With “explainable, industrial-grade AI” the goal is not just predict an issue but understand why.

In this episode of Manufacturing Matters, Berk Birand, CEO, Fero Labs joined TECH B2B Marketing’s Jimmy Carroll to talk about the deployment of AI-based technologies on the factory floor. Leveraging AI in industrial automation can mean more than just visual inspection, as AI tools for process optimization can now help predict when something will go wrong and why, provide positive environmental impacts, and ultimately optimize operational efficiencies and drive revenue.

This episode dives into these topics while also touching on greenwashing in the steel industry, generative AI and LLMs, the digitization movement in the industrial sector, the results of a recent AI survey, and more.

Jimmy Carroll: [00:00:04] Hi, Everybody. My name is Jimmy Carroll. I’m the vice president of operations at Tech B2B Marketing. And welcome to this episode of the “Manufacturing Matters” podcast, where we discuss the trends and technology shaping the manufacturing industry. I have the pleasure of being joined today by Berk Birand of Fero Labs. Berk, thanks so much for taking the time today. I really appreciate it.

Berk Birand: [00:00:23] Thank you. Thanks for having me.

Jimmy Carroll: [00:00:25] Of course. Yeah. So, Berk, for those who don’t know, please tell us about the company and what you do there.

Berk Birand: [00:00:31] Absolutely. I’m one of the founders and the CEO of a company called Fero Labs. We’re a profitable sustainability platform, so we build software that helps manufacturers leverage the latest advances in industrial-grade explainable AI to improve profits and reduce emissions at the same time. So, concretely, what this means is that we allow an engineer working in heavy industries to, let’s say, a process engineer or a chemical engineer, a steel metallurgist who has an engineering background but maybe he’s not a data scientist, they can use our software to pull data from their databases, their data historians, from their control systems, put it in Fero and then use a visual tool to build an AI machine learning model that effectively helps them operate the plant more effectively. It makes recommendations and suggestions on how they could tweak their set points so that they could reduce their odds of scrapping a batch of production, for example. Or they could help . . . the recommendations could help them increase yield, lower emissions, lower water waste, and frequently, in fact, it helps them do all of these things at the same time. And frequently, these different goals come at the expense of each other. And Fero is using these cases to find a sweet spot so that our customers can run their factory in the most optimal way.

Jimmy Carroll: [00:01:46] Yeah, so that’s interesting. There’s a lot of, there’s a lot of follow-up questions I have there, but the term “explainable AI” is fascinating to me. What does that mean to you?

Berk Birand: [00:01:55] Yeah. So we categorize AI in terms of two large families of AI models. On the one side you have black box AI models, black box machine learning models. These are frankly . . . Most models that are used in the tech world or in kind of broader in enterprise world, these are mostly models that were built again by academia or the tech sector for their own requirements. And as a result, these models kind of work in a best effort type of way. So, let’s say a good example, you know, from recent years, of course, ChatGPT where you ask the question and, you know, 80% of the time it gives a pretty good answer. And for most of those use cases, that’s great. The issue is, what happens in those 20% of the time when it’s hallucinating and really, the user and in fact, even the manufacturers, the coders of the model, have no context as to why it’s not working well. It’s a black box. So, this system, these types of models don’t work for industry. In fact, we believe the adoption of black box models . . . The reason why the AI broadly is not adopted in industrial sector is because most people associate machine learning and AI with black box models. The issue, of course, is that if I have a model working in a steel plant, that 20% of the time gives a nonsensical recommendation, nobody’s going to use that, and it’s going to cost hundreds of thousands of dollars in lost production. So, all of our technology is built in-house by our R&D team and my co-founder to be explainable by its very nature, which is if you look at, you know, Six Sigma or typical industrial optimization theories, the real goal is to find the root cause of things.

Berk Birand: [00:03:33] The real goal is to not just to predict that there’s going to be something wrong, but understanding why, what causes there to be something, what causes a quality issue, so that engineers can then fix the issue and then the issue doesn’t happen again, and nobody has to predict it anymore. So, all of the models that we have built in-house are statistical models. The technical name for it would be Bayesian machine learning models. And the, the way that I would describe it is that if a model is making a recommendation, it’s saying, you know, put this much carbon in your steel batch in your steel furnace. It’s not just giving this recommendation. It’s also saying, you should do this because I have learned that carbon has this particular kind of impact on your strength properties that you want to satisfy. And ,and on top of this, you can say, I am very confident in this recommendation. I have a lot of data. I’ve crunched all the numbers. So the model can say I’m very confident. These are all the things that we believe form the backbone of what AI needs to have in order to be adopted in the industrial sector. And that’s what we call explainable industrial-grade AI.

Jimmy Carroll: [00:04:32] No, I love that. I was doing a podcast, I think it was last week, with a systems integration company. And the gentleman I was talking to was the, the head of AI there. And he said, you know, generative AI, large language models, these tools, some of these tools are useful on the plant floor in terms of maybe helping people with ease of programming with PLCs or something like this. But there are these AI hallucinations. And he said, you know, I tested one out and said, give me the names of the 13 dwarfs. And ChatGPT gave, you know, the seven dwarfs, plus other names that were very similar. But, you know, there aren’t 13 dwarfs. So, that’s funny, but on the plant floor, it’s a lot less funny, I suppose. So, explainable makes a lot of sense.

Berk Birand: [00:05:22] Absolutely. Yeah. I think generative AI in that lens will have a huge impact in terms of being a proper interface, a chat interface to do various functions in summarization or reports or even, you know, text to audio for sending messages to people, but it’s not . . . In terms of the core models that really understand the physics and the chemistry of a factory. Our models, of course, are unmatched.

Jimmy Carroll: [00:05:47] So, on our podcast and in general, the people we work with, AI often refers to machine learning, deep learning algorithms that are used in visual inspection. In your case, though, and in the manufacturing space on the whole, AI comes in many forms. So, with Fero Labs, you kind of talk to me already about how you’re leveraging AI, but how does one go about integrating it into existing processes?

Berk Birand: [00:06:11] Yeah. So, we almost exclusively work with factories that already have a data infrastructure in place. So, data historians or they store sensor data, a control system. You know, a lot of times we work in the heavy industries, steel and chemicals and oil and gas being some of our top sectors of focus. In these cases, you have high temperatures, high pressures, fluid-style processes, in contrast to, let’s say, assembly-style processes, for example. So, naturally over the past 30, 40 years, these plants have been becoming more and more driven by control systems. And a lot of this data is available, the control data is available and in fact it is being used. It has been used for the past 20 or 30 years for Six Sigma style-analysis for continuous improvement, kaizen, and so forth. So, the way that Fero integrates with these tools . . . So, first of all, back to your original point, Fero’s models work with numerical data and typically time series data. So, think of the temperature over the course of a reaction in a chemical plant, or the steel chemical composition of a steel batch in a furnace, for example. So, these are numbers. There’s many of them, you know, hundreds actually. In fact, some of our customers have thousands of different sensors in even a small part of their plant. So, the challenge there is to take all these historical production data and then effectively, just by looking at the past data, learn the chemistry and the physics of the factory, and then our models are tuned to be able to do that. We don’t use computer vision directly. Some of our customers take computer vision systems that feed into our models, but we don’t deal with videos or computer vision directly. So, we work with numbers and time series data.

Jimmy Carroll: [00:07:49] Yeah, so it’s interesting, right? Because again, the you know, with the parallel of the visual AI, let’s call it visual AI, visual inspection. Part of the reason that AI has taken off or settled in, I guess, in the manufacturing space, in terms of being able to be partnered with machine vision technologies to do subjective analysis and things like that ,is because of other technologies advancing, right? Industrial computing, you know, CPUs, GPUs, FPGAs, other NPUs, CPUs, these other things. It’s not that much different for you guys on, you know, with your version of AI, right? Because people’s ability to collect data has improved over time. And I guess the cloud probably plays a big hand in that. Would you say that’s the case?

Berk Birand: [00:08:38] Absolutely. And I would say, I would add the sensors becoming cheaper has been a third direction here. So, software infrastructure has improved quite a bit. At some point . . . You know, the type of data volumes that we’re dealing with today would be considered big data 20 years ago, and it required specific clusters or specific pieces of software. Now, of course, we’ve become much better at storing and processing large quantities of data. We have more sensor data initially, you know, probably 20 or 30 years ago, you would only add sensors to things that were absolutely required for you to measure. So, usually inputs to PID controllers, for example. Now, our customers have the luxury of adding sensors just to see if that data point makes a difference. And ultimately, the whole digitization movement in the industrial sector in the past 10 years, 15 years has really benefited and set the tone, set the stage for us to be able to apply our models to improve production in these sites.

Jimmy Carroll: [00:09:33] Yeah, digitization is a term that gets floated around a lot and within the AI space, right? Not just, not just in manufacturing or the industrial space, but AI has been subject to overhype at times. But there is no denying that AI in its various forms can benefit and does benefit manufacturers today. What are some other ways that you’ve seen AI adding value to manufacturing, including your own? I mean, there’s lots of different ways that it’s adding value beyond just, you know, predictive analytics and machine vision. It’s a whole world of. . .

Berk Birand: [00:10:09] Absolutely, I would say . . . so, there’s different directions here. One example is . . . effectively to use AI is a way to integrate horizontally different units of a process. So, typically, let’s say, let’s take example of a steel plant. You have the furnace, the first part where you melt, you know, let’s say electric arc furnace, you melt scrap steel, then you add different ingredients. Then you roll it, you process it. And each stage has its own teams dedicated to optimizing that individual state. And of course, these are hardworking people that have experience optimizing their own parts of the process. But frequently what you can see is that something that’s done in stage one might be the right thing to do as far as stage one is concerned, but actually kind of wasteful as far as the end-of-line quality is concerned. Everybody wants to be on the safe side. Nobody wants to cause any issues. So, as a result, when individual teams optimize different parts of the process, that doesn’t necessarily mean that the process end-to-end is optimized. And what AI allows our customers to do in this case is to take the entire process from beginning to end and just do one single optimization across the board and tell, you know, folks in stage one, here’s exactly the chemistry that you should aim for in your steel product the second time, you know, this is how you should roll this particular chemistry. And as a result, we can still guarantee that each team will be doing a phenomenal job while also making sure that the plant as a whole is minimizing the waste that, you know, in this case, it could be raw material waste and while still hitting their targets. I would say maybe a second component of this that’s really powerful is that, as I mentioned, there’s also different teams that have different goals, right? You have the maintenance team’s goal is uptime.

Berk Birand: [00:11:46] You have the quality teams’ goal is not to have any scrap. You might have a compliance team whose goal is more on the environmental side. And AI is also a powerful tool that can effectively be the connective tissue that connects all these plants, all these different groups, so that all of these goals are optimized. And the way that we do this at Fero is by allowing these different teams to specify a cost function for their, for their part of the world, right? If you’re a quality engineer, you know exactly how much it costs to scrap a batch of production, for example. This is like the back-of-the-envelope numbers. And you don’t want that — that’s bad. If you’re on the raw material side at the beginning of the process, you know how different sources of raw material, raw materials have different quality metrics, different costs. And you want to, you know, clearly you don’t want to overpay on raw materials if it’s not going to help your end-of-line quality. So, by providing, assigning a cost function to the entire production in this way, Fero can optimize the entire factory across different stages, across different objectives in one go, and we estimate that this adds about 8, 9% in additional EBITDA or reduced waste to the plant end to end.

Jimmy Carroll: [00:12:57] Yeah, I was going to ask about that too. From an environmental sustainability perspective, are there any real-life examples beyond what you’ve mentioned and maybe not, maybe they’re not the ones that you can talk about, but any real-life examples of manufacturers leveraging your technology to optimize processes and not only for their revenue but for the environment?

Berk Birand: [00:13:18] Absolutely. Absolutely. We do have definitely several specific examples. Before even getting to the examples, the interesting thing with Fero is that when your goal is to reduce waste, in fact, you can do both of these things at the same time. You can both reduce environmental waste and emissions and reduce your, you know, cost basis and thereby increase your profit at the same time. So, let me give you some concrete examples. One very clear example is around batch cycle times. So, we have one customer is a chemical company that’s using Fero to reduce their cycle times for, for their batches. So, instead of a batch taking, you know, eight hours, using Fero with recommendations and predictions, they can reduce it to, let’s say, you know, six hours, five, six hours, which is a considerable decrease. As a result, now you don’t have to run this entire plant for an additional two, three hours and spend all the energy. In this case, you know, the Scope 2 emissions that come from using, you know, powering the entire plant for those hours is now saved. You can also use that time to produce more if your plant is throughput constrained. So, as a result, this also helps your top line because now you’re producing more and you’re still saving on a per pound basis. You’re still saving on cost as well.

Berk Birand: [00:14:29] That’s one example in terms of cycle time. Another example, maybe again from steel, is in reducing the waste in terms of the raw material additives that they add in steel production. So, in steel again they melt scrap and then they add different ingredients. These different ingredients, these different alloys, have to be mined across the world, have to be processed, shipped across the world to get to the plant. And, you know, it has a different, a specific cost. But it also has a, in this case, a Scope 3 impact as well. So, what Fero does in this case, and this is now operating in about a dozen-plus steel plants, what Fero does is to tell them, actually for this particular batch of liquid steel, you don’t have to add, you know, 40 pounds of this expensive raw material, you can just add 20 pounds, and that’s going to be enough. And that’s going to be, you know, enough to hit your quality goals, and it’s going to save you this much in Scope 3 emissions and this much in dollars, and that you don’t have to, you know, spend to buy raw materials. So, again, these are kind of two examples that highlight how when you reduce waste, you can actually align reduction in emissions and increase in profitability at the same time.

Jimmy Carroll: [00:15:36] Yeah, that’s win-win for sure. You know keeping with the the steel industry, I wanted to ask about . . . You had you? I don’t know if it was you, but somebody at Fero had written something recently about greenwashing in the steel industry and, and sort of highlighting the need for regulations to ensure that, you know, steel manufacturers don’t stretch the truth when it comes to their eco-friendly efforts. Can you talk to me a little bit about this and maybe talk about how AI could factor into it, if it can?

Berk Birand: [00:16:06] Yeah, definitely. We definitely believe that regulation is part of the equation. And especially, this is especially complicated on the green steel side because broadly there’s two families of, types of steel making — one that essentially uses scrap to recycle steel and produce new products, another one to create new steel products from, you know, iron and carbon and creating from scratch. One of them uses very little energy and causes very little emissions. The other one causes a lot of emissions. And over the years, even before the current focus on climate change, different producers have focused on different parts of different production that made sense for their own geographies, for their availability of scrap, or their own customer base, the type of products that they build. So, coming up with a . . . The challenging part in all of this is to come up with a single standard that can work with both sides. In fact, many of the types of producers that used to do blast furnace style, iron and carbon type of production are now migrating to use scrap more, which is a good direction, but then that requires a lot of investment. So, then the challenge is, you know, okay, one group doesn’t have to do anything, so they can just keep doing the same thing.

Berk Birand: [00:17:18] The other one has to spend billions and billions of dollars. So, how do we create one standard that can reward both sides’ efforts and still drive towards a net-zero future? So, that’s the challenge. And I think there’s different philosophies in terms of how to achieve this. But we believe that AI can, for both sides, AI can play a key role, can of course play a key role in allowing for this circular type of production. There’s challenges inherent to recycled steel production. Namely, it’s very hard to have specific, high-quality product types because of course you have, you know, you’re dealing with whatever you have as scrap, which could be washing machine parts and car parts that come into your plant. So, AI can play a key role in allowing recycled steel, electric arc furnace type of steel, to reach higher quality targets. And AI can also help the remaining blast furnace tile production to still become more efficient until, again, eventually, potentially, it’s rolled into, it’s turned into an electric arc furnace type of production. But in both cases, reducing waste is always the right solution. And AI can help with that.

Jimmy Carroll: [00:18:25] It sounds like what you’re saying is AI is not going to destroy us all. It’s going to save the planet, in some ways, right?

Berk Birand: [00:18:33] Our type of AI. Yes. That part we’re very confident. I’m only responsible for our kind of AI.

Jimmy Carroll: [00:18:38] Fair enough. You know, one thing that, that comes up a lot in conversations I have and articles I read and talks I see and things like this is, there’s a lot of companies out there that want to adopt AI simply because they feel like they should or their competitors are, or they’re reading about it, or they’re reading reports about, you know, increased adoption or whatever. But AI is not a tool that works for everybody. People need to consider the desired outcome and do some sort of feasibility study or application analysis ahead of time. So, from your perspective, what are some potential roadblocks or pitfalls to avoid when it comes to AI adoption?

Berk Birand: [00:19:22] Absolutely. Yeah, I would say there’s definitely a lot of discussions around AI these days. And it’s important for potential prospects or people that are interested in AI to have a good sense of whether it makes sense for them to use AI or not. I would say the two things to pay attention to — one of them is data readiness. So, how much data is already available for this purpose? And even beyond just the number of sensors available on-site, can manufacturers track product as it goes through different stages of production? You know, if there’s a container and then at some point everything gets mixed up in a container, and you can’t really trace back the different ingredients to their sources or different previous processing steps, that might be a first place to think about digitizing production, to be able to track and trace things better or add sensors and become eventually, you know, get to a point where one has a solid base, a solid foundation for data availability. In most of the companies that most of the sectors are focused on, this is usually the case that usually factories already have many, many years of production data that has been sitting on-site and that’s being used for different types of analyses already. But that’s definitely an important point. The challenging thing with data readiness is that it’s very hard to say whether a factory has htis data ready by just looking at the number of sensors they have or amount of data that they have because they could be storing a lot of data.

Berk Birand: [00:20:47] But this may not be, you know, the perfect kind of data for the kind of use case that they want to address. So, the best way to test it, to figure out if they get a thumbs up, thumbs down, is actually to just put it in an AI system and see if it works or not. You know, as dumb as that sounds, you know, just try it out. And ultimately, our goal at Fero is to lower the barrier as much as possible for this particular step. We have this feasibility study step in a conversation with potential customers where we do this for them and give them a thumbs up or thumbs down. And ultimately, you know, if the data is not ready, then it doesn’t make sense for us to also continue the engagement and provide recommendations, but we can then circle back when they are ready. And the other thing is to focus on the use case. You know, what is the expected ROI from applying AI? This is the kind of, you know, in the typical Six Sigma, this is the first step, the defining step.

Berk Birand: [00:21:35] What are we doing here? It doesn’t make sense to do AI just for the sake of AI or to apply AI just because you have data to do AI. What is the expected outcome? How good should the models be for these models to drive value? As an example, we have instances where customers are deciding, making decisions on-site based on legacy knowledge, you know, without looking at numbers. So, in that case, having an AI model that’s even 60% accurate is still something. It’s still an additional piece of information. And that can drive decision making in the plant. In other cases, you really need a model to be 80, 90% accurate because they already have good models. They already are running the production in a good way. It’s already stable, and the processes are stable. So, having a sense for, what are we trying to do, what is the outcome, and how good should the models be for this to make sense, and what’s the cost basis for a good prediction or bad prediction and a false positive, false negative?

Berk Birand: [00:22:34] That’s the, I would say, the use case angle. And we have a large catalog of use cases and blueprints on our website. And of course happy to send more for people that are interested.

Jimmy Carroll: [00:22:45] Yeah, well, I would encourage everyone to check that out, but, it’s a good point, right? Because I guess in some ways it starts with data. If you don’t have data, there’s no AI. If you don’t have good data, there’s no AI. It’s that, it’s that cliché, overused term, right? But it’s garbage in, garbage out. It’s not going to serve you if you don’t have the proper data. So, it’s a good point.

Berk Birand: [00:23:08] Exactly.

Jimmy Carroll: [00:23:09] I think you guys did a study recently in industrial AI, right. What are some of the big takeaways there? I think one of the stats said that less than 20% of people were less than 20% of the manufacturers that you surveyed are using AI. Is that is that right? And why do you think that’s so slow?

Berk Birand: [00:23:28] Absolutely. Yeah, I think that’s exactly right. Eighteen percent is the number that came out of our survey of many customers and prospects that we polled. So, the I would say, first of all, there are many problems in the industrial sector that are worth addressing that are not AI and data problems — problems around change management, training, onboarding, labor shortages. You know, there’s a lot of hiring problems. So, there’s a lot of problems for which AI is not the main solution. But there are many problems for which AI is the right solution. Anything where, anytime where we have customers, for instance, that say they’ve been running production with poor quality production at low throughput for a year-and-a-half until they figured out the source of the problem and then they fixed it. But then it took a year-and-a-half of low throughput production to get to that point, for example. This is a perfect situation where I could have helped . . . would have helped. So, I would say the reason why the numbers are so low, first of all, is, well, the black box points that we spoke about earlier is definitely one. Understandably, the industrial sector is not an early adopter for many technologies, which is, you know, natural thing to do when you’re dealing with high temperatures and potential safety issues and potential loss of production, there’s less of a incentive to really do things drastically differently.

Berk Birand: [00:24:46] But on the other side, I think what the pandemic showed everyone, especially the supply shocks that came after the pandemic, showed everyone that things are moving fast, and you need to be adaptive. You need to adapt to different changes in prices, changes in your supply chain. And ultimately, what we like to say is that what made manufacturers successful over the past 30, 40 years is not necessarily the thing that’s going to make them successful for the next 30, 40 years. And AI is an instrumental part of this equation that could help quite a bit if it’s done in a trustworthy, explainable way, which is exactly how we’re approaching things. And ultimately, yeah, I think we, you know, an interesting insight that came out of our survey was that plant managers play a key role in digitization and being enablers of these kinds of technologies. And they sit at the perfect spot where they have full context and knowledge of their sites, they have ownership, and they play a key role. And they are the folks that answered the survey that were successful at doing AI did so usually when there was a plant manager that was supportive and kind of put their weight behind these kinds of projects.

Jimmy Carroll: [00:25:59] So, I guess the part two of that question is in terms of AI adoption being slow, what could lead it to speed up over time? It’ll take time, obviously, but what’s going to get it there?

Berk Birand: [00:26:13] I think that improving the data infrastructure is a key component that has been happening over the past few years. What we’re seeing is that there’s some of the larger kind of manufacturers that have already had a solid data infrastructure, but progressively, medium-sized and even smaller type s of production facilities are also becoming very digital, and there’s a lot of many companies and many startups and large companies alike that are helping this transition. So, ultimately, these folks are going to become, get to a point where, you know, some of the largest chemicals companies already have been for the past few years. And AI plays a key role in both types of production ,on the large chemical side. And we target many of these large companies and work with many of these large companies. They already have data science teams. What they’re interested in is to have their data scientist teams work on use cases of AI that are, that will be their competitive advantage. So, multiyear R&D types of projects that involves first principles and experimentation and R&D. They don’t want to have their data scientists work on building small models, hundreds of small models for predicting the yield or quality and so forth. So, they like Fero for that aspect and the explainability of it. But for small manufacturers, they might not have data scientists on staff. And for them, this is a way for them to get into this digital transition and really use AI to improve their production as well.

Jimmy Carroll: [00:27:35] For sure. I wanted to ask about the labor shortage. And it’s funny too, because like we were talking about before, like other parts of AI — AI deployed in autonomous mobile robots or with vision systems or, or like with what Fero does. It all kind of serves the same audience in terms of manufacturers and logistics and warehousing businesses worldwide. And you mentioned earlier COVID too, right? So, automation technologies and AI of all sorts, these, these technologies really saw a huge uptick in adoption around COVID. And while it would be easy to say that it was because people couldn’t work or didn’t want to work, I think these technologies, and I know I’m not alone in this, but I think these technologies were ripe for growth anyway, especially in certain key industries. And now we’re seeing that there are just people that, that, you know, they don’t want to work in a lot of these jobs, whatever they may be, you know, manual labor, dull, dirty, dangerous, or those, the terms they use for robots. But even, not so much on that side, but overall, with labor shortage, what role will AI play both in terms of what you do and in the larger industrial automation space, in ensuring that these companies don’t suffer from a shortage of skilled workers?

Berk Birand: [00:29:02] Absolutely. Yeah. So there’s definitely, you know, there’s all these projections around the millions of positions that will be left unfilled in the next 10 years. But many of our customers are feeling this even today and are projected to feel it more and more in the next few years. So, we believe AI can play a key role in this in terms of educating a new generation of operators and engineers to drive their plants to operate more effectively. The interesting angle here is, of course, that, you know, let’s say, take an example of a factory that has an experienced staff that’s been running the plant perfectly for the past 10 years. They’ve been collecting data over the past 10 years. So, it’s almost as though the decisions that the experienced staff has been making are implicitly in the data, in the set points that they chose to run the plant at under certain circumstances. One way to transfer this knowledge to the next generation is actually to take all the data that they have been producing, train AI models, and then use the AI models to make recommendations to the next generation. And these recommendations, inherently in them, will have the collective knowledge of all of the previous 10 years’ worth of, you know, experienced staff in them as well.

Berk Birand: [00:30:15] So, that’s one example where we believe AI can play a key role. And the other part that we’re really excited about is effectively to use AI as a co-pilot of sorts The drive to help with diagnostics type of use case. There’s a lot of discussion that, there have been a lot of discussions around prescriptive analytics, predictive analytics. You know, let’s have AI make decisions and run the factory. And we have customers that use Fero to, in this way ,to live optimize the factory continuously. But we also have a separate product which focuses on allowing engineers to diagnose the root cause of problems. So, it’s not in the real-time critical path of the factory’s operations. It’s more helping, you know, the offline engineer answer questions like, hey, what was wrong on this particular point in time? You know, what led to these issues happening and effectively act as a diagnostics co-pilot, diagnostics partner, which can then again help engineers come to the resolution of problems much, much faster in the next few years.

Jimmy Carroll: [00:31:16] Interesting. Yeah, Berk, I want to be respectful of your time, but I do want to ask this question, and we kind of touched on it earlier, but it’s a fun one. You know, when you’re talking to somebody — whether it’s, whatever, at a dinner party or if you have kids at a school event or whoever, somebody that’s sort of outside of the industrial space — and you explain that, that you’re involved in AI, do they get afraid? And how do you explain to them that AI is not something to be afraid of? It’s something that’s here to sort of help serve humanity.

Berk Birand: [00:31:47] Yeah, that’s a great question. I do get that question asked quite a bit. And typically I think there’s a lot of discussion in the tech sphere, mostly, around this, you know, superintelligence, AGI. And, you know, are we trending there, are we are we building, you know, the structure, the software that’s going to cause our own destruction? I think, you know, this is like, this is, to me, I have to take off my tech CEO hat and put on my sci-fi hat. I don’t know. I’m not worried about these situations. And I really think that there’s way more serious threats to humanity. I mean, climate change and nuclear war and so forth that are much more likely than AI taking over. But I think for most people, you know, when they, I think you alluded to this earlier as well, you know, that ChatGPT was a big phase transition where people realized, wait a second, this is, this is actually quite good. You know, this is almost humanlike. But then occasionally as you start using it more and more, it’s giving you some answers, and you’re like, wait a second, this is not that smart. Even, like you can’t even do multiplication reliably. So, is this the thing that’s going to be, you know, the first step in what’s going to cause the demise of humanity? I’m not convinced that the models today are going to be the same models that will be the form of general intelligence. There are many people that believe that, you know, the models today are, you know, almost that kind of are going to start plateauing from now on just because there’s not we don’t have that much more data to funnel into them. And that some major breakthrough needs to take place for us to go from where we are to even something close to a general intelligence. So, yeah, overall, I mean, usually I think, the short answer is, I know this space quite well, and I’m not very worried. I think there’s more important threats, and I don’t think they should be worried either.

Jimmy Carroll: [00:33:29] I couldn’t agree more. Berk, anything else that we haven’t touched on that you, that you feel like you want to mention before we before we close out here?

Berk Birand: [00:33:39] No, I think we touched on most things. I think that, you know, we’ve been very excited to be bringing AI technology to the industrial world. As you know, the industrial world is, you know, the majority of the world’s GDP runs through industry, but it’s a world that most people are not that familiar with. So, I’m grateful for folks like yourself that are, you know, shining a spotlight on the manufacturing sector, for other practitioners, and the rest of the world alike. So, I appreciate you and appreciate you for having me.

Jimmy Carroll: [00:34:04] Of course, I appreciate your time. It’s been fun. If people want to learn more, they can go to FeroLabs.com or check you out on LinkedIn. If you have specific questions for Berk, you can reach out to us at Manufacturing-Matters.com. We’d be happy to pass those along. And thanks for either listening or watching.