Mission Ready AI: From Strategy to Scale

Jeff Kavanaugh:

AI in aerospace and defense only matters if it improves mission readiness, program execution, or decision-making in the real world. Where is the biggest gap between AI ambition and the operational reality? We’ll start with you, Venice.

Venice Goodwine:

Okay. I think it is the desire to move. Do not let perfect be the enemy of good. Move out. And so when I say readiness for the Air Force, what does that mean to you? There is a mission capable statement that we use for our aircraft. That means ready. But it also means that our people are ready also. And so sometimes there’s a gap between what I desire and how to get there, and AI is just one way to do that.

And so when you think about it, for me, readiness through using AI is one way that I can give time back on mission for airmen and guardians. When I have to have my airmen and guardians sitting at a computer doing something administrative, that takes time away from mission. When I have a cyber operator that cannot hunt because he’s trying to, he’s inundated with so much data, when I can use AI to comb through that data and give insights, that’s readiness. And so I think for me, readiness is the key.

But for industry, understanding what that means through which lens is also important. When you’re an operator on the flight line, when you’re maintenance, when you’re flying the aircraft, when you’re in the back office, readiness means something else. But all of that is being measured and counted in the Pentagon by the Chief of Staff of the Air Force and the Secretary of the Air Force that I owe to the Secretary of War on are we ready to engage the enemy and when? So for me, that’s important. Readiness.

Jeff Kavanaugh:

Great words. Erik.

Erik Swedberg:

Sure. So the biggest gap. How many of you believe that there is a full, accurate 3D scan of the Boeing or the Airbus A350 in China right now? It’s pretty likely, right? Why is it that they can’t certify an aircraft? It’s knowledge and know-how. The knowledge on how to make a microphone work at the right volume as well as the knowledge on how to make an aircraft is the real key.

So the biggest gap is we assume that if I have all of the drawings, if I have all of the documents, and I have all of the large language model information, that therefore, I can do anything with AI. That knowledge and know-how is the key. That’s the gap we’re missing.

Jeff Kavanaugh:

Erik, what would be your go or no-go criteria when going from practicing to actually being ready?

Erik Swedberg:

We have an industry that has 50 to 100-year life cycles of its products. We have engineers that have 20 to 30-year careers, right? The key go, no-go for me is around the time on mission. So if we’re working on an area that will stand the test of time, that will meet that, because as those engineers retire, we lose that knowledge and know-how. As we loose that, and so if we’re able to prioritize that, that’s the go for me in terms of that AI use case.

Because we confuse too often today still automation, and I think you were talking about that earlier, and AI. AI is where that stepwise value on time on mission and whether it lasts the mission, being the aircraft lifecycle.

Jeff Kavanaugh:

What about you on that go, no-go criteria?

Venice Goodwine:

For me it is risk. It is what risk is being taken. We often talk about AI and the human. Where is the human in AI? Is the human in the loop, on the loop, observing? So where is the human in that? Well, it depends on the risk of the decision. It depends on the risk of the use case. That is the go, no-go for me when I weigh the risk.

So if you think about AI boundaries. Let’s think agentic AI, for one. Agentic AI will spawn other agents. Where is the right orchestration that makes sure that those agents that are spawned stay within boundaries? That if I had an AI agent to conduct a financial transaction, that that AI agent didn’t do something else. Where are the guardrails that are within the AI? So for me, it’s risk.

And I love using AI, and I think it’s the right thing to do, but I think we need to be smart about it, educated about it. But I think we need to put the AI agent on the org chart, and we need to put performance around it, and fire it when it does not perform as intended. Right? Because it is just that. It is additional manpower.

And the last thing, too, when you talk about risk and go, no-go, is I love it when someone says my AI can increase productivity by 50 times. I say, “Show me the math.” Because that’s the math I want. Because what I’m looking for is, can I... is that cash back in the bank that I can invest somewhere else? And so for me, go, no-go is about risk, but it’s also about return on investment.

Erik Swedberg:

Yeah. And quite often, many people claim the same benefits every once in a while.

Venice Goodwine:

Yeah, yeah.

Jeff Kavanaugh:

Sticking with you Venice, trust, control, and accountability are essential in mission critical environments. What are the governance and operating principles that are needed to scale AI securely while maintaining speed in the mission?

Venice Goodwine:

Responsible, ethical AI. When you put this capability in the hand of our users, there are some that are going to make sure it does what’s intended. But there is going to be someone in your organization that is going to push it beyond the limits in which you intended. What guardrails do you have to know about that?

Do you have guardrails that if I copy and paste my prompt, which is intellectual property by the way, into my own version of ChatGPT on your network, will you know it? Those are the type of things I think is important that we cannot have one or the other.

We need all three. We need trust. We need accountability. Our AUPs, our user acceptance agreements now need to be different. It can no longer just be, I’m going to give you an email account and give you access to my network, and you can go work. I want you to put that last line in that say, and if you violate the responsible AI ethical rules, there are consequences. That could be termination. So I think it’s very important that we highlight the responsibility of the capability that we’re putting in the hands of our workforce.

Jeff Kavanaugh:

Erik, A&D organizations are not short on AI pilots. But how do you turn those point successes into scaled successes, especially given your background as an enterprise software?

Erik Swedberg:

Sure. Thank you. It’s a great question. The pilots, man, we’ve all had thousands and thousands of pilots and some of them I prefer to call proof of concept because they’re not actually pilots, right? They’re not even that far along.

I use a brief little acronym as it relates to this. TICT. So first it comes down to the truth, to make sure that we know the truth. It’s got to be based on physics. It’s got to be based on some reality, human-based, that we understand.

Right? It’s about integration. So the power of this is across that full breadth of tools, and making sure, because the true value of these enterprise projects, these enterprise AIs, is not just with its information in one place. Because some of that knowledge and know-how I mentioned earlier, it’s not just did I optimize the design, but it’s to know that when I optimized the design, I screwed up manufacturing 10 years ago, and so I don’t do it this way. That’s why I don’t do it this way. We too often are giving the data and saying, “Let’s clean it all up,” and the reality is what we need to do is help the AI know this is all the data, and so you can learn what worked and what didn’t work, right? So we have to be more open with that.

But there’s also coherence, so the C. Coherence, which is really we have to be looking at compounding, right? So how do you make these things compound? How do they scale up in value? This is to your point, 50x, right? Too often we’re not thinking about how to make it compound, and that’s expanding beyond that use case of the enterprise, right? With the integration I talked about. And finally is trust.

We have to be able to understand how we trust the data. Where is the sovereignty of the data? The security and the access control, the pieces that my two compatriots here have been mentioning exclusively, right? Those fundamental foundations are how we then move out of that.

Because too often when I see these pilots, we are not asking that question that says, can we prove that we trust the data? Do we know the truth of the physics? Right? We faked the data. We didn’t consider the access controls. So everything worked great in its little pod, but the second we put it in the real world where that real data came in, that’s where those fundamentals fall apart.

Jeff Kavanaugh:

Venice, which is the lever to get right first? Is it architecture, data, talent, governance, or maybe the broader operating model in the ecosystem?

Venice Goodwine:

Talent. Talent every day.

Jeff Kavanaugh:

So it’s an HR problem, right?

Venice Goodwine:

It’s an HR problem. Because here is how you know that your pilot is working or your proof of concept. Because someone in that room has domain expertise. It is the cognitive that that human has that they are lending to that use case. That’s important.

We did this with robotic process automation. Right? Before we used any RPA, we had the person, right seat, left seat with the tool and said, “Tell me your process.” And then we trained the RPA to do the process. And then we had gates. The same needs to be true here in AI. When I think about automation, a different type of automation, when I’m using agents or if I’m using agentic AI, it’s still important to have that domain expertise in the room. So that if we don’t tend to the people and tell them, don’t quit, I am not taking your job. I’ve just elevated you to something that is of higher thinking. It’s an Iron Man suit.

Jeff Kavanaugh:

There you go.

Venice Goodwine:

Right. We need to console them and consult them in this space. Do not leave the talent and the people out of the conversation. This is why people leave organizations, because you bring in AI, but you don’t tell them what their role is in AI. There’s a clear RACI model for implementing AI in your organization, and it’s that domain expertise. So for me, it would be the talent.

Jeff Kavanaugh:

All right, one vote for talent. What do you got, Erik?

Erik Swedberg:

I’m very aligned with that. I would probably just append to that also, though, the discussion we’ve been having about data. Because all the rest of the major topics, the governance and all those things, without that truth and trust on the data itself, and again, remembering it’s not just the good data, it’s also sometimes the bad data, because that’s also helpful to learn from. Right? But that is where most of the fundamentals come from, from my point of view.

Venice Goodwine:

What do you think is the biggest gap? How many of you are normally looking at resumes, getting ready to hire people, and interview them? Right? What do you do? You have resumes, you assemble a panel, they all go through the resumes, they choose people, rack and stack, and then you have a different panel that does the interview, and then you have a different panel. I have 68 resumes, and I was able to dwindle it down to 23, and then down to 12, and then the 6 that I wanted to interview, and then the 3 after that. I did that with AI.

Now, why was I able to do that? Because I have domain expertise, and I knew exactly what I was looking for, for the job I was filling. But now I did that in parallel with having us do our normal process. Have six people sitting on a panel looking at twelve resumes. You do the math. How much time is that? And then six people conducting the interviews. How much time is that? I did it in eight minutes in AI and got to the same results, the same one to twelve list. The same one to twelve list.

That's great. And you would think that's good. But here is what I want to caution us, is the bias. Making sure that there was no bias in the tool that weeded out certain candidates for some reason. The school they went to, the name indicated something. So I still need that human in the loop.

And so what stops us from scaling is, one, I did an exercise that worked for me, but if I don’t share it with others, it’s just my own silo exercise. Now I just saved time, but everyone in my organization is still doing it the old-fashioned way. We don’t scale because we don’t have a use case catalog that everyone in the organization could use. Everybody starts the problem from their vantage point. No, go to the catalog and see if that problem’s been solved, and maybe you can build on that prompt and make it work for you. We’re using these tools for our own individual productivity as opposed to productivity from an enterprise perspective, in my opinion, sometimes.

Jeff Kavanaugh:

All right. Well, that’s kind of the how do you avoid the pitfalls? Let’s end on a lightning round of a positive note. In the next 12 to 18 months, what is the one thing that’s going to distinguish the successful A&D companies that do scale and make it? You first, Venice.

Venice Goodwine:

Me. I think you know I’m going to say the same thing. The right people, in the right places, doing the right jobs with the right skills, levering the tools in the right way. If we can do those things, I think that’s going to be important, and I think we’re going to be successful.

Because remember this thing called the internet that started, and no one said they were going to use it? Remember that thing called the mobile phone, and no one wanted to use it because we wanted landlines because we trusted it? How many of you leave home and turn around to go back and get your phone if you leave it? AI is that for us today. It is everywhere. It is not going anywhere. We must embrace it.

So we need to tell people how to do that, and we need to prepare them, train them, and educate them, and show them the wins, but also make sure they understand the precautions around it. So to be successful in the 12 to 18 months, we’re going to have to do all those things.

Jeff Kavanaugh:

I’m here for it. What about you?

Erik Swedberg:

So there was a study by MIT that in the next 18 months or so, so the end of 2027 or so, that all of the learnable knowledge will be scanned and learned for all the current AI tools. So we’re going to hit a peak of that learning. Right?

And so the companies that are going to be, I think the real movers in 18 months from now, are really going to be the ones who are looking to expand and explode on that. So look for the value, not the volume. So it’s not the number of people. And if you go look, Amazon and Microsoft just communicated that they stopped their ambition to try and drive token usage, right? That they’re not driving that anymore because the usage of the AI for the purposes that Venice described is exactly right to learn about how to learn, not to do AI for AI. But the real piece there, and I encourage you all.

So Dassault Systèmes, we are obviously heavily in aerospace and defense, but we’re also in 11 other industries. And from my perspective, what I get to see and what I really enjoy about my job is that there are things that aerospace and defense does amazingly well, and there are things that aerospace and defense has absolutely no idea about. There is a lot more product sensitivity and opportunity going on at Procter & Gamble. The reason that their profit went up when everyone else’s went down through COVID is because they knew immediately how to shift their product. They weren’t designing it. It was already done. The design is already done for things that might happen there.

And so if you look at that convergence now of all of the knowledge being learned, if you’re leveraging AI in 18 months from now to do the things that you do today, you’re behind. Because the ones that are going to really move are the ones that are going, “Who’s doing this the best, this process? Because if I think about the way I think about it today, I might be thinking about it the wrong way.” It’s why Boeing is trying to make airplanes like Toyota.

Jeff Kavanaugh:

So, it’s partnerships, which is exactly that.

Erik Swedberg:

I couldn’t agree more. I’ve had customers for 40 years. Some of the dirty data is my fault. The interesting that’s coming out though, we have a view and a vision. We’ve just released the communications about our Gen7 capabilities. I encourage you guys to dig a little deeper into that because one of the things that’s really come out of that is we define the different generations from a Dassault Systèmes perspective. What we see as the major milestones. These aren’t versions of software. Okay? Version does create some of its own data challenges, right?

But generation talks about digital mockup, so I had to have the ability to have all the 3D in one space and know it accurately. When I move on to lifecycle management, I needed to know data accurately, but I also needed to know its time and space, right? Was I done with the design? Where is it in time and space, right? If I try to take the data that I stored and I did a digital mock-up of, say, the 777, and I try and compare that to 787, where I did it with full lifecycle management, I don’t have some of the data I need because I stored latest greatest. I don’t actually have any idea of how I got there. And so we consider some of that dirty data.

It’s not dirty. And when you cut through that and you say, “Okay, actually, I’m okay with this data. I only knew this much.” It’s not dirty, but now I have to take and apply learnings to it differently. When you apply that, it actually cleans up a significant amount of what we today have been calling dirty, and that’s massively accelerated many of my customers’ deployments in terms of the latest tech capabilities.

But it’s also set the stage with the engineering departments and the manufacturing departments to say, “Look, guys, if you want to go back and do all of the learning and knowledge about the 777 or the 47, then great. But if you’re not really going to get the value out of it, then let’s just leave it the way it is.” It’s not dirty. It’s just that’s as much as we’re going to get out of it. So there's some strategies there that are coming out, and AI is going to help us massively in this space.

It’s probably one of the bigger growing use cases because that’s the next thing for it to do is to learn how to clean up or how to tell us what generation the data is, and you can only do this. You can't do that, right?

Jeff Kavanaugh:

Everybody, round of applause for our gang.