Latest / Technology Now / Who owns your AI? Understanding data control, compliance & sovereign AI | Trish Damkroger
Transcript
- MICHAEL BIRD Good morning, Sam. SAM JARRELL Good morning, Michael MICHAEL BIRD Hey, uh, you know how I have recently been giving you some really easy questions? SAM JARRELL Oh yeah, the just, the most simple, not, very large scale could write an essay on questions. MICHAEL BIRD Like, what's the meaning of life? Um, well, here we go. I have another one for you. if you wanted to make a sovereign AI, what would you say creating a sovereign AI would entail? Easy question. Easy SAM JARRELL Very, yeah, simple question. Simple question. Not one that governments around the world are asking themselves. I would say, just rattling off the top of my head, biggest thing would be, jurisdictional control, keeping sensitive data and stuff, like within, your own region's laws. probably also, a setup that's disconnected from like the public internet so it's not so easy to get in there, and then like infrastructure that is located within the country that it's operating. MICHAEL BIRD Amazing. Yeah. As, as with last week, thank you so much for listening to Technology Now. What a great answer. Shall we just go home? No. well, today, Sam, we are gonna be exploring the five imperatives of sovereign AI. So five things. Very helpful. Uh, so we can check again if you're right. No pressure. I'm Michael Bird, SAM JARRELL I'm Sam Jarrell MICHAEL BIRD And welcome to Technology Now from HPE. SAM JARRELL So sovereignty is the word of the day or of the year even. It's absolutely everywhere at the moment MICHAEL BIRD Yeah. governments are all over it. Businesses are investing in it. Some people are even demanding it for their own products SAM JARRELL Right, but do they actually know what they're trying to get ahold of and what actually needs to be done to make things like AI truly sovereign? MICHAEL BIRD Yeah, and we've discussed it on the show before. This sort of sovereign word can mean very different things to many different people. So Sam, I think that's an excellent question. here at HPE, we've been talking about the five imperatives of sovereign AI to help add a bit of clarity to the conversation so to find out more about these, I spoke to Trish Damkroger Senior Vice President and General Manager of HPC & AI Infrastructure Solutions at HPE and before we got into what sovereignty is, I wanted to know why someone would choose a sovereign AI in the first place. TRISH DAMKROGER Well, there's a couple reasons that it's becoming, much more of interest across the world. one is the, big push for either all the nations. It's like the new space race, you know, the AI race, right? we don't wanna miss out on what's next, and so there's a lot of interest in the nations, but also in companies to say, "We've got to get AI. We've got to get the return on investment, but we got to worry about our data going out there." I mean, we don't want our individual IP or our sensitive data outside the walls. The second thing, it would be about your regulatory environment. So the EU has put an EU AI Act where basically by August of this year, every European Union country has to prove, show where their data is and the governance of it. So that's a tall order, right? So that's again why you want to control your infrastructure and also your data going in and out. So those are driving it. And then the third area is about just knowing, about your whole data framework and your data management. I think data is just the key here of knowing what is really important for your company or your country. And so this can be whether it's your native language or your scientific discovery IP, so data governance MICHAEL BIRD Because there's an element of, the large language models have all basically been trained on English. maybe most of the internet would be US English but I don't know, if you're France or, Slovakia or, where actually that's not your first native language. So, there's an element of actually you want models trained in your native language TRISH DAMKROGER 100%. And even, some of the interesting stories are indigenous tribes where they are trying to keep their culture moving forward and it's harder and harder when the elders are unfortunately dying off, right? And so they are training the AI models to actually keep that culture, their values, and their language alive MICHAEL BIRD Yeah. So there are, from my understanding, there are five imperatives of sovereign AI. can you just talk me through what they are? And why are there five and not six? TRISH DAMKROGER That is a great question 'cause, you only have five fingers. You got to be able to count them off. I... MICHAEL BIRD That's a good answer for me. That's a good answer for me. TRISH DAMKROGER Okay, so architectural control. This is, where I talked about having the infrastructure and being very secure in that infrastructure, and it's mostly you need that full solution stack. So we talk about AI factory. It's not just the hardware, but it's also the software, the full solution, storage, et cetera. Number two? two, data governance. And data governance is hard. It's really hard. I think it's one of the most frequently asked questions of customers of, "How do I do this?" Right? You get the infrastructure, but how do you keep your data secure, safe, et cetera? So that's number two. Then it's, number three, scalable performance. AI is moving at such an incredible pace where you may buy this solution and it's the latest and greatest thing, and six months goes by and now it's not. So how do you make sure that you can continue to scale that infrastructure so that you're keeping up with everybody else? Four, operational resiliency. you can only, monetize your infrastructure when it's up and running, so it's very important that you put in the right monitoring so that you can keep it up and running. Again, it's money, right? And then sovereignty of intelligence. So this is where, we talked about the models for the different languages or your IP. The imperative is about making sure that you protect that sensitive data, but then also you create the models that have the most value for your company or your nation. MICHAEL BIRD So, how would an organization use those five imperatives? TRISH DAMKROGER So those are the five things that we'll walk through with a new customer of, "Have you thought about this?" And why it's important. There's a lot of hype about the latest GPU or CPU to do it, but not as much about how you do it or why you're doing it. And sovereign is becoming so important because, it's the the control, compliance, and security. AI is amazing, right? Because it allows you to, do new things. It also finds all of your vulnerabilities. And so AI has been able to help hackers get in, and a reason why you would want sovereign MICHAEL BIRD Yeah. That's fascinating. from a sovereign AI perspective, I guess there's an element of talking about keeping your infrastructure closer to you, rather than sending it across an ocean or wherever. is there an element of that also helps to reduce latency, which has a, an impact on AI infrastructure? TRISH DAMKROGER a lot of the regulations are requiring that your data stays close to you or in your nation or, our companies really want to keep it close. So yes, but there is latency in the sense that, a lot of this infrastructure is... out on the edge or across multiple countries or places, it does create some latency. But I mean, HPE is known for running the largest sovereign entities for the last 50 years in our supercomputing space. So we know how to set up these big systems, and we're pretty good at performance and latency, making it, as fast as possible MICHAEL BIRD you briefly talked about the EU AI Act, but can we just talk about regulation? Like, where does regulation fit into this? Because, to some extent there may be organizations who wanna do things in a particular way, but as you mentioned with the EU AI Act, there is regulation there that is sometimes defining how organizations are able to do things. TRISH DAMKROGER Right. So over 60 nations have come up with AI strategy, and part of that now is the regulation side of how do you implement that AI strategy? I think EU is the one that's out in front as far as putting that regulatory framework around it. but even within the United States, the National Institute for Standards and Technology, NIST, has come up with a framework that a lot of our customers are adopting, and it doesn't come with fines yet, where the EU has fines for not, showing that you have control of your data. But it's a good framework and, a way to make sure that you are taking care of your data appropriately MICHAEL BIRD 'Cause, I mean, GDPR was a big thing back in 2018, and, this feels like maybe... It's not necessarily related, but it sort of feels like the next step, like the next build upon it. Because, yeah, GDPR was all about data, particularly customer data, wasn't it? TRISH DAMKROGER Data is still gold MICHAEL BIRD Data is still gold. Data's still gold.Okay. Um, can we talk scaling? of course. um, so modern AI infrastructure is completely different to everything that has come before. So, how should organizations be scaling this up? TRISH DAMKROGER before people would buy big systems and then they'd run them for five years, right? And then they'd buy another big system. That's the traditional supercomputing model. Where, the technology for AI is moving so fast, you need to be able to upgrade as you go, and that's a different concept for a lot of people. and that's why scaling is so important, 'cause if you're not using that mix of technology, you're going to be left behind MICHAEL BIRD All right. So, so final- final question. So from a practical perspective, what's the first thing an organization should think about and do when they're considering sovereign AI? TRISH DAMKROGER So I would tell you the thing that I will tell most organizations I work with is an overall AI strategy, and if you look at, a lot of the customers that are successful, they come in with this overall top-down AI strategy because it is expensive, and if you don't have real use cases, you know, ways that you're gonna take advantage, I see people, fumble for a while. So my number one thing is have that AI strategy and then go through those five imperatives of what does it mean, what is the architecture you're gonna be driving? What about your data? What about, the models you're gonna be doing? Those are the things that are really gonna drive how successful you're gonna be. MICHAEL BIRD Trish, thank you so much for joining us on Technology Now. It's been a pleasure having you. SAM JARRELL Okay, so I just really quickly wanna, rerun through what the five imperatives actually were, 'cause there was a lot that was discussed in between. So first, architectural control. Second, data governance. Third, scalable performance. Fourth, operational resiliency. And fifth, sovereign intelligence. There's so much AI hype right now that, hearing more about these makes me realize that, you can't separate the strategy around it from the marketing. And it's more than just, are we doing these things? But it's like, if you're not doing these things, are you even gonna be able to properly scale and achieve all of the hype that is being, tossed around? MICHAEL BIRD one of the interesting things that I took away from that is that we're sort of moving to this phase of, we actually have to deploy this stuff. And, it's not just on a test bench anymore. We're using our actual, live company data, our, important data that, has company secrets or whatever. so therefore we have to be thinking about, how are we gonna scale it? How, is this, infrastructure gonna be resilient? But particularly, the thing that we just kept coming back to in the interview was just about data. how do you keep your own company data safe and secure? How do you make sure that it's, not going into somebody else's hands? SAM JARRELL I agree, and I feel like we're in,a time period now where, the AI conversation has been going on for a few years now, and I'm seeing more the conversation is switching to, what are we getting out of this? Where is the return on investment? What is the actual, value? it's no longer just "Let's stick it anywhere and see what happens." And increasingly, I think, as part of, that value conversation, it is the security piece of all of this. Because, if the data's not secure, then the AI isn't secure. And also, we're increasingly seeing that with frontier AI, it's got the capability of breaking out of its own sort of architectural controls. And so it becomes a liability if your AI is not sovereign from the perspective of, even being able to be contained within its own walls. that also becomes a problem, 'cause you don't wanna be inadvertently attacking other organizations, governments. MICHAEL BIRD Yeah, and as Trish said, there's this element of y-y-you don't wanna be left behind. if you could be the first to be using an AI model to be doing something really innovative to, do things more efficiently, save resources, you wanna do that before your competitors can do it. But, as we've discussed fairly recently on this show, um, yeah, AI, AI can and is being used for various purposes, both for and to combat against, cyber attacks. So it feels like we're in this pressure cooker a little bit, where everyone's wanting to get the most out of their AI investments, but they're also keeping one eye on the security element of it. What's happening with my data? How am I gonna get value out of this? How is this gonna scale? How can I keep this infrastructure up and running? I wouldn't wanna be in the shoes of somebody who's having to be running this sort of infrastructure at the moment. SAM JARRELL it seems like there's a lot of pressure from multiple sides. I did enjoy hearing, a little positive, AI example during this discussion with Trish, where she was talking about, indigenous languages and cultures and the preservation of that. And I had never considered that you could use AI to basically make sure that something like that is preserved long beyond, after your elders and things pass on and, it's harder to do maybe, specific practice bookkeeping and things like that. It- that is such a cool example to me MICHAEL BIRD Yeah, and that leads really nicely onto the last thing I asked Trish, 'cause I wanted to end on a positive note, and what I wanted to find out from Trish was what excites her most about the idea of a sovereign AI-filled future TRISH DAMKROGER I think the thing that excites me the most about the future of AI, so of any of this stuff, I'm a nerd, I'm an engineer, and I geek out about this technology and bringing these huge computers down, but it's what ha- people do with it, right? So the use cases that they're really driving, whether it's healthcare or the indigenous population and saving their culture or new materials. Mm-hmm. You know, that's where I get really geeked out- Yeah ... about what we can do with this new technology. SAM JARRELL Okay that brings us to the end of Technology Now for this week. Thank you to our guest, Trish Damkroger And of course, to our listeners. Thank you so much for joining us. MICHAEL BIRD If you’ve enjoyed this episode, please do let us know – rate and review us wherever you listen to episodes and if you want to get in contact with us, send us an email to technology now AT hpe.com – subject line? SAM JARRELL Sovereign? Thought it was direct…. MICHAEL BIRD Short and sweet and don’t forget to subscribe so you can listen first every week. Technology Now is hosted by Sam Jarrell and myself, Michael Bird This episode was produced by Harry Lampert and Eva Higginbotham with production support from Alysha Kempson-Taylor, Nik Damarell Beckie Bird, Alissa Mitry, and Jenessa Ayache. Our theme music was composed by Greg Hooper. SAM JARRELL Our social editorial team is Rebecca Wissinger, Judy-Anne Goldman and Jacqueline Green and our social media designers are Alejandra Garcia, and Ambar Maldonado. MICHAEL BIRD Technology Now is a Fresh Air Production for Hewlett Packard Enterprise. (and) we’ll see you next week. Cheers! SAM JARRELL Bye y’all