A bright future of AI in Operational Technology Podcast – Transcript
Merging AI with operational technology (OT) is no simple task, between concerns about robustness and reliability as well as data privacy, there are many challenges to overcome. With that said, the potential benefits make the endeavor worth the effort. Rather than bringing AI into every aspect of manufacturing right away, the key for long-term success will be adopting it gradually through small tasks with immediate benefit that also support broader, long-term goals.
To learn more, check out this recent podcast with Ralf Wagner Senior Vice President of data-driven manufacturing at Siemens, here or keep reading for a transcript of that conversation.
Spencer Acain:
Hello and welcome to the AI Spectrum podcast. I’m your host, Spencer Acain. In this series, we talk to experts from all across Siemens about how AI is being applied to various technologies. Today, I’m joined by Ralf Wagner, senior vice president of data-driven manufacturing at Siemens. In the previous parts of my discussion with Ralph, we’ve examined the applications of artificial intelligence within Insights Hub, the applications of Insights Hub itself, and some of the more advanced technology that’s been coming out of their research lately, and how that’s being applied within the broader context of data-driven manufacturing. But with that said, what are some of the challenges that you’re facing bringing this kind of revolutionary, cutting-edge AI technology to the OT environment, which tends to rely on more traditional technology? Have you faced any challenges of making these AI solutions that you’re offering, especially the ones that rely on generative AI, which is of course prone to hallucination, robust enough for the manufacturing world?
Ralf Wagner:
So that is actually quite a learning process we have been going through over the last few years, Spencer, and I think the first and most important learning is that we are bringing the technology towards the persona that can actually use and get value out of it in a specific way according to our four quadrants. I think this was the first learning to take away all the kind of efforts, which typically the data scientists is going through, and pre-solve this in the out-of-the-box solution for specific user personas. I think this was the first hurdle which we actually took. Since we are way more specified in our product and portfolio development, how we use AI towards the different personas and solutions, we got way better and excited customer feedback. This is exactly what we’re looking for. This is how we can get most value in the most efficient time out of it. So we reduced dramatically the time to value for these kind of technologies for our customers. So that was the first, I think, very important thing for us to go through, and to learn how to adopt these kind of technologies in our product.
The second one, and now you’re speaking also about generative AI specifically, or more generic also for other AI, how we train those models. And one of the very, very important point for our customers is that the training data stays with them. So we are not using customer’s data for whatever, and we train a model and every other customer is actually taking advantage from that training data. Because this automatically creates a kind of, “Oh, no, I’m not sure whether I want that.” So very early in the process, since the very beginning of Insights Hub and it’s pre-successors, we always said, “The customer owns his data.”
Siemens is not doing anything with it other than having some access to the metadata to see how the system is being used. But the data itself is owned by the customer and nobody has access to it. There is no sharing, there is no whatsoever. This was one of the mantras from Insights Hub from the very beginning. Even more important now in AI and generative AI, if we train the model, for example, with the maintenance documents of the customer, that stays only in his tenant. Nobody else has advantage of that he gave this training data to the model. Only that customer tenant has access to this. Also, the underlying model, which we are using is a private model which we are running in the Siemens infrastructure. And this is nothing is going back to open AI or Microsoft or something else. It stays in the Siemens, and we are not training any kind of foundation model with that data from those customers.
That has to be very crystal clear. In the industrial environment, we cannot do this, what the Facebooks and the Googles of the world are typically doing, with these kind of models that they use. You do a little mouse click and then every data is being trained and used for improving the model. We are not doing this. So this is a very important message for our customers, that data stays with them and it’s not used to train the model for others. So that’s the second very important point for us in overcoming these kind of hurdles. The last one, and you mentioned hallucination and a robustness of what the model is actually doing. There, we actually developed in our dev ops environment processes and tools in order to beef up quality validation, which we always do before we release something. It goes through a QV process. But there, we needed to introduce new tools and new processes and new approaches to make quality validation for those AI models, for those new technologies.
So testing the model for consistency. That you ask the same thing slightly different, but mainly the same, that you get consistent answers. That you avoid hallucination with using reasoning in the background. So if the model is not sure or doesn’t have the right answer, it tells you, “I don’t know,” and not come up with something else. So these are mechanisms which we put in place that comes with these new technologies. So we rather don’t give an answer before we give a wrong one. So these kind of guardrails around the systems is just what we built over the last one to two years when this technology came up, because we were pretty aware that everything comes down to the answer and the quality of those answers, the customers are getting out of the co-pilot, in order to trust it, to use it, and to actually rely on it going forward as well. So that’s why this is a major invest for us when it comes to AI and generative AI, that we make sure we have the right guardrails and quality validation processes and tools in place, to give the trust to our customers.
Spencer Acain:
It sounds like this is a very complex issue that I think that you’ve put a lot of thought into, and I think it sounds like are still … It’s an ongoing process even, where what works today might not be sufficient for tomorrow, but you’re constantly staying on top of it and making sure that you’re offering the best quality without compromising your customer’s data. Fantastic. So I think to round things out here, I’d like to ask you, where do you see this kind of AI-driven technology going in the future? The future of manufacturing, the future of AI? I’ll open the floor to you for that one.
Ralf Wagner:
So this is actually a pretty difficult question I have to say, because if you just look at the speed of innovation we’ve been seeing in the last two years, since basically ChatGPT 3.5 was launched in November 2022, this was a roller coaster since then. This was a breakthrough. And the speed of things which we have been seeing basically on a weekly basis and if not, and bigger breakthroughs on monthly basis, was unseen in technologies, I think. So there is always a comparison with Moore’s Law where there is saying the number of transistors doubling every so and so many months. Now, if you look at the innovation speed of AI, that exponential curve from Moore’s Law is way, way more steeper when it comes to gen AI and AI compared to Moore’s Law. So that gives you only a feeling what’s going on in that environment here.
And I think we can expect more and more of those breakthroughs, specifically when it comes to the now reasoning models, which we are seeing here, and not even to mention what will happen. And as I said, we are launching that in end of March as well when it comes to the agents. So that autonomous agents, first of all, agents will be orchestrated at the beginning. But then, agents will become more and more autonomous and will see and find things and take action and inform other agents, these network of agents solving problems independently to problems we don’t even see. Maybe they detect them and then they fix it and we will not even be aware of what was going on. So it sounds at the same time, frightening and exciting, I have to say. But that technology will open up the next level of S curve of improvement in manufacturing on the shop floor.
The OT world is still there to be really improved significantly with IT technology, specifically AI going forward. We have seen a lot of improvement on the IT side of the industry, in enterprises in the last years, and we are pretty sure we will see significant impact, positive impact and improvements, on the OT side with this kind of technology going forward. So for me, I think this is going to be very exciting times, which is ahead of us. So we at Siemens and specifically the Insights Hub team, is there to take advantage of these new technologies, to give this in the hands of our customers. To actually make a competitive difference for them and their manufacturing processes and manufacturing goals going forward.
Spencer Acain:
Yeah. Well, let me just echo what you said and say this is both scary and an exciting time ahead of us for AI. Yes, it’s really just developing at an absolutely incredible pace. But on that note, I think that’s all the time we have for this episode. So once again, I have been your host, Spencer Acain, joined by Ralf Wagner. Thank you, Ralf.
Ralf Wagner:
Thank you, Spencer.
Spencer Acain:
And this has been the AI Spectrum podcast. Tune in again next time to continue learning about the
exciting world of AI.
Siemens Digital Industries Software helps organizations of all sizes digitally transform using software, hardware and services from the Siemens Xcelerator business platform. Siemens’ software and the comprehensive digital twin enable companies to optimize their design, engineering and manufacturing processes to turn today’s ideas into the sustainable products of the future. From chips to entire systems, from product to process, across all industries. Siemens Digital Industries Software – Accelerating transformation.
