have a fight for AI dominance and you need a totally different level of warfare. So if a seven minute war is this this notion a country could have superiority through AI, the worrying phase is when other countries start realizing we have to defeat that AI capability so we can survive. 00:40:29:07 - 00:40:58:11 Unknown Or when my master's degree was in revolutions, military fans, this is a revolution that people are starting to see in the same way as aircraft carriers. Tanks, trains, machine guns, horses and chariots change the nature of battle. We are seeing a revolution which is totally different for the wars that we fight today. And my last comment going back is getting those old ideas out of generals heads is far harder than putting new ideas into them. 00:40:58:13 - 00:41:17:19 Unknown We are still preparing to fight Ukraine as a type of conflict, and seeing that as the future definition dash. Tony, I feel like I can make you work today, so I'm going to give you a little bit of a breather. We're going to do a quick fire round. So if that's the way we're going to time it. So I'm going to ask you 5 or 6 questions that we have 30s to do this. 00:41:18:01 - 00:41:46:08 Unknown Whatever comes into your minds just go ahead and say it okay okay. So let's start. One word for AI today. Nervous. Oh, one words AI in five years controlling AI agents hype or real shift. We all shift. Okay. Biggest risk in AI right now. Leadership, understanding and lack of it. The biggest opportunity for AI right now. Leadership, understanding and lack of it. 00:41:46:10 - 00:42:13:16 Unknown It's that one thing leaders consistently get wrong. I can't say the third answer at the same time. I've started selfish, but for time I'll say exactly that. I think that they fear that leadership understanding AI is all about them, not next generation. Excellent. Okay, and one AI term that you would wish people would stop using, this year again to get sick. 00:42:13:18 - 00:42:37:07 Unknown I think I'm just going to say that, and I think the other one is gen AI as a generic. It's an I think Gen I have excellent. Okay. So, Tony, one of the things that you've spoken quite eloquently on is leadership when it comes to AI. Are there any traits or characteristics that, as leaders in defense, we need when it comes to the use of AI? 00:42:37:09 - 00:43:02:07 Unknown Yeah, I'm a product of Sandhurst, and Sandhurst is very much around traditional leadership values. Apply to modern, challenging situations. But ultimately Sandhurst always teaching a General Slim's concept that leadership is just plain. You. And it's about understanding yourself, understanding your people understand you working with and then being able to lead in those environments. And your AI is changing the AI. 00:43:02:08 - 00:43:19:13 Unknown Suddenly it's it's not just plain you, it's just plain you plus AI. And I think that leaders, especially in initiatives, are struggling with that concept to say, I don't understand AI. I don't understand, you know, I know what I want to do. I just don't know what the toolset helps me do or how to do it or what to do with it. 00:43:19:15 - 00:43:43:22 Unknown And so we end up with these, you know, leaders, not just military, but, you know, leaders, a whole who are very conscious, very worried and think that the leadership about AI is all about themselves and their command in a situation in today. And yet we're seeing AI changing so rapidly, capabilities that are taking six months to evolve from concept to effective capability. 00:43:44:00 - 00:44:04:14 Unknown And so if leadership is concentrated purely on today and themselves and their issue, they're missing the really big challenge of AI. And that really big challenge is about the next generation of leaders. You you know, we are seeing leaders today struggle, but they got their position pretty much by doing know. When I was in the Army, it was one day son. 00:44:04:19 - 00:44:25:08 Unknown And it was sometimes the infantry. One day son or this will be yours. Just do what I did. That's changing. We know that change because the leaders are speaking to us and saying, I don't know how to lead in this new environment. What about the next generation? And so really, the role of, you know, leaders today, military or anywhere, isn't deal with the situation today. 00:44:25:10 - 00:44:49:07 Unknown It's shaped that journey for the next generation of leaders because their journey is going to be so totally different. They are going to be playing with AI from day one. They are going to be using AI in ways that we didn't consider or can conceive. And more importantly, is going to be embedded into what they do and will be using AI to do all the easy stuff, the things that you and I learned and failed at and then, okay, that when they do again, they're going to be automated. 00:44:49:07 - 00:45:06:17 Unknown And so our next generation leaders are not going to have these learning experiences. They're not going to have leaders who are necessarily helping them through that because it's all new to them as well. So the real challenge for AI leadership shape the journey for the next generation, right? Yeah. Do you think that this next generation has almost like an inherent advantage? 00:45:06:17 - 00:45:27:02 Unknown I mean, I the other day my son came to me and showed me on his iPad that he'd built an e-commerce website in Shopify, and he's now selling stuff. Right. And he just did that in a couple of hours. It just becomes it's inherent to him. It's intrinsic. Is there are you seeing that this new generation of leaders that are coming through in defense kind of come pre equipped with their skills? 00:45:27:02 - 00:45:47:12 Unknown Or is there there's still a lot I think we often use shorthand about natives, digital natives, digital nomads and people like that. He. Yeah. Oh they've got an advantage because they were brought up with us. You know, we coded as kids. We saw I played the computer as a kid. We understand how computers worked and people say, oh, well, yeah, it's easy for you because you grew up it. 00:45:47:12 - 00:46:08:15 Unknown I've had to learn all this from scratch. And you know that that concept that, you know, once you hit 40, any technology is magic as opposed to technology, you know, I think is nonsense. You know, I know, you know, 70 year olds who quite happily using code with AI to build exactly the same situation, the same thing. So I can do this. 00:46:08:15 - 00:46:27:10 Unknown And they I didn't know how to code. For now I can do this. And I also know 17 year olds who have got no interest in AI, computers and stuff like that whatsoever. And so to like say it's a generational thing is it is is easy, but I think flawed. I think it's actually positive thing. And it goes back to that mindset. 00:46:27:10 - 00:46:50:00 Unknown How do you create and coach curiosity regardless of your age? I was speaking to a, a pilot from the US Air Force, retired, he flew in Vietnam, and he was talking about how he was using AI to help him in his job. Now is to act as a special advisor. You know, there's somebody he would look at and go, U.S. Marine Corps pilot from Vietnam era. 00:46:50:03 - 00:47:09:02 Unknown Probably not the person you think he's gonna be using AI. And yet he was. So I think we've got to start thinking that in the code generation, we have an advantage. What can I do with that? Yeah, but it's not exclusive. We can all start using AI in those ways instead of thinking forward into the future. Now, I'm thinking 22 years and beyond. 00:47:09:04 - 00:47:31:17 Unknown It's quite hard to predict beyond two years when it comes to AI, right? What do you see as the future for AI? Where is it going to be used? Where is it going to have value? Two years? Cool. I think it's let me give you three answers now. One. Where were you two years ago? Yeah. Whenever I speak to people, I manage lead. 00:47:31:17 - 00:47:54:14 Unknown It's like. Yeah, if you were going to say in two years time, this is where your performance is going to be, where you two years ago, and it's out of curve, you know, am I seeing this hockey stick which suddenly goes like that? And what's the evidence it's going to be hockey stick. I think it's time that I let's look where we were six months ago with, for instance, using AI to to find, you know, zero day exploits, for cyber defense, six months ago. 00:47:54:15 - 00:48:19:01 Unknown Failed to do so today, finding significant amounts, you know, growth methods, all that side coming out and saying, hear us on abilities which have been detected very quickly at pace. You know, the, AI Security Institute study found that, you know, it is successfully penetrating, controlling enterprise network systems, trial versions, which literally had flaws in them just for people start getting too worried, you know, three times out of ten. 00:48:19:05 - 00:48:38:12 Unknown But doing it well should take an expert 20 hours doing it in 20 to 30 minutes. And so I think if you look at six months, that pace of change is so rapid. So, you know, much as I'm an evangelist and love predicting the future and telling you what 2035 is going to be like, I also caution people to say whatever prediction we make for two years time is going to be wrong. 00:48:38:18 - 00:48:56:07 Unknown So start thinking about how do you cope with change. The unexpected kind of goes back to mindset. How do we start thinking about that rather than trying to predict the toolset we're going to be using and therefore saying it's going to do this, how do I adapt and change to it? Adaptability in flexibility is far more certain than having fixed technology. 00:48:56:07 - 00:49:15:05 Unknown It's changing. So that's the first answer. So I forgive me, but I'm going to answer three times. The second time I think that we're often going to see AI doing unexpected things. Now we can argue that that's something that's going to create new jobs, new opportunities, new ways of thinking. But one thing we can be certain about is that we're going to be using AI in different ways. 00:49:15:07 - 00:49:34:02 Unknown And yet, as we discussed, leaders are worried about the situation today. So I'd put that same question where are we going to be in two years? To every leader you meet and everybody that the audience meets and say, what are we going to be like in two years time? What does that mean? What does that go? What does that mean for me as an individual? 00:49:34:04 - 00:49:53:06 Unknown Often talk about becoming more aware about what I can do. I then start asking people to consider a day to day activities. Do they see that being use of AI and then become more engaged? We need to have more of a conversation, not just us, but everybody about where are we taking AI and what do we want it to do? 00:49:53:09 - 00:50:14:04 Unknown And now allow that two year question to be answered much more comfortably, because the whole of society will be answering it, rather than people who are really interested and talk far too much about it. And in the third bit, which I'd answer is I think the AI in two years time is probably going to be on the verge of I don't believe in artificial general intelligence. 00:50:14:04 - 00:50:35:15 Unknown I think that that is possible. It's probably going to be invented, but I in two years time is probably going to be doing things which are significantly harming society if we don't have that debate I've just mentioned. Yeah. So being able to predict it is hard. What I will say is that if we don't start thinking, how do we control it? 00:50:35:17 - 00:51:08:03 Unknown How do we manage it? How do we make sure that people using it have my trust as a member of the public, that they're using it safely, responsibly, with assurance? If we don't have that debate, then it will be doing things that are harmful and people will just be waking up and looking at them. When I was working, in Microsoft, you know, a decade ago, our biggest concern was that we created the equivalent of a stock market crash in 2007, and suddenly it became, we are making decisions which make that possible, not inevitable, but possible. 00:51:08:05 - 00:51:24:06 Unknown And so a third bit is in two years time, I could do really bad things if we don't start thinking about it today and doing something about it. So the lesson here is be aware, be prepared, have the conversation now rather than have it forced on you in the future. Get engaged. Understand what it could do for you. 00:51:24:09 - 00:51:49:08 Unknown Understand how it could impact you. Talk about it. So this video series is of course called AI Changes Everything. With that in mind, is there anything that I doesn't change or shouldn't change? Obviously you spoke a little bit about the preparation two years, but is there anything that comes to mind immediately? I'd be deeply concerned if we have another social media problem. 00:51:49:10 - 00:52:02:23 Unknown I think that my children are old and in their 20s. But I grew up in a social media age, and I think today it's very popular and common for people to go, I didn't realize social media is doing such harm and, you know, so bad for people. And we should ban it for 16 year olds. 00:52:02:23 - 00:52:33:03 Unknown And yet it's pretty obvious. Yeah, we could see that. We could see that this was perhaps unhelpful. It wasn't encouraging people. And with hindsight, we should probably have done more about it. So one thing I hope it does change is we start thinking, how do I use AI safely? Yes. How do my children use AI safely? And I think second thing that it won't change is human ingenuity. 00:52:33:05 - 00:52:53:20 Unknown I think it'll enhance it. I think it will definitely, in part replace it. You'll start giving ideas, suggestions at a pace. So humans have so much better than having to get 20 people in a room and brainstorm something. But I still feel that many things will be resolved by human ingenuity, not by artificial intelligence. Yeah, I totally agree. 00:52:53:21 - 00:53:17:15 Unknown I generally hope that that is the case, right? There is something to be said about the creativity of humans, and we must always preserve that. I product of the week. Tony, I thoroughly enjoyed speaking with you, but before I let you go, one of the questions I always ask is, do you have a favorite prompts, like a prompt of the week that you encourage, listeners and viewers to use? 00:53:17:17 - 00:53:35:09 Unknown I do, it's quite long. It's about 20 lines long, but I do want to share it on on screen. But can you describe it in which case. Yeah. So I couldn't paste it every Monday. It goes across all my emails, all my calendar appointments. And it very simply says, what have I missed? I shouldn't have missed. 00:53:35:11 - 00:53:51:22 Unknown But to do that accurately, it needs a lot of detail and knowledge, but then it prioritize it into, you need to do this now. You need to put some time aside to do this one. And this one you need to be aware of. And you probably need to go and speak to somebody about it. So that prompt has been a essential for me in the last. 00:53:51:22 - 00:54:06:17 Unknown I travel a lot, and so be able to put that in on a Monday morning and just say, what have I missed in the last way? You can prioritize it. For me, such a good. So that's a brilliant prompt. I'm going to use that one for sure. Excellent. As I said. So I really enjoyed speaking with you. We could go on for hours. 00:54:06:19 - 00:56:05:00 Unknown Best of luck going forward. Thank you everyone for tuning in and watching. This is just one of the interviews in the AI Changes Everything video series. If you're interested in watching more of and I encourage you to go to our Qualcomm for Telus AI changes everything. Thanks once again, Tony, and thanks everyone for tuning.

5. 

   24 Jun 2026

   AI Changes Everything: Can You Insure AI Risk?

   Learn how organizations can quantify AI risk, manage model uncertainty, and build trust as generative AI adoption scales across the enterprise. In this episode of AI Changes Everything, see how AI insurance can help organizations balance innovation, governance, and financial risk. Explore more AI topics at http://www.oracle.com/aichangeseverything In this episode of AI Changes Everything, Dr. Michael von Gablenz, head of Insure AI at Munich Re, explores one of the most important challenges facing enterprise AI adoption: how to quantify, manage, and insure against AI risk. As organizations invest in generative AI and large language models, attention often focuses on productivity, automation, and return on investment. Yet every AI system carries uncertainty. Dr. von Gablenz explains why hallucinations, inaccuracies, changing model behavior, copyright exposure, operational disruption, and financial loss must be considered alongside the potential benefits of AI adoption. The conversation examines how Munich Re is helping organizations understand the unique risk profile of AI systems and why insurance can play a critical role in enabling trustworthy AI. Drawing on examples from real-world deployments, Dr. von Gablenz explains how AI insurance can help organizations address uncertainty, evaluate model performance, and build confidence in AI-powered business processes. He also discusses the importance of continuously monitoring foundation models, assessing changing risk exposure, and understanding the downstream impact of model updates. The episode explores the relationship between operational risk management and financial risk management, highlighting why leaders need to consider both when evaluating AI initiatives. Dr. von Gablenz shares how insurance can serve not only as financial protection but also as a signal of AI readiness, helping organizations better understand whether a system is fit for purpose and capable of delivering reliable outcomes at scale. The discussion also covers AI governance, model uncertainty, explainability, trustworthy AI, and the role that chief risk officers, chief information officers, chief financial officers, and AI leaders play in responsible adoption. For organizations looking to accelerate AI deployment while maintaining trust and accountability, this conversation offers valuable insight into one of the fastest-emerging categories in enterprise technology: AI risk management and insurance. ------------------------------------------------------ Episode Transcript: 00:00:00:00 - 00:00:15:04 Unknown The views expressed in this podcast are those of the individual speakers, and do not necessarily reflect the views or policies of their organizations or iCal or its affiliates. 00:00:15:06 - 00:00:29:21 Unknown As we all race to deploy intelligence, how do we quantify the uncertainty of a large language model? Do we know what risk we're being exposed to? And how do we protect against that? 00:00:29:23 - 00:00:53:07 Unknown Hi, everyone. We're here today with, Doctor Michael von Gablenz from Munich. Re. You're the head of Insure AI. I, we're going to be talking today about the. I mean, what you're doing is quite fascinating, actually. You are, as I understand it, looking at the uncertainty around AI models and then figuring out how to insure against that particular risk, is that right? 00:00:53:12 - 00:01:13:00 Unknown Yeah. That's right. Leland, thank you so much for having me today. This is a great pleasure to be here. Very good. Okay. So can we just start a little bit by talking about what is insure AI, please, if you could just describe it. So Intuit is a team within Munich Re. In Munich re is a big insurance and insurance organization. 00:01:13:02 - 00:01:43:16 Unknown And what my team does is that it structures insurance solutions for different forms of AI risks. In particular, the risk that an AI model or generative AI model hallucinates or produces an incorrect output. From our perspective, that's one of the most pronounced risks when it comes to adoption of an AI model, simply because an organization wants to adopt an AI system or change AI system in order to support and decision making, or to automate certain processes or tasks. 00:01:43:18 - 00:02:07:19 Unknown But if you have, the model produces more mistakes or hallucinates more often than expected, then also the processes, the tasks, the executions will be wrong more often, which can lead to financial losses or liabilities down the road. And we believe that the insuring for those risks is very important. Also to drive an adoption of trustworthy AI systems throughout organizations. 00:02:07:21 - 00:02:28:13 Unknown Amazing. I meet chief AI offices almost every day, right? And CIOs, I don't think this is a market that many people realize exists. So talk a little bit about the background on how insurer I came came to be. It is something fairly recent that how long have you been looking at this area? Yeah. So being sure the first AI system in 2018. 00:02:28:13 - 00:02:53:14 Unknown Wow. This was an AI system which was already taking autonomous decisions in autonomous actions. So in the case, that we, that we did, it was basically an AI system designed to detect fraudulent credit card transactions, and users were merchants, who were relying on this AI system to make decisions within milliseconds to say, yeah, either the credit card transaction is fine. 00:02:53:17 - 00:03:20:11 Unknown Now, accept it, or it seems to be a fraud and block it. And from a user perspective. So from a merchants perspective, this poses a risk, because if the AI system accepts more fraudulent credit card transactions than expected, then the merchant as the user has more fraud costs and especially unbudgeted, fraud costs. So this poses pose the challenge for adopting the AI system. 00:03:20:13 - 00:03:53:21 Unknown And the AI provider approached, yeah. Approached uniquely to structure. Yeah. A solution insurance solution to cover, those AI error risks on behalf of the users, the merchants. Very good. And I mean, are there many players in this market at the moment? I mean, it creates huge, but other startups that are coming into this, into the space or any of the other major reinsurers or insurance companies that you also see, we're not seeing major insurance or reinsurance companies, entering the space yet, but we are seeing, several insurer techs are coming into the space. 00:03:53:21 - 00:04:15:13 Unknown So some from the market, some are from. Yeah. From San Francisco Bay area. So it seems that. Yeah, also, the venture community is picking up, on. Yeah. On this topic. And I think that's not surprising because in our mind, insurance is a fundamental. Yeah, instrument has driven many, you know, adoption of, of new technologies. 00:04:15:15 - 00:04:41:17 Unknown So when we're looking back in history, even something like you have steam boilers, the adoption of steam boilers, so they are also insurance was really fundamental. Yeah. Basically. Yeah. Analyzing the safety of those steam boilers and then, providing insurance against major catastrophe in case an interesting boiler blows up. So in my mind, this also has driven then the Industrial revolution forward had a major impact. 00:04:41:19 - 00:05:05:15 Unknown Yeah. It's interesting that you use that as like an analogy or comparable. That's exactly what I was thinking. I mean, this market could be absolutely huge, right? We look at, you know, $2 trillion of investments in AI training over the last two years, the fastest industry in humankind. I'm not going to put you on the spot and ask you to size the addressable market, but where do you see this? 00:05:05:15 - 00:05:26:22 Unknown Munich. We see this going, you know, are there any particular segments that you're focusing on or any particular industries? Can you can you give a sense to where you see the market developing place? You know, from all perspective, every company, globally is looking into adopting AI. And for every I use case where in case the AI. 00:05:27:00 - 00:05:58:18 Unknown Yeah, produces some form of a mistake, and costs occur. Those might be opportunity costs and those might be also actual costs for, for company to for example, we do some work. Well those might also then be liabilities in case an AI model is used, in some kind of customer facing, application areas in all those, those companies, and for all those kind of use cases, there exists a risk exposure there in where the access to risk exposure, the insurance might provide a solution in order to. 00:05:58:19 - 00:06:22:13 Unknown Yeah, take on take off this risk and therefore also drive adoption of this technology. Right. So all these things I hear like IP infringement hallucinations obviously is one of the big one, accuracy of these models, every single one of those is potentially an opportunity for you to provide some level of certainty and assurance of insurance around this, right? 00:06:22:15 - 00:07:08:08 Unknown Yeah. Correct. Huge opportunity. And, I mean, the MIT, yeah, has created a repository around AI risks. I think they have they are now standing at around 1700 different AI risks, which might seem quite overwhelming, but from our perspective, we bucket those risks into two classes. So one are risks which are really driven by an AI model being just at the end of the day, a statistical system with any statistical model, this uncertainty, uncertainty in the output and many of, the risks that you mentioned, the, inaccuracy, the hallucination risk, but also, things like, the copyright infringement risk, they can be traced back to incorrect output by a model. 00:07:08:14 - 00:07:33:10 Unknown And, yeah, this is now, mind a new class of risk of AI, then also new, you know, insurance solution, this new uprising, this new, underwriting considerations with also new considerations around systemic or aggregation risk. Yeah, that that this is needed for this bucket. And the question which I think is going to be on everyone's mind that's definitely in mind is how do you go about quantifying this risk? 00:07:33:10 - 00:07:54:05 Unknown Right. Do you have a team of highly skilled PhDs who are sitting there figuring it out, or how is it done that that's the basis it's pretty much to build up, build up on the role of what we call a research scientist. So those are colleagues with a PhD in mathematics, statistics, computer science. We hired directly from at the from the PhD program. 00:07:54:05 - 00:08:19:15 Unknown So from the universities they were doing research around you have quantification of model uncertainty of related, related research. We hired a those colleagues to put them really on the business side, in insurance in my team in order to help us develop novel quantification methodologies to price, this kind of risk and also to see what is possible to insure and what is not possible to insure. 00:08:19:17 - 00:08:39:12 Unknown So I think this this talent that we need, I think this is really the basis. Are you finding that the domain expertise, the industry, matters a lot? Does it does it factor into your ability to quantify? It plays a role. It plays a role in the sense of, for example, figuring out whether, certain testing regimes are adequate. 00:08:39:14 - 00:09:04:16 Unknown So for the input data that, for example, the model is seeing Turing testing versus Turing production. Is this let's say common or is this pattern reasonable? So in this respect, your domain knowledge can be helpful. That's a unique recruit given that yeah, we entering many different risks from agricultural to yeah. Risks around, you know, certain liabilities and areas. 00:09:05:02 - 00:09:27:23 Unknown We have a lot of different experts in the group, which we can then rely on to bring in this kind of domain than my knowledge. But now perspective, more important is the statistical specific knowledge. So given any kind of black box model, how can we quantify. Yeah. The the uncertainty in, in the output of those models. That's really the key expertise that we have to build up on our business side. 00:09:28:01 - 00:09:48:20 Unknown Does this mean that your insurer AI team has a fairly close maybe working relationship or just observation of all of the frontier model providers? Because I imagine you have to stay fairly close to those. Is that the case? Do you work collaborative with them, or is it more one of you observing them on behalf of your clients? So be observing them. 00:09:48:20 - 00:10:17:18 Unknown And have you also observing how, yeah, our incident model providers, how they are utilizing for example, those foundational models, if they're utilizing them. So many of our clients are also building their own models. But some are also relying then on a foundational model. And for our, for us, it's very important to see how they are also monitoring changes to the foundational model, because we know if the foundation model changes, this will have some downstream impacts. 00:10:17:20 - 00:10:41:08 Unknown So then also the overall models for the fine tuned model, will then also behave differently if the foundation model changes, which can then give rise to additional higher risks or change in risk. From our perspective, this needs to be monitored and controlled. Yeah, this is a really important point because we often work with clients where we've pre integrated the model and we go through their compliance. 00:10:41:15 - 00:10:57:17 Unknown I mean it's almost like one and done. But of course the models are not static. They're constantly changing. There's new versions coming out. They're leapfrogging each other. And I think you raise an interesting point that someone has to worry and quantify. The risk is that every time and these models sometimes are changing on a quarterly basis. Yeah, yeah. 00:10:57:19 - 00:11:29:05 Unknown Do you have clients where you have this ongoing relationship where you're almost having to adjust the uncertainty and the risk depending on the models and how they develop or even the tooling in the applications they're building around the models that is that the case for you? Yes, we require an ongoing monitoring as well as a retraining updating. So from our perspective, in case a new model is trained and should, go into a production, if it performs better than the old model, then of course it's highly preferable that this new model is put in draw into production. 00:11:29:09 - 00:12:00:03 Unknown So then we automatically insure it. However, we always require that there's a testing and that there's a proof that, yeah, the new model is truly performing better for the specific application compared to the, to the old model. And only then it is insured. Let's talk a little bit about the types of risk. Okay. I probably spend most of my time working with clients and partners who think about operational risk, but I'm starting to see to continue. 00:12:00:03 - 00:12:24:23 Unknown Michael, that's just that's not enough. Right. So where do we go beyond operational risk? Yeah. So there's also the different risk component there. And this might be a consequence of a realization of an operational risk. So for example if we have a company which is adopting a generative AI model internally for, for example, extracting information from documents that customers send in this. 00:12:25:05 - 00:12:44:22 Unknown Yeah. The model is now making more mistakes. And also this means that the output cannot be used. And someone so human then needs to redo the work. So that's an operational impact. But at the same time it's in financial impact because this means simply that you're more humans are needed to do the work. More. Yeah. 00:12:45:03 - 00:13:14:04 Unknown Costs occurred because of that. And that's essentially unplanned. And this creates a financial risk for a company. Interesting. So this sort of duality of financial risk and operational risk, it's not evil. It's both again, I think it's something quite innovative that you and your team have developed. What's the reaction you get from clients and tech vendors and partners when you explain these two dimensions and how they now have to sort of go hand in hand? 00:13:14:04 - 00:13:39:01 Unknown Yeah. How do they react? Yeah. So I think it's also many times an educational moment. So as we have said, it's still quite new to see the financial dimension of adopting AI. So many fewer think AI in the future of AI. So similar to adoption of other forms of technology. Yes, but in my mind it's quite distinct. So at the end of the day, will the adoption of a statistical system with all the consequences what this means? 00:13:39:03 - 00:14:05:22 Unknown So we know that, a strong, powerful, statistical system can really improve. Decision making can also lead to companies making smarter decisions. But at the same time, statistical systems are always prone to uncertainty. They're probabilistic systems. So even the best, model, the best AI model with access to all the information universe, that's a mathematical fact that even such a model will make mistakes now and then. 00:14:06:00 - 00:14:26:09 Unknown So from our perspective, it's a matter of assessing the probabilities, then seeing, how can we, yeah. Then deal with, yeah. With this uncertainty is the probability of making an error except for the respective use case. Then do we want to transfer, this residual risk, for example, why insurance is or is it not acceptable? 00:14:26:09 - 00:14:46:20 Unknown And do we need to spend more money, more funds, more effort to somehow reduce the risk by operational means? So I think this duality between operational and risk management, as we call it in financial risk management, I think that's a novel idea for AI and AI governance, but I think it's a very important one. Yeah, truly drive AI adoption forward. 00:14:46:22 - 00:15:07:21 Unknown And yeah, you mean a career. You're providing a service to insurance companies, but also directly to end clients and and to technology companies as well as AI. Right. So technology companies are most of our clients. Right. And yet this year we have started to also provide our AI insurance solutions to other insurance companies. Next, in the role of a reinsurer. 00:15:07:21 - 00:15:36:08 Unknown Yes. Means that insuring those insurance companies for yeah the losses, the risks which arise from our AI insurance products, that's also a preferred way because we would like to bring in more insurance into the space. I risk in the AI adoption is growing, and this means we need more capital providers, more insurance companies coming into the space becoming, you know, comfortable with taking, on, er, risks and insuring those risks, I think. 00:15:36:09 - 00:16:04:02 Unknown Yeah. Many careers, a reinsurer can play a major role in that. Yeah, absolutely. And are you seeing you know, typically if you look at what we've just been talking about, it falls under, the responsibility of the CFO, the CFO. And if a CIO or CIO chief AI officer, are you now seeing these 3 or 4 roles coming together in the same room and having these types of discussions? 00:16:04:03 - 00:16:23:01 Unknown Now, I would add another role, the role of the chief risk management officers. Yes. Then who also has an important role. So traditionally from a, you know, corporate insurance buying behavior, this is done by the risk manager. And you know, when it comes then to I also in my mind the risk manager has an important role at the table. 00:16:23:03 - 00:16:42:20 Unknown So then to say how can we complement. Yeah, the stack governance, process that the company has put in place. But also thinking about the financial wisdom. Yeah. Dimension. How can we utilize insurance to remove uncertainty. And then also this, you know, let's say a trust issue which might come up with the uncertainty of an AI model. 00:16:43:02 - 00:17:05:12 Unknown How can we remove that and therefore make our business units, our service units within the organization, more willing to, experiment with AI or rely on AI and, and drive adoption at scale? So I think removing this trust by, having both operations risk management and financial risk management as a safety net, I think that's quite important. Yeah, absolutely. 00:17:05:14 - 00:17:27:10 Unknown And on one hand, I think, you know, you're really not just at the cutting edge. You're you're driving this market, right? You're creating an entirely new category. But, you know, when you start seeing VCs entering this market, model providers coming to you as clients, it's also a sign that this market is starting to really pick up and reach a level of maturity. 00:17:28:14 - 00:17:42:04 Unknown What does the next two years look like for insurer I, I'm not going to ask beyond two years, because it's almost impossible to predict beyond two years. But, what does it look like for you? Are there particular segments you're going to go after? You know, give me a view as to what that looks like, please. 00:17:42:06 - 00:18:03:21 Unknown Yeah, sure. So we believe that ACI adoption is growing so fast, and also the exposure towards AI risks. And so our risks are then realizing, I think the need for insurance is also growing. So we believe that there's a strong growth ahead of us from an AI insurance perspective. And this will, go over many different industries. 00:18:03:23 - 00:18:22:14 Unknown However, we believe that especially the financial industry, might be one of be. Yeah, where we could see a big, adoption of insurance because the financial institutions are used to dealing with model risk. And then the, you know, the uncertainty which comes with an AI is quite close to the model risks there. So we believe that. 00:18:22:14 - 00:18:42:14 Unknown Yeah. Here insurance and the understanding of, of the different risks that this can be seen quite, quite a bit. You know, quite, quite strongly and therefore that this will also drive a then the, you know, the adoption of insurance as a complement to the governance and AI governance process. Yeah, absolutely. I can see we're a technology vendor. 00:18:42:14 - 00:19:08:16 Unknown Obviously, I could see technology vendors maybe even offering this as a service, potentially with their own technology stack in the near future. With that in mind, is is there any advice you would give to, tech vendors, and clients or other insurance companies? What should they be doing now to prepare for this? You know, apart from picking up the phone and calling you and your team, but how can they compare? 00:19:08:16 - 00:19:33:14 Unknown What should they be doing? Yeah. So I believe having honest conversations about the risks, not just the opportunities. So I believe that with AI there come tremendous opportunities. But this at the same time there's no free lunch. There also comes comes risk. But the nature of an AI system being a statistical model. So I think having honest conversations about what those risks are, I think that's a starting point. 00:19:33:16 - 00:19:57:13 Unknown Then, I think working with insurance companies like Munich Re and my team to see can we now quantify this risk, I think insurance in this respect is quite honest, because the premium is a very nice quantitative reflection of what the true level of risk is. And if, potentially also the premium of one I insurance is too high, this might also send a very strong signal that perhaps. 00:19:57:13 - 00:20:20:10 Unknown Yeah, the AI as it's currently designed, might not be fitted well for the respective use case. And the purpose for that. So I think he also insurance can act as a signal of how risky a certain deployment of AI is, and better than the total costs for the AI. So being basically the usage costs plus the costs of risk, for example, the form of the premium for AI insurance. 00:20:20:10 - 00:20:41:08 Unknown But it's, you know, worth it considering the opportunity. I think having this more honest discussions, I think this will help. Yeah, it's interesting because, you know, you've probably seen the other MIT report where it talks about the fact that more than 80% of AI pilot students of AI pilots fail. They never see the light of production. And my team is obsessed about the quantification of AI impact. 00:20:41:08 - 00:21:03:10 Unknown So it was almost we're focus on what is the value that AI generates. But we never really looked at the risk parts. I seen that we should be. What I think I find quite interesting is, actually what you and your team are providing. I mean, a lot of clients who are very hesitant about taking that leap of faith to invest in AI because it's uncertainty. 00:21:03:12 - 00:21:34:05 Unknown But by bringing you and your team and your service in earlier, they're lowering that risk profile, making it easier to make that decision to invest in AI or not. The case may be, yeah, absolutely fascinating. And just on that, are you where are you engaging in this sort of product development or adoption lifecycle? Are you finding that your clients are bringing you in quite early in the adoption lifecycle, or is it an afterthought or we've launched it now we better communicate and find out what our risk is. 00:21:34:11 - 00:21:58:09 Unknown Where are you typically getting engaged? And so I would say it's a combination of many of the AI vendors. We see that we are getting approached quite early. So there's an AI system already developed, but now they would like to sell, this AI system to many companies and really scale up. And then having insurance, which backs a contractual guarantee that they are giving, towards the AI users around the. 00:21:58:09 - 00:22:22:23 Unknown Yeah, accuracy, rates around the effectiveness of of the AI solution. And this can then be a really powerful. Yeah, a powerful tool. When it comes to the AI uses. So, there we also see that, basically once the AI model is supposed to be more adopted at scale, that then also discussions around risks are truly becoming, yeah. 00:22:23:14 - 00:22:45:17 Unknown A preeminent focus. And there are also, discussions around ensuring this kind of risks. Then also something what we are seeing then where we are getting engaged. Excellent. Okay. Michael, do you mind if I ask you a little bit about your background? Because I'm sure a lot of our, our viewers are sitting here thinking, how does someone like you end up in a role and running a business like this? 00:22:45:17 - 00:23:09:10 Unknown So what is your background, if you don't mind me asking? The journey? Yeah. So I have a PhD in finance and a master degree in, in data science. So it's a combination of, essentially understanding the practical as well as the theoretical, side of AI models. So I've built transformer models myself. Yeah. Which is not always easy. 00:23:09:12 - 00:23:36:16 Unknown Yeah. And at the time, it was still, it's like coding everything in Python and looking up stochastic overflow. And I was stuck somewhere rather than using a byte coding up in the AI system. It's nowadays and then, yeah, understanding the finance world, and basically seeing that when we evaluate in finance any investment project, we always look at the opportunities, but then also the costs of risk to really find out for the net present value of the project is really worth it. 00:23:36:22 - 00:23:58:19 Unknown And I think for AI, I've seen this, you know, the considerations around usage costs, and then the opportunities which come with it, but I haven't really seen yet discussion around the risks and, yeah, from, from my perspective, really combining then the AI knowledge with the finance knowledge. That's yeah, really excites me. And this is what brought me into this role. 00:23:58:19 - 00:24:18:00 Unknown Well, it's almost as well you, tailor made for this role of this business. Okay. It's not often you see someone with a PhD in finance and data science, but I can see it's absolutely work for you. Okay. Definition. Dash, if it's okay, we're going to do a little quick look around. Just a bit of fun. Okay? I'm gonna ask you 5 or 6 questions. 00:24:18:13 - 00:24:41:02 Unknown So the idea is one sentence or one word. Okay. And I think we're going to time this to like, 30s. Okay. Just a bit of fun. So let's go, shall we? Okay. One works I today uncertainty one one word for I in five years statistics I love it. Agents hype or real shift. Real shift I get biggest. 00:24:41:04 - 00:25:10:09 Unknown This is perfect for you biggest risk in AI right now inaccuracies. So my biggest opportunity automation one thing leaders consistently get wrong. Just focusing on the operational risk management dimension rather than the underlying uncertainty. Okay, I've started, so I'll finish last one. One I term you wish people would stop using. 00:25:10:11 - 00:25:35:01 Unknown Atlantic AI systems. That has been the top response from everyone we've interviewed. Very happy about that. So you're oh, you're on the money power okay. Myth versus reality. Okay, Michael. So just sort of building on this fun thing of quickfire. Now let's talk about hype versus reality. Can I give you a couple of statements and then just say hype or reality? 00:25:35:05 - 00:26:07:06 Unknown You got to choose one okay? Okay. So first up I will replace most jobs hype hype okay. Bigger models are always better. Hype AI is already delivering ROI at scale. And so it depends. I think that one I let you get away with. Yeah, yeah, I think so. Every company needs an AI strategy. Reality reality yeah I agents we will place applications. 00:26:07:08 - 00:26:33:20 Unknown Of this depends. Yeah I would say I would tend to be an application. Okay. I think it's very important to think about applications. Yeah. Well, then just let's say the tool itself, because at the end, the value is in the application, not how fancy the tool. Tools. Absolutely. Well said. Okay. You passed. I look for the name of this video series is I Changes everything. 00:26:33:22 - 00:27:03:02 Unknown But is there something that comes to mind that I shouldn't change or couldn't change? So I believe when it comes to true social interactions, that yeah, we are human beings. We also crave social interaction. So I think, yeah, replacing those social interactions, I think that's really difficult. And I believe that we need to have a genuine human being there, not a machine which mimics, human behavior. 00:27:03:03 - 00:27:26:02 Unknown So I think that's that's something which is a key element for me. I completely agree. As you know, as technologists, it's always going to be still a human element. And there are certain things just because we can replace them, it shouldn't we shouldn't. Correct. Yeah. I prot of the week. So Michael, one of the questions I always ask is do you have a favorite prompts or is there a technique that you use when prompting that you'd like to share with the audience, please. 00:27:26:04 - 00:27:44:05 Unknown So I find it quite effective to define the role that the model is, or should, take first, because then I feel that you have the models giving me more precise responses to what I'm asking. Okay. That's interesting. Can you give me an example of the type of role that you would prompt the model to take on? 00:27:44:07 - 00:28:13:04 Unknown Yeah. So for example, if I'm asking about let's say any statistical question, then I would ask it, you know, assume now the role of yeah, a PhD, statistics, student at, say at University of Oxford and now explain this. So you you are having this, this rodent. Now explain this to me in simple terms how you would explain it in your your class if you would be a teaching assistant. 00:28:13:12 - 00:28:31:19 Unknown So that's then let's say, the right level of expertise, which is there. But then from a communication perspective, it's yeah, it's a to see everything. And I find that you are defining those roles and the way of, of communication that this then helps, to provide better, response out of the model to my, to my career. 00:28:32:08 - 00:28:54:13 Unknown That's a great tip, Michael. We've come to the end of our chat. I thoroughly enjoyed it. I found it fascinating. I hope you have two. It was great pleasure. Thank you so much. Thank you so much. And thank you for watching. I hope you've enjoyed it as much as we have. If you're interested in watching more of these podcasts, and I encourage you to go to our Khou.com forward slash, I changes everything. 00:28:54:15 - 00:30:32:04 Unknown You.

Contact AI Changes Everything

Guest appearances

Books guests

Based on episode analysis; this does not confirm that the show is currently accepting guests.