Podcast: The AI you don’t want to use

DS: The first example—the first question I asked you—what was a kind of ridiculous or crazy AI? I have a feeling like we might be able to find something a little crazier than that. I mean, something really out there. What's the most out-there AI pitch you've ever had? The kind of thing where you look at somebody and go, what are you talking about?

PS: Well, I mean, I filter most of that out, so I really don't hear the pitch. I'll give you another one where one of the startups came in and said, through a… virtual metrology—which is a buzzword I've been hearing for the last 10 years—we would eliminate and kill, and we don't need metrology tools anymore.

And so some of the most profitable companies in the world are these cutting-edge metrology tool vendors that make these tools. And I'm like, I think the first thing you need to do is go and pitch—pitch this idea to those guys, because they're spending millions of dollars researching and bringing up new tools. If this is true, right, they should be the ones jumping on it. So then they don't have to make the tools; they just sell us the solution.
So these kinds of crazy ideas always come up. You know, either they can eliminate your manufacturing costs altogether by cutting out operators, or eliminate the need for the tool altogether, right? And so these are the far-out things that I've seen. Like I said, most of the pitchers don't even come to me. I filter them out. It's only the handful of really persistent ones—or maybe if I'm interested in understanding a little bit more—that I spend time with. So I think most of the crazy pitchers don't even come to me.

DS: What is a resource that people who develop AI-based software—people who come to you to pitch—is there a resource, something they should read, someplace they should go to get a handle, even a basic handle, on what manufacturing issues need solving before they go pitching a manufacturer of software? Does that make sense?

PS: Yeah. I mean, there's so much, right? I mean, the semiconductor industry is rich. The SEMI organization has standards with which we operate. And so anyone who's interested can understand what are the challenges we're working on.

We have close collaboration, even with our competitors in the space, where we're working on common, you know, challenging problems. And all of this is well documented. We meet on a regular basis. There are conferences—there's SEMICON, semiconductor alliances, semiconductor conferences—that tackle a lot of these challenges, right? Cycle time, the laws of physics, how do you get things faster, better, quicker?

How do you get the most amount of efficiency through a particular-sized shell? How do you fit in? What's the best layout plan? How do you build a shell, for example? Do you build it all at once? Do you go step by step in a gateway process? And all of this is full-on problems that are out there. Everybody's solving them. There are papers published on a weekly basis, on a quarterly basis.

And so anyone who wants to get into this space, there's so much literature out there, so many areas where they can tap into. I think sometimes there's a little bit of laziness, or maybe a feeling of like, well, we know what we're talking about. You know, it's basically a nail looking for a hammer, and not the other way around.

DS: What are some examples of AI pitches—AI software—where they were almost there? They almost had what they needed, but they were just missing something. Some crucial misunderstandings, some lack of compatibility. What are some close calls, I guess? Let's call them close calls.

PS: I think—yeah, I mean, look, I'll just say with one of the partners that we're working with, they almost fell out of the loop because they came with a pitch that was not there. In fact, we helped them with the last bit of it—like, say, retool it or rework your pitch to fit this angle—and we can start working together. We're working with them, so I won't say who they are.

But they came in and they looked at it. I think they had 90% of the way. It was more on, like, how do we dispatch material? How do we run material in the fab, right? And so they had everything. But what they didn't understand was our tools. What they didn't understand was our tools are specialized—meaning you can't cross boundaries, right? If it's a lithography tool, you run lithography, right? You don't run any other process on it.

And they thought the tools were interchangeable, sort of like if you look at it from maybe an automaker kind of view, where you can retool, retrofit tools to fit your need, right? And you can actually do a lot of crazy things in that space. For us, because the tools are very highly optimized and highly specialized, I think they made a big miscalculation ahead of thinking they could easily retool a tool that performs function X to do something function Y.

And so that was where they fell apart, right? They didn't understand that concept. And so when we sat with them and we explained it to them a little bit more and gave them a lot more detail—by the way, opened our books to them and showed them how things are run—they got it, right? So they're smart people, right? Once we educated them, they got it. They retooled their pitch a little bit. What they thought was going to be the gain was significantly lower, obviously, but there's still gain there. And it's one of the programs that we greenlit—we're actually evaluating it and moving forward with it.

So sometimes it's very— for a trained eye—it's an obvious error, right? But because they're so focused on delivering the solution, they don't really realize what the problem is and how to frame it. And this is where advisors and experts in the field are vital. So those that come with advisors and experts in the field have a better shot, and they really can understand and grasp the challenges too.

DS: This question is aimed at manufacturers like yourself who are constantly fielding a bunch of AI pitches, and they need to find ways to filter things out. So we've already talked about how the idea of saying you're going to replace all operators with AI is ridiculous. You're always going to have some operators on the floor, right? What are a few things that you think AI should never be applied to? Like if somebody knocks on your door selling you a product that uses AI to do this—back right off.

PS: I think anything to do with how we handle our teams, for example—how we handle our employees, the link between a supervisor and the team that they manage. It's almost like—I draw an analogy—it'll be almost like a student kind of researching a paper, not doing it themselves, but rather asking AI to do it.

I think that relationship, that bond between, you know, supervisor, team members, managers, and employees—that's sacrosanct. And I've seen where some people are trying to write, for example, performance reviews, right, with AI. It's okay to have a template. It's okay to have a template that you use—but by the way, you look at me like that, and I know that some people are using it, right.

DS: And I just gave him an incredulous look, everybody, when he mentioned that. I looked at him incredulously.

PS: But everything—you would think it's fairly obvious that you cannot replace this part of the job, right? That relationship between the team, that dynamic. And you can imagine if everybody's getting the same cookie-cutter kind of performance management or performance review, I think that's very disrespectful to the team. So that's one line that I draw hard on the side. And there's no substituting that bond, that relationship that you have to grow with your team—that real authenticity that's needed to bring that. I think that's one of the solid lines.

Everything else, honestly, is fair game. There's room to improve a lot of different areas, even when it comes to handling tough challenges or automating problem-solving or maybe even decision-making. I'm very big on that. But when it comes to dealing with human beings and others, I would not like an AI to be telling me what I need to do better, right? I want it to come from a person, and vice versa. So I want that in my organization as well.

So this is one area where I really push back. Okay with the framework, okay with speeding things up—but the content of it has to come from a human being, right? You cannot replicate that. That's just terrible.

DS: Have you ever had anyone try to sell you software that does that?

PS: Oh, absolutely, all the time. It's not selling it, but even when I look around at where the applications are, you can easily see this. There are companies that talk about how we automate this process completely, right—performance management and all that stuff.

And I think there's a fine line, and I know the intent is good—to make sure that everything's captured, like you said, right? Making sure that notes are captured, everything is clear and concise, and nothing's missed. There's a fine line between that and automating the process completely such that it becomes this mechanical, cold process that has no human— you know—human touch at all.

And by the way, it's not uncommon. I've just spent some time on YouTube, when you talk about how people are hired, fired, and all that stuff—it's becoming a norm right now. People are really advocating that responsibility. You want to break tough news and all that stuff—that's being automated as well, which is insane in my view.

DS: Okay, no manufacturer is going to adopt or really take seriously an AI-based tool until they've really had a chance to look at it, had a chance to talk to the founder. But I'm wondering if there are any kind of positive signs, encouraging shorthand, that if someone's coming to you with an AI-based tool—if you hear a certain thing—you know that they might be onto something. If you hear them talk about this, if you hear them—are there any early signs for you that someone actually has AI that might be useful? Like before they've gone through their whole pitch.

PS: You mean like when someone comes with the concept and says whether there's any value to it? Like even before?

DS: When somebody offers you software—it's like it's early in the pitch, right? They're describing it. Is there a phrase, a keyword, something that if you hear early—it's early in the pitch and they've already covered that—this might be something. Again, this is a tool for sifting through offers. Is there anything like that? Does that make sense?

PS: Yeah, as a global company, there are two things that I want to hear when I hear an original pitch or when someone has an idea, right? One is, you know, reduction in repetitive work—like they’re focusing on something that’s repetitive, that’s reproducible.

But the word that really catches my ear, leading a worldwide organization, is how they can scale across geos. Like scaling—they’ve thought about the scaling piece of it. That’s really clear. Because sometimes the solutions are so customized, they don’t understand, when they’re pitching to a company that has sites all around the world, where data streams may be a little bit different, right? They’ve not thought about the scaling idea yet.

And so when they come to me, I’m looking for that scalable solution, right? And the work that we’ve done with Minds, for example—Minds AI—that was a scalable solution. That’s why we picked it up, right? Because it was agnostic to which fab we piloted in. And then once we saw the gain, they could immediately say, okay, we can scale this very quickly because the architecture was similar. You know, it was agnostic to how the data was plumbed, things like that. And so, to me, once I hear the word scalable, I’m already starting to think about it, right?

DS: And if you want to find out more about Minds AI and what Pradip was talking about, the story is How GlobalFoundries Tames AI for Real Gains. We published it on February 13th. So if you want to learn more about Minds AI, you can go read that there. Is there an issue in manufacturing for which AI could be useful, but no one’s doing it yet?

PS: I think the number one challenge we have is how do we get more productive when it comes to labor? How do we automate for the task? I think a lot of what we see when it comes to AI is in sort of adjacent industries, right? You talk about coding, you talk about simple presentations.

But the sheer act of understanding data flows from point to point, coming out with a solution to propose either fixes or containment along those lines—I’ve not seen anyone come up with an end-to-end manufacturing solution with AI. Because there’s so many different pieces. The tool is running, it’s spitting out gigabytes of data, petabytes of data, which then has to be correlated into another application where you look at SPC charts and FDC charts and their signals and all of that.

Nobody’s looking at it holistically, end to end, right? Everybody’s got point solutions for different parts of the problem. But I think if someone were to really string it together and look at the entire process from end to end and figure out every point where there’s a human involved in decision-making that can be automated—and there’s tons of it—I think looking at it holistically, end to end, would be a really good piece of the puzzle that would solve a lot of it.

Because at the moment, what we have is AI-based solutions in certain areas, right? Some are really good at analyzing charts and spitting out potential fixes or use cases, but not very good when it comes to understanding how the equipment is transmitting the data or what the equipment challenges are.

I think that’s one area—and I push my team, I push the teams we’re working with—to complete that feedback loop. Still, because they’re looking at it at different parts, the engines, the AI engines that are developed, are basically very, very fine-tuned to certain parts of the flow and not the overall picture of it. So I’m looking for that end to end. That’s what’s going to unlock the next level of learning and the next level of productivity. Otherwise, there’s always going to be different pieces everywhere.

About the Podcast
Great Question: A Manufacturing Podcast offers news and information for the people who make, store and move things and those who manage and maintain the facilities where that work gets done. Manufacturers from chemical producers to automakers to machine shops can listen for critical insights into the technologies, economic conditions and best practices that can influence how to best run facilities to reach operational excellence.

Listen to another episode and subscribe on your favorite podcast app