Inside IALR

Practical Tips to Integrate AI in Manufacturing and Business

• Institute for Advanced Learning and Research

Share episode

Host Caleb Ayers sits down with Meredith Gregory, Director of Digital Strategy and Program Management, to explore the real-world applications of artificial intelligence in manufacturing and business. Meredith shares her journey from Northern Virginia to IALR, explains the foundational math behind AI and discusses how companies can adopt AI tools to solve specific problems. From resume restructuring to computer vision for defect detection, this episode is packed with valuable insights for anyone interested in integrating AI into their operations. 

The Institute for Advanced Learning and Research serves as a regional catalyst for economic transformation in Southern Virginia. Our services, programs and offerings are diverse, impactful and far reaching.

Get updates from IALR on other channels:

Meredith Gregory:

It is the tool that will automate the tasks that nobody wants to do. It's that task that someone's like, if I never had to do this again, I would be thrilled. And that is where artificial intelligence absolutely shines.

Caleb Ayers:

Welcome to another episode of Inside ILR. Thanks for being here today. Thanks for joining us. So, first, I'm gonna let Meredith introduce herself and tell us how this podcast helped her get here today. We'll start with that and then we'll get on to what we're gonna talk about.

Meredith Gregory:

So before I came to ILR, I was in Northern Virginia. Um, I was a government contractor as a working as a program manager, leading some pretty advanced uh AI algorithm development work. That said, I have family connections to Southern Virginia and I've always cared a lot about the region and generally pay attention to what's going on uh down in Southern Virginia. And in social media scrolling as you do, came across the institute and started digging into a little bit more. Everyone says, that's in Danville. You see, you see the building, right? And you're like, that's in Danville. And uh so dug into a little bit more and found the podcast. And as I would drive back and forth from Northern Virginia to uh my family's farm in southern Virginia, I would listen. And every time I'd be more and more like, that's so cool, that's just fascinating. How does this work? Uh and over time I had some ideas about just general connections and didn't know who to reach out to. So I reached out to Caleb figuring you were well connected, and you connected me with Jason, and a couple months later here we are having a conversation.

Caleb Ayers:

So that is our number one podcast success story. Daniel, do we have anything better? We don't have anything better than that, right? Okay. Um that's number one. So yeah, you are the director of digital strategy and program management, been here for what, six, eight months at this point?

Meredith Gregory:

February. Yeah.

Caleb Ayers:

Okay, yeah. So what I want to talk to you today, um talk to you about today is kind of artificial intelligence, specifically in the manufacturing world. I know that's kind of your a big part of your role is how do you how do you incorporate those two things? Real big picture. Can you define AI for us? That is a broad question that you can take whatever direction you want.

Meredith Gregory:

That sounds good. Yeah, that's the million-dollar question everyone tries to describe it. My my best way to describe it is through examples at the end of the day. So um artificial intelligence, I think about intelligence, and then you add the artificial piece to it. You think about um machine learning, is typically about artificial intelligence and machine learning, uh, same thing. You talk about learning and how how do you learn, and then you add the machine component to it. And so we're replicating um how humans think and how humans process information and data and learn. Um, and we've just added a computer system behind that to help automate it at the end of the day. Um, so if we talk about some different use cases, your large language models think chat GPT is by far the most common use case today, um, really coming on the scene in about 2023. The math that derive that drives all of this artificial intelligence, your large language models, plus all of the other applications, um, is 17th century math. This has been around for centuries at this point, the concept. Uh, what really launched it into modern day expansion and just the the gold rush, as they call it, that we've seen, um, is this thing called it's compute, which is your graphical processing units, um, which is essentially the the fuel. So that's where on your laptop you're running a CPU. Uh graphical processing units or GPUs are much, much um more robust and can process information far quicker and far more uh far more complex calculations. So those come on the scene about, I don't know, 15 years ago, something like that, and it's just completely accelerated from there. So I got into the space eight years ago, which is far before 2023. Uh so I always try and explain use cases that are not just large language models, because that's what everybody thinks about today. But there's so many more applications uh that you interact with day-to-day already. So, some examples, you've got your your chat bots, your chat GPT, your your large language models. Um I don't think we have to explain what those are. Most people have have seen them interacted with them at this point. Um, beyond that, you have things like recommenders. So your Netflix recommended for you, your um Amazon suggested purchases, Spotify, playlists, things like that. Recommenders are all AI in the background. Until we got to Chat GPT and the big uh revolution of generative AI, no one was labeling features as AI. There's nothing on Netflix that says, here's your AI generated, suggested for you list. It's just a really convenient feature that is nice. You're like, yep, that's that's great. I do want to watch that movie. Um so you have your your recommenders, you also have a lot of uh vision work, vision, computer vision. Um things like looking at your phone to open it is facial recognition. When you go to the airport and you give them your ID now, and they're you know, it's a computer algorithm saying yes, I think it's the same person at the end of the day. Um so there's a a large span of um uh computer vision algorithms and use cases. Uh there's your classic classifiers, so in your Microsoft inbox, you have focused and other. Um that's classifying your spam detectors, uh, all of that is just your a standard classifier. So there's a lot of different AI or machine learning uh examples that are far beyond just the the large language models. Um but sometimes it's helpful to think about those use cases in defining what it is and what it can do.

Caleb Ayers:

Two things. One, can you explain this seventh 17th century math without completely breaking my brain? And two, what is what exactly is the difference between machine learning and AI?

Meredith Gregory:

Good questions. Okay, so 17th century math, um, it's something called neural networks, and it's easier with visuals, so we're gonna go without them or see where all we do. Um it's something called a neural network, and it essentially is a series of um computations. You're gonna have an input that comes in, and our brains, if you think about your neural pathways, we receive some kind of stimulus, and then your brain does what brain things do, and some answer pops out on the other side. You know what to say, you know what a word is, you can read, you can understand context, things like that. Um so neural networks is just the uh mathematical representation of what's happening. So it's um think about it's a think about like a table of numbers. This is a gross simplification, but it works. Think about like a table of numbers, it's all just random numbers. Um, your neural network is just a bunch of random numbers when you're born, you don't know anything. It's just a bunch of randomness, and you start to learn over time. So we're gonna provide some kind of stimulus or provide some kind of data, and then we know what the answer is to uh it's you know, if we have a picture of a dog, we know it's a picture of a dog, and we can tell our model, here's a picture of a dog, and it's gonna run through its random set of numbers, and those random numbers aren't gonna do anything useful, but it's gonna give you an answer at the end that says, I think this is a cat, which is not. So you get amount of loss or error in your model, and then you're gonna feed that back through your model. It's called back propagation, that says, okay, you're not right, change something about those numbers and then feed it through again. And so you do that back and forth, uh, your forward pass and then your back propagation millions of times, so that those numbers become the exact right set of numbers that says when I give you a picture of a dog, you tell me it's a dog. Um, so it's a little bit complex to think about how that works, but um, at the end of the day, you're essentially just optimizing a set of parameters so that it produces the answer that you're looking for based on the data you provide it to break your brain. Yeah, no, no, that makes sense. Thank you. Yeah. So your second question about AI and machine learning. Um, think about big circle, little circle inside of it. So AI is your your biggest circle, machine learning is inside that circle. So there are things that are considered artificial intelligence that are not machine learning, but it is a very small subset at this point. So they're they are often said together, AI, ml, um, because they are almost interchangeable at this point. Okay. Um machine learning is really when you start feeding it data and training based on that learned set of um data. Artificial intelligence can involve some other things, but realistically, when people talk about it today, they are almost interchangeable.

Caleb Ayers:

Mm-hmm. Okay. Thank you. Yeah, that explanation of the how I mean, essentially you were just describing how AI is trained with the forward, you know, giving it, giving it the parameters, it's tell it if it's right or wrong, and doing that, as you said, millions of times. Correct. Um I guess for um your role, I know a lot of what you're doing is trying to help, whether it be internal or or companies in Southern Virginia or even outside of that, be able to effectively incorporate machine learning artificial intelligence into their processes. What are some of the ways that you're seeing companies integrate AI and machine learning, and then what are some of the challenges that come with that?

Meredith Gregory:

Yeah, so the the short answer is slowly. Um we've gotten to a point where AI is common language. When I started doing this, I would tell someone I was working with AI, and they're like, what's that? Which today would sound silly. Um so now everyone has this sense of AI's here, AI's gonna stay, we need, we need to get on board, we need to be involved. Um, but I don't know what that means at all. And if I do, if I have a sense of what it is, it's probably a large language model. Um, and I want to be able to use ChatGPT in some kind of effective way. Um, so there's there's that mindset. Ways that people are integrating AI today mostly is I feed Chat GPT a document and I ask it to summarize it, or I throw in an email and I tell them to make it more professional. Um, so you get these very little use cases, or I want to research something using Chat GPT. Um, there is yet to be just wide adoption of really robust AI solutions across the board. Um, I had a conversation with an external partner last week, and they were like, oh, it's good to know we're not actually that far behind. But you're having this conversation, you're actually a leader in the space. Um, once you're at you really want to dive into a series of use cases and understand what it is that um AI and machine learning can bring to the table. So internally, just some examples that we've already started to implement. Um, we did implement a chatbot on the ATDM website. So if you go to atdm.org now, there'll be a little chatbot, and that chatbot knows all things ATDM and it gets used widely to answer questions, mostly from prospective students, is where we found it to be the we get the most traction, um, most questions coming in in that regard. Um, we've also implemented a tool that does resume restructuring. So we have all of these students in ATDM, they're they're coming in, they need a resume so they can go and do their interviews and go and meet their hopefully future employers. And traditionally, the team has taken those resumes and kind of hand-jammed through restructuring, putting certain ATDM branding on it, making them sound better, making them aligned to a job description. Um, and there are many tools out there today, and um, we've gone with one that will take in a resume, provide a bunch of suggestions, restructure for you, add some branding, and it saves a ton of time. The other thing that that tool will do will be mock interviews with a um to just an AI agent essentially that's asking questions. But it'll grade the student on how well they did. Do they need to slow down? Do you need to speak more clearly? Is what you're saying aligned to the job description that you want? Um, so there's some very specific use cases around ATDM. There are plenty of use cases within the CMA and within the AMCOE as well. We are at the early stages of fleshing out what it is that we're gonna do, but um, even if you think about this is a fairly standard manufacturing use case, we have a bunch of non-conformist reports, your corrective actions, and those are normally some kind of text-based description of something that happened that we need to fix. It's something's wrong with a part. And that's great. And you want to track them, that's great. But imagine being able to visualize uh uh clusters of those uh reports that say all of these reports are all similar. And the worst thing you can do is have multiple similar nonconformances. If something happens and it's never happened before, you're like, okay, we've learned this lesson, it's not gonna happen again. Well, when it happens again, you haven't really learned the lesson, right? So if you can start to just visualize those nonconformances, um that would be a great natural language processing example from an a type of machine learning where you just cluster based on and it's it's text. It's something that traditionally is not easy to cluster. You can't just throw it in Excel and hit cluster and you're done. A couple more steps, but it's not that complicated at the end of the day. Um, so that helps drive uh where you're gonna go, how you adjust your processes, etc. Um from a manufacturing perspective. So there's a lot of different use cases. I think the biggest thing that I tell people is you want a very specific use case, you want a very specific problem and then address that problem. When you go to Google and you say um AI solutions for manufacturing, you get this very hand-wavy, high-level, not specific solution. And that it's hard to see, you know, I'm a manufacturer today and I want to adopt AI. I mean, no, I need to, then I get that answer. But there's a huge disconnect in how how do I get from where I am today to doing that? Um, and there's typically a lot of much, much smaller use cases along the path to get there that are very impactful. Um, you just have to dive into them.

Caleb Ayers:

That last part you were talking about, how you know the the Google answer or the general answer is gonna be very vague and nonspecific. And um I just think about you know, Jason, who's over our manufacturing advancement division, I hear him talk all the time about how we're we're focused on technology we can help companies implement now and and in ways that are actually gonna help them. Um and so what you're saying, you know, making sure that this very specific use cases, maybe even specific problems that they're already having, and how how this tool, because that's what it is, AI is a tool. It's not some end-all be-all solve all your problems.

Meredith Gregory:

Well, I'll make one correction. To me, AI is a toolbox, and then all of your different algorithms are your specific tools. So your your large language models may be your hammer and it's easy to go to and you can do a lot with it, but remembering that there's a whole suite of other tools in that toolbox or even in the garage. Um so yes, it is it's absolutely your toolbox. Yeah, and there's lots of tools underneath that. But yeah.

Caleb Ayers:

So but that idea of you know that there's there are specific problems that this can help help alleviate, help solve, and that's kind of what our whole approach to uh I know Jason, they love to use the word optimization. You know, how do how do we help companies optimize their processes? That this would be another another thing that fits well within that. What do you see as some of kind of the main hesitations or um even drawbacks for for companies who are who are looking to implement this?

Meredith Gregory:

Yeah, there's a couple. Um the first one is a lack of just understanding what it is. Like I said earlier, people understand AI, or they are they know the term AI, and there's generally not a complete understanding of what that really means. And so um they'll they'll say, we know we need to do it, but I have all of these questions and I don't know what to do now. Um, so there's there's not many services that I see, there's a lot of very technical solutions, but there's a lack of um Jason calls it a being a tour guide, taking someone from where they are and helping navigate to where they need to be and what these terms mean and where the pitfalls are and helping avoid those and whatnot. So that's something that we certainly can do it. We're doing internally and we can do externally. Um, so just not knowing all of the steps that are required and having kind of that lack of clarity around the problem. Um the once you get past that, then you have a question about data security. And most organizations, no matter the type of organization, don't want their company's information to just be floating about for anybody and everybody to see. Um we've passwords on our computers for a reason, right? And so there's a fear that AI will expose all of this data, you know, your data's going back and helping retrain models and things like that. And it can at times. So I break them down into there's public, this is all around large language models, but um there's public services, so your chat GPT that you don't pay for, or your Gemini that you don't pay for, you are paying for it in the data that you're providing. Um, so yes, in that case, your your data is being exposed to these companies, they are using it to retrain your models. Um, if you have just very innocuous questions and it's no big deal, it's not really company information, by all means it's a great tool. Then you have private services where you're gonna you're gonna log in, you're gonna pay for a subscription. Now you're paying for that service in a monetary exchange. Um, and in most cases, it's important to read the fine print, but in most cases, your data will stay your data. Um, and then that data security concern is really no longer there. And you can host these models on you know your Gulf Cloud instances and and things like that. There's absolutely ways to make them secure. Um it's just a matter of making sure you're taking the steps to do that. So data security is often where people then are they're like, okay, I got it, I understand it. Wait, but now data security. Um and then they get through that, they're like, okay, I get it, that makes sense. We can do this in a very secure way, that sounds great. Now what? And the next step is around data. Um, so you we talked a little bit about training models. And if you're gonna customize any model in any kind of way, you can't do it without data. And a model only knows what date what it's been taught through the data that you provide it. Um, so if organizations don't have data in any kind of consolidated way, they don't have any data, they don't have the data that answers the questions that they want to have answered through these tools, that's where you get into some deeper conversations about, okay, really what what are the next steps? How do we get there? Um it's a data readiness problem is is gonna be the next big step.

Caleb Ayers:

So on that topic of data and training, um, I know the the old idea of you know garbage in, garbage out, and you mean you see that with you, I mean you see that with ChatGPT. If you ask it a very vague question, it's gonna give you a very vague generic answer. Um or or I should say if you ask it to to generate something for you, whether that be an email or or um whatever, it's it's gonna be it's only gonna be as good as the instructions you give it and the data it's provided. Um how does that I guess how does that idea apply for not just the large language models, but for the other um types of of solutions that you're talking about?

Meredith Gregory:

Yeah, so let's we'll use um a manufacturing example, let's do defect detection as our use case and we'll walk through it that way. So if if I've got um parts that are coming across and we're doing some kind of inspection to make sure that there's not any major defects, computer vision will allow you to go in and identify defects, right? So I can say um, does this part have a defect? Yes or no, or find defects, two different, you've got image classifiers or object detectors as your two different types of models there. Um to train those, you would provide a bunch of data of what a defect looks like, and a bunch of data about what a perfect part looks like. So if I train a model and I've got pretty standard defects that I see all the time, and I have a bunch of good examples, hopefully, it's a lot of a lot of good examples and only a couple defects examples. Uh, and you feed that in, the model will understand what that type of defect looks like. But say all of a sudden a machine crashed, something crazy happened, and the defect looks completely different. The model may think it looks like a defect because it looks more like the defective examples I've given it, but it may not. Um, and it may say, I have no idea what to do, I think it's 50-50, we'll say it's fine. Um, so you just have to think about the data you've provided it to make sure that it is a complete set of what in this case what kind of defects you may run across. Um but across the board, whatever kind of data you provide is the only thing that that model knows. Um so if you leave out a big segment of data that would be helpful to creating the correct answer, um you just cannot get there.

Caleb Ayers:

As you were using that example, I was thinking about uh the a lot of the the large language models, what do they call it, hallucinating, where they'll they'll make stuff up if they don't know the answer and they sound very confident. Yeah. Um does that happen across uh as you're talking about like anomaly detection and things like that? Like does are the do those issues arise in those areas as well, or is that just with that the large language models?

Meredith Gregory:

Um yeah, it's a good question. So large language models fall into this category of generative AI, and with generative AI, you are generating some kind of new content, whether it's text or an image or voice or something like that. Um at the end of the day, you're just predicting the next word or the next pixel, what have you, but um, there's a generative concept involved. So that's where this concept of hallucinations came from, is that it's generating something that's just not correct. Um, when you get more on the predictive side, you can absolutely have wrong predictions. Um that happens uh depending on how your model's trained more than you want. Um so you can have incorrect answers, but because in most cases there is an answer, it's not generating some kind of new text. They're not considered hallucinations, they're just considered errors. So you can have right your false positives or your false negatives, um, depending on the use case, one is worse than the other, but they're an error that you can measure and not just a uh new weird concept or something that's just yeah.

Caleb Ayers:

That makes sense. If you were asked by a company to give your 20-second elevator pitch for kind of why they should consider any type of AI incorporation into their processes, into their toolbox, um what what's kind of your elevator pitch?

Meredith Gregory:

Yeah, I think when it is a tool, there's plenty of other tools out there. You know, my the other half of my title is program management, just good process and program management can go a long way as well. Documenting things can go a long way. Um, so this is another tool and capability to help create efficiencies. When applied correctly, it can have massive impact. It is the tool that will automate the tasks that nobody wants to do. It's that task that someone's like, if I never had to do this again, I would be thrilled. And that is where artificial intelligence absolutely shines.

Caleb Ayers:

Thanks for being here today. Thanks for the rundown on all things AI. Anything else you want to add?

Meredith Gregory:

No, thanks for having me.