The Future of AI in Maritime Warfare with Zac Staples of Fathom5

Speaker A: The most pressing ethical question unequivocally is will artificial intelligence make kill, no kill decisions? Friendly fire is a reality of worth it. An AI autonomous systems ethicist that I worked with at the Navas graduate school for a while told this story, and I thought it was probably the best kind of summation is that friendly fire has been an ever present concern in warfare since the dawn of time. And that's not going anywhere.

Speaker B: Welcome to AI Government and, uh, the Future, a podcast by Corner Alliance. We explore the intersection of artificial intelligence, government, and the future. We work with government to create results. We ignite your agency's mission by helping you to design and implement high impact and innovative federal programs in AI, broadband, cybersecurity, public safety, and, um, more. Being a government ally is at the core of all we do.

Speaker C: Welcome back to AI Government in the Future. I'm, um, your host, Max Romanick, and today I'm thrilled to welcome Zach Staples to the show. Zach is the founder and CEO of Fathom5, a company at the forefront of deploying AI driven solutions for the defense sector, particularly within the maritime industry. With over two decades of experience in the US Navy, Zach has pioneered the integration of AI on Navy warships, making significant strides in enhancing security efficiency and operational resilience. His work at Fathom 5 has helped bridge the gap between cutting edge technology and mission critical defense systems. Today, we'll explore the critical role AI is playing in reshaping cybersecurity for defense infrastructure, the challenges of integrating next generation AI with legacy systems, and the ethical considerations of using AI in high stakes environments like military operations. We'll also dive into Zack's insights on how AI can be used responsibly to ensure security while pushing the boundaries of defense innovation. Welcome to the show, Zach.

Speaker A: Max, thanks for having me, man. It's great to be here.

Speaker C: Uh, excellent, excellent. Well, as a means of getting us started, I'd love to dig into a bit of your fascinating journey. You know, you've been naval commander, founder of Fathom five. I'd love to hear about what sparked your interest in merging AI, uh, and defense technologies, so.

Speaker A: Absolutely. So, you know, I had the distinct honor of serving our country in the Navy for, uh, almost 23 years.

Speaker C: Thank you for your service.

Speaker A: Oh, yeah, absolutely. And I had a, you know, loved most minutes of it, but I've always been a technologist at heart. I think, you know, one of the moments in my life where I realized that I might be on a good path is I was at the Naval Academy and they had Dennis Connor Dennis Connor, who, remember back in the 80s, won the America's Cup a couple times, and they invited him to the Naval Academy to give one of these things they call the Forestall lecture series. Bring somebody famous in and asked him to give a talk. And Dennis Connor showed up, and I'm pretty sure he was drunk, but

Speaker B: he

Speaker A: still gave a great talk. And I don't remember much about the talk except that I. He sounded like he was drunk. A and B, you know, basically he challenged the notion of something that I had been thinking about for, you know, I think I was either a junior or senior at the time. And he challenged the notion that I had kind of in my head, been thinking about for a while. I. And you know, at the Naval Academy, it is a leadership school, you know, centered around a technical education, right, to produce leaders who are morally and intellectually ready to lead sailors and Marines. And because of that, there is a huge push on ethics, responsibility, leadership, and all of the other tenets you would associate with that. But I've been a tech geek my whole life in one way or another, whether it was. Whether I was. I was five years old and then I was asking my grandfather 10,000 questions about his tractor until he's just like, son, just shut up and put the hay bale over there. But I've always had that natural inclination. So, uh, you know, I remember being at the Naval Academy and thinking, you know, we talk a lot about leadership, but the man or the woman sailing the most lethal vessel probably wins, right? And so Dennis Connor was the first person to come to the Naval Academy and say, success for anything that sails is technology first and everything else second. And so I found it so ironic in how many leaders and everything else we had come through the Naval Academy. It was the sailing guy who was the first one to speak to my perspective about what actually matters in creating and sustaining a navy. And so later, in later years, I had this example that you could take John Paul Jones and his crew that were, you know, amazing versus the contemporary technology, uh, of their day. And if you put him up against a Ludang frigate from the People's Republic of China, he loses a hundred out of a hundred times because a sailing vessel with smoothbore cannons does not be a modern frigate, right? And so that's a pretty wide. That's a 200 year technology gap. But, you know, another great example of that is I just signed up to go to the Naval Academy in 1991. I was a senior in high school in January 1991. You know, when in February 1991, uh, I'd already signed the paper, so there was no backing out at that point. Right. And that's when the Gulf War. And if you recall back, if you're around the time or you can reach.

Speaker C: Oh, yeah, older than I look.

Speaker A: Yeah. So that was the most lopsided land conflict in human history. Right. So you had tens of thousands of casualties on the sides of the adversary. And so. And across the whole coalition, about 300, like Alexander the Great, never won at that margin. Caesar never won at that margin.

Speaker B: Right.

Speaker A: There was something fundamentally different about that fight. And what was different was we had prepared what is now. We got kind of thought it was the second offset technologies for 30 years. So in preparation to fight the Soviets and World War iii, that, thank God, never happened, we created some amazing systems. We created digital stars so you could take a kid from Plymouth, England, or Plymouth, Virginia, and, And drop them in the middle of the desert and they didn't get lost. That was like the first time in human history you could ever do that. We had missiles that you didn't have to pick which building had. You could pick which window you wanted them to go through. We had balls in space doing amazing things in support of ground combat operations. And so collectively that yielded the second offset set of technology that resulted in the most lopsided land campaign in the history of human morph that we know of. And so I watched over my career when I graduated in 1995 through, when I retired, and you, uh, know, in 2018, what felt like peer adversaries catching up. You know, there I have, uh, there's. There's a. Just a wonderful defense innovation thinker named Pete Newell, who runs a company called BNNT out in Palo Alto. And Pete has this quote I've heard him say a couple times that I just love. He's like, the United States has the most impressive tech demos of anybody in the world, and the Chinese build more ships.

Speaker B: Right.

Speaker A: And so there's just this stark reality that over my career in the Navy, everybody else caught up. And so now I wonder if we don't figure out how to integrate the third offset technologies the way two generations prior to me figured out how to integrate and synchronize the second offset set of offset technologies, we might not just have a, uh, close one. We could very well lose a war that has more political and existential meaning than any of the, not to diminish it, that I thought in them. So I think I can critique them. The skirmishes that we've had in Afghanistan and Iraq. And so I, I did that long time. I did several things, you know, innovative things that I really enjoyed. And you know, while I was on active duty and, and then I'm like, I'm building the company to get after this, how to actually build and deploy this set of third offset tech and make it work with the existing systems. And so hence fathom thought.

Speaker C: Yeah, that's perfect segue. You know, AI and combat defense systems is really becoming that sort of transformational tool. But I would love to just sort of dig in from your perspective. You know, with those big transformational tools typically comes risks. What do you see as the biggest risks that we are facing in trying to deploy AI in defense applications?

Speaker A: Yeah, that's a fantastic question. Let's divide this up into two risks and try and tackle those kind of one at a time. Risk one is we don't get it done and China does. Then risk is we lose a really big war. So risk one is speed matters. And the other risk on um, that kind of, in that block of we could lose a war that, that is existential, that we very much care about is we aren't going to out produce out industrial, you know, industrially produced warships faster than China. That literally did, you know, to use the pun, that ship is sail. The trillion dollar manufacturing deficit between what China is doing and everybody else combined right now in the world means that it would be a fool's era to think that we can make more warships than the Chinese. So then we got to ask ourselves, well, what's left for us if we're not going to get more ships than them? Well then we better figure out how to take the ships that we have. Ships, submarines, airplanes, war fighting systems, swarms. Like, let's put all that in one big group and say, let's take our, our maritime war fighting system of systems and ask how can we make that system of systems exponentially more capable through digital modernization? That's a fight we can actually win. Right? So I think part of what we need to think about in small groups like this and then in large groups, kind of in a geostrategic context is the most important thing we do in preparation for a war, or more importantly in deterring a war, is set the terms of a future engagement where we have or could generate advantage. And so a place where the United States still, you know, could ostensibly win is building the pathway to digitally modernize our maritime war fighting systems faster and more effectively than China, who ends up. So in the best case scenario, they end up with Steel ships and we end up with a digital war fighting ecosystem and we get the lopsided outcome that quickly resolves whatever political situation creates the conflict. Right.

Speaker C: Trying to turn those steel ships into smoothbore cannons as quickly as possible.

Speaker A: There you go. There's the analogy. Right? You read my mind. We can turn the. Yeah, yeah, you built more steel ships with electronic fire control than us. Doesn't uh, matter because what we've built isn't really just a shipping. What we've built is a manned node inside a learning ecosystem of maritime power that surrounds, subsumes and overwhelms your ships with electronic fire control.

Speaker C: So compared to things like uh, nuclear proliferation or cybersecurity, sort of more modern military advancements, where does this set of risks fall on that spectrum?

Speaker A: Yeah, that is a fantastic question. I also do want to get back to kind of the AI ethics question.

Speaker C: I got lots of questions for you there.

Speaker A: Yeah, yeah, yeah. So let's keep going on this theme of like, opportunity, risk. Like if we don't capture this opportunity to step hard on the one advantage that's open to us, we really would be at strategic risk for a war that we might lose. So your question was where does like traditional technological advancements like nuclear or cybersecurity fall in relation to this AI? Right, yeah.

Speaker C: Like, are we looking at the introduction of like you say, you know, third wave technology here or is this a better mousetrap that we're trying to figure out how to put into the field?

Speaker A: So let me give you an example for a company that's out of business. So it's a self fulfilling example, right? It sears with a website is not Amazon. Right. So there is a way that you're like, hey, we're going to go get an AI algorithm on a ship and what you end up with is sears with a website. And there is a way that you rethink the entire digital experience of maritime command and control, employment and engagement. And it's Amazon, it actually rethinks the entire retail experience from a digitized platform. So I think one of the other large strategic and unanswered risks, and this one is real, is can the United States do militarily what we've, you know, what we've outpaced the world doing in the E commerce and digital information site. Right. So we built the Internet and then the DoD under optimized for IT and commercial industry built some of the biggest, most successful businesses of, of our generation. If you look at the 10 most, the 10 most wealthy companies in the world right now, now versus 50 years ago. You don't find Microsoft, Apple, Amazon, like all these companies, right? And then if you ask yourself did the US under the US Defense industry, in the US defense larger ecosystem under optimize their invention of the Internet versus commercial industry? I think you'd have to say oh yeah, uh, bad. They badly under optimized for that. And that's crazy because the United States invented the first airplane and we found a way to both build, to both have the country that has Boeing, the largest commercial aircraft manufacturer in the world. And we ended up building the Joint Strike Fighter, which for all its problems in the F22 are the best fighter jets equipment. So somehow over the hundred years between the invention of flight in the early 1900s to 2020, we managed to both successfully commercialize aviation and militarize aviation for the defense of our nation. It has been. The Air Force keeps track of this when they use it in lots of slides. But it has been like decades since anybody dropped a bomb on a U.S. marine or U.S. soldier. Uh, it's been a long time since any. I think the last time anybody actually literally had an airplane above our folks and dropped a bomb on was like in Korea, right? And think about all the places that marines and soldiers have fought and they never had a bomb dropped on them one time. That's because we crushed it when it came to taking an American invention flight and figuring out how you militarize that technology.

Speaker C: What I really like is that you're setting up this sort of like one of my questions for you is like, okay, so how do we mitigate this risk? And with most risk I like to think about it in terms of, you know, short term, long term kind of mitigations. Like in the short term we didn't do things with the invention of the Internet like regulated into irrelevance. We allowed it to pretty laissez faire develop out into whatever's going on. Not until a, uh, real harm presents itself do you then step in with a regulatory mechanism and try to control it or stop it or find some governance structure. And then long term it's now been able to blossom into this giant thing that empowers so many industries. I'm m sort of curious if you are sensitive to a similar sort of dichotomy here where there's some short term things we can do to mitigate the risk and then what the long term sort of side of that might look like.

Speaker A: So uh, you know, kind of keep it both sides of the risk equation in mind. Right. The biggest risk that we have right now relative to AI and other digitization, we can kind of branch out what that might look like more in a second is continued under optimization to the point where our forces become irrelevant. Right. If we know we're not going to build more of them, that each one has to be better. And the best thing to do to build them better is to figure out how to create an ecosystem, a digitally enabled ecosystem of national power. And not doing that puts us at existential risk because it gives an adversary confidence that they might win, so they'll try. Right.

Speaker C: They only got to get lucky the once.

Speaker A: That's right. So I continue to hold out that that is the largest risk. The largest risk for large scale conflict in the world is failed deterrence on the part of the United States because our adversaries, they can win conventional war. And so that's the number one risk. And then there is. Okay, well, Zach, and what you're saying is we just got to get after this AI stuff that could go crazy sideways with all of the risks, you know, you know what, what is your P do? Right. It kind of questions, right? So kind of teeing up, uh, we must do this. Like not doing it increases the likelihood of a war that would impact the lives of millions, if not create that many casualties. And then there is, okay, if we must take action, how do we mitigate the risks of, uh, how that action proceeds? So I always like to think, like, what works and what doesn't. And one of your other episodes that I think we were talking about a Little Bit, the June 2024 episode where you really took a deep dive on the defense intersection with AI and perhaps an AI, uh, regulated environment because of defense concerns. My takeaway from that episode of your podcast and is entirely applicable here, is when you do what you're already pretty good at, you could just kind of expand the scope of something you're good at. Like, I have a strong hunch that I pitched a wiffle ball at Babe Ruth. He still would have knocked it out of the park. Right. Because it's close enough to a baseball that it's his thing. So what we got to do, we got to ask ourselves, what are we doing good that we could just expand the scope on a little bit and it actually help us with this problem. And there is a ton of bureaucracy involved in the Department of Defense's OT&E Operational Test and Evaluation and DTE Developmental Test and evaluation. But we have really well established framework for OT and DTNE and more so than A lot of people. And we built those frameworks because when we buy a piece of technology, whether it's fighter jet that we've been talking about, a new submarine or whatever, most of the cost is caught up in the last 5, 5 to 10% of the performance. And so we've built really robust, rich test and evaluation firms in order to measure out. Are we getting that last 5, 10 or 15% of performance that, uh, it's representing the taxpayers investment in this technology. What I have not yet heard of is how do we take that that's working really well and use that as the basis. You know, the measurement criteria are going to change, but the process and framework and importance and primacy and order in which we do OT and E and DT and E is a great tool set. And when I go out and I look at commercial industry, they have none of that, right? So what they have is AB testing. We're going to try two things at the same time, and whichever one, you know, and sometimes the metric has proven to be bad. Which one of these things gets teenage girls to watch this social media thing more is the one we're going to just double down on showing them. And those sorts of AB testing has created broad societal harm because they had no framework for operational test. And so they just let their users be their guinea pigs in what I like to call a completely unauthorized social science experiment.

Speaker C: Can't say it better myself, but we

Speaker A: have really good frameworks. And so I actually think there is a merger between the Department of Defense's uniquely rigorous operational developmental test and evaluation frameworks and A.I. uh, maturity and safety that would actually allow the Defense Department to contribute that. You know, that's a toolkit for measuring the safety, performance, efficiency and effectiveness of exceptionally complex things. And we're able to apply that framework across everything we buy.

Speaker C: I really love the way you're describing it. I have more experience on the civilian side with the Department of Homeland Security, but a lot of my work there is in that RDTNE sort of landscape. And we do operational and functional exercises for law enforcement, fire ems about introducing a new piece of technology. The purpose of those final examinations, though, is not so much to determine whether or not your widget works as how do we work your widget into our governance and command and control structure so that we can get the max benefit of your thing and do it safely to our personnel. This feels very applicable to what you're talking about. It's just that the DoD happens to mostly specialize in things that are far more dangerous and far bigger than what our state and local first responders have to deal with. But the general wisdom is really there. And like, this is getting out ahead of one of my questions. You know, it's like we talk a lot about, oh, do we have human in the middle? What's the accountability? What is the control gonna be? Like, what governance structure should we use? I mean, we do probably need to do some structured, scientifically relevant experimentation to figure out what some of those answers look like. And to your point, this is one of the better arguments I've heard. We have a framework for how to do this. Why are we not using it rigorously and actively and enthusiastically?

Speaker A: Yeah, there's a couple great examples that I like to fall back on of, uh, when DOD investment, even people don't realize it's DOD investment, like benefits the commercial space. And so a couple that come to mind are there was a slide and I've got it somewhere I might find to see if I can't dig it up. But about 80% of the technology in the original iPhone was seed funded by Beauty. And so. And Steve Jobs did an integration, right? Uh, so whether it was scratch proof glass from Corning that was originally rifle optics glass that couldn't be scratched, whether it was GPS that the Naval Research Lab built so submarines could pop up anywhere and know where they are, whether it was, you know, cell phone technology which the army built for comms compression.

Speaker C: So.

Speaker A: Well, there was the Internet that was on the original iPhone that was built by arpa. Right. The DOD has been a contributor for decades. That's our best model is when DoD takes something and says, hey, society, this is something that we really reduce the risk and the basic science out of. Let us introduce this to you as part of the framework you use. And then you go do something cool like you integrate an IFE jobs. So I think the OT&E and DTNE framework is something that could be kind of the radio D could help. The other thing is, and just a Navy example that I like to pull on, we coming up the end of World War II, where we had just been driving so hard to create a nuclear weapon. And then you, uh, had a guy in the Navy named Hyman Ricko who was not the first, but was certainly a pioneer and said, well, you know what? We could use splitting atoms to boil water, and we could use that boiled water to create steam to drive submarines. And then they don't have to have diesel engines to get around there. Imagine the fear and the concern when almost Everybody's exposure in the 50s and 60s to nuclear power was it's just a bomb. And he's like, no, no, no, it's cool. I'm going to put 200 people underwater and they're going to sleep next to one. It's going to be fine.

Speaker C: Right. Big metal tube. What could go wrong?

Speaker A: That's right. And the answer there is rigorous engineering applied by serious people with clear and unambiguous objectives created a technology that works, and we have not had one lost ship to a nuclear accident in 60 years. Because that philosophy of rigorous engineering applied in a dangerous scenario that doesn't have to be risky because we actually had very talented people thinking through the risk analysis across all the different vectors and a system where they could be heard and impact the design to end up with something that was safe makes total sense.

Speaker C: I mean, you know, it's a really profound application of that methodology you're talking about. You know, obviously for sort of like call them hard problems versus soft problems. Hard problems, engineering, the science, soft problems, the perception of the engineering and the science. You know, applying something like this, like, let's get straight into the heart of the debate, you know, AI presents some ethical dilemmas. Where do you think this sort of technique can help with that? What are the ethical dilemmas that you see as the most pressing for our current situation as we forge into this frontier?

Speaker A: Yeah, the most pressing ethical question unequivocally is will artificial intelligence make kill, no kill decisions? You know, and friendly fire is a reality of worth it. Let me kind of pull this thread for me because I think it's really important. And uh, an AI ethicist that I worked with, Autonomous Systems ethicist that I worked with at the opens graduate school for a while, told this story and I thought it was probably the best kind of summation. This, right, is that friendly fire has been an ever present concern in warfare since, since the dawn of time. And that's not going anywhere. But somehow we feel that, you know, you can be a commander and write a letter to the mother and father of a fallen soldier, sailor, airman and moria, and say, you know, we regret to inform you that your son or daughter laid their life on the altar of freedom due to a horrible accident that can never be made right. And it kind of goes along those lines, right? And they're going to feel pain and anger, but at some point they're going to accept that, you know, their son had either volunteered or was serving with honor and that this tragic thing happened. No one has gotten a letter yet. No One wants to be the first person to write one that starts off with your son or daughter laid their life on the altar of national defense because Windows crashed and the AI screwed up. Oh my God. Like human beings don't have a framework for accepting that tragedy at this point. And so certain, the largest question in AI is how close is an artificial intelligence agent going to be to the fire controller? And then you've got a couple of what I would consider very interesting kind of syllogism, logical and philosophical syllogisms around that question. The first is if we know that there is some percentage of uh, friendly fire incidents and let's just say it's 10% in a particular type of casualty and the AI in every simulation we could possibly put it through is about 6 or 7%. Don't we have a moral obligation to deploy it for the 3 to 4% of people that it wouldn't kill because it was making the right decision? And then at the same time we have a moral dilemma that we don't really understand how large language models work. They're not explaining at this point. And 5 or 6 or 7% of the people that had made targeting decisions on were our own folks. So the mortal stakes on that question, you know, have compelling arguments on both sides and horrible ethical choices to be made.

Speaker C: Right? I mean it's the trolley problem on

Speaker A: steroids, it's the trolley problem on sterilites a hundred percent. So then you're like, well that's almost debilitating when you think about how equally balanced those two moral philosophic arguments are. And then so where we come at this problem is, okay, what's a proxy problem that we can use that almost has no more M that has an objectively better outcome than any of those situations, right or wrong, that we can use to inform the theoretical debate with actual scientific data and evidence. And so we are being critical of social media's deployment of AI. I think there are other places where we've seen over the last decade that have been kind of, why was it equally unethical in the big search engines and online E commerce's deployment of AI for preference, generating about what you want to see or what you want to search for, what you want to buy and how we've been using AI to drive excessive consumerism and pushing people in the debt and all sorts of stuff, Right? So our stance is, hold on, we are probably deploying A.I. uh, in the wrong place. This brand new super powerful technology, we probably ought to let it like break machines instead of people for the next 10 years and then see how that goes and decide which AI is safe and appropriate to use in any sort of situation that involves people who may or may not be aware that an AI is interacting with. And so we have a couple of what I would call kind of tenets about what is a good safe environment to learn about AI's potential in. One of those is, does the harm show up, uh, quickly? And so let me give you the counterexample. We didn't realize that social media AI, uh, was causing emotional and psychological distress in teenagers, particularly young girls, until it had already been doing it for about 10 years. And then we had the longitudinal studies that showed us that if we deploy AI, uh, machine optimization and the compressor on the air conditioning brakes, that's going to happen a lot quicker. And so machine optimization AI like will show results quicker and it won't be systemic harms, it will be repairable harms that have no impact on society. So our first question is, do harms show up quickly in this testing? The second characteristic is if those harms manifest, is it a societal impact or a system impact? System impacts can be remediated by better systems engineering. I had no idea how to fix societal impacts. Philosophers and historians have been thinking about that for centuries and don't have a good answer. My brother is finishing a, uh, PhD in divinity at Harvard this year. Right. And he's trying to figure that out for his dissertation.

Speaker C: Man, those holiday discussions must be fascinating.

Speaker A: I know, right? Yeah. So does this deployment of AI, uh, create a prevent a systems engineering US or societal assessment challenge? Right. And we want AI, uh, the is, you know, where harm is redes, redressed by better, a better systems engineering. And so when you say, okay, hey, this is a safe sandbox, it's hard to do anything really bad in this sandbox like predictive condition. So we have a program that we support for the Navy for condition based maintenance. We love it. Not because it's just a really cool project, because it is, but it's because the sandbox is safe. So like, go as fast as you can. Build the most powerful successful maintenance, prediction and readiness assessment analytics in the world. And while you're doing that, we can figure out a OT&E and a DTNE process for that AI to approve one algorithm over the other. And then we can do transfer learning to pull an AI word, right? And say, oh, this is how we evaluate machine AI. Now we're going to go over and we're going to look at some of the stuff on the Combat system side. But people aren't doing it for the first time from a process and engineering perspective, while they're also wrestling with the horrible moral question.

Speaker C: Yeah, it's a really innovative approach because it adds specificity to the answer that you get back all the time. You know, I've asked a hundred people this question and I get the same answer. Put a human in the loop. Okay, where? When? Yeah, what you're talking about starts to provide data to answer the where is the human in the loop supposed to be to be the most effective, to get the maximum benefit out of the AI system without unnecessarily or artificially restraining it. And also, when is their presence in the system the most important? You insert a person in the wrong place, like, hey, don't bother closing the barn door. The cows are already out. Like that doesn't matter. Or if you're too early and your cows never get out to feed and they all starve to death, you've also messed it up on the other end of the spectrum. But like, what you're talking about is using this ODT system to be able to start narrowing down where some of those variables are and actually getting to something that looks like a governance structure. It's pretty freaking cool. Thanks, man.

Speaker A: Let me give you a very practical example in addition to kind of what you said about where to put the human on the loop. There are a lot of advocates that say, no, we don't put the human on the loop. The whole reason we're doing this is speed. And then you've got the people who say, well, we need a human on the loop for safety. Right. I think that whole structure of the argument, let's run the head to head on a destroyer. I've got four air conditioning plants. Let's let AI plan and direct the maintenance for two of them, and let's use our existing maintenance plan for the other two, and then let's just ping the one that works the best.

Speaker C: Right. And I mean, you still sort of semi have a human in the loop. It's just on audit. They're looking to see which one performed better. And then through a good audit, you would also determine whether or not it started doing stuff it wasn't supposed to do, which would be the other big concern. But like, you're talking about setting up the test criteria in safe environments so that you have test criteria to apply to unsafe, safe environments. When we determine that's necessary, like that's. There's more sanity to that than I've heard just about anywhere Else, I'll take that. Yeah.

Speaker A: Let me give you, uh, kind of one other thing, because we do need to get after the combat system stealth soon. But I think there's a sane way to go there too. So let me give an example. While I, uh, applaud the people that are thinking, you know, a decade down the road about fully autonomous AI culture, I think a lot of people would just rather have a better pair of glasses, right? So I think back about in the 18th century, I'm 50, I gotta wear these readers, right?

Speaker C: Right.

Speaker A: Human beings of eyes have always crapped out at about 45 or so, right?

Speaker C: Eyes, knees, other things I'd like to take up with our creator. Couldn't you come up with a better knee? I mean, come on, man, 45 and my knees don't work anymore. I still have to walk another 40 years.

Speaker A: That's right.

Speaker C: Yeah.

Speaker A: Yeah. But, you know, in the 18th century, we came up with an amazing new technology called eyeglasses. And now we could take, hey, this human deficiency and this human acuity that tends to degrade. We now have a technology that restores that acuity too. There are. There are a ton of problems in the kill chain where all I really need is better acuity, right? So when in the times where we thought about the adversary's aviation order of battle, like, how many things could they fly at us, right? Maybe a dozen, maybe two dozen. No, we got a radar that can track that times X, you know, so we can track a lot more. So we can track all the stuff they can fly at us. Well, how many things could they fly at us now that. It's a big number, right? And so now it's like, oh, which one of those things actually is big enough to carry a bomb? Um, big enough to hurt. And which one of those things is just a real flying thing? But it's just clutter, right? The AI could provide a lot of acuity to that problem. And so I often. So we're going to. And we should get after the combat systems problem, most of the AI, uh, problems that ever were. When we got beyond the conceptualization, we actually went and did the systems development. It's almost everybody wants to talk about the algorithm, and 80% of the work is the data engineering. And so our, uh, combat systems ethics problems are years away on the other side of a data engineering problem. And so we need to get after the data engineering problem that delivers artificial intelligence capabilities that just provide operator acuity. And then we get the accelerated decision making because the choices have been made clearer. Right. And you do that for the next two or three years in the combat systems world. You focus on these more open box problems in the uh, industrial control space and you end up with a really good framework for tackling your combat systems. You know, thorny problems once you're actually capable of teeing them up for a test because you've got the data engineering

Speaker C: pieceweight, that makes a ton of sense. You know, it also seems to also fall in line with one of your other bread and butter areas, which is augmenting legacy systems with these new capabilities. Talk to me a little bit about that. I'm no cybersecurity engineer, but I know enough that when you're going to add stuff to an old existing system, most of the time you've succeeded in expanding your threat fabric along with some other areas. But like, where does that all come to play too? Because we're not talking about sweeping it all aside, replacing it with all this new whiz bang stuff. We're talking about trying to make what we have work a lot better until we get to the really, really whiz bang stuff in that interim period of time, that transitional period of time is going to be really important.

Speaker A: Yeah. You know, tons of senior leaders have, huh, talked about the Pacific 2027 concern. Right. And so we will have mostly the fleet we have now. We will have mostly the Air Force wings that we will have now, mostly the marine and army units in 2027-2030.

Speaker C: Wm.

Speaker A: Matt. So what do you do about Matt? Let me give you the best example I've heard and again, uh, I'm a good listener and so I take most of my good stories from other people. But Admiral Darryl Cottle is the commander of the U.S. he's the commander, of course, and a couple of weeks ago he was explaining it like he was like, I've got about 300 ships and suckers right. When you, when you add them all together, he goes, at any given time, 100 of them, um, are overseas, are out at conducting four presence operations that are kind of 100 of them are taken apart and in some various stage of, you know, maybe get a whole new reactor put into them or something that's going to make them pretty hard to get there in the next 90 days to six months. But that leaves 100 ships and summer that are in some intermediate phase where they're not taken apart but they're not ready to go because they've got a couple little things we're doing or the crew isn't quite trained up Right. And we've had a cycle of doing forward deployed operations just to keep that 100 forward always ready. But we've never really come through as a nation or as a navy to pull the thread to say, what exactly is it going to take to get the other hundred out? So what we've done, again at the highest strategic level is we told the Chinese, well, you can read in James how many ships and submarines we have, and you've really only got to get ready to beat them out of third. That doesn't deter confidence because beating one third of the United States Navy and beating 2/3 of the United States Navy at the same time are, you know, it's not just twice as many. There's a lot of synergy comparing apples to hammers. And so the question is, how might we design AI, uh, and decision support tools in that safe sandbox that do something that is militarily consequential, like get that other hundred ready to go. So now you start thinking about, what if I could do supply chain optimization for a critical part that's keeping that engine? What if I could do shipyard planning, modernization work, package development? So I actually schedule the work through the shipyard with the various trades to get that out of the shipyard two months earlier. What if, while the ship is just coming back in that hundred deployable and about to transition into, you know, a work package, the ship could predict the maintenance that it needed from all of the sensor readings that are on it actively. And now it pulls in. And instead of just doing some standard set of work, I have this very tailored, focused list that's built off the measurements. And so we're turning it around in a month and it's back in the tube. Right. So when you actually look at what is AI's greatest possibility in the near term to make a meaningful impact in preventing war, it's actually not shooting missiles. Um, it's actually changing the entire strategic calculus where China is looking at twice as much, which is more than twice of what they would have to do to do something. So ironically, industrial and machinery optimization is actually the vital AI role to prevent and win. A pacific concept.

Speaker C: That's something else, I guess. In the near term, should there be something that comes to a head before a lot of this stuff is really fully deployed? You're looking probably at a situation where there's going to be some AI augmented defense and some traditional military. Do those strategies coexist? Or are we talking about needing to replace one with the other? Like, how would that even look in a transitional period?

Speaker A: What you're talking about is so the former chief engineer of the Navy and after he was the chief engineer for about four years, he became the chief of Naval Research and he ran onr. His name's Laura Selby. His kind of foot stomper for his last several years was what he called the hedge strategy. And so let me lay this out for you because he and I collaborated on it before, we're continuing to collaborate on it now that he's out. Here's the general concept. Going into every war ever, you always have the thing you think is going to be decisive. And then the historians write about the thing that actually was decisive. The one that's probably most familiar to folks is, you know, everyone in the Navy was convinced that battleships would be the decisive element of a Pacific fight in World War II. And it turns out it was aircraft carriers and submarines. The important thing from that very short analogy is that on December 8, 1941, when all of the battleships were sitting on the bottom of Pearl harbor, we had a bunch of aircraft carriers and submarines and they went out and got it done. Why didn't we have those if everybody believed battleships were going to be the answer? Right. Why were those there in the first place? They were expensive. What it meant was, is that the leaders of the nation and of the Navy in the 20s and 30s were willing to invest in. They didn't call it a hedge strategy at the time. Right. We think about hedge fund as alternative assets funds now. So it makes sense in 2025 that we use the word hedge strategy, but they believe in having more than one choice. So right now we have the large capital platform strategy from naval perspective. We have aircraft carriers, lots of variants of F18 fighters being augmented by joint strike fighters. So high end, expensive individual unit cost fighter jets. We have Virginia class submarines which are amazing. And we have Aegis destroyers and cruisers. Right. And all of them are super expensive. And no matter how hard we wanted to, we couldn't change. The production rate is very difficult to change. We're going to change a couple of those by onesies, but we're not changing it by orders of magnitude. So that is the thing that we believe is going to work. The large capital investment, the hedge is in the small, the agile and the mini. We could actually afford to buy a thousand of something and we could afford to deploy a tactical path to run the software for that thing and deploy lots of different AI capabilities on it. So what your question was, what's the interim strategy? I think the interim strategy is we need to invest in the hedge and get after it. And then we need to figure out uh, how do the high end capital platforms and the hedge force actually interoperate? Could create that enveloping, overpowering maritime system of systems. But until we've got the submarines and the carrier sitting in Portland, in Hawaii, like we've only got one choice. So what we need to do is get after building the head strategy and the truth is there are so many smart men and women in the operational forces, they're going to figure out the synergy piece. We just got to get supportable kit that works and that represents going back and just pulling it through on some of the analogies you've used. That represents Amazon, not Sears with a uh, website. Right. We need to get them capability that actually represents continually integrated, continually updated software delivery to the tactical edge with a tight MLOPS loop. So the AI is always.

Speaker C: Yeah, I mean it's a fascinating space to be in and we're watching it all play out in real time. It's very fun stuff. Zach, we have gone and covered a lot of ground on our conversation. I always like to end these things the same way. A lot of our listeners are legislators, policymakers, work in the federal government, people that need to be aware of all of these sort of things. Final thoughts here. What advice would you have? Where should they be putting their attention in the coming months?

Speaker A: Tactical platform as a service. So let me put just a kind of quick explanation on that and we can wrap it up. If the United States AI capability is going to operate in the hyperscale cloud, Azure, aws, some of the other big cloud providers, then we just need to work on policy and governance and deploying it. I don't believe that. I believe that our ships, submarines, aircraft carriers, soldiers, sailors, airmen and marines who are deployed forward will have tenuous access to over satellite and other links back to that environment. And we will want AI, uh deployed with them on those platforms that continually updates from in situ data that it's actively collecting. There is a really well defined technical reference architecture for how you build a data center and how you deploy AI workloads in that data center. And that's evolving, but it's got its own momentum. There is no tactical paths, there is no software stack on ship, submarines, unmanned systems, aircraft, et cetera that allows us to replicate the cycle of deploy, uh, AI, measure its performance, retrain it and redeploy it onto a tactical platform. And so the measure of how good is an AI is not actually a measure of how good is an Algorithm. The measure is how fast can I train and redeploy that algorithm every time I get a new interesting thing in my training data? And so a tactical PAAS that is built for cyber war is what we need, and it doesn't exist. The runtime environment on all of those things is some collection of usually, uh, like a Red Hat operating system and VMware and some distribution of kubernetes. Sometimes that's if you're doing really good. Lots of times it's not even that. And it's just a hodgepodge. So we need to consolidate on a tactical infrastructure. This supports the ML Ops loop. We need to accelerate the learning of our, of our learning systems. And last kind of last quote, right. The admiral, early birth in World War II, he said, uh, he said the difference between a good officer and a great one was about 10 seconds. And which was, which was about. He thought of a good officer could figure out how to aim and shoot about 10 seconds faster than the other you usually want in naval conflict. So I don't know what the number is now, but the one thing I know is that our acuity analytics, our decision support analytics, and our people will need to be on a very fast cycle time. And so we got to figure out, uh, how to measure that time and see how many days or seconds it is for the AI and then go after with a vengeance, getting the number down.

Speaker C: Interesting. Well, thank you so much for sharing your wisdom with us today and for all of our listeners. We'll see you next time.

Speaker A: Thanks, Max.

Speaker B: AI, Government and the Future is brought to you by Corner Alliance. To find out more about Corner alliance and how we work with government to create results, visit visit our website@corneralliance.com and then make sure to search for AI government future, uh, in Apple Podcasts, Spotify and Google Podcasts or anywhere else podcasts are found. And click subscribe so you don't miss any future episodes. On behalf of the team here at Corner alliance, thanks for listening.