AI and DoD Technology Adoption with guest Bob Beaton
Our Nation’s Military has always relied upon technology adoption to keep it ahead of peer adversary forces. Today Artificial Intelligence and a host of related and supportive technologies are front and center as our Nation works to remain strong in the face of peer competitors, China and Russia. Expert DoD technologist, Bob Beaton, joins Jim and I to discuss the state of play for AI and related technology adoption.
Webinar Transcript
Marv: [00:00:02] So welcome, everybody. This is our ninth webinar, and we’re pleased to have Bob Beaton as our guest today. Bob and I have been colleagues for a long time, starting at DARPA back in the late 90s, where Bob was already well ahead of me from his days of building new capabilities for the military. So, Bob, thank you for joining us. And let me let you expand on what I said about your background a little bit, and then we can get into the discussion about the technologies that you’ve been working on to help improve our DOD.
Bob: [00:00:35] Sure, thanks, Marv. Yeah, well, let’s see, I’m an aerospace control systems guy by education background. And back in the eighties when I first got out of school, that was the second wave of AI and everything that was going on was automated vehicles and getting started in that area and automated planning system. And there were lots of interesting things that DARPA was funding at the time and so on. And so I kind of got pulled in that direction and quickly got pulled out of the aerospace community and into the bigger automation AI Enterprise Command and control kinds of things, and sort of started going down that line. As you said, Marv, I got a chance to work at DARPA in the late 90s, which was a lot of fun. And we did some really interesting things. And since then, I’ve been consulting on my own, mostly for DOD and intelligence community customers in the government, but an occasional commercial venture here and there. And in the last couple of years, I’ve been spending most of my time supporting the Joint EHI Center and doing some work with Disha Emerging Technologies Group as my primary activities.
Marv: [00:01:51] So I remember Bob back when we were in the DARPA Information Systems Office Division. That was when natural language processing was being developed by that organization to a significant extent, and also the automatic image recognition technologies were being developed. And then, of course, AI went quiescent and nobody would say the word because it had sort of been overblown, I guess. And now we seem to be, as you said, in the second wave of AI that’s now promising to do everything for the world. So, let me just ask you, what technologies do you see that are critical for going forward given the peer competitor structure we’re dealing with today in China and Russia?
Bob: [00:02:43] Well, I’ll tell you, it’s interesting, My view is that from what I’ve seen now and in the last three or four years is that technology is really not our problem. There is tons of technology out there, commercial worlds developing it like crazy and all sorts of different areas. And our big problem, I think, is going to be how to be able to digest it and bring it in to our mission operational activities in light of all the very complex processes we have around security and acquisition and data sharing and all sorts of things that are just presenting really, really big obstacles to us being successful. And in my experience here, working in probably the last two or three years, I would say security by far becomes the biggest challenge. We probably spend 80 percent of our time trying to work through all our security constraints and all of the processes and issues there. And to me, that’s probably the biggest challenge of all, how do we how do we have a breakthrough and how we deal with security so that we can we can get some of the good stuff that’s out there in place, let alone try to develop new stuff like DARPA’s trying to do.
Marv: [00:03:54] Yeah, an interesting story that goes back a long way, to when I first met Jim Pietrocini. He was the guy that introduced me to World Wide Web when I first got into the DASN job in the early 90s. So, of course, now we’re up in the business of cloud and trying to adopt a commercial cloud into our war fighting military infrastructure. You might want to say more about that, Jim, but I think that’s a significant change that’s ongoing right now.
Jim: [00:04:26] Yeah. So, just back from the JADC2 conference at Texas A&M, and it was the technical director from Google Federal that kind of said we need to change the way we have multiple classified networks where we, can’t control our data by just putting them on different classifications of networks. And if you really want to have data to reach across all the services, you’ve got to change that. And so I think that’s, you know, from a cyber point of view, I guess the question is, do you think the DOD will ever go to one or reduce the number of different networks they have?
Bob: [00:05:05] Well, you know, we’ve been talking about trying to do those different ways for a lot of years and still haven’t seemed to make much progress. Know, I don’t have a good answer there. I think the technology’s there or if we have the will to do that. You know, it’s interesting. I was just talking to some folks a couple of days ago in this whole network area, I guess, Admiral, I’m sorry, General Skinner. At DISA has decided that the way we’re dealing with all our network interconnections here at scale is become such a problem that he’s starting a major new initiative called Thunderdome to try to kind of completely reengineer the whole Internet to government interconnections and how we start to do across a bunch of these networks. And we’ve got to do something on that scale, it seems to me, because we’re we’re kind of really old and patchwork system that just has to be totally reengineered. So, hopefully we can do that. We’ll see how that goes. But I would like to add that it’s interesting when it comes to data sharing, we’re even having major challenges sharing data on the same network.
Bob: [00:06:12] We have lots and lots of organizations, let’s say, that are working with the guys I’ve been working with, mostly doing work at the unclassified level on NIPRnet. Just sharing data at the same classification level has been a nightmare with all of the tendency to have people wanting to not let their data go to all of the rules and policies they can point to for why they shouldn’t share. And then even if they are willing to share, then we have our security authorizing officials who are reluctant to either let data out of their networks or accept somebody else’s data into their networks, because maybe that will cause a problem and somebody’s going to get in trouble. So, I don’t know. We could do a lot, a lot of benefit if we could just figure out how to share data within a single classification, in my view, let alone the bigger problem across. And I think you’re absolutely right. I don’t know how we’re going to be successful in JADC2, too, if we don’t really figure this out.
Marv: [00:07:10] I actually like to pull that thread a little bit more, Jim and Bob, because I’ve thought for a long time that this three physical or multiple physical networks for Security Classification Network find one perimeter was your only way that you try to mitigate the cyber security solution. But unfortunately, it minimizes your ability to share data, as we’ve been talking about. And these authorizing authorities can never get on the same page when you’re trying to get an authorization to use multilevel security or multiple security domain, cross domain devices and et cetera, et cetera. So, the net result is we’re holding back the military capability. We’re minimizing our ability to move forward in any kind of rapid way. And in the meantime, we have software defined networking that allows us to break networks into thousands of networks and do all kinds of different things with them right in the same pipe. And we don’t take advantage of it because we’re locked into this archaic, historic reason for having the physical networks on everything we do. And I think is crazy.
Jim: [00:08:15] Yeah, I think that the conversation at the conference was, you know, Netflix users, digital rights management. And then we have things like attribute-based control and role-based access control. So, there’s technologies out there that you could put everything on the same network, have one network, but you limit who can see it by other things other than physical separation. Right. And that’s a culture shock and education level. And like Bob said, you have this legacy bureaucracy of processes. And I think the comment was from a senior person who worked in the Pentagon. He said just to get jaywalks in the Pentagon took me two years. So, it’s definitely an issue. It’ll be interesting to see if it maybe JADC2 becomes the effort that tries to fix this. But we’ll see.
Marv: [00:09:07] I don’t I don’t think JADC2 is trying to really fix it. Nobody tries to go back to that very same problem and foundational problem and fix it. And the problem, as you just stated, is the productivity is just completely down the drain for anything we try to do, whether it’s just getting a connection drop or whether it’s operating on the systems to make something happen for the mission activities. Go ahead
Bob: [00:09:30] Yeah, well, I was going to say that the one area where I could see some hope right now is in the commercial cloud environment, the big providers and I won’t get into specific providers by name, but for the most part, they are putting data on the same physical network and they’re limiting access to it through software controls of various kinds. Right. Some deeper than others. And, you know, if you go to one of those cloud providers now and you know, you can’t you really can’t get a physically separate network for your data, you have to live with their software separation’s and they’re very secure and they put a lot of effort into it and they’re very strong. And it’s really comes down to what people are used to. And you’ve got folks who just are used to and comfortable physical separation and they just don’t want to to reconsider. But as that becomes more and more ubiquitous, both in the commercial world and as we start to bring clouds into more and more into the government world, I think that’s going to change it. Just unfortunately, changing a lot slower than we need to change.
Jim: [00:10:39] Yeah, that’s a great point, like Amazon doesn’t have a physical JWCS cloud and a physical SIPRnet cloud and they use software technology to do the separation. Right. And you’ve got your different DOD impact levels, the same physical cloud and hardware, so that there might be a great way to maybe get some interest.
Bob: [00:11:04] Yeah, they should get better over time, but again, will we get better fast enough? We got other adversaries that are being as tentative and. Slow as we are, and that’s not good.
Marv: [00:11:16] So let’s move over to AI for a minute. How do you see a plan in the future of the building, and what from your vantage point, where do you see happening?
Bob: [00:11:25] Well, I mean, you know, that’s the really hard to say, right? I mean, it’s obviously an incredibly high priority and it’s got people’s attention and it’s got people running scared, not just at the DOD level, but from an overall national competitiveness perspective. Right. You have political leadership thinking we’re in this big existential fight for the high ground with other countries. So, it’s certainly got a lot of attention and a lot of resources. I do worry a bit that it’s you know, we may see another wave like we’ve done in the past where it gets overhyped and then people, you know, it comes back to reality a little bit. But, you know, there are some really big breakthroughs here in this third wave that we’re in relative before, largely driven by what we didn’t have in the first and second wave, which was really cheap, massive computing at scale through the cloud and a massive, massive data storage. And I think that’s probably been the key, right, is we now can get big enough data sets and process them at scale that we can start to find and do things with deep learning and things that we just didn’t have the horsepower.
Bob: [00:12:25] So that’s really the big change and it’s having a really meaningful impact. And you can see the commercial guys are making huge investments and, in the attempt, to keep up. So, I think it’s here to stay. I think the challenge for the DOD, again, as I’ve said, is, is less a technology one and more of a cultural business process reengineering. And can we change the way we operate to to account for the way you’ve got to do? And Mark, we’ve talked about this. There are a lot of really good of the most advanced AI enabled organizations. They really make the case that it’s you have to make an organizational level philosophical change to be successful in AI It’s not a little gluon. Add this on to your current process. You really got to really think about changing the whole way. You think about things and to be successful. And that’s, I think, going to be the bigger challenge for the department. Right. We know how hard it is to change the culture, the change processes and change mindsets. And can we do that? It’s going to be it’s going to be a real big challenge for us.
Marv: [00:13:29] What I think is interesting is that unlike the days when we were in the middle of this business, we now have a whole hardware layer that’s radically changing. So, we have tensor processing units, neural processing units, GDP’s, I mean, CPU’s. So, we are causing the hardware solutions to overwhelm some of the complexities along with those large datasets that you’re talking about. And one thing that that I’ve thought about a lot is I think that in the DOD side, we tend to be thinking about AI to solve problems that it’s really not ready to solve. For instance, command and control problems, because you need to have explainable AI That helps the commanders understand why the machine is recommending what is recommending before. He’s going to try and do that and put it into the battlefield where real time activities that are going on. But the other thing that is really good at is it could easily be a AI model that’s taken information straight off sensors and pulling out the exact right information and passing it to the right places much faster than we could do it any other way. And it could also eliminate the amount of data that we have to move around from our mobile platforms like ships and airplanes, et cetera. And then the other I had this experience recently. I wanted somebody mentioned to me that I ought to put a written transcription of these Web webinars onto the site. So, I looked around how to do that, found a new commercial business Web page that for a very reasonable price, was able to take my audio files and transcribe them into written files. And they come out phenomenally good and they even identify which speakers are speaking. I just have to go in and edit the name of the speakers because they call the speaker one, two and three. So, I knew immediately that they were using an AI capability to translate these kinds of verbal discussions into a written language that’s so perfect.
Jim: [00:15:28] Yeah, I know, I know, like some of the early I would say probably machine learning is catching on a little bit more aware. I do think daily folks are starting to see that like reading logs, you know, after the Solarwinds incident, like computers can read logs better than humans. And from a cyber point of view, like, you know, the old days, we’d have 50 people sitting at a sock pretending to look at monitors. And so I think I think anomaly detection is going to be the first place. And, you know, image, you know, intel analysts are starting to see that computers can do some of the mundane stuff better. But we still have a long way to go, I think.
Bob: [00:16:07] I think you’re right on, Jim, that that’s a really great area, there is tremendous amounts of data in the cyber space that’s generated all the time, and it’s too fast and too painful to have all but the Super seven people. There are some people who can read it like the code in The Matrix, and it’s just almost intuitive. But for 99.9% of us, it’s a real drudgery thing and there’s some really good opportunities there. I know one of the things I’m doing with this right now, which is kind of interesting, they just are bringing on board what they call a cyber asset inventory management commercial tool from a small company that’s going to generate a tremendous amount of information on over one hundred thousand devices on the GST network. But then the next question is, OK, great, you’ve got all this data. How in the hell are you going to possibly figure out what’s going on at the right speed without some sort of AI type of help?
Bob: [00:17:06] We seem to be really good in DOD about buying more and more sensors to produce more and more information. But we aren’t so good about spending money on the things that make sense of that information and turn it into something useful. Well, remember the one of the classic cases, it’s not a direct AI effort. But Marv, I remember we were working the thing with the Navy and all the money they were spending on on P-8’s and the the Triton’s. Right. You know, hundreds of billions of dollars on these aircraft. And you couldn’t get them to spend any money on the talks that would collect and store and process the data, relatively speaking. So, how do we how do we balance that out? It’s going to going to be a big challenge. But I know the cyber area is a great area, probably one of the best. And also another existential problem for us right now. Right. With all the businesses getting ransomware, it attacked all over the place. So, I definitely like to see some work and focus in that area.
Marv: [00:18:05] That’s a good point you brought up. I thought about it for a while, but you’re referring to the data study we did on how much data is going to be available. And we did this back in the 2014 timeframe, if I remember right. And we have determined from from all of the meetings and reviews, we know that the Navy was going to be producing petabytes of data per day by 2018. So, obviously we’re producing that much and more, and as you point out, we don’t have enough capacity on the ground sites to bring all the data down and stored or do anything with it more than throw it away if they didn’t get something off of it in real time. And that goes back to my point about we really could use much more more effectively the AI tools that could be applied to that streaming data coming off those sensors. But here’s an interesting thing about cyber security. The Navy just ran three AI cybersecurity challenges, so they ran a cyber security challenge. It said which company has a product that can help protect the endpoints with AI? Then they went on to say which companies and technologies could protect the network? And now they’re finalizing one on which I can protect the SOC. And the interesting thing about the awards and the findings that they’ve done so far is the AI systems that have been brought in and tested and then have been awarded the award. They’re not doing any better than the non-AI capabilities for that kind of cyber detection of malware. So, we’ll see. I’m sure it’ll get better over time.
Jim: [00:19:41] Yeah, yeah. Another area is there’s all this data, but there’s a new buzzword in the commercial world called index lifecycle management, where, you know, out of all the data that you collect, what’s hot, what’s warm, what’s cold and even what’s frozen like, how much data do you really need to keep and analyze? Because most of the commercial companies are dividing up the information in the segments. Right. And if I only need to look at logs for the first couple of days and after their five days old, I just archive them. And so that’s a challenge to get the government to kind of when you ask him how much data do you collect? First of all, they sometimes don’t even know how much data they collect. And then they don’t know how much is hot, warm and cold, so so they can’t get any kind of cost analysis. And there a view at the Getsy conference that like to collect all the data. The government can’t afford it, you know, if they if they want to use commercial tools. So, it’ll be interesting how that evolves.
Bob: [00:20:42] One of the things I think an interesting area of I want to just throw out there while I’m thinking about it is, you know, we. We’re going to leverage commercial tools and there’s a lot going on in the commercial space to improve the odd business practices and things that sort of line up with the commercial world. But we do have some really interesting, unique problems out as we get out into the tactical, you know, warfighting domains at the edge. And, you know, one of the things that’s been going on is there’s been this big push the last couple of years while I’ve been working with this mindset for a lot of the folks coming in doing AI, that we’re going to bring all the data back to the central cloud and crunch away and do great ehi there and then push out EHI models. And we’ve been really trying to convince folks that there’s a lot there’s going to be a lot of data out at the edge that can’t come back. We don’t have the tactical networks, or they’re being used for more higher priority things. And we’ve got to figure out a paradigm that may be different from commercial where we can push out the AI, what I call the AI workload out to the data and figure out to do AI in a much more distributed way where it’s not just all about moving data around, but the right combination. Sometimes maybe you move data to the compute. Sometimes you might move the AI training workload to the data and figure out how you how you make that all work. And that’s not something that’s as high a priority for most of the commercial world. So, that’s something that maybe the DOD can be focusing on on some of those areas.
Jim: [00:22:08] Bob, we had my last previous company. We used to say send a question out to the data and get the answer back. You don’t need to send the data back to get the question answered done. And the question and answer are are lower sized files. And with that culture that you’re going to have nodes out at the edge and then you have things like cross cluster search and cross cluster replication is a commercial concept that the government still they still have this mindset, you know, send all the data back to the intel center so that, again, culturally, it’s new stuff that you have to get educated on. So, we’ll see how that happens.
Bob: [00:22:49] Part of the challenge, I think, is a security driven model. Right. So, we actually have a guy here who just brought on board the Jake who had been doing a lot of work with Tesla. And I talked to him about, you know, he’s talked about how companies like Tesla are pushing a big cloud in their case out to the cars. Right. The data is at the core of millions of cars. You don’t want to bring data back from millions of cars, but you can send some AI training out. And what you have to do is you can’t train one model on one car. You’ve got to train on lots of cars and then figure out how to aggregate the results and sort of the best model. So, there are some challenges, but the nice thing that they have is that they have a security model where they allow themselves to do training on an operational system, their equivalent of the tactical environment. Our security rules do not allow for us to do any sort of development or training activity on a production or tactical system. Right. That’s against the rules. And they were put in place for good reasons at the time. But now, as we get to a world where to take advantage of AI and to move quickly enough, you’ve got to be able to start to blur that distinction between a production environment and a development environment. And you have to allow development to happen in production environments and real production data to be available in development environments. We’re breaking our security paradigm rules and we don’t fit that and solve that. We’re really going to limit our ability to be a foreign leader and AI and
Jim: [00:24:17] We’re doing we’re doing dev stickups. But right now the office is still separate. But, you know, I think part of the stickups is supposed to be continuous monitoring. Right. And then you continuously update the software and they’ve done some demos with, like Joint Strike Fighter. But again, it’s still a mindset or an education that, you know, you are going to update that operational system in real time.
Marv: [00:24:42] So, again, it’s a critically good point, Bob, but I’m actually in this particular area a little bit more optimistic because of the work on virtual twins. So, as you know, the Aegis system is working on a virtual twin. And the idea is they preload all the new software onto the twin, which is plugged into the excesses of all the sensors and controllers that they’re trying to control with the software. And it can test it there before it moved in and used as the uploaded software version for the operational software that runs the ship or runs the combat system. And that’s not unlike what Tesla does. So, when Tesla does, they call it a shadow compute system or shadow image. So, the shadow is actually learning and trying to improve the AI while the operational system is used in supporting the car to do self-driving or cruise control, whatever they’re trying to do. And the interesting things that they’re also doing about it, which points to your earlier discussion or, Jim, discussion about Heart-warming called data what they care about for the full self-driving, which is. Camera driven only right now is they care about the outlier points, they care about the points when what the car already knows with its training algorithms is a new thing to it, that it doesn’t get quite right. So, the shadow computer is really looking for the outlier things that are where the car’s doing something different than they would have expected it would do or should do. So, I think we could do the same thing with our virtual twin kinds of things as we train systems. But what hasn’t going on is everybody doesn’t think about having a virtual twin architecture. It’s in the Aegis combat system right now. It’s not in the airplanes right now. It’s not in the command of control systems right now. But with the compute capacity we have today, there’s no reason not to have insurance all the time.
Bob: [00:26:35] Yeah, no, that’s a good point, Marv. It’s certainly doable. And I’m familiar you’ve been briefing me on that in the past, and that’s really exciting. But I would say that’s the rare exception right now than the rules. So, we got it. We’ve got to start to educate people that that’s feasible to do. But, yeah, absolutely. That’s that’s the way to go.
Jim: [00:26:54] So so I have a question for both of you. Just because you two are very, very smart with your backgrounds. General John Murray, who’s the four-star Army Futures Command, they’re planning to bring in software developers into the active duty Army over the next five years because their mindset is we have to have a developer and a coder at the edge. We can’t systems are going to change so fast that we can’t rely on contractors or even the labs to do, you know, monthly or six-month updates. They’ve got to be fixed in real time. So, He came with a strategy to acquire these developers and recruit them and bring them on. But then a lot of people in the audience said, well, how are you going to retain them? So, I was just wondering your thoughts on where do you see that going in the next 20 years?
Bob: [00:27:47] You know, the whole the whole talent area. And how does the government and the military get the right talent? And that’s a huge challenge there. You know, the National Security Council, a group that’s that’s been chartered, has put a lot of effort into looking at that problem, not just from a DOD perspective, but from a national perspective as well. And it’s a tough nut. You know, there’s a lot of interesting things we can try. I think we’re going to have to be experimental on. Some of the ideas I thought were interesting is I’ve had a chance to work recently with a lot of the AI activities, most of the combat commands with some of the groups that I’ve been running for, the Jake and the where I see them doing the best work is they have military officers that have the military experience, that have retired and gotten interested in data science or AI and gone back and gotten those degrees. Some started them while they were still on active duty. But a lot of them actually went out after they retired and did that. And they’d come back as sort of military savvy data scientists or AI people. And they can actually now have both the domain knowledge and the data science knowledge to do essentially real time work for the combatant commands or in some cases further down the chain. And that’s where I see the by far the most successful progress. Now, these are little islands kind of scattered about, right, that they’ve kind of done this on their own. But maybe there’s a way that we could do that. Maybe there could be some programs where retiring people could go in and the government might help them with the degree and bring them back for some period of time if they’re interested in getting into that field.
Bob: [00:29:30] The other thing I’ve always kind of been scratching my head with is how do you get the real superstar talent to come into this, especially now with industry throwing so much money at people? Right. There’s no way the government can compete on salary with startup money and stock and all that that’s going on. But the one area that I kind of find is interesting in my older son field of law is that right now the very, very best talent in the world coming out of the very best law schools, they fight for all over themselves, trying to get the lowest paying job you can coming out of school, which is to be a clerk for a federal judge or a Supreme Court judge and make thirty thousand dollars a year for one or two or three years. And the reason they do that is because that has been made into such a prestigious position that they’ll do anything to get that. And I’m just wondering if there’s a way, we could somehow have very prestigious opportunities for the best talent coming out of school to maybe come into these jobs, you know, not permanently. They’re not going to stay for. Forever, but something like that to combat the pure financial perspective. I know there’s there’s probably a lot of other good ideas, but I think we’re going to have to really think creatively if we’re going to if we’re going to attract that kind of talent in there.
Bob: [00:30:52] The last thing I would say there, too, and I’ve talked was interesting. My younger son’s a software guy and he was graduated recently. I think I mentioned this to Marv and was actually interested in looking out, getting into government service, working some software kinds of things. And he ended up. Pursuing at the at the job fair is when everyone came in, he was a Georgia Tech guy, talk to commercial guys, talk to government guys from several of the organizations. And the commercial guys were getting back to him in a week or two with, you know, come out and visit us or with offers. The government took six to eight months. In some cases. He got he got an actual call from a government organization over 12 months later after he had selected a job, moved to another state, got an apartment and been working for six months. And then for the first time, he heard back from this government organization. Now, maybe this is a typical I hope it’s not. But my sense is that government has to move with the pace of its commercial competitors or at least be close because these kids aren’t going to wait around for six or 12 months because the government’s waiting for a rectangle. But if you want to go get talent, you what commercial guys to have a way to go get talent and bring them in and then figure out where to put them. Don’t sit around for two years and waiting for some guy to have a rep before you go get a guy. It’s too late.
Marv: [00:32:17] Is great discussion about it, Jim. But Jim, you know well that the entry system, as you worked on and Admiral Tuttle, who was the operational fleet commander, that was the example that worked very, very well of what you’re talking about. But it was done with the industry folks writing ships, writing code during their waking hours on the ship to increase the command and control capability. And of course, we’re still living with the basis of that technology today. So, rather than what the army is trying to do, I agree with Bob. We’re probably better off if we bring in the experts from one form or another where you can. But industry reform and you don’t have to worry about the government salaries. And it could be very prestigious for an industry guy to go get a position where he’s working on something as critical as that and applying leading edge technology to the front lines. And that would help his career as well as help the government in a different way.
Jim: [00:33:15] Yeah, I think there’s some we do some work at space camp and, you know, there’s some interesting things like Air Force has done with Kessell run in space camp that know there. I just was at its base camp two weeks ago. And like the the Air Force captain wears khakis and T-shirt and a ball cap and they’re in a we work like facility. And, you know, and I don’t know, Marv, I was I was I was thrown out the idea like, do we see bonuses like we saw in the nuclear power world where, you know, not not not a huge amount of money, but enough money to separate them from you know, they were specialized, right? If you are a nuke, you got your extra 50, 60 K after five years. And that was enough to retain my brother-in-law as a COBRA pilot. He got a $65K bonus and that’s what retained him to twenty years.
Marv: [00:34:04] You know, I think you’re making good points. So, we’ve actually gone probably thirty minutes, maybe a little bit over. This is a sort of a wrap up discussion. Bob, I wanted you to just come on what your perspective is on the China race. You already mentioned it once, but where do you think that’s going to take us over the next five, 10, 15 years?
Bob: [00:34:27] Well, 15 years is tough. Let me start with five, maybe I can handle that. I mean, clearly the next five years this is going to be a big push on both sides. And I think the wave of momentum behind this will carry this along for at least five years. Right. And with the budgets that government budgets and so on, that alone is going to carry this forward. And we’ll see what we see in the past. Right, is after it’s one thing to go the first two or three years on on sheer panic. But at some point, people are going to want to start to see things paying off and some progress happening. And I think we get to the five- or six-year point. If we’re not showing real progress, then like the second wave and the first wave, you know, there might be some ratcheting back in terms of how much we’re spending here. And so, we’ll see. I it’s hard to predict. Again, I think, unfortunately for us, it’s not so much a technology issue. It is a cultural change and process improvement. There’s lots of things we could be doing right now quite quickly that I see stifled and impeded just purely by the bureaucratic obstacles in front of us. And, you know, that’s what we’re going to have to change. And Jim, you mentioned the Kessel run and the and the Air Force base camp and the thing. And one thing I will say is that with the folks I’m working with, they have they understand that goodness of the Air Force work in there. And we’ve partnered very carefully, very closely with them and are trying to to look at some of the desktops automation. How do we extend it to the full lifecycle? And there’s a very strong partnership there with the Air Force.
Bob: [00:36:07] So there’s some I think people are looking at the right things, but we still have a long way to go with respect to the security processes and so on that know keep us from getting all the way there. I like to actually see I was thinking about this. I’d like to see us somehow have a multi service, multi policy, technical, you know, some sort of real focused effort at a very high level in DOD to go to essentially build a new security approach for AI for the department that’s not pressured by immediate need to deliver a specific product. Because I what I see happening a lot is. People who are charged to build a capability. We’ll be at the same time trying to reinvent a new process that they need to be successful while they’re also trying to build the mission, a capability that they’re being asked to build. And those are those are too much. And what happens is the process reengineering falls by the wayside at some point while you seek to make progress on your core mission AI application. So, if we could get the real team of real empowered people to sort of essentially what’s what comes after RMX, the risk management framework, what’s that got to look like? And how does it have to work for AI and do that for the department as a whole and really crack that nut? That would be, I think, the biggest thing that we could do that would benefit everyone across all the services and that commands the whole nine yards. So, that’s what I’d leave you with a push for that. We need to make that happen.
Marv: [00:37:43] That’s an excellent point. And, while you’re talking about that, we didn’t mention that AI opens up the whole new security or cybersecurity domain of AI spoofing, which is way different than just normal cybersecurity. So, thank you again for being here. Jim, did you want to say the last word?
Jim: [00:38:00] No, I just thought maybe you just maybe say something about Admiral Clemens and maybe we honor this podcast for him.
Marv: [00:38:09] It’s a great idea. I was telling these folks before we started the recording that I attended Admiral Clemens funeral this morning. In fact, I’ll post the memoriam that was created by his wife, Marilyn, for him, it’s a very strong tear-jerker that represents his thirty four years in the Navy. And fortunately, he’s followed by his son, who’s in the Navy, still as a Navy commander. And he’s got four beautiful grandchildren and his wife still lives in Boise. But he was a great man and he died at the age of 77 well before his time. So, I’ll put that memoriam up. So, thank you. Yeah, thanks. Thank you, Bob.