Interview transcript

Terry Gerton We’re gonna talk about military supply chain and logistics. A lot of defense leaders are still warning that supply chains are fragile despite some modernization, despite post pandemic recovery. From your perspective, how can AI play a role in building the resilience of those defense supply chains?

Jon Garrity I think AI plays a critical role. And the reason is that supply chains are intrinsically very complex. There’s a lot of information. It’s hard to grapple with all of it. You have end users who have some visibility into what they might have at a Ford supply point. And you’ve got the manufacturers who are receiving signals from higher level commands and from headquarters. And there’s a lot of latency between those different points of signal. And so AI presents an opportunity to effectively coordinate and orchestrate across that supply network, optimize decisions on everything from purchasing to stocking to mobilization, transportation, all different elements of supply chain logistics operations. So it really is a new paradigm the way that supply chain logistics can be orchestrated.

Terry Gerton The Defense Logistics Agency tells us that they’re now rounding dozens of AI models to monitor supplier risk and to forecast disruptions. When you think about the options out there, what do you think the most promising AI-driven approaches are to identifying vulnerabilities early and fixing them before they become problems?

Jon Garrity So a lot of the traditional approaches to using machine learning and supply chain applications come into things like demand forecasting or identifying patterns in the data. And obviously the zeitgeist now is all around large language models and how those can be applied. But the challenge is when you look at the initial attempts, you know, you’re focused on a very small set or subset of the logistics enterprise, which again could be demand forecasting, but that’s just a piece of the puzzle. And on the flip side, you have these very general models like a large language model that can answer very general questions in incredible ways. But they aren’t grounded in the realities of supply chain. They don’t have ways to reason over or optimize over such a large-scale system. So the the new opportunity and one we’re focused at with Tagup is building a world model, it’s called. So basically providing structure such that you can reason effectively over these large scale systems, right? What is the relationship between the receipt of materials and the the use of those materials, of transportation from a manufacturer and expiration of material downrange, right? Hard questions to reason about at scale for humans and even for legacy approaches. But now with these world models, you can effectively reason at that scale.

Terry Gerton We’re talking about supply chain here, but we’re also hearing from the Air Force and the Army who are investing heavily in AI-enabled predictive maintenance to cut down on downtime and improve readiness. How do predictive analytics, sort of like what you were just talking about, your world model and supply chain, how can that apply to traditional maintenance models?

Jon Garrity Yeah, very directly. So predicting a specific component failure on a specific asset is is very difficult, right? Even if you have it very well instrumented, it’s a hard problem. But it’s a useful demand signal. And certainly when you aggregate it over a fleet, you can get very high levels of accuracy and understanding, okay, in aggregate, how many replacement transmissions will we need for this program, right? As an example. And the reason is it’s it’s similar to you know picking stocks. It’s hard to pick the one stock that’s going to get you 100x returns. So you buy a portfolio and you’re going to have one that’s going to give you that high return. Similarly, if you have a fleet of vehicles or or a fleet of aircraft, you don’t know which one is going to have that catastrophic engine failure, but you know across the fleet that prediction that of that event will occur with some level of certainty, and that can inform upstream in the world model what decisions are being made. So, in short, it’s a very useful input to the overall optimization, but it’s just a piece of the puzzle.

Terry Gerton When you think about the array of military equipment, I mean everything from water filtration units to F-35s and aircraft carriers, how complicated is it to scale an AI model to cover that scope of inventory?

Jon Garrity Yeah, great question. The answer is that there are, regardless of what asset or platform you’re looking at, there’s a lot of commonality, right? They all have parts that are installed on them, the replacement parts, they have distribution around the world, they have certain patterns of utilization, maintenance requirements. So the world model can encode these basic rules that relate these elements and then can take the data that exists today. And I think that’s a big, big opportunity, right? When you look even at predictive maintenance, there’s so much data that exists in existing IT systems related to service requests, parts required, requisition, all kinds of things. And there’s a lot of signal in it. But to employ that signal effectively, you need to be able to provide that structure. And so in short, in spite of the scale, it actually is the strength for these AI models, because if you can apply that structure, the models can get better much faster because of the scale of the enterprise. So that’s that’s the opportunity with that scale.

Terry Gerton I’m speaking with John Garrity. He’s the founder of Tagup Inc. One more maintenance question just because maintenance is near and dear to my heart. Congress and DOD are pushing right to repair provisions to let service members fix their own equipment using digital manuals and 3-D printing. How do you see AI intersecting with that approach?

Jon Garrity In a couple of important ways, right? One we’re we’re looking at now is changing the way that you — so take 3-D printers as an example. Where do you put 3-D printers? Right. There are now platforms where they have 3-D printers and containers, that can be moved around the world. But where should you put those sources of supply, right? And so if you have data from maintainers as to what parts are required, you can ensure that, via an optimization, you locate the sources of supply, the advanced manufacturing, other capabilities close to the point of use. And so I think that’s one way that AI will improve the maintainer experience is making sure they have the tools and the parts that are necessary closer to the point of use and and basically ultimately reduce non-mission capable rates due to supply and and and maintenance.

Terry Gerton Okay, let’s take one step back from the military units themselves and talk about manufacturing capacity. Companies are using AI and advanced manufacturing to compress production timelines and reduce parts counts. How do you see these technologies then changing surge capacity, industrial base opportunities? Sort of, what’s next on the front there?

Jon Garrity And actually I think it is intimately tied in with military as the ultimate customer, right? What’s what’s critical is avoiding whiplash effect and having visibility upstream so that manufacturers have access to what demands are they and how can they be satisfied. And I think the opportunity is and it’s coming to play now where if you can track use of equipment, of parts at the tip of the spear, you know, downrange and provide visibility back to the manufacturers across the enterprise. Now the OEMs, the manufacturers, can understand how their equipment that they manufacture is being used, where it’s being used, where there are issues, and that can inform their investments. if you could — right now the purchasing for these programs is done in the current fiscal year, right? There’s not visibility over many years into the future. That’s a huge lost opportunity, both for negotiating better prices with the manufacturing base, but also ensuring that the capacity exists into the future. So I think that’s one of the most exciting opportunities for these sorts of systems at scale is giving visibility to the manufacturers as to what are the demands in out years.

Terry Gerton And that seems to take us right back to where we started, which was with supply chains and making sure that not only the military supply chain, but the commercial supply chain stays safe, secure, predictable.

Jon Garrity [The] most interesting applications in my view of these new AI systems is the ability to identify systematically what links in the supply chain provide the most risk against readiness or other operational requirements. And so you can be very systematic in identifying, okay, how can we store up capacity or ensure that we have redundancy and supply to avoid getting just hammered on our readiness rates. And I think that’s one of the, again, the ultimate aim, in my view, from these sorts of systems is to be able to tie tactical use of the end users, aggregated at its scale, and provide that visibility and signal back to the industrial base all the way back to the supply chain. And by doing that, we can strip out a lot of the inefficiency in the system and drive higher quality service to the end customer.

Terry Gerton Real time decision making inside the system that you just described has always been sort of the holy grail of military logistics. As you think about the future of integrating AI into each of those pieces and parts, are there any concerns that you have about the cultural obstacles or individual training obstacles that might make it difficult to optimize the value?

Jon Garrity I think there’s been — it’s certainly a paradigm shift in sort of the user experience, right? So yeah, I’m sure you’ve long, history and experience, Terry, in military logistics and and in the private sector too. You know, ERP systems have been with us for many years, right? Enterprise resource planning systems, they track what we have and where we have them. And you know, you have to know a lot of different codes and keyboard shortcuts to be able to efficiently navigate these systems and then ultimately, you know, you put those into reports that you can summarize up to your commanding officers. And what’s changing now is that, as you’ve maybe seen with ChatGPT related technologies, now you can you can just ask questions directly and get answers. The challenge of course, operationalizing those for defense applications and supply chain is grounding it in the reality of the situation, making sure you you’re not hallucinating, right? You can’t just ask chat GPT for docking policies. So there there is risk that needs to be mitigated to make sure, and that’s something we provide explicitly by providing that structure in the world model that you can make sure that the answers you’re getting, the COAs that are recommended are grounded in reality. But I do think there’s going to be a bit of a learning curve that I think will be tempered by the fact that a lot of people are using large language models now. There’s a familiarity in private use with these sorts of technologies. I think in many ways it will be a more intuitive user experience, but it will be a transition from the historical way of interacting with IT systems and log IT systems specifically.

The post AI is stepping into the fight for supply chain resilience and battlefield readiness first appeared on Federal News Network.