AI Wars and Industrial AI Battlefronts - A Conversation with Colin Masson and Jim Frazer Part 2 of 5

Welcome to another insightful episode of The Sustainability Podcast. Today, we are thrilled to bring you Episode Two in a collaborative series with ARC's Industrial Digital Transformation Podcast. Our guest is Colin Mason, Director of Research in Industrial AI at ARC, who joins us to delve into the theme of "AI Wars and Industrial AI Battlefronts."

In this episode, we continue the discussion from our first episode, exploring whether the impact of AI on the industrial sector is evolutionary or revolutionary. Colin shares his perspective on the various battlefronts in the AI landscape, including data center hardware, general-purpose AI software platforms, and the emerging edge AI technologies. He also highlights the role of industrial grade data scientists and the importance of strategic partnerships in navigating these advancements.

 Tune in as we unpack the complexities of AI's role in the industrial sector, discuss the major players and their strategies, and look ahead to future developments. Whether you're an industry professional or simply curious about the intersection of AI and industrial innovation, this episode is packed with valuable insights.

 Join us for this engaging conversation and stay tuned for upcoming episodes in this series, where we'll cover industrial AI use cases, industrial grade data fabrics, and the role of AI in powering sustainability. Don't miss it!

--------------------------------------------------------------------------

Would you like to be a guest on our growing podcast?

If you have an intriguing, thought provoking topic you'd like to discuss on our podcast, please contact our host Jim Frazer or Our Producer Tom Cabot

View all the episodes here: https://thesustainabilitypodcast.buzzsprout.com

Transcript

Jim Frazer  0:02  

Welcome again to another episode of the sustainability podcast. This actually is episode two in a collaborative effort between AR C's industrial digital transformation podcast and the sustainability podcast, both of course by aarC and as in our first episode that was entitled industrial AI evolution or revolution. I'm again thrilled to welcome Colin Mason, Director of Research, industrial AI here at aarC. Welcome, Colin, how are you today?

Colin Masson  0:36  

Great, Jim. Looking forward to this one.

Jim Frazer  0:39  

Great, great. This is one of our more SEO friendly and exciting titles. Episode two in our series is AI wars and industrial AI battle fronts. So let me just start with whether you view the business outcomes in the industrial sector as evolutionary or revolutionary, as we discussed in our first episode, there certainly is a lot of chatter, noise, conflicting opinions, and old all things about AI. And here they are see, we have used the analogy of the AI wars to contextualize what what's going on and what we're thinking. I know you are at the forefront of that. So please share and take your time.

Colin Masson  1:33  

Yeah, where did we end up on revolutionary or evolutionary?

Jim Frazer  1:40  

Our last bit of a hybrid answer for both of us? Yes,

Colin Masson  1:45  

I think it was, I think the impact on the way we interact with software generally, I think is I think we're starting to see could be revolutionary. But in terms of Gen AI techniques, it's really just another tool in the industrial AI toolbox. So, I think that's where I stand on it. That's kind of evolutionary in terms of the overall AI toolbox, but potentially revolutionary in the way that we interact with technology in general. And that's really the analogy of the cell phone, right? Yeah. And I'll touch on cell phones a little bit. Again, and this analogy of AI was and to your point. The reason it's, it's useful is I'm obviously a maybe a special case, because I subscribe to all sorts of AI related topics, not always knowing I'm doing so but I am. So, my inbox and my news feeds are full of news on AI. And it isn't always clear. Is this a significant AI announcement? Isn't a me to AI announcement. What kind of AI are they talking about? Is it one of those evolutionary steps or revolutionary? So, I think the idea of framing what's going on in AI as a war with lots of different battle fronts, skirmishes going on? And, and all sorts of propaganda going on, right? Is I think, a useful one just for simplifying or categorizing all the news into right, which Battlefront is that. And then you can take a more measured approach does not get overloaded with all the news that's coming in. So, let's talk about that analogy a little bit more. Jim? Because I think the battle fronts I kind of I kind of split them into kind of AI battle fronts, and then there's some specific industrial AI battle fronts, but let's talk about the big AI battle fronts first, because I think there's, we all tend to think of Nvidia as right now the biggest beneficiary of the investments that are being made in AI technology. And a lot of that is concentrated in the data center and kind of building out capacity both for training view, foundation models, and of course, deploying those models and we'll talk about the Energy Impact and, and water impact in a future episode. So, let's not dig into that just yet. Jim, but I think that'll be a later episode that we'll, we'll go into, you've clearly got a battle For the data center hardware, and while Nvidia is the stock that is doing best out of that right now, because they clearly leading they have the best performing technology for those data centers. There's also a concern that it's a bit of a monopoly. And so, you've kind of seen the hyper scalars, you've seen Google and Amazon, AWS, and Microsoft will come out with their own hardware for the data centers, because they understand and to some extent, or to mitigate that dependency they have on in video. So that's one Battlefront is kind of the data center hardware. And that's probably, if you're investing in stocks, that might be the most obvious ones. Everyone sees how well they're doing. Another battle front is really, for general purpose, AI software platforms. And you can obviously break that down into more granular battlefronts. But I mentioned the hyper scalars are trying to mitigate their dependence on Nvidia for the data center hardware, but they're very much at the forefront as well of the building the platforms and trying to get everybody to adopt their particular platform. They're also kind of expanding beyond in the case of Microsoft, obviously, the most popular, open AI as the is arguably who started the hype, but also nowadays, only one of the large language models that you can really provide are a foundation models, not just live language, but image mixed mode models that you can leverage to build AI solutions on top of and we're not specifically talking about industrial AI, we're just talking about general AI technology that will potentially revolutionize daily lives. But if you look at AWS with Amazon bedrock, and a collection of models that you can choose for each particular problem, Microsoft's kind of followed them down that path, but essentially, they're trying to make sure each of the hyper scalars and I can't leave out Google, I maybe not quite as big a fan of Google in the industrial space. But that's got more to do with the fact they don't have the IoT platform anymore. And so there's a set of use cases where I don't necessarily think of Google as being the leader. But they started a lot of this with the transformer technology, this latest wave. And, of course, with their access to massive amounts of consumer devices. They very much feature in the building out of these general purpose AI software platforms. And of course, you've got the open source platforms themselves and the tensor flows and, and hugging face community sharing of AI models and things like that, that could be lumped into this category. But there is a big battle for what is the platform, right, that you are going to build your library of AI solutions on top of and that's essentially the hyper scalars. And some of those long standing open source platforms and probably the most obvious and they kind of spread across spread across the Gen AI platforms. But then a lot of what Microsoft and Amazon and Google have had for some time, it's kind of a library of machine learning and neural net deep learning technologies that go beyond the kind of latest focused on Gen AI. So really putting a platform where you can it is a toolset. It is an AI tool set and you pick the right AI tool to train a particular skill. That's really what they're all battling for right now. And then there's another Battlefront would be which is much more of interest to us in the industrial sector is we haven't really seen the edge AI Battlefront really explode yet right. I think we're starting to right. And this is going to be a combination of yes industrial grade hardware at some point. I But let's start with what's already happening. And showing that there is a Battlefront there for edge AI, with some of the recent consumer developments. So think about what I think was pitched as it was going to be an absolutely incredible AI WWDC or the Apple developer conference ended up being Apple intelligence, right. And so it is their take on AI and how the new hardware or hardware that they have with iPhones and the M series chipsets have NPU capabilities that can run pretty large models on the on their devices on the edge. And so a lot of what they're promoting now as the fact that you can run these models and an increasing set of use cases on the edge and take advantage obviously, of some of the most broadly deployed cameras. So for example, we'll see a lot of computer vision. Use cases become a lot more feasible and practical. But perhaps just as interesting was the way that they were pitching, protecting that data, protecting your privacy doesn't all have to go to the cloud and round trip or if it does need to, it is encrypted,

Colin Masson  11:31  

all the way. And that followed the windows, or sorry, Microsoft announcement by a couple of weeks of their copilot plus AI PCs, right. And so they have also now tried to embed capabilities on edge hardware. So they have these view copilot plus PCs have 4040 tops of capabilities in NP use. And so that's the minimum spec, the Apple devices by the way, I kind of ran about the same I think there are 38 to 40 talks as well of NPU performance capability which can run a very wide range of models. And so you'll see edge AI hardware and software evolve to take advantage of that hardware that they're really trying to standardize is kind of the minimum spec for running AI models on the edge. And of course we will, ultimately in the industrial sector, there is a large amount of commercially available hardware that gets deployed on the edge. But we'll also get into a little bit of the edge hardware Battlefront that there's different chipsets I've chosen to highlight a couple of the big consumer announcements because that is scale, right. And that kind of starts to give you an indication of how quickly these edge AI software capabilities are likely to expand and emerge or emerge and expand as, as that capability is mainstreamed in the devices that we carry around in our pockets or as our main way of doing work on a on a daily basis. So that's really, it.

Jim Frazer  13:42  

We talked about noise and just this past week, we had the or last week, the Apple Developers Conference. And soon after Elon Musk had a number of points of remarks about that development. So there is a lot of noise, like

Colin Masson  14:02  

what are the stakes, the stakes are high, right? The stakes are high. They are battle fronts, it is war. And I was having a discussion with a client who didn't want me to use the AI wars analogy in a keynote I was doing for them because they felt like that's not our culture. We don't believe in winners and losers. And it was like I respected the client. But I think the stakes are high here. There are going to be winners and losers.

Jim Frazer  14:32  

Again, I'll go back to some of my comments my closing comments on our first episode about bow the elegance of a user interface and the smartphone. And, you know, no one when they've unpacked my phone today needs an instruction manual. You just use it. And there still are many applications in the world today where there's a long ramp up time, a long training cycle. And one of the greatest impacts of AI is going to be to minimize that. You know, imagine we, you know, the world has talked about Google and its search engine, will we need to read pages and pages of snippets? Maybe not? Maybe not. Colin, I think one of the battlefronts I'm interested in and I think you talked about it a bit obliquely. But what about, there's all these announcements, technology and business announcements, and it isn't AI war, and I'm very interested in alliances in general. And then in particular, the collaborations between industrial automation multinationals and the Data Fabric companies. Can you

Colin Masson  15:52  

touch on that a bit? Yeah, I was, I think I started with the AI, battle France, right. And as I kind of start to talk about the industrial AI battle France, then I can really expand on the question you're asking, because that's very much about alliances. But let me just address your comment about Elon Musk. And I'm not going to say whether I like him or not, right. He's obviously highly respected by some and, and has huge influence in the mainstream technology sector. But there is a Battlefront, which is analogous, if you like, to the propaganda wars that we've seen, right? There's lots of lobbying going on, and so on, in, so take some of those comments and reactions to various announcements as part of that. propaganda war, if you like. They're all fighting. I mean, Musk has his own AI model, right. And there's a lot of competition going on. And that's just one of the battle fronts, right. So I did list kind of AI lobbyists campaigns, as one of the battle fronts to keep an eye on a lot of announcements will be the battle for mindshare, or the battle for to preempt or avoid certain legislation or shape legislation governing AI. And so that's, that's another Battlefront. And, you know, talking of some of these characters that we have out there leading a lot of these companies. There's also a Battlefront for the AI gurus, you know, who is the next. Sam Altman is obviously a well known name, CEO of open AI, but who is going to be the next Sam Altman leads the next breakthrough. So that's one battle. But it's also a good point to transition to the industrial AI Battlefront, because we've got an even tougher battle frontlets emerging, which is really a battle front for industrial grade data scientists, right. So the people that can actually build industrial AI solutions that understand both the AI domain or the data science and AI, domain and the industrial domain, so they understand time series data and mission critical needs that we have in our particular set of use cases, not all of them are mission critical and use time series data, but using those as kind of illustrations of additional complexity that that we have. So there's a battle for AI gurus, which is really the, you know, the hyper scalars and the largest software and hardware companies in the world fighting for those gurus, we have a slightly more mundane, but nevertheless, it could end up being the real differentiator between those that succeed and those that don't in the in applying industrial AI is really access to industrial grade data scientists and it's kind of one of the battle fronts. Your question kind of framed? Really what we're hearing from industrial automation and software vendors who can't avoid those larger battle fronts because they depend then very often it's about taking that commercially available cloud or edge hardware and industrializing it or making industrial grade with their particular skill set, and they are already in the battlefront for industrial grade data scientists. I mean, I talked with I mentioned, I did a podcast with one of the first chief AI officers that I had actually talked to was at Schneider Electric, and he has a team of 400. Or at least he did when I, I talked with him, that he is a master across Schneider Electric kind of focused on solving and it's a blend of data scientists and domain experts, but gradually is going to building that, that that industrial grade data scientist capability in his organization. But then I was at another event recently where Siemens is doing the same thing, right, they have 1500 spread across their organization that they are starting to, so they're already amassing, if you like, their, their soldiers, their platoons for the battlefront on industrial AI.

Jim Frazer  21:27  

To head actually to Mike X. No, it's fine. Fine. It's intuition. So I was prepared to ask, so how our industrial organization suppliers, standards organizations, trade associations in the industrial domain? How are they getting pulled into? Or are they actively creating their own real estate in in the AI domain?

Colin Masson  22:00  

They are, I would, I would kind of break it into because the other Battlefront. So I see a really around there is that Battlefront where I'm now seeing probably more on the vendor side, but I'm expecting to see some form of it emerge in in industrial organizations who specialty isn't necessarily they're not in the software or automation game they are making or making products or, or producing energy. So those organizations I think, will also start to have their digital transformation teams kind of expand into having a chief AI officer or some role that really zeroes in on the combination of skills and how to govern AI, how to form partnerships. And actually I cover a number of those topics. By the way, Jim, in that leader laggards report I talked about it's about how leaders are using industrial AI to widen the digital divide. And as I looked at that, and looked at the characteristics from 525, industrial organizations, we did see some practices and that included setting up a center of excellence for AI, having policies in place about ethical AI, also some interesting characteristics that kind of hinted and really precipitated some of my thinking on industrial grade data scientists and that Battlefront, which was that they were more likely to rely on internal skills. And we're investing in building internal AI skills than the laggards who are more likely to go to a consultant or a system integrator and just lean on their skills. And it doesn't mean the leaders are not also using system integrators. But it was clear that they put a heavier weighting on building internal skills and capabilities, had already invested in governance models around AI. And we're taking, you know, ethical considerations into account as well. In that particular case, the vendors are being relied on. And so when I look at the industrial software and hardware vendors that are supplying the industrial organizations, I see that partnership deeply. Because there's all because they do have these on years of industrial grade data scientists then understand the problem, they understand the technology. And industrial organizations are turning and turning to those partners in deepening relationships. And so they're not necessarily they don't want to be in the general AI wars, they really want to lean on their industrial software and automation vendors to navigate that, and to bring them the best technology, the right tool for the job and keep pace with the emergence, you know, there may not if you use the rule of threes, in most markets, you know, there won't be one winner. In in pi, I mean, rule of threes will still apply, you probably still have in each industrial use case, you may still have three solutions that that make sense. I mean, that's as you markets mature you, you normally see that consolidation and get down to that. But I think in the meantime, that chaos they really want industrial organizations really want their key strategic industrial software and automation vendors to navigate that for them. Right. They have the skills they've they can afford those skills, and they understand the domain. And so that's the pattern I see emerging. I thought it said your full question live chat.

Jim Frazer  26:40  

This has been another fascinating half an hour. Now I think we might close with one submitted question. Can you expand on the key protagonists in the industrial AI Battlefront world? Versus the general AI? Battlefront? Are there different sets of the protagonist and arguably antagonist?

Colin Masson  27:05  

Yeah, look, I think we, one of the eye, I expect to see and are already a handful of more than a handful. But I'm already aware of quite a few innovative startups that recognize that gap. Right? That really in the general AI wars, it's all about large language models or foundation models, Frontier models and getting to the next evolution of those general purpose AI technologies. But startups in there's a lot being invested in startups in a fair number of them are industrial, AI startups that are focused on autonomous AI or a specific domain problem like chemical synthesis. And so we've got one called chemical.ai, or at our IRC forum, we had Kent Sanderson from composable, who specializes in autonomous AI and, and solving some of the most challenging use cases in multivariable control. And beyond or Kelvin, who've been around for a while and do tackle a lot of those edge AI problems. So you'll see a lot of these startups, I think we'll see a lot of money go into those startups for specific edge AI use cases or another way of thinking about it as narrow AI. Right. So you'll have all the, the hyper scalars and the big AI invest, innovators that are getting some cases, billions of dollars of investment to get that next frontier model, because the stakes are so high. They will continue to, to focus on Gen AI and maybe making the next breakthrough to get towards true general intelligence. But for us in the industrial sector, most of the use cases I would still saying most of them have an edge component. And so the battle fronts and the protagonist are about solving the edge to cloud. Use cases that we have, and so there'll be no startups that know how to deal with. We talked about it for time series data or how to develop small, small models. Train them for very narrow use cases, but deploy them at scale. And that's where the That kind of cloud may come in as the mechanism for helping you deploy lots of edge instances of those models. And then, of course, I mentioned already, that really industrial organizations are recognizing their dependence or the opportunity to deepen those relationships with whether it's Rockwell Automation, or Schneider, or that could be careful, because I won't mention somebody but you get the point. It's those industrial automation giants there, they have to double down and show that they can actually deliver. As they're expected to AI embedded in their solutions or enhancing their solutions, the new UI for their solutions, or augmenting them with the assistance, which is another way of thinking about the investment in the UI. So I still expect to see the big automation vendors. But if they don't move fast enough, then I think we'll see a lot more of the startups and smaller companies that are more nimble, more narrowly focused on specific verticals or use cases should do pretty well out of this. The other category, we tend to forget, Jim is that industrial organizations don't just have factories or refineries and distribution, pipelines and grids, etc. So it's not just the operations side of the business, they still have sales, they still have servers, they still have supply chains that they have to manage. And so the enterprise vendors, we're also seeing pretty quick because they don't quite have the same set of problems in terms of diversity of the solutions in the factory or in production. It's kind of much more standardized, which I think was part of your question as you move kind of from the factory floor on the shop floor, up to the enterprise supply chain, dealing with customers dealing with your suppliers, that data is much more standardized, much more transactional, and the software solutions there already have high data quality. And they're jumping at the opportunity to embed AI, how they going to monetize it, we're not quite sure yet, but we're already starting to see a kind of mix of add on solutions that you have to pay for and then some teaser capabilities, if you like, where they enhance the URL column.

Jim Frazer  32:58  

That's a great a great segue to episode three, which is industrial AI use cases. So not to cut you off. But we have so much information to cover. We've got five episodes, I don't want you to run

Colin Masson  33:17  

out of steam. Give us give us the

Jim Frazer  33:19  

punch line too early. Yeah. So once again, again, this is this is episode two in our industrial a AI series at aarC advisory group. It's broadcast on both the sustainability podcast and our digital industrial digital transformation podcast. The title of this episode was Ai wars and industrial AI Battlefront. And once again, my guest has been Colin Mason. Colin, I believe your email address is C M A S S O n at ARC web.com. And he's very active on LinkedIn. So feel free to reach him there there. Colin, any last words for our audience today

Colin Masson  34:05  

today. Just enjoyed the conversation again, I'll let you speak a bit more and next time. But just for those that kind of want to get a deeper dive into the AI wars analogy, it is on LinkedIn. So I kind of updated it with some feedback that I got from a number of clients. So look for it there might go a little bit deeper into some of the hardware that's powering AI I dig a little bit into FPGAs and GPUs versus GPUs, and ASICs technology which a lot of custom hardware is, is being built for AI and that could be an interesting way we'll see the edge hardware evolve in the world of industrial AI. So, to get more dig into that, but the next one will take quite a bit deeper into what are the AI use cases we kind of mentioned it a few times, but

Jim Frazer  35:13  

in our, in our next our next few episodes, coming next is an industrial AI use cases. Then we have industrial grade data, fabrics and AI. And we'll wind up this series with industrial AI is role in powering sustainability. So, thanks to everyone for joining us today. Colin, thank you again for all your insights and we'll see you again on another episode of both the industrial digital transformation podcast and the sustainability podcast and very soon. Thanks for joining us.