Digital transformation, artificial intelligence (AI), and industry 4.0 have all been industry buzzwords for some time. However, in the past few years, these buzzwords have come to fruition as they start bringing real value to the manufacturing space. In this live episode, Bryan DeBois, director of industrial AI at RoviSys, joined TECH B2B Marketing’s Jimmy Carroll to explore how advanced technologies are shaping the future of the manufacturing industry. They discuss how the market is evolving, the power of smart manufacturing, and how industry leaders can adapt to stay competitive.

Jimmy Carroll: [00:00:06] Hello, everybody. My name is Jimmy Carroll, and welcome to the “Manufacturing Matters” podcast, where we discuss the latest trends and technologies reshaping manufacturing and industrial processes worldwide. Today I’m joined by Bryan DeBois of RoviSys. Bryan, first of all, thank you so much for taking the time. I really appreciate it.

Bryan DeBois: [00:00:23] Yeah, thanks for having me, Jimmy.

Jimmy Carroll: [00:00:25] Of course. Yeah. My pleasure. So I’d like to start by asking you: Tell us a little bit about your company and what you do there.

Bryan DeBois: [00:00:31] Yeah. So as you mentioned, I work for RoviSys. So RoviSys is a system integrator. And we focus basically exclusively on manufacturing and industrial customers. We use the term a lot, OT, so for your listeners who may not know, OT is short for operational technology. It was a term coined by Gartner back in 2006. You know, everyone knew what IT meant, but they needed a term to refer to, kind of, our world, that manufacturing industrial world where, you know, that typically the technologies we work with have some kind of impact on the physical world. They move things, they make things. You know, they… And so that OT system integrator, that’s really what RoviSys is. It’s been around since 1989. I’ve actually been with the company for my whole professional career. So about 24 years now. And what I do is——my title is director of Industrial AI——and so effectively what I do is I typically am sitting down with customers that have a vision of incorporating AI into their organization. And typically that’s part of their broader digital transformation goals, which we can kind of talk about today. But, you know, they’ve got this vision for AI, and they want to know what’s the state of the art? What’s kind of pipe dream, you know, out there versus what can we do today? So part of my job is kind of education and navigating through all the hype around that, and then actually coming up with tangible projects that can have a real impact on the day-to-day operations of that company.

Bryan DeBois: [00:02:11] Part of my job, though——I sometimes say I also am kind of the bearer of bad news because part of my job is, I’ll get called in and there will be a lot of excitement around AI, you know, with that company. And we sit down and we start to look at some of the enabling and technologies, some of the readiness around that and realize, okay, well, we’re probably not ready yet. So then I kind of pivot. And one of the nice things about working for a system integrator is that we’ve got all that capability in-house, too, so we can pivot and we can say, “well, let’s start talking about OT data strategy” and “let’s start talking about infrastructure and readiness,” which we can talk about today as well. But those kind of enabling technologies so that we can get you to an AI future.

Jimmy Carroll: [00:02:52] Sure. Yeah. Yeah. There’s a lot that I’d like to dive into there. I guess I’d like to start, though, with sort of the general manufacturing space and just this common thread of manufacturers and businesses worldwide, their needs evolve. And so that places a pressure on technology vendors to adapt their offerings. So I guess, with that in mind, how is the marketing or how is the market evolving overall? How should companies adapt? And I’m thinking of buzzwords, but they’re more than buzzwords, right? When it comes to things like digitization and smart manufacturing and smart factory.

Bryan DeBois: [00:03:29] Yeah. Yeah. You know, it’s interesting because that “smart factory” has been a buzzword for a while now. I guess my take on it is this——there’s a lot of talk about smart factory. I’ve got customers that are really big into “lights out factory.” I don’t know if you’ve heard of that term, but lights out factory——where effectively it’s an unmanned factory and raw materials come in and products go out the other end——is just has basically a skeleton team monitoring that factory. And I mean, that’s fine, I guess. Like, that’s a nice kind of vision of what the future could look like. But, how about just a smarter factory? Let’s start there. Right? So for me, my message typically is: think in terms of continuous improvement. When you actually go into these facilities, you know, again, a lot of times it feels like sometimes the people who are creating the hype and creating, you know, these terms aren’t the same people who actually walk into these plants on a regular basis and see… I mean, most plants still have a lot of manual processes. They’re light years away from a lights out factory. So let’s talk about smarter manufacturing. Let’s talk about continuous improvement. And really what I see is, AI is the biggest, um, one of the biggest tools that we have right now to make that big impact on manufacturing today.

Bryan DeBois: [00:04:53] What can we do? But even with that, when we talk about AI, it’s not some pie in the sky type of thing. We’re very tactical at RoviSys. And so one of the things we’ll talk about is use cases. And so we’re really big on starting for… Leading with use cases. Don’t let technology lead. Lead with use cases. Technology should follow. And so let’s find a use case,a very specific, you know, maybe even somewhat narrow use case where we can apply AI and have a 1%, 2%, maybe even double-digit percentage improvement on yield or improvement on scrap, reduction in scrap, improvement on energy utilization. When you start talking about those types of percentage improvements, that’s millions of dollars typically every year. Let’s work on that. Let’s focus on that continuous improvement and what we can do today. Yes, let’s work towards, you know, the future. But let’s instead of worrying so much about that hypothetical future, let’s really work on those types of improvements that we can do today, and those types of improvements that also are going to have a benefit on the lives of the folks who actually operate these factories.

Jimmy Carroll: [00:06:04] Interesting. So if I were to ask you to distill what a smart factory is and why it’s necessary today, what’s your elevator pitch for that? And maybe even like a secondary or combined question with that is, how do you explain that to somebody who’s maybe not within the industrial space?

Bryan DeBois: [00:06:20] Yeah, so typically the way I talk to them is, you know, you kind of have to look at the history of manufacturing, right? And so it started out and it was all manual. And so you’ve seen these pictures from the turn-of-the-century where you have just rooms full of people and they’re doing very menial tasks over and over again. And then, you know, it evolved through the different iterations of automation. And now we have pretty sophisticated automation. And that’s been around now for, you know, 30-plus years. Now, really you layer on top of that the volume of data. So when I started in my career, the typical piece of equipment might be able to give you 10 or 20 data points. Right now, everything on the plant floor is smart. Everything can give you hundreds of data points about how it’s operating and things like that. Now, what can we do with all of that data? And what I’m saying is, is that we talk in terms today of these industrial revolutions, and a lot of people will say that we are in the middle of the fourth industrial revolution, but you don’t know until you get to the end of the revolution what it was all about. When we look backwards, we know what the third industrial revolution was about. The second one, we can look back and say, oh, that was leading up to this. You know, it was leading up to pneumatics or was leading up to relay, whatever.

Bryan DeBois: [00:07:35] What I see with this fourth industrial revolution is… I strongly believe that AI will be the thing. We look back and say, “oh, that’s what we were working towards.” Because when I look back on the arc of my career, the last 24 years and when I started, we were still preaching the value of saving all that data, recording all that data, capturing all that data from the plant floor. And we we had customers at the time, they were like, why am I doing this? Why am I capturing all this? We just throw the data out. Why do I need to capture all of this data? And now when you fast forward, now we have all that. Now we’ve got these AI algorithms and these analytics that are hungry for data. Now we have all that data because we’ve been spending the last two decades capturing that data. So I do believe when we get done with this revolution, we’re going to look back and say, “oh, we were working towards AI.” Whether or not we knew it at the time, that was ultimately where we were trying to get to. So when I when I talk about smart factory, that’s typically what I’m talking about, is capturing all that data, the huge volumes of plant floor data and doing something to add value, doing analytics, leveraging it for machine learning, doing something that’s going to add value and improve the operations of that plant floor.

Jimmy Carroll: [00:08:46] Yeah, that’s really interesting. And I’ve heard other people kind of echo a similar sentiment, right, with the AI being the focus of this industrial revolution. I think it was Andrew Ng, in fact, I know it was Andrew Ng, of Google Brain and Coursera and Landing AI and other ventures, who said AI is the new electricity. And, you know, it makes you it makes you pause and think. But the other part that’s really interesting is the mention of data and the ability to collect all this data to train these algorithms in visual inspection, for example. It’s super cliché, but garbage in, garbage out is correct, right? If you don’t have a means to collect vast amounts of data or data on what’s the defective part, what’s not a defective part, you don’t have a useful algorithm. So it’s just kind of a reflection of all these different types of technologies advancing together to solve new problems, not just the algorithms, but, you know, industrial computing capabilities and cameras and whatever else.

Bryan DeBois: [00:09:48] Absolutely. And one of the things that I’ve been at… We created this industrial AI division five years ago, so I’ve been at this for a while, and I’ve been having a lot of these conversations with customers. And a lot of times, like I said, this may be the first time they’ve really looked into AI in any depth, and so they don’t really know what the state-of-the-art is. And one of the challenges that we see in this industrial AI space is that these AI algorithms were really built in the IT space. They were built with very clean, very correlated data. You typically had millions of rows of data. The state-of-the-art for training one of these machine learning models is that if you get to the 100,000th row and there’s a field missing in that data set, or, you know, there’s some kind of bad data in it, you just throw that row out and you keep on trucking. Well, in our world, you may have just thrown out a batch that you’re not going to make again for a couple weeks.

Bryan DeBois: [00:10:43] Right? So, that world of very clean, very correlated large volumes of data, millions of rows of data… You know, our world in the OT space is much different. Our data is highly siloed. It is not in any way, shape or form clean. And, it’s not correlated. So you’ve got in these different silos, I might have a column called “prod ID” and then over here it’s “PID,” and over here it’s “product ID.” The AI doesn’t have any clue that those all refer to the same thing. And so we actually spend a lot of time on the data science side of this, making sure that we have clean, very correlated data. The other thing that we deal with all the time is, I mean, these sensors go bad. So now you’ve got bad readings in the data. So we have to spend a lot of time, I feel like a lot more time getting the data and the data sets in a cleansed, correlated way so that we can actually use it to train those ML models. So, yeah.

Jimmy Carroll: [00:11:39] So, I mean, you know, today I have the luxury of being able to talk to somebody who’s worked on these plant floors his entire professional career. Right. So, you know, marketing hype can say maybe 7 or 8 years ago, deep learning was sort of introduced. It wasn’t introduced. It’s been around for a while, but it was being hyped within the industrial machine vision, industrial automation space as something that’s going to change everything. And, you know, eventually it settled in as an incredibly useful tool. But for sure, what you’re saying before, it’s a bit overhyped for someone like you. Where have you seen the most success with AI adoption? And, you know, I mentioned visual inspection before, but not just there. I mean, whatever, simulation, digital twins, data analysis, predictive analytics, anything.

Bryan DeBois: [00:12:26] Yeah. So where we’re seeing the biggest impact right now——and actually it’s funny you mentioned deep learning. So deep reinforcement learning, DRL, came out around 2016 with DeepMind, which was that Google spinoff that they basically set about to create a new ML algorithm. And what they came up with was deep reinforcement learning and it’s unique amongst machine learning algorithms in that it can build long-term strategy. It’s a machine learning algorithm that can actually build long-term strategy and actually can respond to situations like a human can. And so it’s unique in that regard. We actually have been using deep reinforcement learning. We call it autonomous AI. So that’s kind of the broad umbrella term for it. But we’re actually leveraging autonomous AI on the plant floor to have that huge impact. The million dollar-plus a year ROI type of impact on these types of situations. So as an example, I’ve got a customer right now. And they make glass bottles and they have come up with a new cutting edge way of making glass bottles that will revolutionize the industry. However, the problem is, is that it’s a very finicky process. So it requires an operator that really is an expert, knows what he or she is doing.

Bryan DeBois: [00:13:50] And really the operator has to go like this, you know, and it’s a drifty process. So after they get it all kind of dialed in and they’re making good bottles, even if they walk away for an hour to go to lunch and they come back, the process has drifted and they’re not making on spec bottles anymore. And so it was like, we’ve got this amazing new way of making bottles, but only these expert operators will be able to ever operate this process. So they approached us to specifically develop an autonomous AI. A neural network trained this way is called a brain. An autonomous AI brain that could operate this process as well as their expert operators could. And so we actually trained a brain to do that. Now what does that look like? Part of that, as you mentioned, SIM, part of the difference between deep reinforcement learning and other ML algorithms is that it learns by doing, not necessarily by data. So it actually has to have some kind of simulation, some kind of playground where it can try all kinds of things and get better, incrementally better and better and better.

Bryan DeBois: [00:14:51] And so we built that. Now, one of the things that I hear from customers when we talk about that is, well, we don’t have simulation, you know, widespread simulation deployed at our facility. That’s okay. Even if you don’t, as long as you have data, we can build what’s called a data-driven model, which is effectively a simulation built off of the historical record. So it’s a simulation built off of what actually happened. Now, there’s some tricks that we can do to kind of extend that state space so that there’s plenty of room for it to try a lot of different things and get the feedback that it needs. But that autonomous AI approach, that really is having that impact that folks are looking for, the big dollar impact. And where we’re starting in terms of finding use cases for that is, we’re first starting and saying, “okay, where today do you have manual processes, where do you have humans in the loop because you’ve never been able to do it any other way?” We’re really looking for those unsolvable problems on the plant floor and then applying this autonomous AI technology to try to solve it.

Jimmy Carroll: [00:15:58] A lot of questions I want to ask and follow up. But in the opposite end of the first question is——and you brought this up sort of organically before——but a lot of people might come in and say, “all right, we need AI in this process,” but not really know where to start. So where have you seen people falter with AI adoption? Like, what are some pitfalls that people need to avoid when it comes to deploying?

Bryan DeBois: [00:16:21] So, there’s lots of areas where there’re pitfalls. So the first is if they lead with technology. So if they first grab that hammer, everything starts to look like a nail. That’s typically the wrong way to do it. So we heavily emphasize that you need to lead with use cases. Because here’s the thing. Once we lead with use cases and we start generating those ideas——and I’ll talk here in a second about how we do that——once we start generating those ideas, it may be that you don’t need the most cutting edge AI to solve that problem. It may be you just need a little more visualization there, or you may just need a historian deployed, or you may just need some traditional optimization techniques. So all of those things are in our toolbox as a system integrator. We don’t have a product, so we just partner with all these vendors. And so we’ve got access. We’ve got a toolbox. We’ve got access to all of these different technologies and vendors. So for us we’re happy to implement that project with whatever technology makes the most sense. Now, how do we get to that point where the customer has use cases to start to tackle? We typically do what we call a workshop. It could be an AI workshop. It could be a digital transformation workshop… Wherever the focus is for that particular company.

Bryan DeBois: [00:17:35] But that’s where our experts actually come in and sit down. And we do a half-day with them. We talk about what the state of the art is. We show them what other customers are doing, [what] their peers might be doing. And then we go through this brainstorming exercise. As a result of that brainstorming exercise, we oftentimes get 20 to 40 executable project ideas that are generated from them about where we can make the biggest impact in their organization. Once we have those project ideas, we can start to work backwards and figure out, what are the technologies that are going to solve this? So that’s a big pitfall. There is leading with technology instead of leading with use cases. I would say another big area, particularly in the AI space, is that this industrial AI space is filled with a lot of snake oil salesmen. You know, they’re slapping AI in every single product, whether or not it’s fundamental to what it does or tangential to what it does, they’re putting the AI branding on it. So there’s a lot of snake oil, there’s a lot of vaporware, there’s a lot of startups. Navigating that can be pretty hairy. There’s a lot of IT vendors that have ventured down into this OT space and don’t really understand what it takes to put, you know, something like this… To operationalize it, to put it into action on a plant floor where it has to run 24/7, the organizational change management that’s required to get the operators and the supervisors to buy into this system that they don’t necessarily trust.

Bryan DeBois: [00:19:03] There’s a lot that goes into that. So that’s a pitfall oftentimes. And then I would say the OT data strategy——if I was talking to a semiconductor manufacturer and gave them the whole AI pitch and they’re super excited about that, and it’s like, okay, great, let’s go. And then, so I go, let me ask some questions. So what’s your OT data infrastructure look like? They’re like, “well we have a lot of disconnected skids. And each one has a little sequal database attached to it. And none of it’s really networked or anything.” And I go, oh, okay. And so then they go, “these AI algorithms, do they know how to go out and pull all that data from all these different databases?” I’m like, no, not at all. So then we have to pivot and then we’ve got to talk about OT data readiness and what kind of infrastructure upgrades we need to do to be able to get them to where they can get, to where they can realize these AI dreams that they have? So, that OT data strategy, I would say, is another big pitfall that we oftentimes are talking to customers about. Yeah.

Jimmy Carroll: [00:20:09] And it is interesting to hear it from your end——your end being the integrator end, and not necessarily like the vendor or technology provider end about snake oil salesmen and not just in the industrial space, right? Like I was buying a fan recently and part of the marketing said, oh, it’s built with a smart AI chip. And I wanted to say to whoever, you know, whoever wrote that, “what does that mean to you?” Right? What exactly does that mean? Anyway, another topic that I have to ask about, I’d be remiss to not ask about is large language models, ChatGPT… A lot of interesting stuff going on there. These are improving over time. And I feel like, even as recently as a couple of years ago, the technology was being dismissed by the industrial space. But now there are some massive companies in the space that are using it in small——I shouldn’t say small——but using it for very specific tasks like helping with ease-of-use programming and things like that. But, the concept of using generative AI in the industrial space–what’s your take on it?

Bryan DeBois: [00:21:21] I think that we want to walk a fine line there. We don’t want to just dismiss it outright because, as you said, there’s some very clear benefits that generative AI is going to have in our space. And that ability to generate code I think we’re finding——even more so than its ability to generate poems or something like that——it’s ability to generate code is actually really interesting and I think maybe underplayed compared to its other capabilities, the impact that that’s going to have, because up until the point that generative AI existed, which again, we take it for granted, it only came out in November of 2022. I mean, they were obviously working on it for years, but that was its big coming out, was OpenAI in November of 2022. So I mean it hasn’t really been around for two years at this point. But that ability for it to generate code is really interesting, because we don’t even know what impact that’s going to have. Up until that point, the only person that was really capable of generating code was a programmer, right? And so that programmer was the person that would stand in between the requirements from the user, and then would create code to meet those requirements. Now, with generative AI we have this other tool that can do that. So yes, it’s going to have a huge impact. We’re already seeing it. You mentioned Siemens, I think Rockwell is working on it, I think all the major vendors are working on the ability to generate code——PLC code, typically——DCS type code be able to generate that code.

Bryan DeBois: [00:22:51] And, you know, be able to run it on their controllers and things like that. And so that’s going to have a big impact for sure. Now, the place where I think people are trying to shoehorn it in that I see over and over again at these conferences is they go, “well, what about maintenance? We could have this be our maintenance guru and so we could feed into it all of our work orders and all of our SOPs and everything we know about maintaining this equipment on the plant floor. And we could create this oracle that if anything breaks down on the plant floor, we could say, hey, how do I fix this?” That I’m not comfortable with right now. So I’m not comfortable with using it in that regard on the plant floor. And here’s why: It’s LLMs still have a very nasty problem called hallucinations. And what that is, for your listeners who may not know, is when the generative AI makes stuff up and it unfortunately makes it up in a way that is completely convincing, completely compelling. It seems like everything else that it’s said up to that point, which was true, it will make something up and it will sound just as convincing. Here’s an example. And I mean, I just tested this, I think a month or two ago on ChatGPT, I said, “name for me the 13 dwarves in the Disney classic Snow White.”

Bryan DeBois: [00:24:04] Now, we all know that there were only seven dwarves, right? And so it created a number that said, “okay, yeah, I can do that.” And it created a numbered list. And the first seven were the seven dwarves that we’re all familiar with. Then it created six more. And all the names were completely convincing. If I didn’t know any better, I’d say yeah, okay, there were 13 dwarves. I must have misremembered. Right? And that’s called a hallucination. And that’s… It’s funny when it happens in that context; when it happens in the plant floor, that’s where you could kill somebody. So I’m not comfortable yet in deploying these things now; it’s not like the LLM vendors don’t know about this. They’re actively researching this. There’s an up and coming technology called Rag that looks very promising. Effectively, what that means is, is that the LLM has to cite its sources. So it’s not enough for it to just say, “yeah, to fix this, you have to tweak this bolt or torque this bolt. You got to do this, you got to, you know, grease this up.” It has to say, “and oh, here’s the page number in the SOP that I got that from.” Right? Everything has to be cited back to the source. A lot of promising, but until that gets there, I’m not comfortable putting any of this on the plant floor yet in the generative AI space.

Jimmy Carroll: [00:25:10] Yeah, that makes sense. That’s an interesting thing I hadn’t heard about, citing sources. It’s almost like, you know, that there’s been that issue of deep learning and AI in the industrial space referred to as the black box problem.

Bryan DeBois: [00:25:24] Yes.

Jimmy Carroll: [00:25:25] Maybe that helps you kind of get inside of it a little bit.

Bryan DeBois: [00:25:27] Right, yeah for sure, for sure.

Jimmy Carroll: [00:25:31] We kind of touched on it earlier, but one thing that maybe doesn’t get discussed enough in the realm of AI or in the realm of industrial automation, manufacturing in general, is like, yes, AI has been advancing because of the work being done on these algorithms. But there’s a lot of other technologies that are sort of advancing in lockstep, right, to help enable this. And I’m thinking of things like industrial compute and in the cloud, right? So like in the cloud, for example, a couple of years ago, it seemed to me like people were a little bit hesitant to want to put their data in the cloud. Is that issue going away? And in what other ways have these technologies helped AI advance?

Bryan DeBois: [00:26:14] Yeah, it’s funny because, again, you don’t know until you get to the end of the revolution what it was all about. But in a weird way, all these things kind of seem to have been working towards that. So a perfect example is the adoption of cloud, right? We probably wouldn’t have been able to accomplish everything we can do right now in the industrial AI space without the compute capabilities of the cloud, and so it required everyone to start moving that way, whether or not they realized it or not, to get to this AI future. So, you’re absolutely right. The adoption of the cloud in the industrial space is accelerating. I’ve got more and more customers that are saying, “look, we’ve got a big OT on-prem data footprint, we’ve got on-prem data centers and things like that. And, the software is aging. We’ve got to do major upgrades, and the hardware is aging. And honestly, we don’t want to maintain this on-prem data center footprint anymore.” And so they’re moving all of that infrastructure to the cloud. So that’s a big driver of it. The other big driver that I’m seeing is that a lot of these projects nowadays… when I started my career, the only folks in the room were the OT folks. It was a bunch of OT engineers. And you know, and us, the SI, who were really spearheading these projects. Now, when I look at the makeup of the room, at least half the room typically is IT.

Bryan DeBois: [00:27:34] So a lot of these projects are actually being driven by IT. They’re already very comfortable with the cloud for the rest of the businesses data. They’re basically saying, “look, we’ve got no issues with with putting this OT data up into the cloud. And moreover, not only do we have no issues, but since IT’s the one driving this project, we’re telling you, we’re forcing you to move this data up into the cloud because we want access to it.” You know, one of the big… about seven years ago, people started talking about this IT/OT convergence. And one of the big aspects of this IT/OT convergence is that IT wants access to that data on the plant floor. That data has been a black box for them and they know that there’s untapped value in it and they just haven’t been able to access it. So a big driver is to move all this data up into the cloud where they now can have access to it. But all of that was important, but then, as you mentioned, the cloud compute autonomous AI——going back to that——requires a lot of compute during the training process. We’re typically spinning up not just one simulator, but a hundred simulators in parallel. It’ll run for up to a day or so to train a brain. And so we need that compute horsepower up in the cloud.

Jimmy Carroll: [00:28:48] Yeah, interesting. I have to ask this question of you just because, you know, we’ve talked so much about AI and your title and everything else, but AI and the general public, right? When you tell somebody outside of the industrial space that you’re the director of industrial AI, what kind of reactions do you get there? And I know that some people probably have reservations about that, right? Because we talked about it right before this call, like Skynet, Terminator 2, Ex Machina. Movies like this. Like, how do you explain to people that AI and robots will actually help serve humanity and not the opposite?

Bryan DeBois: [00:29:22] Right. Yeah, I think that… it is a question I get a lot and the concerns typically fall in one of three buckets. So let’s just address them all. So the first one is that concern of putting people out of work. The reality of it is when I go into these plants and I’m talking to these plant managers all the time, the reality of it is, the last survey I saw from the Bureau of Labor Statistics, I think, showed a half million shortfall in manufacturing jobs. So they are looking for still a half a million people to fill those jobs. They don’t have enough people. What has happened, and it’s a generational thing, is that the folks coming in now into the workforce, they’re just not interested. They have high turnover in these positions. They’re just… These folks are just not interested in spending the next decade and a half of their career learning how to operate this line——it just doesn’t exist anymore. And we could wish that it wasn’t the case, or we can just accept the fact that that’s the way it is and we have to figure out how to move on. So there’s no issue about putting people out of work. They don’t have enough people. And, in fact, I was talking to one plant manager at a vinyl extrusion plant and he said, “I’ve got two expert operators. They’re the best in the world at running this, this vinyl extrusion line. And they’re both 5 years from retirement.” He’s like, “I’ve got no bench behind them. I got one 6-month and one-year operators. I got high turnover. They’re not sticking around.”

Bryan DeBois: [00:30:44] For him it’s existential. Like he’s either going to adopt AI or he’s never going to run the line as well as it does now with those expert operators. So it’s not a question of putting people out of work. They don’t have enough people to do this. So that’s the one thing. And I typically try to dispel that fear. The second one that I get is, well, if the AI is controlling these machines, how do we know it’s not going to blow up the plant or something like that? So again, I kind of always have to say, “well, okay, compared to what? So what you had before you put this AI in, is you had, again, maybe a six-month operator who had a couple of weeks of training and he or she could do whatever they wanted with those knobs. And you trusted them not to blow up the plant.” Now, with my AI, I can actually restrict the capabilities, the operating window of it even more than a human can because it’s all digital. So if the human was able to move that dial between 1 and 100, and you could tell me, “well, I’m not quite comfortable, can we just make sure it can only go from 20 to 80?” Fine.

Bryan DeBois: [00:31:39] Done. Right. So we can actually restrain the capabilities of that brain far more than even the humans can. So that’s also not a concern that it’s going to do something. There’s also all of the typical protections in the control system are all still there, all the interlocks and those types of things that are going to protect the equipment, protect the people who work there. So that’s typically not a concern. And then the Skynet thing. So I can tell you with confidence, because I work with these AI algorithms every single day, while they are very good at the very specific tasks that we give them, this is not general AI by any stretch of the imagination. It can’t think for itself. Even, you know, the most advanced human-looking AI, which I would say is this generative AI that we have now, it doesn’t really understand the words that it’s saying. It’s just a really good autocomplete. It’s  really good at picking what the next word is based on all the context before it. It has no idea what the words mean. It’s just stringing them together. And it looks intelligent when it’s done. We’re nowhere near the Skynet, so everyone can rest easy. We’re okay there. So those are typically the three kinds of areas of concern that I get. And yeah, none of those are something you need to lose sleep over right now.

Jimmy Carroll: [00:32:51] Yeah. Just a couple quick things I want to follow up on there. I don’t want to… I want to be respectful of your time. I could ask you a million questions, but, one of them is, I’ve got a systems integrator friend who, awhile ago, when the concepts of deep learning and AI and manufacturing really began to pick up, like you said——whatever it was, 7 or 8 years ago——around the deep learning, DeepMind, right? Is that what it was?

Bryan DeBois: [00:33:18] Yeah. DeepMind.

Jimmy Carroll: [00:33:19] And he’s like, well, the thing is about AI is that it doesn’t really exist. It’s a collection of different technologies that you might– A better way to say it might be, you have an AI-enabled software or AI algorithms, but AI itself, it doesn’t exist. And it’s a concept that it took me a minute to wrap my head around, but it’s probably helpful for a lot of people out there, especially in the general public. These things are only as good as the data they’re programmed with, and they’re not…  there’s no general AI, right?

Bryan DeBois: [00:33:56] There’s not. That “intelligence” word in AI is still very misleading because I mean, again, when you’re close to it like I am… And, you’re right. AI is an umbrella term and machine learning is really the enabling technology behind that. But I mean, you can even put analytics… There’s some old optimized linear optimization has been around forever, you could put that under that AI umbrella. I mean, these are things that can give glimmers of intelligence in certain circumstances. But machine learning really is the enabling technology, and it’s those advances in machine learning… So in the case of, you know, LLMs, there was an advancement called transformers. And that algorithm was really the thing that made that all possible. Deep reinforcement learning was the thing that made autonomous AI possible. And that’s really what makes all of this possible. But yeah, the intelligence part, they’re not very intelligent. I mean, they’re very good at very specific tasks and they can get really good. I mean, we’re seeing human-like capability out of autonomous AI, but just at running that one part of it. They’re not going to run the plant. They’re not going to make the business decisions that it takes to keep a business running. Like, that’s not what they do.

Jimmy Carroll: [00:35:01] Yeah, totally. The other thing I wanted to mention that I thought was interesting is——well, it was all interesting, don’t get me wrong!——but I just want to bring it up because I’m glad you mentioned labor shortage. We have this persisting labor shortage. And it kind of speaks to the importance of industrial automation, especially, obviously, for large companies. But even for some of these small- and medium-sized companies, it’s no longer like a nice thing to have. In a lot of ways, you need to have it. I mean, there are some stats and I don’t remember them off the top of my head about, you know, the majority of welders in the United States are going to be retired within X amount of years. Luckily, folks in that space have not shied away from adopting robot and collaborative robot, cobot technologies and in collaboration with machine vision and AI technologies to do automated welding. So that’s just an example of, like, you could let that industry die or adopt technologies that let you maintain productivity. Bottom line.

Bryan DeBois: [00:36:04] Well, I think it’s important. And, I mean… I was born in the United States. So obviously my area of focus is kind of the U.S. Right? We’ve got global customers. But if we look at just our nation and if we want to compete in the global stage in this manufacturing space, you’re absolutely right. We’re going to be going up against countries that do have just masses of people that they can throw at the problem, so they can just throw crowds of people at solving these problems. If we want to be able to compete with that, we have to adopt AI, we have to adopt automation and digital technologies. You’re right, it’s not “nice to have” anymore, even for the smaller side of the manufacturing scale. They’ve got to be looking at what they can do with the resources that they have to adopt these technologies. I will say this, for me, I’m very… I’ve spent my whole career in manufacturing. I’m very bullish on manufacturing in general. I do believe strongly that manufacturing is the biggest lever that we can pull to get to achieve prosperity for everyone, right?

Bryan DeBois: [00:37:09] So that’s the biggest lever that humans can pull, manufacturing is, to achieve prosperity for everyone. So what does that mean? It means that when you have access to high quality things, you know, from drugs to food to consumables, whatever it is, when you have access to plentiful, high quality products that are cheap, they’re affordable, and they’re everywhere. That is what prosperous prosperity looks like, right? That’s not going to make you happy, but that’s a whole philosophical question. But if you want to look at, you know, reducing or eliminating poverty, when you want to look at improving quality of life for the most people that you can, manufacturing is the biggest lever that we can pull to do that. And so I get really excited when I go in and we implement one of these AI projects, and we see these huge improvements, that they’re seeing reductions in energy usage and increases in throughput and decreases in waste. I get really excited because, you know, in whatever small way that I can, we’ve made that better and we’ve improved that. And marched everyone incrementally closer to prosperity in that, and that’s what I get really excited about.

Jimmy Carroll: [00:38:17] I appreciate that response. Bryan, I wanted to call your attention, we had a couple questions here from the audience. On the right side of the panel, first of which asks, “do we really need AI to automate?” And then the second one says, “plus AI is dependent on data. What’s the right way for manufacturers to share their data to make it useful across industry?”

Bryan DeBois: [00:38:39] So let me… Two different questions here. So let me address them separately. Do we need AI to automate? No. We’ve been automating manufacturing systems for decades now without AI. Is it going to accelerate our ability to automate it? 100%. So just like anything… I mean, we don’t always need these tools. But if you want to get those accelerations in automation, then that’s, you know, these tools are what… AI is going to be one of those those big, big tools that’s going to help us do that. As far as sharing data to make it useful, I love that question and it’s not asked a lot. One of the challenges that we see in manufacturing is that everyone is so concerned about their own data that we don’t get the sharing that other industries get, and there’s a lot of benefits to that. I don’t want to drop any specific vendors names or anything like that, but there are some vendors right now that are focused on building a community. You know, being able to easily share your production data across different boundaries and with your partners and with the public and things like that, particularly when you have companies who service, you know, maybe municipalities or something like that, and they’re required to provide their data to the public.

Bryan DeBois: [00:39:55] That’s becoming a lot easier. And I would continue to encourage that, because I think that’s how we’re going to be able to make these big strides. One of the challenges we see right now in applying AI in the manufacturing space is that because everyone’s kind of so secretive, and because their data is their data, is we don’t get a lot of repeatable use cases yet. We don’t even know necessarily what the different use cases are and how we can apply them, because everyone’s so secretive. So we’re getting over that. But eventually, you know, we will need to do better about sharing data, sharing results, having customers go to trade shows and actually speak about the benefits that they’ve gotten and stuff like that. If they want to see the… Again, rising tide raises all boats, right? If we want to get to that AI future, we’re going to have to do that.

Jimmy Carroll: [00:40:47] Bryan, anything else that we haven’t discussed today? For example, the labor shortage thing you mentioned toward the end, which I’m glad you did. Is there anything else that we haven’t talked about that you want to bring up?

Bryan DeBois: [00:40:59] Yeah, and I know it sounds self-serving, but one of the things that I typically close… So I speak a lot at trade shows and things like that on these topics and, how I typically close is find an SI that you can trust. I know it sounds self-serving, but this is like I said——there’s a lot of snake oil. There’s a lot of, you know, the vendors themselves are creating this uncertainty on purpose. They’re leveraging the AI label because they know that it’s got such a nebulous definition that you can’t disprove that it doesn’t have AI in it. Right? They’re creating this fog on purpose. And I really do believe that a strong system integrator partner, if you can find one, can help you navigate this, and can bring ideas to the table that you wouldn’t have necessarily thought of yourself. So, find a good SI that you can trust and then work alongside them to help develop that roadmap and then get started. That’s typically the other thing I tell people, tell customers, is it’s okay if you don’t want to go full, you know, whole hog on AI right now and you don’t want to jump in with both feet. Fine. But let’s start taking those small first steps. Maybe we focus then on infrastructure. Maybe we focus on readiness. Maybe we focus on getting everything set so that when you are ready to start tackling these AI projects, you’ve got the data readily available to be able to go and attack those.

Jimmy Carroll: [00:42:24] So if people want to learn more about RoviSys. I think it’s just RoviSys.com, is that right?

Bryan DeBois: [00:42:28] RoviSys.com, and if you put in RoviSys.com/ai that’ll take you to the industrial AI section of our website. And then they’re always welcome to connect with me on LinkedIn, Bryan DeBois on LinkedIn. And happy to connect with anyone.

Jimmy Carroll: [00:42:44] Well, Bryan, I really appreciate the time. And if anybody has questions, we’d be happy to pass them along as well. You can visit us at Manufacturing-Matters.com or TechB2BMarketing.com. Bryan, it’s really been a pleasure. Thanks so much for taking the time. Really appreciate it.

Bryan DeBois: [00:42:58] Thanks, Jimmy. Appreciate it. Thanks.