Understanding the rise of Industrial AI
The Future-Ready Podcast: Industry & Beyond ·
Show notes
AI isn’t a new technology yet, today, AI is taking off like never before, finding countless new opportunities not just for consumers but in the much more demanding worlds of business and industry as well. Reaching that point is not simply coincidence but a confluence of factors culminating in the Industrial AI landscape of today and tomorrow
In this episode, host Conor Peick is joined by guests Matthias Loskyll and Samir Desai to answer the question of why now? As well as understanding what it took for AI to reach where it has today, the broad spectrum of technologies that fall under the AI banner, and what that all means for industry.
In the episode you’ll learn about:
(0:32) Why Industrial AI now?
(3:13) Consumer vs. Industrial AI
(7:34) Managing non-deterministic AI results
(9:10) Understanding the different types of AI
About the voices:
Matthias Loskyll is the Senior Director of AI and Robotics at Simens Digital Industries with a PhD in Production Automation.
Samir Desai is the Senior Director and Global Program Head for Data and AI at Simens Digital Industries with more than 30 years’ experience developing and managing industrial software sweats.
Conor Peick is a Marketing Professional creating forward-looking content for the Thought Leadership team at Siemens Digital Industries Software. Conor collaborates with industry experts and executives to produce impactful content exploring the challenges companies face and the technologies that can provide solutions.
Transcript
So, welcome to the Future Ready podcast. My name is Connor Pike. I'll be monitoring our discussion today. And I'm joined by Matthias Lasko and Samir Desai. We have 2, there are 2 excellent AI experts from Siemens here, so I'm really excited to uh jump into a discussion today. So, to get us started, I would love both of your guys' perspective. Maybe Matthias, you can start for us. Um, basically, why industrial AI and why now? Why are companies, uh, really looking at, at adopting these solutions? Uh, what sort of what problems are they facing? What are they looking at industrial AI to help them solve? So if you look at industrial companies today and where they are, it's, it's like a perfect storm, you know. They still have the requirements they've been facing for decades. It's about high quality of their products. It's about efficiency and cost saving and production. But in addition, now they are facing a lack of skilled labor as a big challenge. So in the end labor shortage, workforce shortage, um, but also they have to get their production more resilient because of all the supply chain crisis going on, like on a regular basis in the meantime. And they, they really have to get more adaptive in their production to unforeseen situations, to additional um product variants that they have produced. So it's really this, these old challenges, known challenges, but also the new ones coming up. So it's really the time to act for industrial companies now to use AI to really get to the next level of productivity. And so, Matthias, you bring sort of an automation perspective to this, and Samir from your side, it's more of the engineering perspective, um. So what do you think from an engineering perspective, maybe uh some of the same challenges that Matthias just outlined, or yeah, we kind of see some of the same and, and but with a different flavor on it, right? So, like, let's pick the demographic issue. We're seeing a lot of change in the type of users that our customers have. We're having the older generation move out, the newer generation coming in, and the newer generation is very impatient, right? And for them, they wanna really quickly learn the tools and do their tasks. So, How do you really bring these new users on board? Some of those are the challenges they're running into. They got data today that is completely fragmented in lots of different systems of records, if you would. And when they're building their products, this data from different systems need to come together. So they're they're kind of seeing that, and the last part that we see is hypercompetitiveness. Their competition New, not just local, but global competition is moving really, really fast, and they're saying, how can we cut short our Development times, our engineering time so that we can take our products much more faster, right? So, all these things are coming back together, a perfect storm for our customers, right? And, and, and that's where they wanna be able to use AI, yeah, a perfect storm. So maybe some of the, the, the innovation cycle speeding up, I mean, perhaps similar in the automation world as well. So maybe as we, we can jump into a bit of an AI introduction here for some of the listeners who are, who maybe are not quite as familiar with the technology as, as you guys, um. Maybe really quickly, Samir, you could give us a rundown on consumer versus industrial AI, some of the different, uh, requirements, the different needs. Um, yeah, sure. So, um, you know, when we say consumer AI, go back December 2022, chat GPT, right? Kids to their grandparents, everybody's using it, but what are they using it for? It's primarily around text, videos, images, things like that, right? Which is ubiquitous all around the place. We were using Google and YouTube for that. Now we're using AI for all those things. But when we look at Industrial side. What our customers do. Yes, they have documents and they have images and they have videos, but that's not their primary type of data. They're dealing with engineering, design engineering, manufacturing, simulation, manufacturing operations. That is a very complex type of data that has to be dealt with, right? And so that's, is where the big difference comes in. And we don't see In the consumer marketplace, what are known as those modalities for this complex engineering data available. So that separates us and which, which is where we start getting into our strategic elements around industrial foundation model. It's all about building models that deal with the engineering, the manufacturing, the operational data that can then be used to build AI features on top of that, right? So that's kind of like the real big difference in terms of What is the type of data? That our customers use for their products, right, I see. And Presumably there there's also a question of risk involved as well. I mean, in the production environment, yeah. Something, uh, an AI doing something that that you're not expecting or not ready for, could be potentially extremely costly or, or damaging to a, a, a production facility, presumably. Yeah, actually, besides the topics that Samir mentioned, so the different data modalities, we need models understanding the, the language of engineers, so to say, or the language of the industry, that's, that's an important aspect. But besides that, just think about where AI acts in the industry. It doesn't act in the, in the digital world only, it also acts in the real world. We're actually operating factories using AI. And if, if you think about this, this means there can be safety issues because of wrong predictions of an AI model. There can be massive downtimes or standstills of a production line, which can cause the damage of millions and millions of dollars. So it's a different thing than I always always try to explain to my family what I'm actually working on. So I always say, just think about recommendation of your favorite movie or, Using AI to translate some text. If something goes wrong, it's not a big damage. If something goes wrong in the industrial space, we have a huge issue. So it's all about reliability, it's about deterministic systems that engineers are used to, and how do they then interact with an AI system which is by definition non-deterministic. So how do you really integrate that into a reliable, trustworthy manner that can be used for industrial production use cases. If, if I could just add one thing on top of what Matthias just said, you know, there's this key difference of language of the internet versus language of engineering and manufacturing, right? And, and that's where the, the main separation happens on the consumer versus industrial AI. The risk elements. Engineering and manufacturing is a science. It needs to be deterministic. If something goes wrong, then the car or the plane or the machinery that our customers are manufacturing, there's going to be a huge problem there, and we cannot risk it. That's the Certain I mean, certainly, and, and I, Matthias, you mentioned this aspect of it that AI is, uh, in some ways is inherently non-deterministic. So, it's a fascinating way of thinking about how do you Build it into systems in a way that um that. Manage that risk, I suppose for you. Um, yeah, it's a, it's a really interesting marrying of this of this technology that can maybe be very powerful and predictive and um help you maybe knit together patterns you couldn't see otherwise, but, um, doing so in a way that you're not going to open yourself up to too much danger of a of a wrong prediction or something like, like you mentioned. And this really depends on the use case, how critical it is. Sometimes it's OK to have a certain non-determinism and and work on statistics and predictions of an AI model and work with that. It's, that's fine for many use cases, but there are some that are really, they go into the actual control of a machine or of a production line or even AI based safety applications. That's also what we see coming up more and more. So then we, we really talk about and embedding into I would say something like a sandbox or into some guard rails um that that are more rule-based maybe even, um, or that at least can be certified to a certain level and you can make sure that the outcome of the eye model is more predictable and more deterministic to really be able to to use it in such critical environments as well. It's similar to autonomous driving actually. We have similar challenges there in in industrial production as well. Speaking, I've been thinking of that of that safety, safety criticality, and um, yeah, yeah, making sure that the the system doesn't step out of its bounds or something to that effect, yeah. So, um, Maybe as we kind of keep thinking about this AI introduction in a way, um, there's obviously lots of different AI models, maybe different AI applications that that can be used in engineering and production environments, um, things like Maybe people will have heard of predictive, generative, um, agentic, physical AI, all these different terms, right? Um. Maybe you could give me just a quick, Samir, maybe a quick rundown of, are there different benefits, or maybe different advantages of these, of these various models or or systems. Um, yeah, how do they inform, uh, perhaps uh engineering process differently? Yeah, fair, fair question. So I mean. AI first of all, has, is not new, right? Generative AI is new. Predictive AI has been there for 3 to 4 decades, right? And if you look at in the Siemens portfolio, we have been doing predictive AI for a long time. Now, there is an increasing value going from predictive to generative to agentic, right? Predictive, as the name is suggesting is it'll say, hey, this is what could happen and this is in there, but generative is saying, well, it's not only we are predicting it, we'll help you create that, we'll help you generate that information for you. That takes it up to the next level, and then beyond that, the agents are coming in with agentic AI and saying, We don't, you know, we will assist the engineer in creating that, right? So now the engineer is free to do their primary task, which is doing engineering and design engineering and manufacturing engineering, and it's taken away repetitive actions and tasks that have to do. So, there's always an increasing level of benefit that they get out of it. Now, we use this at, in different ways on the engineering side of our software portfolio. Um, all of the above elements are used in a way that when a design engineer is going into our mechanical or electrical design software, we got co-pilots in there that they can quickly start chatting with them and say, hey, I want to do ABC. How do I do this thing? I want to create an electrical diagram. I want to create a 3D geometry with so and so. How do I do it, but not only how do I do it, can you go create it, do it for me, that I'm gonna review and say yay or nay, or I'm gonna edit it in there, right? So we, we kind of start doing that across our portfolio. There's several examples of how we take advantage of all these pieces for us. Fascinating. I think that there's one thing which is really important. It's, it's not always about following the hype and not everything should now be agentic or generative. We see a lot of use cases actually out there in the industry where predictive AI works best because these are specialized models, solving specialized use cases. So that's a perfect fit and there's no need actually to go for, for the, the hype topics. But that's something that where, where we really also have to consult our customers sometimes to clearly say, OK, for this problem, you don't even need AI. For another problem, you could use predictive AI which will work perfectly fine. You just have to train it based on certain data. Um, to make predictions then, and for others, there's really the need for generative and engentic AI approaches. That's, that's a really good point, matching the matching the tool to the problem that you have, right? And it's, it's the right tool for the right problem at the right time, you know, um, and you don't just sort of Throw the AI blanket over every problem that you have and expect that it's gonna, gonna work out. But as interesting as this has been, unfortunately, that's all the time we've got for this episode. Please tune in again next time as we continue our discussion with Samir and Mathias on the Future Ready podcast.