FuturePrint Podcast
FuturePrint is dedicated to and passionate about the power of print technology to enable new opportunities and create new value. This pod features deep-dive discussions with the people behind the tech as well as market analysis, trends, marketing and storytelling!
FuturePrint Podcast
#302 - AI at the Edge: Why Print Cannot Afford to Miss Out
In this episode of the FuturePrint Podcast, Marcus Timson is joined by Dr Peter Brown, Chief Commercial Officer, and Jamie Jeffs, Director of Industrial Instrumentation at 42 Technology – a UK Cambridge-based consultancy that bridges bleeding edge technology and real-world manufacturing.
Together, they demystify what AI really means for industrial print and packaging, with a particular focus on edge AI: running intelligent models directly on or near machines rather than in distant data centres. Jamie explains how advances in edge silicon and neural processing units now allow complex tasks such as machine vision, anomaly detection and even small language models to run locally – delivering faster response, improved cyber security and more control over sensitive production data.
Peter brings a physicist’s pragmatism to the AI hype cycle. He argues that every project should start with a simple question – “What problem are we trying to solve?” – and a hard look at the underlying data. In many factories, institutional knowledge, paper records and patchy logging still dominate. Before any AI can add value, sensing, data capture and basic analytics must be put on a solid footing.
The discussion ranges from practical use cases – predictive maintenance, stabilising complex print processes, smarter vision systems – to the strategic threat of global competition. Both guests stress that AI is unlikely to remove humans from the loop; instead, it will augment operators, capture expertise and help mid-sized businesses compete with better-resourced rivals.
If you’re curious about how to start an AI journey in print without being overwhelmed by the hype, this episode offers a clear, sober and encouraging guide.
Listen on:
Apple Podcast
Google Podcast
Spotify
What is FuturePrint?
FuturePrint is a digital and in person platform and community dedicated to future print technology. Over 20,000 people per month read our articles, listen to our podcasts, view our TV features, click on our e-newsletters and attend our in-person and virtual events.
We hope to see you at one of our future in-person events:
FuturePrint TECH: Industrial Print: 21-22 January '26, Munich, Germany
I can see myself in the background. Reflecting off the reflecting off the whiteboard. I can see my scalp shining.
SPEAKER_02:Right. So I think we are recording.
SPEAKER_00:Okay. Countdown? Three, two, one.
SPEAKER_02:Started by you. Let everyone know they're being recorded and transcribed. You are being recorded and transcribed, Marcus.
SPEAKER_00:Brilliant. Welcome to the latest episode of the Future Print podcast. And I'm really happy to have back with me Dr. Peter Brown from 42 Technology. Thank you for joining us, Peter, again.
SPEAKER_02:You're welcome. Nice to see you again, Marcus.
SPEAKER_00:Yeah, and Jamie Jeffs from 42 Technology, who hasn't been on the podcast before. And we're going to find out more about Jamie and his work, and we're going to have a really interesting conversation because they always are with 42 Technology. They're unique with all of the partners that we've worked with because they do some fascinating work in a variety of industries helping people and businesses innovate successfully. And I know the theme of this talk will be revealing, educational, and exciting. So, anyway, welcome to the podcast, gentlemen. Thank you. Thank you, Marcus. Great to be here. And today's talk, we were going to focus a little bit on the um possibly one of the most hype technologies that I've ever been sort of witness to is around AI. But before we get into that, um, I want to start with introductions. Peter, would you give us an in brief about yourself and your background and what it is you do at 42 Technology?
SPEAKER_02:Sure. So, well, uh fundamentally, I'm a physicist. Um, and uh uh if you cut me through the middle, that's what it'll say all the way down. Um currently I'm the chief commercial officer at 42 technology, um, and uh which makes me um responsible for all the sales in the business. But if you if you ask Jamie how I go about doing my business, I'm still still a physicist, and that's how I do it. Um and I've been in technology consultancy all my career. Um, and I've probably been doing digital printing uh and coting for a good two-thirds, three-quarters of that.
SPEAKER_00:Yeah, yeah, absolutely. So it's starting physicist and then into printing, um, to the two P's, um, um which is um which is interesting. I know that um you've you've got a lot of insights around technology for printing and innovate and everything. We'll come on to that. Jamie, uh we haven't spent before. I'd be interested in a little bit of your backstory and and what it is you do for 42 technology.
SPEAKER_01:Sure, thanks, Marcus. So um, yeah, Jamie Jeff. So I've been with um 42T just under three years. My background is 20 years really in in instrumentation and sensing businesses. Um, that's right through from sort of RD, manufacturing, sales, marketing, uh, working with largely industrial clients, um, fundamentally helping people use the data that these sensors produce to be able to develop insights, to be able to protect their processes, and and and quite a lot of work in sort of environmental sensing. So doing some really interesting things, helping uh manage pollutants, gases, particulates, VOCs, things like that. So lots of different technologies uh spanning sort of spectral analysis, uh radar ultrasonics, uh, and and everything in between. But uh yeah, 42T, I uh like Peter work largely in the commercial team, um Director of Industrial Instrumentation, so focused uh a lot on our product development work, uh, but also supporting our work in what we call manufacturing innovation. So this is helping companies um using sensing and instrumentation and the data it produces uh to give uh you know improvements to what they're doing in their production processes. And of course, with that comes the data management. Obviously, we've had IoT, and now really it's about how are our clients wanting to leverage AI and some of the capabilities with uh AI in their products, but also in their processes. So, yeah, I'm really interested to uh obviously join the chat today, but uh hopefully I can give a slightly different insight to things.
SPEAKER_00:Thank you, Jamie. That's really useful. And um we did have a little chat before, and and you talked about this thing called edge AI. Could you give us a little explanation in simple terms what edge I actually is and why it matters so much for manufacturing and industrial printing?
SPEAKER_01:Sure, no, and I think it's a really, really interesting area of of development, uh, particularly in the world of sort of sensing instrumentation, but machinery and and the industrial space uh in its entirety. Um obviously edge processing, uh which is the idea of having data that's managed close to the source. So if you're from my background, you're talking about sensors that are processing data close to a particular production process or a big digital printer, having that data processed close to where it's where it's occurring. Now, the the comparison is between really edge and cloud. So cloud is the concept that you're taking data you're producing and pushing it up into the cloud, and for that we mean big data centers, which for everybody understands from the news, you know, is a massive expansion with AI. Edge is really saying you can do that processing now close to the source, and in many cases without needing a connection to the internet, and therefore not sending that data into the cloud. So, why is that important? Well, there's some technical benefits there around things like latency, so you can get sort of faster response. You can also look at it from a you know perspective of um cybersecurity and saying that you've got the data uh staying within the premises, which for our industrial clients is a big challenge. And obviously, there's been a lot in the news about that over the last year or so. But really, then you introduce the concept of AI and being able to process models uh on the data that's being generated close to source. This, I think, is the real uh exciting area for edge AI, as we call it, or AI at the edge, which is now transferring some of that model capability that these um machine learning, deep learning, generative AI, even now things in the sort of natural language space, which are getting really exciting for companies wanting to do more locally, uh both within their products and within the processes uh that they're being deployed in. So, yeah, really exciting development, I think, in the last sort of 12 to 18 months where people are starting to explore what they can use edge AI for, uh particularly in industrial applications.
SPEAKER_00:And for the the lay person who's not as informed, anywhere near as informed as you are, edge AI is significantly different to large language AI.
SPEAKER_01:Well, it's an interesting point. I think the development of edge has has been really largely around data processing, but actually, there are now new technologies and new capabilities within the edge processing world. Um, largely we're talking about neural processing units that can actually inference the data that you're capturing locally. Um, there are now some interesting developments in something called transformer capabilities within those, where you can start to see the use of things like small or medium-sized language models at the edge as well. So something we can touch on later, I imagine. But there's a huge amount of fast-moving evolution of the technology, which is driving even more and more interesting use cases for people at the edge.
SPEAKER_02:Yeah. And I think that the the what I would say, Marcus, is that instead of thinking about is it large language models or is it AGI, it's AGI or AI done in the cloud or in large data centers. That's that's the thing that differentiates it. Whereas the large language model is just a version of AI. You could have machine learning or deep learning equally as part of it. So it's more about where are you doing the data analysis? And the point about AGI is it's at the point at which you've measured it. So you get data security, you don't need to pump it up to the cloud and wait for someone else to do the analysis and then give you results back.
SPEAKER_00:Yeah, fascinating. Thanks for that. Um, Peter, we'll stay with you for this. A lot of companies clearly are very excited about AI, but they're equally as perhaps mystified about you know where to start. When you first meet a client, how do you help them figure out whether AI is really the right solution for the problem they're trying to solve? How do you how do you work? That's one of the advantages of working with forget two technologies, isn't it? You're not pushing a specific product.
SPEAKER_02:So give us a that's right. Yeah. It's almost a weekly occurrence now. Um people calling us up and saying, we think we need to use AI, largely because they've read about it somewhere, um, but they don't know how to employ it. So 42T, can you help? And of course we're delighted to. And we're a very pragmatic organization. We're with this is we're not doing this because this is a hype. This is real, it's happening, and it is going to change the world, quite frankly. But we very much, because we're a pragmatic real-world organization, we're very much problem statement led. What problem are you trying to solve? Is usually the first question we ask our clients. Because um it may well be that when you dig into the situation of that specific client, you know, uh are they having um product quality issues, are they having uptime issues on their machinery? Are you know what is it that they're having challenges with? And you dig into it, it could be that AI actually is the right solution. But if you go and analyze the situation and you measure things properly, and then uh you you kind of uh improve their processes um in a way that would be reasonable and sensible. It may be that you don't need AI to do what it is that they want to do. Um, and we're not in the business of selling someone a cheap watch just because they asked for a cheap watch. Um, we're in the business of making their business better. And if and and so uh we don't want to be pushing AI on people just because they've asked for it. It might be that they're absolutely right and they're well well informed and and they know exactly what they're doing and they do need AI, in which case we'd be delighted to help. So the foot the first answer I uh first question, sorry, I would ask people when they ask us about AI is what problem are you trying to solve and be led from that? Um and and ultimately whatever solution they propose has got to be commercially sensible. It's got to provide a return. Um that's so that those are kind of our our guiding principles for all of this.
SPEAKER_00:Yeah, yeah. It's like you say, interrogating the data and really understanding what the key problem is, deploying AI if necessary, if it's right, but otherwise solving the problem is the priority, which makes sense.
SPEAKER_02:Well, that's right. I mean, we've all lived through various hype cycles, right? You know, in in the 1990s, it was all um nanotechnology. Every everyone had to have nano on it. It was like in the 1980s, everyone had everyone had to have turbo written on it, like your sunglasses or something like that. And there's nothing to do with uh the air intake on your engine. And you know, it was nanotechnology in the 1990s, and the it was internet of things for a very long time. Everything was called internet of things, and even though it has it absolutely has a real definition and a real use, there's plenty of people that were using it for for reasons that were nothing to do with um uh the internet or even things, um AI is the same, but AI is AI for me is an industrial revolution, it it is gonna have a massive difference. But we want our clients to understand what they're using it for and what returns they can get when they implement it. And if you do your data collection properly, if you do your sensing properly, it may well be that conventional data analytics will give you the answer quicker, cheaper, easier than then you need. Let's find out.
SPEAKER_00:Yeah, yeah. So it's a good process to go through, and it's it it's good that you're not pushing the hype, you're pushing the solution, which is uh which is fantastic. Um, Jamie, you mentioned in your intro um about multimodal sensing and machine vision both evolving quite rapidly. Could you share perhaps any examples of how combining different data streams at the edge is transforming process or unlocking value?
SPEAKER_01:Absolutely. And it goes to Peter's point just there, really, about the quality of data that you have when you're trying to think about how you can use that data to improve or enhance the process. And certainly, then if you want to use AI or machine learning tools to do that, um, I think the advancement really is the fact that you can now do a lot of that processing processing at the edge across multiple devices. And historically, you would have a central sort of uh data collection point, and then that data might have to be sent up into the cloud to be processed, you know, with large swathes of data that has to be tagged and modeled and understood. Obviously, doing it at the edge means that whether you're using conventional sensing, whether you're using um simple sensors or complex sensors, it doesn't really matter. It's about being able to combine those different data sources, process them to get the viable, valuable insights that that that particular sensor or data point is giving you, and then processing that data collaboratively to give you more than what you would have otherwise. So multimodal is not just uh different types of sensing, it's also different types of processing. So again, Pink to said, we're a you know problem statement-led business. If we're working with a company that's got lots of different challenges across the manufacturing process, lots of different sensors and data, they could be vision systems, they could be time series data, it could be audio data, it could be all types of, as you say, multimodal sensing, having the capability to process that locally and then be able to translate that into something meaningful. Let's say it's about vibration coupled with uh temperature spikes on a particular piece of equipment, combining those two bits of data and being able to translate those into something that you can act on, you know, and not always about automation. Automation is a big driver for manufacturing industry, but as Peter says, it's got to be commercially viable. Sometimes you need to start by understanding what are our teams that are already, you know, providing valuable insight to the machines and processes they're operating. What extra data or support can Edge AI in this case give them? And can we use the data across multiple locations to do that? And I think that is quite transformative because you are being able to do that in a scalable way, in a way that doesn't require as much necessary investment as some of these massively large-scale automation-based processes. You can do it locally with local hardware that's being able to be spoken for your particular application and do it in a way that can be rolled out to multiple locations. So, yeah, it's a it's a massive subject, Marcus, but hopefully that sort of touches on it slightly. But it's it's one of these areas where actually we think one of the big gaps between, let's say, some of the experimental AI and certainly some of the work that's being done in a lab base, translating that into a real-world application means can we demonstrate how this can be scaled across multiple locations, across multiple plants, uh, and use that data to give viable, valuable insight that's commercially uh viable for the people that want to deploy it. So, yeah, really exciting area, but there's a pragmatism, as Peter touched on earlier, that is important about these things. Um so yeah, it I think one of the other factors is we're talking about a lot of historical sites as well. So being able to retrofit some of this technology into existing plants is also a really big benefit. So, yeah, lots of exciting areas from both those uh examples, but something that we're yeah, really excited to be working on.
SPEAKER_00:Brilliant. And you've you've described this before as 42 technology being a kind of an expert bridge between technology and and and your customers, which I think's hugely advantageous for clearly for your customers and helpful and so on. But also in in order to do that, you've built some and you have some great partnerships, haven't you? Um, I'll let you mention them in more deep detail. These these sound incredibly powerful. What do these collaborations allow you to do that would be really difficult or impossible for a customer to do it alone?
SPEAKER_01:That's a great question, Marcus. Thank you. I mean, I'll I'll start with that, but Peter's been a big part of this as well in terms of building these partnerships. I think the first thing to say is, as you just said, 42T has an amazing capability uh and some amazing people in-house that are able to work across all the sort of RD disciplines and some of the work that we do in in manufacturing innovation. But recognizing that AI requires uh a number of different uh skill sets, capabilities, technologies in it, we've started to build quite a powerful partnership across the supply chain. Um, so we've been working for about a year now with a company called Synaptics who develop the edge AI silicon that drives a lot of these solutions. Um, they, alongside a number of their ecosystem partners, have uh what we call kind of foundational model partners, people that are developing the models that can be supporting some of these applications. But also then working with uh independent software vendors who are developing the platforms under which you can then deploy these solutions and start training models and deploying those into sensors, into products, into production processes. We partnered with a company called GRIN, who have been working in Nedge AI for probably as long as anybody in terms of developing the sort of system or module devices that these things can then be deployed on. Uh, but we're also working in some really exciting um new methods with a company called Inatira, a recently announced partnership, who are doing something really exciting in something called spiking neural networks. Um, this is the concept of chips that act in the same way that a human brain might respond to events or occurrences, which gives exciting opportunities around ultra-low power, uh always on sort of sensing, but only really alerting to something when there's something to react to. Um, and then one of our most exciting partnerships at that foundational model level is with Fujitsu Research Labs. And perhaps, Peter, you want to talk a bit about this relationship, but this is really looking at some cutting-edge model uh technology at a research level that gives us the opportunity to work in a um iterative way and developing some of these models to fit the specific requirements of our clients. Um, Peter, I don't know if you want to touch on that.
SPEAKER_02:Yeah, no, um the point is that it takes a village um to to create one of these AI systems. The chip is just a component. Uh you don't go out and buy an arm chip and say, I've got a mobile phone. You know, you don't you don't uh you also need the sensors, but you can't just go out and buy a camera and say, Oh, that's it, I've got an AI system. You can you need the chipset, you need the sensors, you need the algorithm, which is where it likes the Fujitsu come in. But someone has still got to bring all this together into a system, um, an integrated system, that does what the end user wants to do. And that's what 42T is. We're we're the wrapper and the integrator and the system developer that brings all these components together to deliver what the client needs. So it's a one-stop shop. Uh now clients, of course, big clients might want to do all that in-house, but most people can't. I mean, to start to get to where we are now is such a steep learning curve. Why would you want to do that? And we build these partnerships, they're non-exclusive, so 42T retains its independence. We would still only recommend the best route as we see it for our clients. But if you need something that requires synaptic chipsets, which are the best in the business at what they do, then that's great. Or if you want something incredibly low power, we would recommend Inatera. Again, the best in the business in that area. Um, if you want to a brand new algorithm fit just for your purpose, maybe we can pick up the phone to Fujitsu and say, hey guys, we've got an interesting project for you. Do you want to develop this AI algorithm to go in the system that we're developing to deliver the value that our client needs? We'll do that. And that's that's the real power of all these partnerships. We're ready, we're ready to go, we're ready to deliver for the client.
SPEAKER_00:Yeah, it sounds like you have a complete toolkit there with different access to the top-level products and expertise and that and that you deploy the right tool for the right job. Simple as that.
SPEAKER_02:Bingo, bingo, and that's how we work everywhere. I mean, we've been doing this for decades, literally. Um, AI is no different. Best tool for the job with the best suppliers, but you come here because it's easier for you, Mr. Clyde or Mrs. Clyde, that you don't need to start from scratch. We'll we'll help you.
SPEAKER_01:One of the really important things I think that's driving a lot of the uh acceleration of some of these solutions is the drive towards more open development environments and more open access to these types of models that allow companies like 42T to do what we do best. Um, and I think that's increasingly going to be a pressure on the wider ecosystem for people developing the chipsets and and and developing on the on the AI tools, is that the expectation is this becomes more accessible to people to build their own bespoke solutions around, which is where we come in. But those partnerships, I think, as you say, are tools in the tool bag, but also we're also learning and developing on the latest sort of technology developments with those partners as well. So it's both an immediate access to the best and brightest, but also continuing our role as innovators and understanding new technology and deploying it um for our clients in the real world. So yeah, a really good mix, I think.
SPEAKER_02:Yeah.
SPEAKER_00:Sounds fantastic. And um, you mentioned the word accelerating there, you know. I I remember growing up and there's this thing called Moore's Law, right? And um incredibly exciting because it I think it was doubling every seven years or something, or I uh you probably know better than me.
SPEAKER_02:Doubling every 18 months, yeah.
SPEAKER_00:Oh yeah, okay, I've got that wrong. Um, so AI has kind of broken the rules there, hasn't it? And and uh you you accelerating, it's difficult for you know the regular business person or even somebody who's technically a suit to keep up with that. Um it's almost too fast to measure. How do you help clients stay ahead then? I I I guess in a way you've answered that really, but without that feeling of uh being overwhelmed or making a mistake, that whatever they build might be actually outdated in a few months.
SPEAKER_02:Well, I'd I I really wouldn't worry about it being outdated in a few months. Uh that that's that's not something I see likely. I mean, these guys, our partners are innovating all the time, but that's why that's our job to stay on top of the rate of change. And that and that that's something that that we focus on quite hard. I I mean the funny thing is, Marcus, is that you're saying, oh, uh clients are might be anxious or overwhelmed about the rate of change, and therefore it might be better to wait to do something. Um I would really flip that on the head. Um the world is changing rapidly and uh a lot of uh future print um members and a lot of people listening to this podcast will be in a highly competitive market. Do you think for one second the Chinese are sitting around doing nothing with this? I mean, just for one second, do we think that that country of a billion highly innovative people are doing nothing? Of course they're not. And we're already competing hard against them in many of our industries. Um the fact is that the real risk here is doing nothing. Now, a lot of industries that printing is in, quite frankly, the margins are small. Um if you work go into a print workshop or somebody's running some of these printers, or even the people who are manufacturing the components, the printheads or the inks or whatever, on the factory floor, manufacturing uptime is everything. Yeah, costs are minimized. Um you couldn't possibly think of stopping the line to run an experiment to maybe make things a bit better because the guy who's running the line is under extreme pressure to keep things going. And because margins are tight, businesses might think, oh, I don't have enough money to really invest in uh you know something that as highfalutin as AI that may not deliver the uh the returns I want. I'd flip that on a head on its head and say, you can't afford not to. I'm afraid, because we are gonna get out competed in the West by the Chinese, the Indians, the people in the Far East who are doing this, they're doing it right now. And if we don't start doing it in our factories, we're done. You might as well turn off the lights and go home. So I say to I I would say to people, it's not a question of can you find the investment to do this? You must.
unknown:Yeah.
SPEAKER_02:Simple as that. And and I think you should be overwhelmed by the wave of competition that's gonna come and not be overwhelmed about the speed of change of AI and wondering whether you should employ it or not.
SPEAKER_00:Yeah, too much time prevaricating was gonna have a place you at a disadvantage. And I know that I had a a recent podcast with Amir as the LHP, he said, and I'm hearing this, that OEMs in the printing industry are coming under increasing pressure, competition from China, partly due to the impact of the US tariffs. But it's a real thing, and it's it's it's not like going to happen, it is happening. And as you said, the innovative culture is quite amazing. So Europe really has to kind of up its game there, really. Um, Peter, drawing on your experience and expertise in printing, how do you see Edge AI really supporting these next generation printers? Whether it's, I don't know, you mentioned predictive maintenance, process control, uptime, stability, you hinted at it already.
SPEAKER_02:Yeah, I I think it's really interesting uh that the printing industry, because I think there's lots of different ways in which AI could help. And we're not really seeing a lot of it yet. But if if you think about it, printing is difficult, but it still is. Yeah, you you you know, there's a there's a particularly the biggest printers that you can buy. Um end users have to employ small armies of RD experts to keep these things up. Um and you know, there's plenty of printers out there where uptime of 70% is cause for massive celebration. Uh because they're hard to run. They're they're fiddly, they're finicky. Anything that can go wrong usually does go wrong. Um and that causes problems. Um, which means you've got a very you've got a noisy environment, you've got a highly variable environment. You can have the same settings one day after the next, and the printer will behave sufficiently differently that the output is goes from acceptable to unacceptable and back without you actually doing anything. Um that for me is a perfect place for AI, particularly perhaps a machine learning aspect, to be implemented to start to control. I mean, we're we're talking about quite a uh you know an extreme end of it, but can you get sufficient data, and it almost certainly means you're not measuring enough stuff in the first place, sufficient data into an AI core to do some real-time um settings changes to keep the printer working um perfectly. Now, um one area which I really think that this could happen is in vision systems. Um, anyone who knows anything about vision systems for for digital printing or even just like conventional printing, um, it measures vast amounts of data. And then in order to get real-time information back from conventional algorithms, most of that data is thrown away. It's averaged over large areas or it are, or it's just or it's just simply discarded in order to um get the amount of data down to manageable size that you can get a conventional algorithm running on it. Um uh I strongly believe that AI could be really used there. I mean, we're we're using vision systems all the time right now to compare um man uh products coming off a manufacturing line with the ideal. Um, well, that that's exactly what a vision system is supposed to be doing for printing. So there's running the printer, there's um the vision systems, um, there's uh I would say that there's an element of uh uh changing the interface um to the printer to using uh natural language. So that that's perhaps using an LLM model to in order to get the printer to do what the operator wants it to do, but in a more human way. Um so there's there's a few aspects right there. And and that's before you drill down into your condition monitoring and so on, which might be a lot more low-level and still extremely valuable. Some people might argue that conventional algorithms will be good enough there, but but maybe not. Well, that's just running the printer. The manufacturing processes of the components themselves are quite complicated. I mean, we all know print heads are fairly complex beasts. Um, they might be quite uh conceptually simple, but they're hard to make. The tolerances are tiny, and and even then, the the variableness of the input materials is can be quite high. You're you know, pHelectrics are not no batch of pHelectric is the same uh over the other. So the manufacturing processes of of printheads need to be controlled quite highly, and and manufacturers work extremely hard and worked extremely hard in order to get those tolerances down, but there's almost certainly scope for AI to be used there. So you go back and back and back. What about color matching of your inks? The manufacturing processes for inks, say you've got highly variable inputs there from pigments, which are natural materials sometimes. How do you you know put ink manufacturers or coatings manufacturers or paint manufacturers have to work extremely hard to make sure that every ink is identical? The list is endless.
SPEAKER_00:Yeah, so the opportunity is as well. I guess that's the key, right? That that that uh the options and opportunities, both on a strategic level, if you're in a production operation and on an innovation level too, but equally whether you're creating print technology or not, really. So it's it's kind of touches everything or can touch anything. Obviously, you mentioned Europe, culture is always an issue, isn't it? Culture, receive wisdom, institutional thinking. Um many production environments rely heavily on that, don't they? That kind of tribal knowledge, if you will. How can AI help capture that expertise without replacing the human element? I don't know whether this was for you, Jamie.
SPEAKER_01:Yeah, it's a really, really important point because I think it's absolutely true in the printing industry, as as you and Peter have mentioned. I think it's true across a lot of industrial processes because we know that these processes are hard, they're variable, there's a lot of experience in the industry, but that's also a threat. Because I think the threat is obviously that companies rely on expertise of individuals. And as we know, you know, labor markets are challenging in different parts of the world and always challenging when it comes to things like operational cost, which is a driver towards automation. But I don't think it's a question of removing people from the process. I think it is about how can AI assist those in the process. And I think there's a generational point there as well, because as more people come into the industry, they're used to using devices in their day-to-day life that is driven by things like AI assistance. Increasingly, we're all surrounded by in our day-to-day use of technology. So you've got, I think, two challenges there. One is people with that institutional knowledge that that know the very complex settings of a particular print machine in that example, um, and those also then coming into the industry that expect to have the kind of tools and uh AI assistance that they see in their day-to-day lives. From a technology perspective, I think there is an absolute drive with what Peter touched on earlier, is this concept of natural language, using the capability of AI um data capture to be able to send huge amounts of information about a complex thing into a model that's capable of um computationally understanding that and then translating that back into language that can be you know discussed, interrogated by even non-experts to be able to give information that comes back to them. That's that concept of natural language. So I think that evolution is really important. I think a lot of that exists today, but it's it's largely being cloud-driven by large language models that require data center processing using speech-to-text or text-to-speech or text-to-text outputs. The edge processing that we touched on earlier is where I think this for machine builders and people that want to implement this into their production line becomes much more viable because once you can start translating that capability into a local system, a local process, a local product, suddenly you're then able to start putting the boundary around it. So you're feeding information about a particular device based on the manuals, based on uh perceived wisdom, knowledge that you can feed into that model, and then translate that into something that you can then interact with. And Peter touched on this earlier in terms of the user interface. These are complex machines with lots of different potential risks and points of failure that individuals in a in a business or a process understand. If you can capture that information in a way that can then be translated back to other users, you're expanding the knowledge and capability in-house and addressing that concern around institutional knowledge. But you're also potentially getting greater insight from the analysis that those tools are able to perform that you may not have had from conventional inspection or conventional uh understanding of the process. So, again, it's a massive area uh of exploration for companies and an area that we're seeing the technology is now evolving to the point where you can get these kind of models down to a small enough size to run at the edge and run locally. Doing that then for really specific challenges, such as the print industry, I think is opening up some really exciting opportunities for people to not just address the threats that institutional knowledge has, but benefit from it, find ways to capture that data and turn it into something that can be a benefit to you through largely what we call AI assistance rather than automation.
SPEAKER_00:Yeah, and it's so much about having a mindset, and that's that's the comforting thing. If if you've got that curiosity, that ability to want to learn and and to employ the right people to focus on this, you know, the potential is huge and the learning is massive, and um it could take your business on an entirely new level, can it? Um just thinking about companies that perhaps are at the beginning of their AI journey, what advice might you give them in terms of your first step? Um, particularly if they're unsure about the maybe the quality of their data, I believe. Frankly speaking, AI only really delivers if the input is great at the beginning, or the scalability even of their systems. Yeah. Um Peter?
SPEAKER_02:Yeah, yeah, uh well, you're absolutely right. I mean, that the first advice is what problem are you trying to solve? Yeah, you've got to be very clear about what problem is you're trying to solve, and what's the what what's the return you'd expect from solving that problem? Um, like, like, and and that is how you do any form of innovation, in my opinion. Yeah. But what problem are you trying to solve and what returns can you expect from if you solve it? Um, and then you're absolutely right. We would then say, right, show us your data. And uh you'd be amazed, uh, Marcus, that the range of answers we get to that, to that, to that question. The number of times I walk into a factory and people say, our data is written down on bits of paper and stored in a warehouse, and we don't even keep all of it. Or the next one would be, oh yeah, we save all our data, um, but when our server gets full, we overwrite it. Um, and then you start saying, Oh, really? So you record which machine you've done this particular process on. You've got five machines, which machine did you do it on? When's the last time it was serviced, or et cetera, et cetera? And then then you find out that actually they're not recording all the data at all. And all these things are important. Um, and then you've got to go and ask yourself as well, how are you measuring this data? Um, uh, are you using the right type of sensor or instrument? Um, what's this resolution? What's its dynamic range? Um, can it cope with anything that gets thrown at it? How much human input is there into the measurement? Um, and so on and so on and so on. So that, and then before you know it, you've built up a really good picture of the workflow, the process flow, what you're measuring, how you're measuring it, where are you saving it? Um, ultimately, there becomes an IT infrastructure question. Um and and there is a real question to be asked is should you be doing uh your analytics at the edge, which could be AI, or should you be doing it in the cloud, where you get better horsepower, um, but there's data security and there's latency, as Jimmy was describing earlier. So that's how we get started. Problem statement, the expected return from solving the problem, looking at the data, thinking about how you might improve the data, and then figuring out, well, what algorithms can I use? And if it's AI, then perhaps we could start doing experiments on small amounts of data, and then building up as the quality of the data gets better and better and better. So you can bootstrap your way up. But those are the kind of the basic logical steps that we would go through. And in an ideal world, you do an awful lot of that analysis in one phase, one step, and at the end of it all, the output of that work is what's your plan? And that's really valuable to anyone running a business. What's the plan? Um, you've got a better idea of what it's going to cost, what it's what you you know, how long how long it's gonna take and what it's gonna deliver.
SPEAKER_00:Makes sense. Thank you for that. Any any any does make sense? But it's about the the the human thought at the beginning, isn't it? It's about thinking first, focusing on the problem, understanding what might help, and then deploying AI, I suppose, or the right AI. Any uh any thoughts on that, Jim?
SPEAKER_01:Yeah, I think it's a really interesting point in the terms of what's the problem statement, because I think I think I've seen some stats. I don't know if it was possibly from MIT recently that surveyed a number of companies deploying AI, and they were saying something like 80% of AI initiatives globally are failing. Um and obviously with the vastness of solutions coming out, they're largely you know driven by the sort of large language model, Chat GPT type uh approach. So that'll be a big part of it. But I think a part of that is also is people are assuming that AI is something that they can get started with really fast, move through really quickly, and it's going to help accelerate development. Now, there are undoubtedly tools to help do that, and we're working with those where they're appropriate for the business, appropriate for our clients. But I think also there is a rigor and a human nature, as you say, Marcus, to going through the right process at the earliest stages, as Peter's just described, to give that kind of best opportunity for these things to address the problem statement that it's been designed to address and to show the returns and the scalability that Peter mentioned earlier. So, from our perspective, I think there is a rigor around applying AI. The first question is you say, what is the problem it's trying to solve? And is therefore AI the right answer for it? But there's also then this absolute need to be able to go through the right steps in the right way to be able to ensure success. I think that stat about 80% failure is quite striking when you think about the potential. And we talked about the hype curve earlier. You know, those things are natural, I guess, for something like this that's so transformational. But we're trying to help bridge the gap by doing it in a risk-managed way that addresses the main purpose of developing something that is done in a way that actually returns something of value at the end. And uh yeah, we think that's the right approach. But um, equally, we're we're working with really innovative clients, really innovative companies that want to push on the on the on the boundaries of what's possible. So, of course, there's always that chance that things uh don't work out in the way that you envisage, or that you learn something new on the way. But it's um yeah, incredibly exciting. But equally, I think there's a there's a good opportunity to be able to give yourself the best chance of success by following the approach that we're advocating.
SPEAKER_00:Yeah, absolutely. It mitigates risk to a large extent, which makes a lot of sense and saves time. Um, so thanks both for today, just leaving people with uh looking ahead as we end the year or come close to the end of this year and look into early next year. Peter will be speaking at the AI for industrial print conference in Munich on the 22nd of January. So if anybody's listening and would like to follow this conversation up in person, do come along to that event. So, looking ahead, what it what excites each of you, perhaps, about where AI is heading, particularly in relation to manufacturing and printing technology. What should our listeners be watching closely in 2026 and beyond, maybe?
SPEAKER_02:Well, it's a very interesting question because it it this is one of those industrial revolutions where if you look at it from day to day, nothing much will change, and then you look back in three years and think, wow, it's kind of it's suddenly almost everywhere. Um I'm expecting a lot of what I discussed earlier, actually, it is what people are going to start doing. I can't obviously can't reveal the conversations and the uh that we're having in the work that we're doing. That that's highly confidential. Um, but let's just say it's not a million miles away. Um, you mentioned HP earlier, they're they're already quite publicly um do uh uh AI in large aspects of their work. But HP's a um a multi-gazillion dollar gorilla in the marketplace. The interesting thing is what are the the mid-sized companies and uh going to be doing and the smaller companies who don't have the resources that HP do? They are gonna have to start employing it so that they don't get left behind. Um, and it's it's where they employ it first, it's a really interesting thing. Um and I think you'll be surprised at at how esoteric some of the applications of AI are. Uh printing is a very complex process in many ways. We all know there's a thousand one ways in which a printer can stop working perfectly. But if you go back in the supply chain, at some point in time you reach a point where natural minerals or whatever are being dug out of the ground and they need to be refined and processed and turned into something interesting, like an ink or a PXOelectric transducer or a silicone chip. Um, and at some point in time there, these highly variable processes need to be turned into something extremely uniform. Um and that's quite difficult. So it'll be applied both at the running a complex machine part of the process, but it'll also be right back at the top end of the supply chain, which people might think are a bit more uh grunt work, but actually it needs to become more and more refined to get better and higher quality materials.
SPEAKER_00:Yeah, so you see a fast evolution of um improved application. And you mentioned user interface before as well. And I I'm hoping some of the things that you're talking about actually allows the human to function easier and not require a 100% in in physics, Peter, to operate things. Um it's quite handy, but a lot a lot a lot of people don't.
SPEAKER_02:Well, yes. Um putting that to one side. No, not everyone needs a PhD in physics. You're absolutely right. I think what's really uh there's one point I really want to make, and that if you're not careful, you can you can find a lot of the discussions around AI with really miserableists. Oh, it's gonna put everyone out of a job, oh that computers are gonna do everything. I mean, we heard this before in the 1970s and the 1980s, right? Oh, computers are gonna take over the world. Well, no, it just changed the jobs that people did, right? I really don't subscribe to this notion of AI. I do not accept that it's gonna put everyone out of job. It will change jobs, but it's not gonna put us out of a job. It's gonna allow human beings to do more. It's gonna enable productivity gains. You know, those of us based in the UK are well known the productivity discussions we have in the UK economy. Frankly, that our productivity has not been going up as it used to historically. Um, lots of reasons for that. Let's not go into that now. But AI is one of those things that's gonna allow us to get the productivity to go up so that people doing jobs can generate more revenue per unit time than they are right now. That's a really positive outlook on AI that I subscribe to.
SPEAKER_00:Yeah, very positive. Jamie, do you concur with that? Are you any on it? Final thoughts looking into next.
SPEAKER_01:Absolutely. No, absolutely. And Peter's absolutely right, and that and that message I think is is clear and it's reflected in the conversations we're having, as Peter said earlier, maybe every week now with with clients across our different range of industries. Um, I guess I think in terms of your question about 2026 and and what to look out for, obviously, I'm looking at that from a technology perspective. There are some really exciting things that I think will enable, and Peter mentioned earlier, maybe those medium-sized companies to start playing at this space. I think some of the big companies, so HP, obviously, from a printing perspective, if I look into a sensing and automation space, you've got a number of the major corporations have been uh deploying these solutions, but there's been a bit of a barrier, I'd suggest, for the sort of smaller to medium-sized companies to really start leveraging AI. I think the evolution of edge technology, as we talked about earlier, but also the the open nature, more increasingly open nature of these models, making things more accessible, making things easier to operate. The concept that you don't need to be a physicist to run printers is applied to you could also run AI models if you don't need a data scientist to do it. And I think a lot of the technology that's coming forward is going to enable companies to like us and our clients and the people we work with to be able to do more without necessarily having um, you know, huge capabilities in the data science world, which quite frankly is hugely in demand across the world. So that evolution, that open development, that ability to be able to deploy things, you know, in we haven't really touched on it today, but that that ability to deploy things in industrial manufacturing settings where it is a lot more challenging than your home or your car or some of the consumer electronics applications where things are starting, uh have already uh established themselves. So I think those are the things that are gonna help drive that transformation in in industry over the next, not necessarily just one year, but you know, five, ten, fifteen years. And um, yeah, of course, there's lots of interesting model developments that are going to support that as well. And we're really looking forward to getting involved in all types of applications and and challenges like that. So yeah, lots to to look forward to, but um also a lot to uh to be optimistic about, I think, as Peter said, in terms of the deployment of AI in the future.
SPEAKER_00:Yeah, and just going back to the Industrial Revolution, the key here really is to not be a Luddite. Right to embrace, move forward, see the positives and the upside, learn, engage. Exactly. Incredibly exciting, and um the productivity gains, the uh you know, virtually endless. So I appreciate you both joining us for a really excellent podcast today. Appreciate that. Looking forward to seeing you, Peter, in in in Munich. Delighted to have met you, Jenny. Congratulations on your work. I'm looking forward to re reconnecting in the future and um catching up with AI again as you continue to push it forward. So thank you very much for joining us.
