The Next Industrial Revolution is Industrial AI with Barbara Humpton, CEO at Siemens USA and Olympia Brikis, Director of Industrial AI at Siemens USA

ssible before for someone working in a factory to be able to predict when a machine was likely to break? Would it be possible actually to go from mass production to mass customization? These are the kinds of things that are now possible because we have the ability through AI to manage large quantities of data in the blink of an eye.

Richie Cotton: That's really interesting, the idea that you're putting the concept of an AI revolution, the successor to the previous industrial revolutions. You talked about trains, you talked about electrification, and I like that it's one in the sort of long line of technology improvements, helping the world.

But I'd like a little bit more concrete motivation. Can you gimme some success stories? I'd like to hear a little bit about how you've been helping real businesses to, benefit from using ai? 

Barbara Humpton: Sure. And let me tell one story and then I'm gonna turn to Olympia for another.

The, we have an automotive customer who had a problem with doors coming off the production line. They had people visually inspecting those doors as they came off the line. As you can well imagine. It's hard to detect errors, and when you do, then who knows how many other doors have already had those same errors.

They sought help from consulting firms who, one after another said, oh, for a price of X or YI can help you find one. Or I'll find two errors, for x plus Y price. And Siemens Advanta Consulting actually came at the problem differently with AI said to the customer. What if you define perfect and we actually test every door against the standard of perfect.

So with a visual it's inspection, cameras plus ai, now that same manufacturer is producing perfect doors and as soon as there's an error it's detected and they can make changes on the production line. But honestly, I think Olympia's got a great example where she's had firsthand experience.

Olympia Brikis: And that's in one of our own plans because we don't only build this technology for our customers, we also apply it in our many own manufacturing plans for electrification, automation equipment. And one of those plans is our plant in Amba in Germany, where we have worked with the same amount of people in the plant for the last decade.

And this had come at a pace of rapid transformation for the plant. So where it had in the past not been automated at all, it is now fully automated. And you can imagine that the staff working there now, instead of ma doing manual inspection jobs, instead of doing assembly manually, instead of doing quality checks or predictive maintenance operations, they focus in on.

Other tasks on tasks that are more cognitively challenging. For instance, when a piece is detected by an AI system to be faulty, they are the ones figuring out how can I rework the piece to lose as little, scrub as little material as possible and keep the throughput that I'm trying to achieve for that day.

And similarly today, we are introducing generative AI assistance into these type of plans so that the operators don't only have to make that decision on their own, but they're actually giving assistance from documentation from past experiences that we have or from maybe there are very experienced counterparts that have been at the plant for many years, but there is another plan where we have been innovating quite a lot.

Barbara, I know this is one of your favorite examples from the us from Fort Worth. 

Barbara Humpton: Yes, at Fort Worth, we had a hyperscaler customer who said, I need as much electrical switchgear as you can make. Data centers, especially AI data centers, consume a lot of power. That power has to be distributed throughout the data center and Siemens makes that low voltage switchgear it's called.

And so we needed a new production line and the team went to work right away. First creating a digital twin of the ultimate. Product. They wanted to produce a standard product. And then likewise, the production line that would produce it where our Siemens real estate team found an existing industrial plant an industrial park where there was space available.

And in 15 months they were able to stand up that production line. And now we have about 500 people working there. It's interesting because yes AI brings a lot of power, but just as Olympia said. The people now are using higher order thinking. It's all about making sure that they're accomplishing what they're intending to accomplish.

It's not the grit and dirt of our grandfather's manufacturing environment. And by the way, Richie, one of the things in that. Plant in Oberg and now, here in Fort Worth, that big question of what, what will happen to people? Will automation take jobs? What I've heard is that in Oberg, they have the same size staff that they had decades before, but they're producing something on the order of 10 x the amount of product that then was ever possible before.

Through the aid of this kind of automation. And then likewise, at Fort Worth, we currently have 500. We expect that we're gonna expand to 800 as we add more lines, and as we just have an insatiable demand for electrification here in the United States. 

Richie Cotton: Lots of great examples there. I love the idea of the the car example where you were looking at quality control and making sure that it was a lot easier to solve those issues.

I like the idea of increasing productivity per worker and also just getting some of those boring tasks. Certainly inspections doesn't sound like a very interesting thing to do. I know you make use of the term industrial ai. Can you talk me through what that involves? Olympia, do you wanna take this?

Olympia Brikis: Great question. So think of it this way. Consumer AI might be the thing that you use to get the recommendation for the best pizza place in your town. Industrial AI is the AI that will make sure that the pizza comes out perfectly crisp out of the factory, that your oven consumes the right amount of temperature and saves energy while it bakes that pizza.

And of course, it doesn't burn it. So in other words. While consumer AI is built for applications that you and me use in our day-to-day for our own productivity improvements, industrial AI is built to run in industrial environments and support engineers and operators in their day-to-day tasks. So those can range from operating a power plant or an industrial facility an electrical grid to design and engineering processes that we have when we wanna come up with products that we're building.

And technically the major difference between those two forms of AI is where the data comes from, that we're using it. So consumer AI is typically fueled by the large amount of data and information that we have out there on the internet. While industrial AI is fueled by industry specific data, which is usually proprietary, it means the internet does not know how.

One of our automotive customers produces their cars or what the geometry of a certain product is that you're buying. So that data is the data that fuels ai, industrial ai. And this data is quite different. It's not only just proprietary, but it's other modalities. So it's not just text and images like the data sets that we commonly use in our personal lives when we deal with ai, but it is also time series.

Sensor values, diagrams engineering drawings, and all of these special type of data points that require a lot of knowledge to understand and to interpret. And the last difference I wanna point to is that with consumer ai and with other applications of AI that we have in our day-to-day.

Usually the accuracy that we are looking at and the preciseness of the AI that we target is not as high as it needs to be in an industrial space. So we have here much higher requirements in terms of the accuracy because the processes and the decisions that we're making in the industrial space usually have much higher risk rates and much higher cost if we take on decisions.

Richie Cotton: Okay. That's fascinating. The idea that using different types of data. So I guess, yeah, sensor data's gotta be incredibly important in a factory or manufacturing setting. And then, yeah, that's interesting that you mentioned, diagrams as well. Does AI help with like the design phase then for manufacturing world?

Does it help you set up plants? 

Olympia Brikis: Absolutely, yes. So it can help you design better parts. So for instance, if you wanna optimize for a certain characteristic, like aerodynamics of a part, or you wanna optimize for material use, you can have AI that supports you with that. So we, we use this AI inside our Siemens design tools to support you as a design engineer or as a simulation engineer, depending on what step of the process you're in.

But then it can also help with other types of engineering processes like actual engineering, like systems engineering of an entire plant or an entire facility, 

Barbara Humpton: Olympia. Let me build on that then with a story about one of our customers at CES. This year we introduced Jet Zero. Jet Zero is an aerospace company.

What they've been focused on is producing a blended wing aircraft, one that would be 50% more energy efficient, and quite frankly, more comfortable. So they actually used our design tools just as Olympia described. You may know, Richie, that over the last several decades, Siemens has been pulling together the best of the best in the companies who have been developing the underlying technology we now refer to as the digital twin.

So this idea that with the right design tools, you can actually make a virtual model of just about anything you might want to build it. It turns out Siemens has the most comprehensive physics-based digital twin in the world. We recently actually acquired Altair. Complimenting our own skills with those of Altair engineering.

Adding even more power to the fingertips of engineers who are busy in the design process. Imagine Jet Zero working on this innovative breakthrough in aerospace. They have had the ability to use not only classic design, an engineer saying, please, model me this, but also generative design.

Hey, I've got a problem. I need to solve a problem, but it's gotta fit within these energy constraints at this weight. And the generative design tools that Siemens has built into this portfolio enable them to then do things that perhaps any, any of the engineers might not have produced themselves.

I've seen these kinds of debates amongst engineers, oh no, we make need to make that horizontal. No. We need to make it vertical. And then say, hand it over to the design tool. And you discover that using. Generative design AI enabled, right? We end up with something that frankly looks a lot more like something Mother Nature would produce.

So it's fascinating to see these tools in action and here, jet Zero has put that to work not only in building their pro, in designing their product and getting ready to build it, but also then in designing the manufacturing plant that will produce that plane. So look for more news from Jet Zero.

Richie Cotton: That's such a wide range of different use cases there. So something like I guess doing predictive analytics on some sensor data to see when a component could fail. That seems like fairly simple building, like a whole digital twin of an aircraft. That sounds like a big undertaking.

And I can imagine it's very cool that you can plant different scenarios, work out how to optimize the, this blended wing airplane. But talk me through, how do you go up building something like that? 'cause it seems huge. Where'd you get started? 

Olympia Brikis: Yeah so you're definitely right coming up with a comprehensive digital twin that really describes all the characteristics of your plan.

Not only the geometry, but also the physical behavior and different type of conditions is a process that takes time. So what we usually do is go, engineers will go step by step through it, so we guide you step by step through that as well. And one of the things that we see happening right now, which is maybe an interesting kind of future and forward looking change in the technologies, we see AI helping with that entire process as well.

We see actually AI agents stepping in and taking on various tasks in this entire complicated chain of. Things that you need to compile to come up with that digital twin. So we might have an AI assistant or an agent that walks you through your geometric design. And then we have another one that can take that input and already come up with a first setup of your simulation that you might need to come up with aerodynamic behavior of that car, for instance, that you are, that you just designed.

And so we see that AI workflow automation is helping us more seamlessly connect the different steps that you would have to go through manually in the past to come up with that really holistic and comprehensive digital twin. 

Barbara Humpton: Richie, if you think about, if you think about the way things were designed and built in the past and think about, oh, the Mars r.

NASA and Jet Propulsion Labs have worked together for decades on amazing missions all over the place, including Mars. And often in the past that was very much a manual, experiment, build, try, fail, build again, and particularly for designing a Mars Rover. The challenges were especially large because, yes, we could construct something here on earth, but how would we know how would perform in the environment of another planet?

And it turns out this is the power of our digital twins. The team was able to obviously capture the digital twin and your question about how do they go about doing this in. In the case of the way NASA and JPL work, of course they've got subsystems, right? And there are teams of people who own subsystems.

And and then the question is, how do those subsystems come together? Of course, doing all that virtually is really powerful as app and far less expensive than doing it with real materials in the real world. I'm sure we could spend a whole episode just on the Mars Rover, but the magic of this was then being able to simulate how it would perform in a.

Different environment, different gravitational forces, et cetera. They had to model the Mars rover landing. Do you remember the seven minutes of terror when the rover lost contact with mission control as it came through the Mars atmosphere? It took, thousands of permutations, thousands of tests to settle on the right angle of entry, speed, et cetera.

The right use of boosters to ensure that it would be a soft landing on the planet. And lo and behold. Lo and behold, after all that simulation, it was successful and those rovers have really performed beyond anyone's expectation. I it's interesting how as people, we accomplish these incredibly complex feats, but it's a matter of building things up.

Bit by bit. 

Richie Cotton: Yeah, certainly some amazing accomplishments in spaces. Yeah, those Mars rovers were pretty incredible. And it does make you feel anytime you think my job's pretty hard. At least I'm having to send a rover to Mars. It's it helps helps you feel good about yourself.

But, okay. Suppose you are running you're running a factory. You're like, okay, I want to get all in on, on ai. What do you do first? What's step one to start introducing AI into your processes? 

Barbara Humpton: Yeah. Let me start from the what I'll give you as the business perspective on this, because Siemens many of our facilities are going through exactly this process now, right?

As all of us who've been manufacturing for. Centuries and search certainly for decades. We've got equipment in our manufacturing environments that are old. Maybe they're analog. So you know that AI operates on data. The real tough part of this whole transition is actually harnessing the data.

And in fact, one of the big breakthroughs at Siemens is now bringing together the data fabric. Recognizing that by deploying sensors on older equipment or connecting new equipment that is actually already quite digital, we now have the ability to pull together data from these many sources. Once we have data, it's actually not that difficult to take the next step to begin to apply the tools of artificial intelligence.

Olympia. You wanna pick it up from here and talk about the technical steps? 

Olympia Brikis: Yeah, absolutely. I would add that the next thing that you have to think about. Because it will inform what type of AI tools you select and what type of data you apply them to is what is your ultimate goal? What is the most important thing that you're trying to improve?

And we'll see that manufacturers in different industries have different things that they are worried about or that are concerning to them that they wanna automate. This could be throughput. Industries with very high demand. This could be quality in very specialized industries. It could be the time it takes to rework or to inspect, or it could be material waste or energy waste that is happening, or consumptions of materials in general.

So we see that depending on what type of manufacturer you are. Different things will matter to you. And I think that's the next step after making sure your data is in order and it's able to be used, is to figure out what is your, what is the KPI you are trying to optimize? 'cause that will allow us then to select, like Barbara said, from a wide variety of tools that are available that are not the tricky part of the process anymore, which one to achieve the goals that you are trying to achieve and to really get that successful implementation.

And I talked about the business KPIs on the one side. So what is your main target that you wanna achieve? And then the other side that I would wanna mention since we are getting onto the topic of KPIs is adoption KPIs. So no automation technology really can be successful if it's not adopted in the plant by the people in the plant.

And so what we will always recommend to our own facilities as well as to our customers when we work with them, is to think about that side of the pie as well. And already from the beginning. Come up with metrics and come up with goals that we have in terms of the adoption. This is even more important when we're talking about ai, that still requires the human in the loop for either feedback or for prompting or for initiating the AI's use in the manufacturing process.

Barbara Humpton: Richie, I'd love to bring this to life with a story. Siemens had begun working with a startup in the vertical farming space. And when they put up one of their farms, think of it as a, it's actually like a data center, but instead of bits and bytes, it's just producing bites.

That's the joke. But anyway, the the, these folks in the east. High stacked greenhouse type environments under very controlled conditions. They can produce plants meeting very specific specifications just by controlling the light, the water, when nutrients are added, et cetera. But at the end of the process, they package up their goods.

And we again, talking to this. Customer of ours. And they had not actually collected any data on their packaging line. Once they did, they discovered that every container they were shipping out contained about 40% more. Product than the net weight advertised on the packaging. Think about that.

Something as simple as that. They were able to actually get 40% more product out the door by packaging at the appropriate weights. These are really simple automation steps to take. I really think the magic happens when you empower the people on the manufacturing line to start asking those, what if questions?

Once people realize and begin to adopt the tools and discover that this isn't scary, right? This isn't something where, oh gosh, you're gonna have to go back to school and get a degree to, to be productive in this environment. No, the technology that we're talking about has user interfaces that can feel more like playing a video game or.

Or are frankly simply like talking, using natural language with interpretation behind the scenes via a copilot. I'm excited that this is very approachable and that if manufacturers will think first about as Olympia says, what's the problem I'm trying to solve? Then go solve that data problem.

Apply the right technology for solving the problem. I think what they'll discover is their own teams can be pulling the technology into the environment. 

Richie Cotton: That's a great story and certainly, yeah. Seems very important to put the right labels on the packaging of how much we're actually selling. So I relent that approach there of think about what's the business problem you're trying to solve first and then worry about the data and then only worry about AI after that.

So that seems like a sens board of things. I'm curious as to how you need to go about worrying about getting the technology there. I know when you've got manufacturing plants, often you've gotta think about building physical machines or putting actual sensors in place. That can take a while.

Talk me through, how do you think about that, that physical hardware technology side of thing, and how does that affect your adoption of ai? 

Barbara Humpton: We've got obviously, and especially right now in the United States, we, I think the last tally I saw said $7 trillion of planned construction of manufacturing or investments in the US in all kinds of critical.

Critical technologies. So we know that where companies are looking to expand their operations in the United States, often they're looking at Greenfield. And when they look at Greenfield, what we hope is that they're taking a look at the kind of already built in technology features that are available today.

A more highly automated line so that the data collection is already much easier. What's interesting, Richie, is that. Mo, that's not actually the bulk of the manufacturing that gets done in the us. 98% of all manufacturers in the United States are small and medium enterprises. They're actually existing enterprises.

You can talk about them as Brownfield. And so the question there is what does it take to get those manufacturers into this new era of manufacturing? How can we make them more productive? And we're actually thinking through the non-technical barriers to entry. The technology exists, it's here, it's accessible, but what would it take, what kind of capital is needed?

To make the shift. So we've got Siemens financial services busy working with the small business administration and other capital partners, looking for ways that we can make capital available so that for instance manufacturers can get automation as a service. They can pay for it as they use it so that hey, as their own operations improve and their own, they can grow and become more profitable.

And yes, indeed, it will recoup the expense of that automation over time, or likewise, the people who are needed in these growing operations. And here, what we're looking to do at Siemens is make our learning modules accessible. 

Richie Cotton: Okay. Alright. Do you wanna tell me a bit more about the technologies?

Because when I think of ai in manufacturing, I'm thinking it's a lot of like computer vision and then a lot of, I guess like other cancer sensors or as well heat sensors, pressure sensors, all that kind of stuff. Yeah. Just talk me through what kind of tech is involved particularly like you mentioned a lot of it's like small businesses or existing businesses that haven't adopted the latest round of technology.

What sort of tech are we talking here to make this work? 

Olympia Brikis: So yes, rich, you're right. One of the big technology staples in the industrial AI space is computer vision. For instance, how we approach that at Siemens is we give you a bundled up solution that consists of the physical equipment that you need to install an inspection system.

So meaning the camera, the IPC that runs your AI model, but we also give you the AI model, an AI model that already saw a lot of industrial defects. And a lot of types of pieces from plastic to metals, so it's familiar with the industrial domain. And then you are walked through a nice user interface through the adoption process of that model to your own production line.

And this adoption process is very stimulus meaning you don't need to create a lot of additional data. You just need to show the model a couple. 10 to 20 maximum examples of what your desired output is. So what is your part look like if it's perfectly produced? And then the model can train in the background on that, adapt on that information from your own production line, and be deployed essentially in in a day to your manufacturing line.

So that's one example. We also have a portfolio where we allow you to come with your own model that you wanna deploy at scale. This is particularly interesting for manufacturers that have already invested. In developing models that are specialized to their manufacturing process. So then we come with the inference services and the monitoring services for these models to allow you to deploy them in one factory and then also easily scale that to other lines or other manufacturing facilities that you might be operating that would use that same kind of model.

And the third, and I guess for me personally, very exciting technology that we're building right now is generative AI more. The manufacturing floor. And this is here technology from us that we built into our industrial copilots. So we basically take the capabilities of large language models and other foundational models such as vision action models and bring them to shop floor applications.

So we bring them to our robotic arms, we bring them to our HMI, so that users can interact with this technology and get assistance. Live as they're operating and as they're troubleshooting issues. So one example for instance of where we use this technology is for maintenance. So if you have maintenance engineers today, the way they usually work is they will go up to a machine after they receive a maintenance request from somebody on the floor, and they will have to troubleshoot that issue.

If we're lucky it's something they saw before and they can fix it quite quickly. If we're unlucky, they need to do quite a few steps to get to an answer. They might have to look up the manuals of that machine. They might have to look at operational data of the machine. Maybe they have to even go back to their desk to a laptop to actually be able to we and investigate that data.

And then thirdly, they might need to take something apart, physically apart to be able to fix it. So that means that downtime issues can actually take a long time to resolve, especially if there's, if we're in production environments that have a lot of variables that can influence the the production process and what we do now, how we assist these.

Of these maintenance engineers today with the help of generative AI is that we allow them to take an assistant with them, basically on a tablet or whatever device they're using. And that assistant, that generative AI assisted is connected to all of these data sources that they might need to investigate the issue.

So that means it's connected to the documentation of the plant. It's connected to our operational data. That could maybe be coming from a SCADA layer or a data lake that operator already created, or it comes directly from the PLC. If we have not done that aggregation yet, and the chatbot or the assistant now is able to pull from that data on requests of.

Maintenance engineer, the maintenance engineer can formulate those requests in natural language. So this can be as simple as say, okay, what's wrong with this machine? Why is it, why did it stop? It can get more complicated if you want to actually figure out what were maybe common issues that you saw on this machine in the past couple of weeks?

Is there anything I should be aware of that my predecessor and shift one did to, to this machine? So this assistant basically has all of this knowledge and can serve as an interface, as a natural language interface to our maintenance engineers. So this is one of the, one of the more recent technologies we have been working on specifically for manufacturing.

That, yeah, we're excited about. We're starting to use it in our own factories. And we have, yeah, quite a few stories that I could share of like how people are liking it, how they're starting to use it and how adoption is shaping out. 

Richie Cotton: But yeah, customer service chatbots are one of the big use cases for generative ai.

I suppose. It shouldn't surprise me that they're also coming to the factory floor as well. So yeah, I like the idea you can have your bot on a, on a. Tablet or something, just take it round with you and help diagnose problems with machines or problems with the assembly line.

Barbara, before you were saying how half, the secret to this is about empowering your workers to make process changes I'd love to know a bit about that. I suspect there can be a lot of workers, whether oof, I'm not sure about making changes, having AI replace my job. Can you talk about overcoming AI hesitancy and making sure that.

All your workers are comfortable in updating things to make them more productive? 

Barbara Humpton: Yeah. First of all, let's talk about the jobs because I think, all audiences need to understand that there are more jobs to do than we have people to do them. E, especially in manufacturing, what we actually need is the technology to expand what's humanly possible.

What we wanna do is make every person more productive in the manufacturing environment. So you think about something like our Fort Worth facility. I'll go back there. And by the way, on social media, you'll find all kinds of great stories about how we approach this in Fort Worth. It was very much a decision by the leaders closest to the problem they were trying to solve, who said?

Where are we gonna go to get 500 people? And it turns out their first hires were retired high school administrators and teachers. What those folks did was actually create an onsite training program. Now, this is for a brand new workforce. It turns out that if you're looking at a workforce that's already in place and rocking and rolling right along, and you're asking yourself, how do I get people to up their adoption rate, it could look a little different.

What I tend to believe in is multi-generational teams. In a manufacturing environment, you have veterans of that environment who are masters at their craft, who know deeply the work that needs to be done. But you bring a digital native onto the manufacturing floor, give that person some tools that they can literally just begin to interact with the more experienced manufacturers and, and start brainstorming together about what needs to be done.

In Siemens USA, we spend $37 million a year on training. But the real magic isn't time we spend in classes and looking at formal training, our research shows that's only 10% of the formula for building true knowledge. The second part of it is what we think of as a social network being around people who are masters in that craft.

So the digital natives who know how to use the AI tools, the experienced manufacturers who know how to train in the actual manufacturing environment, and then ultimately the best teacher is on the job experience. So ultimately I think that manufacturers can start to think differently about the workforce they're trying to build.

What we've decided is we're looking for people who show curiosity and initiative, and what we found is that if they have a growth mindset. They can learn anything. 

Richie Cotton: Yeah. Certainly training is very dear to our heart at Data Camp. I'd like to know a little bit more about this. Are there any particular skills you think are very useful to make use of industrial ai?

Either if you are working in a factory or if you're interested in the the technical side of things? Yeah. Olympia, do you wanna take this? 

Olympia Brikis: So I think on the, if you are a non-technical person coming into the field, so you're not a computer scientist, you're not a, maybe a digital native coming in with that technology, I think the most important skill and the most important mindset you gotta have is the curiosity and is the willingness to learn and to try things out.

So that's gonna really take you very far with getting to know the technology and figuring out ways. In which it is assisting you. On the other hand, if you are somebody who wants to become more active in the development of technology, I usually think of it there being two pathways into the field.

One is of course the one that we usually think about first, and that's the one is a computer scientist. As an expert in ai, in software and technology, you can take that path. But the world is not close to you if you did not take that path. There's another way in, and the other way is through the domain that you are talking about or that we're applying the technology to.

Especially in the industrial world, because the industrial context in which we're operating is so complicated and very few people have knowledge of that. Having this knowledge can make you an very attractive and important partner in shaping out this technology. So I always encourage people that have that practical experience to not cut themselves out of the creation process of these technologies because they're important partners to somebody like me who whose background is in AI and computer science to really build technology that works in the day-to-day.

I have a maybe interest funny anecdote to share for how a lot of this comes together then in, in practice from one of our own plans in Langan. So this plan was one of our early adopters of the industrial copilot for operation, so that's our generative AI technology for operators. And we accompanied the rollout of this technology with a study that we did where we asked the users of the technology periodically how they're feeling about it, how they are whether they find it useful, and whether they how they perceive the technology changing their feeling of job security and stress at work.

We saw some very promising trends at our own factory trends of higher job satisfaction and less stress in the day to day, but as well as higher productivity. But at the same time, we noticed something that we couldn't explain it first, and that was we saw people in the night shift using the industrial copilot for operations way more frequently than our operators in the daytime shift.

So we were like looking at the data, we're double checking it. We knew that this was really happening and we couldn't configure out why. And so we went back to the plant and we asked the people from the night shift, why are you using this copilot so much compared to your peers? The others are not using it as much.

They said that's very simple. During the night shift, the senior operators, the very experienced operators that Barbara talked about, the veterans of the plant that know how to fix things quickly because they have been around for a long time, they are at home sleeping, and so are the process engineers and everyone else that does that has the classical night to five hours and their work schedule.

So they're all at home and of course there is numbers to call and there is people will answer and emergencies. But what the staff at night really appreciated about the copilot for operations is that they did not need to wake up their colleagues. They were now able and empowered to solve the problem on their own, and they were super happy to take that opportunity.

So they were not discouraged at all. They saw it as an added value to them and to the work-life balance of their colleagues and their ability to create a good and stress-free work environment for everyone involved. 

Barbara Humpton: So I love that story. I love that story. And the day shift gets a good night's sleep.

Richie Cotton: Alright, so it sounds like a brilliant set there. If you are not a technical person, all you need to worry about is being curious. Understanding some of the use cases for AI at a technical level, having that domain knowledge as well as some sort of AI skills, that's gonna be really useful.

Alright just to wrap up, what are you most excited about in the world of industrial ai? Barbara, do you wanna go first? 

Barbara Humpton: Yeah, I'm excited about, again, I'm gonna go back to where we started, what's possible that we're not even imagining today. I'm hearing things from innovators about, being able to develop new materials.

Being able to solve medical questions that have never been solved before. I am, I'm excited that innovators everywhere now have a new set of tools in their hands and they're off and creating a future. I can't wait to live it. 

Richie Cotton: Having new tools to play with to play with in order to make better things.

That seems a wonderful thing. Olympia, how about you? What are you excited about? 

Olympia Brikis: Yeah I love that Barbara. I'm on the same page. I think this technology is very exciting for innovation. It's also very exciting for inclusivity. In the sense that providing information more broadly by democratizing the access to certain tools, like we spoke about on the manufacturing floor, the operational procedures, the troubleshooting instructions, we are allowing a much broader.

Range of people to come in and to get access to certain jobs, get access to certain technologies that might be on the manufacturing floor that might be in innovation and in engineering. And I think that is also tremendously exciting that it's on you to be curious and to be to be excited and then you can get into many more fields easily, more easily than you were able to in the past.

Richie Cotton: Yeah, I suppose manufacturing historically has not been like very broadly inclusive in some senses. So I like the idea that technology is opening this up to to make it available to everyone. Wonderful. Alright. And just finally I always want recommendations for people's work to look out for.

Who's work are you most excited about at the moment? Who should I be following? 

Olympia Brikis: Maybe I can start on the technical side. So what I would be looking out for is the next wave of AI models. So I think we have seen large language models trained on the internet, foundational models, trained on the vast amount of imagery and video out there on the internet.

What we have not seen yet is where our hidden data. Sources, data silos that are maybe not publicly available, that are maybe behind closed doors and what will happen if we build AI models on those type of data sources. So I think we are going to see a next generation of models emerge that are gonna be more specialized in a way or more domain specific.

And I would be, yeah, watching out for those that might come in this industry, might come in other fields. There's many domains that are currently looking into that. I think that's an exciting. 

Richie Cotton: Technology. Alright, brilliant. Yeah, lots of exciting things to come. Yeah. Thank you so much for your time, Barbara.

Thank you so much for your time. Olympia 

Barbara Humpton: Richie, this has been a pleasure. Thanks. 

Olympia Brikis: Thank you for having us.