Hi, I'm Paul Scharr, Director of the Technology and National Security Program here at C&S. Here for another issue of our podcast series on Artificial Intelligence. Today, I'm joined by Captain Mike Kanin from the U.S. Air Force.
Welcome. Thanks, Paul. I appreciate it. Thanks.
So, you know, let's talk about this. I'm going to ask you to say your job title because it's kind of a mouthful. Can you go ahead and explain for us what you do for the Air Force? Yeah, that's kind of hard for the course at times for us in government, but two functions.
It's on the core team of the Algorithmic Warfare Cross-Functional Team, or Project Maven, as it's better described or more simply. And this is a team that's there to establish a pathfinder for artificial intelligence machine learning, integrate with industry and academia, and at the core, turn our enormous volume of data into action at the speed of relevance. In addition to that, I'm under the Director of Air Force Intelligence as the Air Force ISR lead for artificial intelligence machine learning. But more exciting on this topic is I'm a 20-year-old captain, right, and a millennial, and in par with our messaging from the U.S.
Air Force's Unleashed Airman. I think that's compelling and it's exciting for us. I'm glad to have you here on our podcast, Unleashed. So let's start with what does the Air Force do with artificial intelligence and machine learning?
So I think that we should start with why we're doing things at the core of it all. And what we realize is there's just simply no such thing as sustained comparative advantage. And in this digital age, the acquisition or more traditional acquisition or approach to more things, more sensors, just simply won't continue to act as the answer to a problem in which the character is fundamentally changed over the course of the past decade. Traditional visions of supply and demand just won't cut it.
And why is that the case? And we call this the infinite dilemma that we're in. And I can pull a peel down on your back for a minute. We break this down.
We live in an infinite world. Infinitely growing situations, infinitely growing capabilities, infinitely growing amounts of data, and through that, an infinite number of possibilities at the core, theoretically. And we have to meet that with finite resources over a finite time, doing a finite prioritized response. And even if we did live in a world where supply could equal the demand, it's unlikely that we could afford a sufficient number of experts to take that on anyways.
So what we realize behind machine learning and the why is, is the race is on for this. There's only a small window of second place. Who's second place to Google? Who's second place to Amazon?
Who's second place to Apple? They don't exist. So we want to do this at speed and scale because it's a necessity to us. You said your task is to move on at the speed of relevance.
The D of the acquisition system is rarely described as swift. So how's this progress working out so far from incorporating AI and machine learning into the workforce? So obviously there's some certain aspects that I can't speak upon because of operational security and what we're doing. But I think it starts with our culture and how we're starting to think about artificial intelligence and machine learning and taking it out from this esoteric research and development world into something more of a discussion about fielding.
And that's exciting. But I want to talk about our culture a little bit from the eyes of those great competitors we have out there. So let's talk about 1997. That moment when the Russians learned a subtle but vast implications of machine intelligence writ large when Garry Kasparov lost IBM's Deep Blue.
Before then, it was widely accepted that not some cold and calculated machine could beat the nuanced imagination of Garry Kasparov, one of the world's best. And when they did lose the machine, they recognized the significance behind that. Let's fast forward 20 years later from there in 2016. In an even more rocking moment of cultural recognition, that DeepMinds AlphaGo dismantled the world's best in the game of Go.
And that's a 2,500-year-old cultural pillar that fell, and it disappeared. And they realized that. And as an aside, we can all remember there was a mark of Western society in each of those moments with IBM and American flag in the corner of the screen that they were playing the game of chess on. And furthermore, in the corner of the screen with AlphaGo, there was British flag.
Well, we know DeepMind, the company that created AlphaGo, is now an American company. So what we've learned, and this is all about culture here, is that the organizations that can harness and wield machine intelligence do so because it's a natural reflection of their border and how they see things. We want to build a learning culture on this topic. Nothing quite spurs learning like pain and necessity.
Pain in two ways. Number one, it's something that brings you to your knees. Or the way in which we are approaching it, is that our airmen want to pain forward so badly to get to this space where we're appropriately wielding artificial intelligence, machine learning, that they want to fail forward, that they believe again. Because the necessity, frankly, is there as you look around everything that's happening in industry and government.
And we're making strides in the right direction. We have a vision. We have appropriate messaging. We're working on our understanding of what machine intelligence is, what it isn't, what it should be, what it shouldn't be.
We're implementing it, and we're iterating upon it. That's different than sustainment. We're working to make it better all the time. Because this is all about new with new.
And that's the way that we're looking at it. New approaches, new ideas, with new implementation upon it. Because those who don't in this world, whether it's a government or military or an academics institution or a corporate entity, will see to the high ground. And it's important to us to generate cultural urgency on it.
And I can sum it up in why this is a nuanced approach simply with this, this millennial sentiment. You're an apple. I'm a droid. We have to get past that and start sharing the data in which we have to do machine intelligence appropriately.
So what applications of AI machine learning for Air Force patients and priorities are you most excited about? So I think we can break this down into the near term, midterm, and long term. Let's start with cloud. And frankly, when it released to machine learning, I think the first question is to always ask, well, what are you trying to solve for?
And not doing AI for the sake of AI. But near term cloud. And by cloud, I mean the thoughtful and purposeful integration of our data, whether that's working on infrastructure as a service, platforms as a service, or airmen can develop software as a service, and making our core competency to connect and share with others in ways we've never done before. Now, for Air Force intelligence, I can break it down pretty simply.
Collection. What I mean by collection is we put our assets in the right place at the right time to provide an outcome to a decision maker. This thing we also do called processing, exploitation, and dissemination. How we observe those assets.
Make sense of the situation to, again, provide an outcome or an assessment of the situation on the ground in the air subsurface, whatever that may be. Analysis. When we look back upon what we've done to say, well, what are these overarching trends? What are we noticing so we can be a little more predictive in nature?
And then assessment. To look back and say, well, how would we perform across the entirety of the spectrum? Because we're going to use it to generate kinetic and non-kinetic effects. That's the bottom line to our business.
As for the Air Force's strategic implications, I think we can look at industry and see how they've been affected, because we do many of the same things. And we can break that out. We make sure we're putting the right people to the right jobs. That's perfect for AI.
That our operations are within scope, and they're effective. Perfect for AI again. We can have a profound impact on the way that we move our material and manpower. We can measure budgetary processes with AI in ways that we've never did before.
So I don't think there's any aspect of our operations that AI shouldn't be woken into the fabric of. But uniquely for AI, we'll discover new insights, discover brand new outcomes to accomplish our goals. And that's a shift for us. That's from the days of identifying outcomes from what is essentially a very specific point of view and experience.
And now we can look across the enterprise, and as our chief staff of the Air Force in our goal, it means us best, so we can connect, share, and learn. So you mentioned processing, exploitation, and dissemination, or DOD-speak pay. This is something that some of our officers may be familiar with, but maybe not all of them, but a big deal for the Air Force. So this will certainly be one of the idea that the Air Force flies planes in the sky.
That's their job. Many of these planes nowadays are unmanned or uninhabited or drones, or they're, for the term, remotely piloted. What folks may not be aware of is that for the Air Force, one of the biggest personnel burdens is not flying the airplanes, but it's managing all the data that comes off of these airplanes, doing this processing, exploitation, and dissemination. Let me talk a little bit about how the Air Force is seeing AI as a tool to help manage this personnel burden in terms of doing the PED for all of this data collection.
That's a great question and a great point to bring up. We will work to keep a human on the loop, and that's our messaging. That's to observe and to help us along making decisions and acting upon the information which we have. But what we need to do better, and I think that every industry or entity in the world, is to get our analysts out of the business of what and into the business of why.
Let me explain that for a moment. We collect a lot of things in many unique ways. I think that all of the listeners can say, well, we look with videos from the sky. What we want to make sure that we do exceptionally well is not count things coming and going, but asking our analysts why are those things there.
When we're listening, we want to make sure that they aren't identifying the what, but they're being provided the time and space to tell me about the why. To that end, that's how we want to use AI machine learning to generate insights that otherwise took the time of human. And we need to break that down into nanoseconds of the speed and scale of what the globe currently looks at and what more will look like in the future. So that really seems like an example of this kind of blended use of AI and humans that makes an awful lot of sense.
And we've seen, of course, AI and machine learning systems being able to do object recognition and beat humans at benchmark tests. So using them to identify objects in these video feeds or other data feeds makes a lot of sense. But one of the vulnerabilities we've also seen is that many of these current machine learning systems are vulnerable to adversarial data inputs. People can send in particularly manipulated digital signals, whether they're physical objects or images or other signals, maybe for auditory recognition systems to be sound, that would come in and smooth the system.
And AI systems with a high degree of confidence will misbelieve in image. How would you guys deal with that problem? Because obviously, the military environment is an adversarial one. We have to contend with people trying to manipulate and trick your systems.
This isn't something that goes lost upon us. And this idea of relying upon an algorithm or a computer, whether our adversaries do it or those competitors go after that idea, or whether we do, it's going to provide opportunities, it's going to provide challenges that we have to meet with an intellectual foundation on what AI is, what it isn't, where it goes, where should it, and where's the appropriately place to put the human. The only way that we're going to combat the vast amount of data that is to come, that is coming in information, is to make sure we respond with commensurate technologies, the same technologies that generate these things, graphical processor units, neural networks that monitor whether something was computer generated, whether it wasn't. So the way that we think about that is we can't rely solely on our analytic backbone to be overwhelmed one day.
So we have to place it at the appropriate edges where it can sift through and find patterns of speed and scale that we were probably going to find out anyways. But to the specific problem of some of these manipulated images or spoofing attacks, how are you thinking about the vulnerability of your AI systems to be tricked? So there are certain aspects that we could keep close held on how we approach this. But for what it's worth, what's so important behind AI is it gives us an opportunity to academically exercise some things, to act as our best selves.
So when we go off and we go to train data and provide labeled data sets from the foremost experts in the world, U.S. airmen, at whatever that mission is, to generate training data sets, it's an opportunity for us to act as our best selves. So we're excited by those things. And we realize that we're going to face those same manipulations just as any competitor will face the same manipulations from us.
So it's an approach we take. I'm excited that our airmen have so much buy-in to make sure whatever we are creating, whatever neural network we are training or platform we're operating on, that is a representation of their best selves. What about your ability within the Air Force to tap into where much of the innovation is happening, which is outside of traditional defense companies? So the new national defense strategy has, I think, a notable shift in terms of how it talks about industry, talks about a national security innovation base rather than a defense industrial base.
We'll sort of broaden the aperture for companies that we might think about due to needing to tap into and draw upon their expertise. But there can be a lot of red tape in the way. So how's that working in terms of actually reaching out to companies that are doing some of this really groundbreaking work in AI? I can tell you that I've had tremendous interactions with industry, and those who are in this line of work are also having those same interactions.
So personally, I've spent a significant part of this past year in Silicon Valley engaging with all the companies in which one would imagine. Because for the U.S. Air Force and for the department writ large, we offer something very cool. We offer a purpose.
And that purpose is to work with the best people on the hardest problems, growing the best technologies for the greater good. We're finding a tremendous amount of excitement with these nontraditional companies that are identifying with our airmen, our best connecting with their best, because they come from the same backgrounds, largely, working on hard problems together. And it's an exciting prospect. And for what it's worth, that triangular relationship between industry, academia, the military, or government has delivered some transcendent moments.
And I think we're in the same place right now. Well, this is good. I'm glad your force is unleashing you and many others on this problem. Let me just close with asking you, as you're thinking about kind of the space, and it's very dynamic and quickly, what kinds of things keep you up at night?
So, Paul, I can tell you this. I sleep well. Our servicemen and women are the best in the world, demonstrably so, and have done so for a long time. They were born in innovation.
The best part is there's genuine momentum behind unleashing them, with the best technologies that we're talking about right now, working with the best people in industry and academia on the most meaningful problems, for the good of our citizens and for our allies. We focus on win. Our airmen want to win. For what it's worth, losers focus on winners.
So our airmen are all in. If America or our vital national interests are threatened at all, we'll fly and fight. Our top priority is to be ready at any time. And with this vision, and especially with these technologies, it only makes them more lethal than ever before.
All right, well, thanks very much for joining me today and for sharing what the air force to do in AI. Thank you.