Latest / Machine Dreams / Your Employees Are Telling AI Your Company’s Secrets. This CEO Says It's Time to Bring AI In-House
Transcript
- speaker-0: For the last few years, companies have been racing to use AI. A lot of them have tested models, launched pilots, and many now are increasingly starting to experiment with AI agents. But there's a really big difference between asking AI to summarize a document and giving it access to your company's data systems and you know, potentially the ability to allow that system take actions. on your behalf. And I believe that that raises a much bigger question for us right now. And that question is how much control are we actually prepared to give AI? Well, this is Machine Dreams, where we talk to the people building AI about what's working in the industry right now, what is not working, and where all of this is going to take us. My name is Kalawali Samuel Adibayo. I'm a technology analyst working at the intersection of AI, cybersecurity, business and society. And of course as always, I'm joined by the one and only remarkable, delectable co speaker-2: Hey everyone, I'm Leia Stern, Global Communications Strategist and former venture capitalist. Hey Sam, how are you doing? speaker-0: Very great. You look fantastic. speaker-2: thank you. So do you and I'm loving your new backdrop. speaker-0: Yeah. I think all the praise should go to my wife 'cause she actually suggested that, so speaker-2: She deserves the praise and remember happy wife, happy life. Today we're joined by Yevgeny Vallis, co-founder and CEO of Shakudo. Shakudo builds the infrastructure companies used to run AI inside their own walls. So that that's the data that never leaves. Their customers include Lobla, Canada's largest retailer, and before founding the company, Yevgeny was doing this work inside big organizations at BMO, Borealis AI, and Georgian. Yevgeny, welcome to Machine Dream. speaker-1: Thank you, Leah and Sam. Pleasure to be here. speaker-0: All right, great. so let's just get right into it. One of the major things that you do is around sovereign AI. And that's a term that I've been hearing a lot lately. You know, just recently I was at the United Nations AI's AI for good summit, and one of the major themes at the conference was sovereign AI. but as you know probably, there are so many definitions for what sovereign AI is. So many that it doesn't seem to to have A consensus definition. So I'd like to ask you, what really is sovereign AI? speaker-1: That's a great question, Sam. And you're right. There are many valid ways to define it. The way I would approach it is asking the question of why is it even important? Why is it important that AI is sovereign and and that we think about the definition? We're used to thinking about data as backward looking, data and analytics and machine learning. Taking some of your existing data sets, producing a report, producing an audit, producing some analysis of what happened in the past. But like you said, today data with AI on top of it is used for the actual decision making of the business. And it's not just decision making around, you know, what color should my website be? I'm not just talking about the the successes that we've had in coding code generation with AI. I'm actually talking about more major decisions that are strategic, that influence the financials of the business, the the employees, the customers, investors. Everyone I and know, y y you guys can correct me if if that's not the case for you, I'd be very surprised. But everyone these days, when in doubt, goes to Chat GPT or Claude and just drops the question in should I go this way or this way? Should I, you know, invest in this company or that company? Here's the profile, here's everything I know about them, tell me what to do. Everyone kinda does that these days. And so When you think about AI in those terms, when you're actually outsourcing your most critical thinking, your most critical decisions to the AI, which makes sense. It baselines you, right? It gives you a baseline of here's here's what's what something that is smart enough would potentially do. Doesn't bind you to that decision, but it gives you a baseline. So you're not completely off track, or at least you're aware of kind of what the baseline is. Think about what that means to give that away to someone else at scale. Right? Personally, of course, you know, you you can kind of think about it from a personal perspective. Do I want to tell OpenAI all of my deepest personal thoughts and secrets and ideas? But at a corporate level, you are giving away your employees, your executives, and your directors. most inner strategic thoughts about risks, about opportunities, about what you're actually doing to a third party. And that third party is able to take all of that and aggregate it at scale and draw conclusions at scale without having people look at the data. It's software that consumes all of that, analyzes it, and knows how to infer and understand what's happening based on what all of the employees in. your company, your corporation are are sending in. So back to your question of what is sovereign AI, I would define sovereign AI as a as AI that is in the control of the decision makers of any entity, whether it's a country, a corporation, a military. Whoever's making the decisions should be able to fully put under their control the AI that the members of the organization are using. Does that does that does that make sense to speaker-0: Yeah. Pretty much makes makes a lot of sense. speaker-2: So between BMO, Borealis and Georgian you saw this from inside a bank, inside a research lab and inside a fund. What was your vantage points that told you the most about why this stuff fails? And perhaps it was the moment you decided you had to start your own company all around it. speaker-1: Yeah, so Shukudo started in twenty twenty one. So that was before the the GPT craze, before kind of everyone got into AI. So we were we were always doing sovereign data in AI. We just didn't call it that because that's a new buzzword that the industry has has come come up with. back then the problem was related but different. A lot of the innovation labs across the industry do experiments but then put them on a shelf, right? They they kind of they'll build something really cool and then we'll just go somewhere in some storage to kind of collect dust. eventually the technology becomes irrelevant and that effort goes away. The way I see it, a lot of enterprises, a lot of non-technical organizations invested in the since two thousand really since dot com have been investing in innovation labs in preparation for doing something with that innovation, but never actually kind of pulling the trigger and and taking that innovation to market. And so when we started Shikudo, and I've seen this d to your point, Leia, I've seen this a few times. I've seen this at Georgian, you know, working with with the portfolio companies. I've seen this at RBC, at Borealis AI and and Bimo, where I kind of built the the central AI team. And Georgian was, you know, a bit different because these are startups. So we were kind of working with later stage startups and advising them on how to scale with machine learning. Back then it wasn't people weren't really calling it AI. But in the banks, it's always the same challenge. How do you actually benefit the business? with the innovation that your team comes up with. And so what we realized is one of the bottlenecks is the teams do not have an end-to-end environment where they can do everything from inception, from idea, all the way to full production deployments of their solution. And it has to be connected to real data. You can get access to sandboxes online, but the banks are pretty good at Controlling their data flows. They've gotten that built into their system architectures. They're good at making sure things don't leak. What they're not very good at is making sure that we can take full advantage of that data. And so we started Shikudo to build a system which is end-to-end, fully self-contained inside your organization, so sovereign, running on our customers. infrastructure, whether it's their cloud or on-prem, private cloud, where they can do everything end-to-end. So start from the most sensitive data in the company and build a full cutting edge s innovative solution. And that that's been our business. I believe we're still the only ones doing this in the market. We can talk about kind of what we run into, but that's you know, there's no one that's directly doing what Kuder is doing. And we've had some really good traction with With what we call critical infrastructure businesses, finance, healthcare, defense, oil and gas, those types of businesses, nuclear. speaker-0: Yeah. Yeah, can you we'll come back to what you said about sensitive data in a moment, but at the heart of sovereign AI is a lot of conversation around control safety, building guardrails for for AI systems. But you know, lately, and I'm sure you must have seen that, this like increased rhetoric around how AI might be the end of us all, right? the CEO of Anthropic is talking around that. a lot of people are talking about that. How does sovereign AI play into all of this? speaker-1: So so first I'll say AI is definitely changing the the world, changing the industry. I think we're very far from it being the end of us all. the the CEO of Anthropic was saying that a the AI that existed before GPT was going to be the end of us all. And it was clearly not the case, right? So I think we're we're c we're very far away from that. it doesn't mean it can't happen, doesn't mean that AI eventually cannot become a risk, but I think we're we're far away from that. now how does sovereign AI help though? So well the biggest way in which it helps is decentralization. Imagine that AI sits with a duopoly like Anthropic and OpenAI. There's only two companies globally. That have the strongest, largest AI models in the world, and everyone's inner thoughts are funneled to those machines and to those servers, you're creating a risk where you there's a there's a black box that's constantly being evolved by those two companies where all of humanity is sending their thoughts to those to that box. Now that's an obvious risk, right? If AI If AI ever gets to a point where it's dangerous, having it concentrated in one place makes it so much more dangerous. And so Sovereign AI helps with that, of course, by making it more decentralized. There are different AIs, they're tied to different organizations, different thinking, and potentially can de-risk each other when it becomes powerful enough to be a threat. Now, again, all of that is hypothetical. I think we're very far from that, but it's not impossible we'll get to that. speaker-2: mean you mentioned decentralization as a de risking tool. On all of this AI panic, what is regulation worth when software crosses borders and China seems to be writing its own rules? speaker-1: So China has done a great job building open weights models and putting them out in the market. Yeah, it's it's interesting. I was listening to another podcast, you guys may know it in In Invest Like the Best if you've if you've heard of that one. They've had the CTO of Palantir there, and you know, he said something pretty interesting, which is China pushes these these companies to to do things that are not commercially beneficial, but they benefit China. politically. And so having these AI models published globally, that then they're competitive from a performance perspective. They're competitive with the American models, but they're available to everyone disintermediates the US re regulators in this scenario. So to your to your point, you you can regulate all you want, but if you ever anyone can download AI that's as powerful as Fable or Mythos or O you know or or GPT you know Astra, the the regulation kind of loses a bit of its enforceability. So so I don't think regulating AI in in the sense of limiting it is the right way. I don't think we should slow it down. I think the slowdown requests that are coming from Anthropic and OpenAI are more driven by their desire to IPO. Than actual kind of risk to humanity. But there has to be some way to regulate its impact. So if you're going to use AI to make decisions around loan adjudication, mortgages, welfare, healthcare, if you're going to actually influence people's lives in regulated industries, within those industries, the use of AI can be regulated. You cannot tell. a a a person on the street, you should not download an an open source model and use it. It's just not enforceable. speaker-0: Yeah, just just off the back of Leah's question about this r decentralization, which I think is actually really important. you know, I speak to companies all the time and there are even big companies that you will think would know this, but they don't. a lot of them put like some of their most valuable information into AI systems built and operated by other companies in a way they're giving over their intellectual property n without even knowing. And So my question really is, if these businesses give away more control than they realize and what should they be doing for businesses that are doing that right now? speaker-1: So that that is that is exactly the point I I was I was kind of emphasizing earlier. It's not just intellectual property that you're giving away. You're giving away your current view of your own business and the market. 'Cause your employees are going to Chat GPT and and Claude and typing their business problems in there to get advice. So you're giving in inside knowledge to external organizations of your business and of your industry and your competitors, keeping that in-house is instrumental. It's it's baffling to me that people would not clamp down on this faster because by the time it becomes a clear problem that everyone's kind of aware of, a lot of the industry has already given cool detailed internal snapshots of their internal business and their competition to an external party. You you gotta bring the AI in house to enable your employees to get that advice. They need the advice, right? But you can't send be just kinda sending it unfiltered to to third parties. speaker-0: Does bringing AI in house automatically make everything safe? speaker-1: You gotta have your own controls. You can you have to do it right. You have to have the right team in place that will know how to operate it and to make sure that your your employees and your contractors and and your partners all use the in house sovereign AI. But if you've done it, if you've brought it in house, it certainly makes it safer than sending your innermost secrets out. You you need to be careful about the guardrails. Right. So one of the things that you have to watch for is if you're operating your own AI, is the guardrails have to be in place as well. Now, the the models today are are trained pretty well. They're trained n they're not they don't destructively break things as they used to, but it's not impossible. It's you have to have guardrails in place if you're bringing AI in-house. You also need to just be aware that the geopolitics play a major role here. So the models you're getting from US are trained differently from the models you're getting from China. You don't actually know how they're trained. So a lot of our customers in the defense space, for example, they insist on using American models, which makes sense for them. I would say taking that the kind of the geogra the geo origins of the models. And the AI is something that a lot of folks who who are bringing AI in house have not yet started thinking much about and I think it's time to start thinking about it. speaker-2: Very, very interesting stuff to think about. And with our final question, we will ask you, imagine we're having this conversation again five years from now. What do you think AI will be doing inside companies that most business leaders say would still be uncomfortable allowing it to do? speaker-1: Yeah, a a AI will be running a lot of the business processes, the internal processes. doesn't mean that the people will not be there. I think their jobs jobs will change and evolve and will become more productive, but as they all become more productive, all businesses will grow as a result. So I do think that the economy will benefit substantially from the presence of AI. A lot of the competition will be around how good your in-house AI is. Because once everyone internalizes that you cannot give all your internal secrets to a third party, running your own AI will become imperative. It's like having a good CEO, right? It's like having a good team. and so the a lot of the work that's operations, finance, HR, sales, engineering, all of the team all of the teams, all the functions will be dramatically higher output. Through AI. I don't think AI will eliminate the the need for people in those organizations. It will just make the jobs very, very different. And the scale, the scale of what's possible very will be very different. There's the yeah. So yes, what would be uncomfortable? I think the biggest uncomfortable conversation around AI will be the the use of AI for measuring the performance of the business and of its employees and of its leadership. It can be compelling because it's it can be objective. If you agree on the rules, you feed them into the AI, day I will just execute on the rules. but but that makes people uncomfortable. It takes it takes the human element of are we doing well or not? more out of the qu out of the picture, out of the question. And i yeah, that's I I can s I can foresee a lot of cultural changes in the industry around how AI plays a role on You know, who gets hired, who gets fired, how the company scales, what are the strategic decisions that the company makes. speaker-0: Yaf Gan Ivalis, co-founder and CEO of Shakudo. Thank you so much for joining us today on Machine Dreams. It was a great conversation. speaker-1: Thank you very much. It's it was a pleasure. Thank you. speaker-2: And thank you to everyone watching and listening. More conversations with the people shaping what comes next in AI are on the way. See ya next time.