Latest / Technology Now / What is sovereignty in AI?

Transcript

What is sovereignty in AI? · Technology Now

• MICHAEL BIRD Hello and welcome back to Technology Now; a weekly show from Hewlett Packard Enterprise where we take what’s happening in the world and explore how it’s changing the way organisations are using technology. We’re your hosts Michael Bird... AUBREY LOVELL ...and Aubrey Lovell. This week we are diving deeper into the subject of sovereignty in tech. • We’ll be asking what sovereignty actually means and what this looks like in practice... • We’ll be exploring how artificial intelligence and data sovereignty often go hand in hand... • And we’ll be asking one of the most important questions of all: why should you, our listeners, care? MICHAEL BIRD Yes we will! So, if you’re the kind of person who needs to know *why* what’s going on in the world, matters to your organisation, this podcast is for you. (oh) And if you haven’t yet, subscribe to your podcast app of choice so you don’t miss out. Right, let’s get into it! AUBREY LOVELL We know security and trust are vital when it comes to using technology. No one likes to give up control if they don't have to. However, when it comes to online storage for our work, our data, or even our intellectual property, we need to have trust that it is safe and cannot be stolen. This is an important topic that governments regulate how personal data can be used, but this also is known to have caused a few snags when it comes to cooperation between countries. The EU has a number of laws to protect people's data. The EU has a number of laws to protect people’s data, but because they don’t gel nicely with the data laws in the USA something entirely new had to be drawn up: the Data Privacy Framework . This agreement between the US and EU would mean that organisations wishing to send data across the Atlantic would all have to adhere to the same set of rules – on top of any local rules already being followed! This data, being subject to the laws and regulations of the country where it’s generated is known as Data Sovereignty. MICHAEL BIRD But what if we expand the concept? If data can be sovereign, why not the models that use it? AI models must, like anything created in a country, adhere to the laws of the land. Controlling not just the data, but the infrastructure which uses it can be critical to national security. This means that not just the data is sovereign, but the AI must be too. And this doesn't just benefit the country from a technological perspective, multiple countries around the world have sovereign AI strategies which will require a local ecosystem of workers which should also create local jobs in the industry. AUBREY LOVELL But what exactly makes a sovereign AI different to a regular AI model in practice? Well, when Michael visited Hanover Messe, he didn’t just spend his time in a bath in the middle of a conference centre... MICHAEL BIRD I just want to clarify that was recording a podcast episode. ANDREAS GEISS Indeed ...he also spoke to Andreas Geiss, the Vice President of Automotive and Manufacturing at Aleph Alpha to find out more about Sovereign AI and why it should matter to each and every one of us. Let’s take a listen to their conversation. MICHAEL BIRD can we start just by defining what we mean by “data sovereignty”. ANDREAS GEISS Looking specifically at data sovereignty, the definition is that you can have your data at any place with any, let me say, possibility you would like to have, be it on-prem, be it on cloud, and you have always the full-time control and ownership. This means data sovereignty. MICHAEL BIRD So what is data sovereignty in the context of the AI revolution? ANDREAS GEISS Yeah, that's a very good question. It's the goal and as geopolitical landscape has changed a lot, this is enforcing that also companies make up their mind in the era of gen AI what sovereignty means. You own the data, you keep the ownership, and you can always decide where to put data or get data out without any login. It's about the operational sovereignty that you decide also where I would like to run, maybe my GenAI application. And this is getting decisive because you have a lot of domain expertise, which is getting reflected out of the data. The value sovereignty that you also know what the value or the outcome that you can decide what the business case and all the value what you get out of maybe an application is looking like and last but not least, it's about government sovereignty. Governance sovereignty means that you can describe to which standards, internal and external, you would like to adhere that GenAI is taking part in your business enterprise on your journey. And this is, I believe, a core difference between GenAI and sovereignty in this context and maybe other type of general sovereignty. MICHAEL BIRD Cause the concept of data sovereignty has been around for a while. I remember when I was in the IT industry and the cloud landed and sovereignty was a big conversation. So how is the conversation changing now we're in the age of the... in the age of AI? ANDREAS GEISS In the age of AI, I think we are talking of massive data which has been looked at, which you couldn't look at in the past. This is more related maybe to also unstructured data. When you mentioned we always have been talking about sovereignty, absolutely true, but it was sovereignty in terms of looking at time series data, at structured data, we have maybe also machine learning algorithms, stuff like this, and that was also... 20, 25, 30 years ago. Why does sovereignty matter when we talk about Gen AI? It's that all customers are worried about what is happening to the data, which has to deal, number one, with trust. What you trust, Michael, if you ask a GenAI application something twice and you get twice a different answer to rely that your factory maybe depends on this or that your product design is depending on this. It's a bit risky and therefore industrial GenAI has again a new dimension which is coming to the table. MICHAEL BIRD So it sort of feels like we've had this rapid expansion of AI and now we're maybe coming to a phase where we're not necessarily regulating it but trying to make a bit for like how can we use this in industry, how can we trust the information that we have. I mean are we seeing like the next stage of the AI revolution now? ANDREAS GEISS The next stage of AI revolution. So number one, it's of course having the courage. Many, many people would like first to define what is the real impact? What is the outcome? Before I maybe decide and invest into some use cases. And I believe it's in most cases the wrong way. Because customers... don't exactly know what the outcome might be. GenAI can excite you about results which you never thought about and knows also answers to questions you never have been amazed about and therefore it's important to just take the step and give it a try, which is easily done. We can have a compute environment spun up in a couple of hours, get something running. You are smarter in a couple of days. It doesn't take long in order to decide what the impact will be and then decide on the onwards journey what you will do in terms of operational scalability and how to bring that in use in order to create that business impact everybody's looking for. MICHAEL BIRD So why would industries in particular care about data sequencing? ANDREAS GEISS Wow, that's a bold question. At industries, it's important that all domain knowledge is of course within a supplier, within a manufacturing company. And in order to be on the forefront when it comes to innovations, you would like to secure your IP, that your data is there where you want it to be, that you always know who's looking at your data, that you have clear structures. What is also happening if you have certain partnerships who are dealing with your data, that this is going in the right way and that you never have a type of leakage, what puts you maybe under pressure or where a cool new innovation is getting overtaken by someone else and this is what matters at industrial AI and industrial sovereignty when it comes also to data and therefore customers need to have that trust number one first before starting off the journey. MICHAEL BIRD Wouldn't it slow, I doesn't it slow industries down? to sort of have a base. know, making sure I know where my data is. Does it stifle innovation? of soften a conversation. ANDREAS GEISS No, I don't believe so. So every customer has to be clear where he's putting certain workloads, if it's on cloud, if it's maybe on a hybrid architecture, or if it's on a private cloud instance, or maybe also on-prem. So all possibilities are there. What matters is that you have a clear data semantic structure of your company. Given that, I think things are spun up very easy. And if you look to new use cases, the most compelling observation I made the past couple of weeks and months, customers would like to have a safe environment they can trust, with partners they can trust, they have NDAs in place and then you spun up a use case and you instantly get insights to value what you can discover MICHAEL BIRD So we're in the EU at the moment and there are different countries, different rules and regulations. How does different legislation from around the world affect data sovereignty in industry? Particularly because guess many of our businesses are in multiple different countries involved in data ANDREAS GEISS Yes, yes, absolutely. So it does affect. It's of course a matter if you are talking of a use case in US, if you are talking the use case in Europe, or if you are talking also of a use case in China . So important is to adhere always the local standards which you must adhere given by law. And therefore it's also important to know that whatever you choose in terms of technology, terms of GenAI capabilities, that they reflect the legislation and also the regulation in that country where you would like to get the use case spun up AUBREY LOVELL Thanks Andreas. It’s so important given the speed at which technology is advancing to have people working at the frontiers of innovation doing all of that behind the scenes work to make sure that we, and our data, are kept safe. MICHAEL BIRD Now it’s time for “Today I learned”, the part of the show where we take a look at something happening in the world that we think you should know about. Aubrey what do you have for us this week? AUBREY LOVELL Well Michael, the fancy glasses are out again because we are diving into the world of computer chips, and quantum mechanics. So before I get into the story, I need to teach you Quantum Tunnelling 1-0-1 which you might already know about... Michael Bird Let's say I do, but I'm also going to get my notepad and pen out. Hang on. Aubrey Lovell Okay, get ready. This is a free educational lesson. Michael Bird That's a piece of paper out and a pen. Okay, here we go. Alright, I'm listening. Okay. we go. AUBREY LOVELL The math is complicated, so we are going to ignore that bit completely, but the outcome is pretty simple: once things get small enough, sometimes you find them in the wrong place. If you put an electron in a box, and you make the box small enough – *sometimes* that electron simply... well... isn’t in the box anymore – it has somehow managed to “tunnel” out of the box. Are you following me, Bird? This is a relatively well-known concept within quantum mechanics, and the reason I’m talking about this is that computer chips are very small. The transistors on them are unbelievably small with the smallest transistor made - as of the date of this episode - so tiny that you could fit thirty-five thousand of them across a single human hair. But here’s a problem: when transistors are this small, engineers have to work around quantum effects. If a transistor gets even smaller, the current flowing through it might simply ignore the “on/off” switch and “quantum tunnel” straight through. This would stop the transistor, and potentially the entire chip from working. MICHAEL BIRD Oh my goodness. my goodness. Right, how- Okay, carry on. AUBREY LOVELL So how do we get around this? Well a paper published in Nature by a Silicon Valley startup has revealed that they have developed a new type of computer chip which uses beams of light instead of electronic signals to move information around. This is known as photonics (think ‘electronics’ but instead of electrons, it uses light – or photons). The thing that makes this chip particularly special is that these chips can be used for neural networks which is kind of crazy. So, why should we care about this new chip? Well, these chips could potentially be a two birds one stone situation. Using light would mean that data can be transferred more quickly which would increase processing speed but on top of that, the chips would be able to run using just a fraction of the energy needed for current computer chips. Although the chips may take a decade to become mainstream, in theory they could provide a faster, and more sustainable future as AI demand continues to soar. MICHAEL BIRD Wow. Isn't technology amazing? I mean, I'm not even sure where to start with that. I mean, everything is just so fantastically small. Unbelievably small, isn't it? AUBREY LOVELL It's really crazy thinking about that like smaller and so many of them to fit on a piece of human hair like I still can't get over that. That's incredible. MICHAEL BIRD And it's worth saying, we at HPE, I know in our Hewlett Packard Labs, we're also doing photonics research at the moment, so do you know what, I think in a future episode we should get somebody from Hewlett Packard Labs to come in and talk about some of the cool photonics stuff that HPE are doing. AUBREY LOVELL Yes, I think that is an amazing idea. They do some really cool stuff over there and would love to hear it. Okay, well, while we continue to try and process the madness of what can happen when making a computer chip, let's return to Michael's interview recorded a few weeks ago at Hanover Messe with Andreas Geist to find out more about sovereign AI. MICHAEL BIRD So how do you build data sovereignty into an AI model? ANDREAS GEISS Number one is that you decide who are your partners you would like to engage with in order to connect with your data. Step number two is that you choose on technology providers who make sure that their technology is adhering to the local standards and also to their maybe internal governance policies which the company would like to set forth in order to have reflected in all their GNI engagements. And third, it's important that you decide about how you scale your solution later on. MICHAEL BIRD So would a sovereign AI interact with humans using it differently? ANDREAS GEISS So if you interact with AI, you instantly found out depending on the type of questions you ask, how sovereign this AI is or how, let me say, perfectly set up is an AI infrastructure. MICHAEL BIRD Would humans using the model know that the AI was different? ANDREAS GEISS they would know because it's the quality of the reasoning, the quality of the answers and the results you can get out and with that you can of course judge. And again, judge what means judge? Judge is again the super brain behind, the human who's judging the outcome. And so my answer is clearly yes, they would know and they would figure out. MICHAEL BIRD So how does data sovereignty affect business models going forward? ANDREAS GEISS My claim is without sovereignty, innovation is merely not possible. Having said this, if you would like to embark on a new journey to reach more with less, to come to better qualified decisions, to come also to new business models, it's ultra critical that of course data sovereignty or sovereignty at all is adhered. And having said this, it's also very decisive that of course in the whole context of driving innovation, how's innovation happening? Innovation happens if you have trust in that what you do, basically speaking. And this applies also to general applications. So you need the trust AI is behaving in the ethical and sovereign way as you designed, as the user, as also the company and the owner. And that brings me to an important point. You should remain the ownership. Own your AI. Own your AI destiny MICHAEL BIRD is data sovereignty always going to be a reactive measure? Or do you think it could be proactive in future? ANDREAS GEISS I think in the past, looking a bit, was by majority reactive. Most of the companies, most of the customers I have been talking to, they have made a safe bet with partners, which turned out, since a couple of months, not being a safe bet and a paradigm shift in the future. And therefore the geopolitical landscape has changed, of course, as we see and with that, you need to rethink whom you rely on and how sovereignty will look like, maybe also here in Europe. We need answers but we should also have an answer if uncertainty kicks in that we are not cut off and that we can try forward in reaching new business data centric business models also with a autonomous technology stack here out of Europe We see that disruption is taking place. We see that maybe also the load of model weights might differ. How much training intense cost you have to fund upfront depends a bit on the technology and also on the R&D. And again, coming back to innovation, what you are driving. And this brings me back that we have to have a firm answer also when it comes to Gen AI, also when it comes to sovereignty here in Europe. That's getting decisive. MICHAEL BIRD So why should our listeners care about data sovereignty? Why is it important? ANDREAS GEISS because it decides about your future. if you will be here in five years or if you will not be here in five years. And that's a question looking back at the Fortune 500 list, how many companies disappeared the past 24 years. And I mean not disappeared from the list, but disappeared as a company and they got disrupted and the same, if I look now in the future in 2045, I think the same is true And this is due to the fact that you need to keep control how you drive innovation, how you keep the ownership of your IP, how you protect also IP. And IP infringements are increasing. I mean, there are specialized firms and entities within companies just looking at infringements where they are happening. Because innovation is something what is the lifeblood, I would call. What is the guarantee that I can bring something to the table of my customer? What is decisive and what's also taking care of the wealth of my company and ultimately of course also of the wealth of a society. AUBREY LOVELL Thanks so much Andreas. I hadn’t even considered that building the AI to be sovereign could mean that you are cutting yourself off from other markets around the world which could be risky. On the other hand, if we are building sovereign AIs, at least we can know exactly how they are using data and that trust is really critical. For companies especially, I can see why they would feel more comfortable being completely certain of how their clients’ or employees’ data is being stored and used, especially if it involves training up AI models. It really is fascinating, and if you want to find out more on the topics discussed in today’s episode, you should go check out both the show notes, and our previous episode “How do you cut off cloud computing from the internet?” where we dive into other ways of keeping our data safe. AUBREY Right, we’re getting towards the end of the show which means it’s time for This Week In History, a look at monumental events in the world of business and technology which has changed our lives. Now Michael... you provided last week’s clue .... MICHAEL Yes I did, so, the clue last week was: It's 1983. Ring ring. That's the sound effect. You'll never guess where in the house I am. This is of course the first time a cordless phone would go on sale in my native UK [i]. We’re all used to phones these days being… well… mobile, but we all know that wasn’t always the case! It was way back in 1876 when the phone was first invented, and it took over a hundred years to get from the original design, to one which you could walk around your house and use. Now, this is a cordless home phone which is not the same as a mobile phone which if you’re a long time listener, you might remember we covered a while ago! Now the cordless home phone was introduced by British Telecom[ii], the cordless phone had to be charged for 24 hours before it could be used and people could “Walk and talk on the Hawk, the mobile cordless InPhone” from anywhere in their house as this new, exciting phone was “connected to the base unit by radio signals”. The phone could be used up to 100m - that’s a little over 300 feet – from it’s base unit. Let’s take a moment to think about the fact that it took only a further 15 years for the first commercial satellite phone to be created which could cordlessly connect to a satellite around 780 kilometers – that’s about 485 miles – floating above the surface of the planet. Now that is progress. Aubrey, did you have a cordless home phone growing up? AUBREY Oh, I definitely did. The one that I remember the most. Yeah. And I think we've talked about this before, like the golden age of the 90s. Do you remember when we had like colorized wireless phones? . And they were just the big ones, but they were so fun because they were a certain color and you could see all the components inside the phone. I felt so cool back then. Kids today won't know. They won't know the joy. MICHAEL BIRD I've got multiple siblings so the cordless phone was a was brilliant because you could sort of hide it So you could only you could have the phone and speak to your friends AUBREY LOVELL So true, So true. It really is quite amazing. And I know we really have talked a lot on this show about how certain technological advances happen in such a short space of time. So I'm glad that we, you know, have obviously fast charging nowadays. Could you literally imagine if we, you know, still took 24 hours to charge your phone or it has to take several minutes to, you know, hook up to the internet? It's pretty crazy. MICHAEL BIRD I think if it took 24 hours to charge my phone maybe I would, maybe I'd look at it less. Maybe that's a good thing. AUBREY LOVELL Yeah, might be actually. MICHAEL BIRD Anyway, Aubrey what's the clue for next week? AUBREY LOVELL Okay, so we're jumping back in time a little bit. Our clue for next week is: It’s 1961 and modern computing is about to be born… Huh... MICHAEL BIRD I mean, has to be the transistor? AUBREY LOVELL Yeah, that’s a good guess. MICHAEL BIRD Maybe I've got the timings wrong. 61. 61. Yes, okay. I'm to go for the transistor but we shall see you next week. MICHAEL BIRD That brings us to the end of Technology Now for this week. Thank you to our guest, Andreas Geiss And of course to you too... Thank you so much for joining us. AUBREY LOVELL Technology Now is hosted by Michael Bird and myself, Aubrey Lovell. This episode was produced by Harry Lampert and Izzie Clarke with production support from Alysha Kempson-Taylor, Beckie Bird, Alyssa Mitry and Natasha Naik. MICHAEL BIRD Our social editorial team is Rebecca Wissinger, Judy-Anne Goldman and Jacqueline Green and our social media designers are Alejandro Garcia, and Ambar Maldonado. AUBREY LOVELL Technology Now is a Fresh Air Production for Hewlett Packard Enterprise. (and) we’ll see you next week. Cheers!