The following is a conversation with Kiyoki Jackson.

He's the CTO of Lockheed Martin, a company that through its long history has created some of the most incredible engineering marvels human beings have ever built, including planes that fly fast and undetected, defense systems that intersect nuclear threats that can take the lives of millions, and systems that venture out into space, the moon, Mars, and beyond.

And these days, more and more, artificial intelligence has an assistive role to play in these systems.

I've read several books in preparation for this conversation.

It is a difficult 1 because in part Lockheed Martin builds military systems that operate in a complicated world that often does not have easy solutions in the gray area between good and evil.

I hope 1 day this world will rid itself of war in all its forms.

But the path to achieving that in a world that does have evil is not obvious.

What is obvious is good engineering and artificial intelligence research has a role to play on the side of good.

Lockheed Martin and the rest of our community are hard at work at exactly this task.

We talk about these and other important topics in this conversation.

Also, most certainly, both Kiyoki and I have a passion for space.

Us, humans, venturing out toward the stars.

We talk about this exciting future as well.

This is the Artificial Intelligence Podcast.

If you enjoy it, subscribe on YouTube, give it 5 stars on iTunes, support it on Patreon, or simply connect with me on Twitter at Lex Friedman, spelled F-R-I-D-M-A-N.

And now, here's my conversation with Keoki Jackson.

I read several books on Lockheed Martin recently.

My favorite in particular is by Ben Rich, Carlos Conkirk's personal memoir.

But from that I was reminded that the engineers of Lockheed Martin have created some of the most incredible engineering marvels human beings have ever built throughout the 20th century and the 21st.

Do you remember a particular project or system at Lockheed or before that at the Space Shuttle Columbia that you were just in awe at the fact that us humans could create something like this?

There's a lot of things that I could draw in there.

When you look at the Skunk Works and Ben Rich's book in particular, of course, it starts off with basically the start of the jet age and the P-80.

I had the opportunity to sit next to 1 of the Apollo astronauts, Charlie Duke, recently at dinner.

And I said, hey, what's your favorite aircraft?

And he said, well, it was by far the F-104 Starfighter, which was another aircraft that came out of Lockheed there.

It was the first Mach 2 jet fighter aircraft.

They called it the missile with a man in it.

And so those are the kinds of things that I grew up hearing stories about.

Of course, the SR-71 is incomparable as the epitome of speed, altitude, and just the coolest looking aircraft ever.

that- That's a, yeah, intelligence, surveillance, and reconnaissance aircraft that was designed to be able to outrun, basically go faster than any air defense system.

That's really what took me ultimately to Lockheed Martin.

And so Lockheed Martin, for example, has been essentially at the heart of every planetary mission like all the Mars missions we've had a part in.

And We've talked a lot about the 50th anniversary of Apollo here in the last couple of weeks, right?

But remember 1976, July 20th, again, National Space Day.

So the landing of the Viking, the Viking lander on the surface of Mars, Just a huge accomplishment.

And when I was a young engineer at Lockheed Martin, I got to meet engineers who had designed various pieces of that mission as well.

So that's what I grew up on, is these planetary missions, the start of the space shuttle era, and ultimately had the opportunity to see Lockheed Martin's part, and we can maybe talk about some of these here, but Lockheed Martin's part in all of these space journeys over the years.

Do you dream, and I apologize for getting philosophical at times, or sentimental.

I do romanticize the notion of space exploration.

Do you dream of the day when us humans colonize another planet like Mars, or a man, a woman, a human being steps on Mars?

I haven't given up yet on my own opportunity to fly into space.

But from the Lockheed Martin perspective, this is something that we're working towards every day.

And of course, we're building the Orion spacecraft, which is the most sophisticated human rated spacecraft ever built.

And it's really designed for these deep space journeys, starting with the Moon, but ultimately going to Mars and being the platform from a design perspective, what we call the Mars Base Camp, to be able to take humans to the surface and then after a mission of a couple of weeks, bring them back up safely.

And so that is something I want to see happen during my time at Lockheed Martin.

And I think once we prove that's possible, colonization might be a little bit further out, but it's something that I'd hope to see.

So maybe you can give a little bit of an overview of, so Lockheed Martin has partnered with a few years ago with Boeing to work with the DOD and NASA to build launch systems and rockets with the ULA.

What's Lockheed's mission timeline and long-term dream in terms of space?

What's the engineering challenges and what's the dream long-term?

Yeah, I think the dream long-term is to have a permanent presence in space beyond low Earth orbit, ultimately with a long-term presence on the Moon and then to the planets, to Mars.

Sustained and Sustainable presence in an economy, a space economy that really goes alongside

With human beings being and being able to launch perhaps from those, so like hop.

a lot of energy that goes in those hops, right?

So I think the first step is being able to get there and to be able to establish sustained bases and build from there.

And a lot of that means getting, as you know, things like the cost of launch down.

And you mentioned United Launch Alliance.

And so I don't want to speak for ULA, but obviously they're working really hard to, on their next generation of launch vehicles to maintain that incredible mission success record that ULA has, but ultimately continue to drive down the cost and make the flexibility, the speed and the access ever greater.

I mean, I think you know this, or you may know this, you know, there's a lot of ways to accomplish some of these goals.

And so that's a lot of what's in discussion today.

But ultimately, the goal is to be able to establish a base, essentially in cislunar space, that would allow for ready transfer from orbit to the lunar surface and back again.

So that's sort of that near term, I say near term in the next decade or so vision.

Starting off with a stated objective by this administration to get back to the Moon in the 2024-2025 time frame, which is right around the corner here.

So- I'll give an engineering challenge is that.

I think the big challenge is not so much to go, but to stay, right?

And so we demonstrated in the 60s that you could send somebody up, do a couple of days of mission and bring them home again successfully.

Now, we're talking about doing that, I'd say, more to, I don't want to say an industrial scale, but a sustained scale, right?

So, permanent habitation, you know, regular reuse of vehicles, the infrastructure to get things like fuel, air, consumables, replacement parts, all the things that you need to sustain that kind of infrastructure.

So those are certainly engineering challenges.

And those are all things that we're going to have to work through.

You know, the other thing, and I shouldn't, I don't want to minimize this.

I mean, I'm excited about human exploration, but the reality is our technology and where we've come over the last 40 years essentially has changed what we can do with robotic exploration as well.

This seems like old news now, but the fact that we have rovers driving around the surface of Mars and sending back data is just incredible.

The fact that we have satellites in orbit around Mars that are collecting weather, you know, they're looking at the terrain, they're mapping, all these kinds of things on a continuous basis, that's incredible.

And the fact that, you know, you got the time lag, of course, going to the planets, but you can effectively have virtual human presence there in a way that we have never been able to do before.

And now with the advent of even greater processing power, better AI systems, better cognitive systems and decision systems.

You put that together with the human piece and we really opened up the solar system in a whole different way.

We've got OSIRIS-REx, which is a mission to the asteroid Bennu.

So the spacecraft is out there right now on basically a year mapping activity to map the entire surface of that asteroid in great detail.

But the idea then that, and this is not too far away, it's going to go in, it's got a sort of a fancy vacuum cleaner with a bucket, it's going to collect the sample off the asteroid and then send it back here to earth.

And So, we have gone from sort of those tentative steps in the 70s, early landings, video of the solar system to now we've sent spacecraft to Pluto, we have gone to comets and intercepted comets, we've brought stardust material back.

So that's, we've gone far and there's incredible opportunity to go even farther.

quite crazy that this is even possible, that, Can you talk a little bit about?

What it means to orbit an asteroid and with a bucket to try to pick up some soil samples

Yeah, I mean so part of it is just kind of the you know these are the same kinds of techniques we use here on Earth for high speed, high accuracy imagery, stitching these scenes together and creating essentially high accuracy world maps, right?

And so that's what we're doing obviously on a much smaller scale with an asteroid.

But the other thing that's really interesting, you put together sort of that neat control and data and imagery problem.

But the stories around how we designed the collection, I mean, as essentially, this is the sort of the human ingenuity element, right?

That essentially, had an engineer who had a, 1 day he's like, oh, starts messing around with parts, vacuum cleaner, bucket.

And that was what led to what we call the pogo stick collection, right?

Where basically a thing comes down, it's only there for seconds, does that collection, grabs the, essentially blows the regolith material into the collection hopper and off it goes.

So what is in those, we talk a little bit more about space, what's the role of the human in all of this?

What are the challenges, what are the opportunities for humans as they pilot these vehicles in space?

And for humans that may step foot on either the moon or Mars?

Yeah, it's a great question because I just have been extolling the virtues of robotic and rovers, autonomous systems, and those absolutely have a role.

I think the thing that we don't know how to replace today is the ability to adapt on the fly to new information.

And I believe that will come, but we're not there yet.

And so, you know, you think back to Apollo 13 and the ingenuity of the folks on the ground and on the spacecraft to essentially cobble together a way to get the carbon dioxide scrubbers to work.

Those are the kinds of things that ultimately, and I'd say not just from dealing with anomalies, but dealing with new information.

You see something and rather than waiting 20 minutes or half an hour, an hour to try to get information back and forth, but be able to essentially re-vector on the fly, collect different samples, take a different approach, choose different areas to explore.

Those are the kinds of things that that human presence enables that is still a ways ahead of us on the AI side.

Yeah, there's some interesting stuff we'll talk about on the teaming side here on Earth.

And in space, let's not leave the space piece out.

So what is teaming, what is AI and humans working together in space look like?

Yeah, 1 of the things we're working on is a system called Maya, which is, you can think of it, so it's an AI assistant.

And you think of it as the Alexa in space, right?

But this goes hand in hand with a lot of other developments.

And so today's world, everything is essentially model-based, model-based systems engineering to the actual digital tapestry that goes through the design, the build, the manufacture, the testing, and ultimately the sustainment of these systems.

And so our vision is really that when our astronauts are there around Mars, you're going to have that entire digital library of the spacecraft, of its operations, all the test data, all the test data and flight data from previous missions to be able to look and see if there are anomalous conditions and tell the humans and potentially deal with that before it becomes a bad situation and help the astronauts work through those kinds of things.

It's not just dealing with problems as they come up, but also offering up opportunities for additional exploration capability, for example.

So that's the vision, is that these are, you're gonna take the best of the human to respond to changing circumstances and rely on the best of AI capabilities to monitor this almost infinite number of data points and correlations of data points that humans frankly aren't that good at.

So how do you develop systems in space like this?

Whether it's Alexa in space or in general any kind of control systems, any kind of intelligent systems, when you can't really test stuff too much out in space.

Yeah, that's the beauty of this digital twin, if you will.

And of course, with Lockheed Martin, we've over the past 5 plus decades been refining our knowledge of the space environment, of how materials behave, dynamics, the controls, the radiation environments, all of these kinds of things.

So we're able to create very sophisticated models.

They're not perfect, but they're very good.

I spent part of my career simulating communication spacecraft, missile warning spacecraft, GPS spacecraft in all kinds of scenarios and all kinds of environments.

So this is really just taking that to the next level.

The interesting thing is that now you're bringing into that loop a system, depending on how it's developed, that may be non-deterministic, it may be learning as it goes.

In fact, we anticipate that it will be learning as it goes.

And so that brings a whole new level of interest, I guess, into how do you do verification and validation of these non-deterministic learning systems in scenarios that may go out of the bounds or the envelope that you have initially designed them to.

So it has this system and its intelligence has the same complexity, some of the same complexity a human does.

It learns over time, it's unpredictable in certain kinds of ways.

So you also have to model that when you're thinking about it.

So in your thoughts, it's possible to model the majority of situations, the important aspects of situations here on Earth and in space enough to test stuff?

Yeah, this is really an active area of research and we're actually funding university research in a variety of places, including MIT.

This is in the realm of trust and verification and validation of, I'd say, autonomous systems in general and then as a subset of that autonomous systems that incorporate artificial intelligence capabilities.

We're working with startup companies, we've got internal R&D, but our conviction is that autonomy and more and more AI enabled autonomy is going to be in everything that Lockheed Martin develops and fields.

And it's going to be retrofit, autonomy and AI are going to be retrofit into existing systems.

They're going to be part of the design for all of our future systems.

And so maybe I should take a step back and say the way we define autonomy.

So we talk about autonomy essentially a system that composes, selects, and then executes decisions with varying levels of human intervention.

So this is essentially the human doing the task.

You can think of effectively partial autonomy where the human is in the loop.

So making decisions in every case about what the autonomous system can do.

Or remotely, exactly, but still in that control loop.

And then there's what you'd call supervisory autonomy.

So the autonomous system is doing most of the work.

The human can intervene to stop it or to change the direction.

And then ultimately, full autonomy where the human is off the loop altogether.

And for different types of missions want to have different levels of autonomy.

So now take that spectrum and this conviction that autonomy and more and more AI are in everything that we develop.

The kinds of things that Lockheed Martin does, a lot of times are safety of life critical kinds of missions.

And so we require and our customers require an extremely high level of confidence.

2, that we're going to, that these systems will behave in ways that their operators can understand.

And so this gets into that whole field again, you know, being able to verify and validate that the systems have been, you know, that they will operate the way they're designed and the way they're expected.

And furthermore, that they will do that in ways that can be explained and understood.

And that is an extremely difficult challenge.

I don't mean to bring this up, but I think it's a good case study that people are familiar with.

Boeing 737 Max commercial airplane has had 2 recent crashes where their flight control software system failed.

And it's software, So I don't mean to speak about Boeing, but broadly speaking, we have this in the autonomous vehicle space too, semi-autonomous.

When you have millions of lines of code, software, making decisions, there is a little bit of a clash of cultures because software engineers don't have the same culture of safety often that people who build systems like Lockheed Martin do where it has to be exceptionally safe you have to test this on.

So how do we get this right when software is making so many decisions?

Yeah, and there's a lot of things that have to happen.

And by and large, I think it starts with the culture, Which is not necessarily something that A, is taught in school or B, is something that would come, you know, depending on what kind of software you're developing, it may not be relevant, right?

If you're targeting ads or something like that.

So, and by and large, I'd say not just Lockheed Martin, but certainly the aerospace industry as a whole has developed a culture that does focus on safety, safety of life, operational safety, mission success.

But as you note, these systems have gotten incredibly complex.

And so they're to the point where it's almost impossible, the state space has become so huge that it's impossible to, or very difficult to do a systematic verification across the entire set of potential ways that an aircraft could be flown, all the conditions that could happen, all the potential failure scenarios.

Now maybe that's soluble 1 day, maybe when we have our quantum computers at our fingertips, we'll be able to actually simulate across an entire, almost infinite state space.

But today, there's a lot of work to really try to bound the system, to make sure that it behaves in predictable ways, and then have this culture of continuous inquiry and skepticism and questioning to say, did we really consider the right realm of possibilities?

Do we really understand, you know, in this case, you know, human and machine interactions, the human decision process alongside the machine processes.

And so that's that culture, we call it the culture of mission success at Lockheed Martin that really needs to be established.

And it's not something, you know, it's something that people learn by living in it.

And it's something that has to be promulgated, you know, and it's done, you know, from the highest levels at a company of Lockheed Martin, like Lockheed Martin.

Yeah, and the same is being faced at certain autonomous vehicle companies where that culture is not there because it started mostly by software engineers.

Is there lessons that you think we should learn as an industry and a society from the Boeing 737 MAX crashes?

These crashes obviously are either tremendous tragedies, they're tragedies for all of the people, the crew, the families, the passengers, the people on the ground involved.

And It's also a huge business and economic setback as well.

I mean, we've seen it's impacting essentially the trade balance of the US.

And these are the kinds of, We've seen similar kinds of questioning at times.

And it is, I think, always important to remind ourselves that humans are fallible, that the systems we create, as perfect as we strive to make them, we can always make them better.

And so another element of that culture of mission success is really that commitment to continuous improvement.

If there's something that goes wrong, a real commitment to root cause and true root cause understanding to taking the corrective actions and to making the system, future systems better.

And certainly we want to, we strive for, you know, no accidents.

And if you look at the record of the commercial airline industry as a whole and the commercial aircraft industry as a whole, there's a very nice decaying exponential to years now where we have no commercial aircraft accidents at all, right?

It was through the regulatory agencies, FAA, the airframe manufacturers, really working on a system to identify root causes and drive them out.

So maybe we can take a step back and many people are familiar, but Lockheed Martin broadly, what kind of categories of systems are you involved in building?

Lockheed Martin, we think of ourselves as a company that solves hard mission problems.

And the output of that might be an airplane or a spacecraft or a helicopter or a radar or something like that.

But ultimately, we're driven by these, what is our customer?

What is that mission that they need to achieve?

And so that's what drove the SR-71, right?

How do you get pictures of a place where you've got sophisticated air defense systems that are capable of handling any aircraft that was out there at the time.

And make sure it gets out and it gets back.

And that led ultimately to really the start of the space program in the US as well.

So now take a step back to Lockheed Martin of today.

And we are on the order of 105 years old now between Lockheed and Martin, the 2 big heritage companies, which were made up of a whole bunch of other companies that came in as well.

General Dynamics, kind of go down the list.

Today you can think of us in this space of solving mission problems.

So obviously on the aircraft side, tactical aircraft, building the most advanced fighter aircraft that the world has ever seen.

We're up to now several hundred of those delivered, building almost a hundred a year.

And of course, working on the things that come after that.

On the space side, we are engaged in pretty much every venue of space utilization and exploration you can imagine.

So I mentioned things like navigation, timing, GPS, communication satellites, missile warning satellites.

We've built commercial surveillance satellites.

We've built commercial communication satellites.

So everything from human exploration to the robotic exploration of the outer planets and keep going on the space front.

But a couple of other areas I'd like to put out, we're heavily engaged in building critical defensive systems.

And so a couple that I'll mention, the Aegis Combat System, this is basically the integrated air and missile defense system for the US and allied fleets.

And so protects carrier strike groups, for example, from incoming ballistic missile threats, aircraft threats, cruise missile threats, and kind of go down the list.

See all Lex Fridman transcripts on Youtube

Keoki Jackson: Lockheed Martin | Lex Fridman Podcast #33