Lockheed Martin's CTO on Autonomy, the Moon, and the Culture of Not Getting It Wrong

Lex Fridman sits down with Keoki Jackson, Lockheed Martin's chief technology officer, after reading Ben Rich's *Skunk Works* and a stack of other books on a company that has spent a century building machines that fly fast, hide from radar, intercept missiles, and leave Earth. Fridman is frank about the gray area: Lockheed builds military systems that operate in a world without easy answers between good and evil. Jackson does not open with a product catalog. He reaches for the stories that made him want the job.

He had recently sat next to Apollo astronaut Charlie Duke at dinner and asked for a favorite aircraft. Duke named the F-104 Starfighter, Lockheed's first Mach 2 jet fighter, nicknamed "the missile with a man in it." The SR-71 still sits in a class of its own for speed, altitude, and looks: an intelligence aircraft designed to outrun any air-defense system of its day. Jackson, though, calls himself a space junkie. That is what took him to MIT and then to Lockheed Martin, a company that has had a hand in essentially every planetary mission, including the Mars flights.

The date he remembers is 1976, July 20, when the Viking lander reached the surface of Mars. As a young Lockheed engineer he got to meet people who had designed pieces of that mission. Asked whether he dreams of a human stepping onto Mars, he does not hedge. "I haven't given up yet on my own opportunity to fly into space." The company is building Orion, which he calls the most sophisticated human-rated spacecraft ever built, aimed first at the Moon and then at Mars. Internally there is a design called Mars Base Camp: a mothership in Martian orbit that would take people to the surface for a couple of weeks and bring them back up safely. That, he says, is something he wants to see happen during his time at Lockheed Martin.

The long-term dream, as Jackson frames it, is a permanent presence beyond low Earth orbit: a sustained, sustainable foothold on the Moon, then the planets. Fridman presses on timeline. Near term, the picture is a base in cislunar space that lets you transfer from orbit to the lunar surface and back, with an administration goal of returning people to the Moon around 2024 or 2025. How big an engineering challenge is that? "The big challenge is not so much to go, but to stay." The 1960s showed you could send someone up for a couple of days and bring them home. What is on the table now is permanent habitation, regular reuse of vehicles, and the infrastructure for fuel, air, consumables, and spare parts. There are budgetary challenges as well as technical ones, and he does not pretend otherwise.

He does not want human exploration to crowd out what robots have already changed. Rovers driving on Mars and sending data back still strike him as incredible, even if they sound like old news. Satellites in Martian orbit map terrain and weather continuously. Time lag is real, but processing power, better AI, and better decision systems add up to a kind of virtual human presence that did not exist before. Put that together with people on the scene, he says, and the solar system opens in a different way.

The example he reaches for is OSIRIS-REx at the near-Earth asteroid Bennu. The spacecraft had already spent about a year mapping the entire surface in detail, flying itself. Next would come a short collection: what he describes as a fancy vacuum cleaner with a bucket, in contact for only about four seconds, blowing regolith into a hopper and leaving. The sampler, nicknamed a pogo-stick collection, started as an engineer messing around with vacuum-cleaner parts and a bucket. From tentative 1970s landings, Jackson traces a line through Pluto flybys, comet intercepts, and sample return. The ingenuity he wants to underline is human, not algorithmic.

If robots are this capable, why send people at all? Jackson's answer is specific. What we do not know how to replace today is the ability to adapt on the fly to new information. Apollo 13 is the canonical case: people on the ground and on the spacecraft cobbled together a way to make the carbon dioxide scrubbers work. There was no procedure for that. It was invented in the moment. The second argument is time scale. Mars communications can lag by twenty minutes, half an hour, or an hour. If a probe sees something interesting, the window may be gone by the time a command comes back. A person can re-vector immediately, take a different sample, choose a different patch of ground. Those, he says, are still well ahead of what machines can do.

The teaming picture in space is a system he calls Mia, which he describes as Alexa in space. It sits on top of model-based systems engineering and a digital tapestry that runs from design through build, test, and sustainment. When astronauts are around Mars, the vision is that they have the entire digital library of the spacecraft: design data, test data, flight data from earlier missions. The assistant watches a near-infinite set of parameters and correlations that humans are frankly not that good at, flags anomalous conditions before they become a bad day, and helps the crew work through them. It is not only a trouble-shooter. It is also supposed to offer unasked-for exploration opportunities.

You cannot flight-test that cheaply on orbit, so the development method is the digital twin. Lockheed has spent five-plus decades refining models of the space environment, materials, dynamics and control, and radiation. They are not perfect, Jackson says, but they are very good, and he spent early years of his own career simulating communications satellites and missile-warning spacecraft in every kind of scenario. The new wrinkle is that the thing you drop into that loop may be non-deterministic. It may learn as it goes. That raises verification and validation to a different height: how do you prove a learning system will still do the right thing in scenarios that wander outside the original envelope?

Jackson treats that V&V question as an active research program: university work including MIT, startups, and internal R&D. The conviction underneath it is that autonomy, more and more AI-enabled, will be in everything Lockheed develops. It will be retrofitted onto existing systems and designed into future ones. The working definition is a system that composes, selects, and executes decisions with varying levels of human intervention. No autonomy means the human does the task. Partial autonomy keeps a person in the loop for every decision, in the cockpit or remote. Supervisory autonomy has the system doing most of the work, with a human who can stop it or change direction. Full autonomy takes the human off the loop. Different missions want different rungs, and a lot of Lockheed's work is safety-of-life. Customers demand extremely high confidence that the system will behave in ways operators can understand and explain. That, he says without varnish, is an extremely difficult challenge.

Fridman brings up the Boeing 737 Max crashes, in which flight-control software was implicated, and the clash of cultures in a world of millions of lines of code. Jackson does not defend Boeing and does not dodge. It starts with culture, which is not something AI is taught in school and is not even relevant if you are targeting ads. Aerospace as a whole has spent decades centering safety of life, operational safety, and mission success. Modern systems have also become so complex that the state space is almost impossible to verify systematically across every way an aircraft might be flown, every condition, every failure combination. Quantum computers might one day simulate an almost infinite state space. Today you bound the system so it behaves in predictable ways, then you live inside a culture of continuous inquiry and skepticism: did we consider the right possibilities, did we test enough, do we understand the human-machine decision process?

Lockheed calls that a culture of mission success. People learn it by living in it, and it has to be pushed from the highest levels of the company. The Max crashes were tragedies for crews, families, passengers, and people on the ground, and a huge economic shock as well. Jackson reaches back to Challenger for the same reminder: humans are fallible, and the systems we create, however hard we strive, can always be made better. Mission success includes a commitment to true root-cause analysis and corrective action. Commercial aviation's fatal-accident rate has followed a decaying exponential, he says, not by accident but because regulators and airframe makers kept driving causes out.

Asked to describe what Lockheed actually builds, Jackson refuses the airframe label. The company thinks of itself as one that solves hard mission problems. The output might be an airplane, a spacecraft, a helicopter, or a radar, but the driver is the customer's mission. The SR-71 came from a need to photograph a place whose air defenses could handle any aircraft then flying. The answer was a reconnaissance jet faster than those defenses. In his shorthand, they built a nice flying camera and made sure it got out and got back. That logic, he adds, also fed the start of the U.S. space program. Between Lockheed and Martin, the heritage is on the order of 105 years old.

On aircraft, that means the most advanced fighters in the world, with several hundred F-35s delivered and production running at almost a hundred a year, plus work on what comes next. On space, the list runs across GPS navigation and timing, communications satellites, missile-warning satellites, commercial surveillance and communications birds, human exploration, and robotic missions to the outer planets. Defensive systems get equal billing. Aegis is the integrated air and missile defense for U.S. and allied fleets, protecting carrier strike groups from ballistic missiles, aircraft, and cruise missiles; a land-based version, Aegis Ashore, is deployed in a couple of places. THAAD, the Terminal High Altitude Area Defense system, is meant to protect relatively broad areas against sophisticated ballistic-missile threats, and international customers were looking to buy it. These systems defend population areas as well as military ones. The first public use Jackson points to is Patriot in the Gulf War. Underneath sit radars, command and control, and human-machine interfaces that have to work in strenuous environments.

Fridman wants the milestones that still sound insane: stealth, and flying at three, four, five times the speed of sound. Jackson starts with Skunk Works, which he says is as much an idea as a place, driven by Kelly Johnson's fourteen principles. He will not list all fourteen. The core that he thinks any engineer on a hungry small team will recognize is this: a handful of very capable people who want to work on hard problems can do almost anything if you shield them from bureaucratic noise and build a tight, shared vision with the customer.

The stealth story he wishes he had witnessed is Ben Rich rolling a ball bearing across a desk to a general officer and asking whether he would like an aircraft with that radar cross section. Jackson calls it probably one of the least expensive and most effective marketing campaigns in the history of the industry. Asked how you make a plane look as big as a bearing on radar, he sticks to what is in Rich's book: the geometry of how radar reflects, and materials that either reflect or absorb. Then he stops. "It's a cat and mouse game." Radars get better, stealth gets better, and it is continuous improvement. He will leave it at that.

Speed of sound, he notes, is about 750 to 760 miles an hour. Supersonic is the band from Mach 1 up through about Mach 5; hypersonic is past Mach 5. Lockheed and others have been in hypersonic development since the late 1960s—the X-15 and the space shuttle are the examples he names. What is different now is the threat environment. Russia and China have said publicly that they have hypersonic weapons that could threaten U.S. and allied forces and, in the claims, get around existing defenses. Jackson hears a real sense of urgency from people such as then-Under Secretary of Defense for Research and Engineering Mike Griffin. Lockheed has spent R&D dollars in the field for years and has programs with several U.S. government customers. He expects more news to come from those customers. He will not add the details.

The defensive problem fascinates him more: perceiving things that are extremely hard to see. He winds the clock back to the 1980s Strategic Defense Initiative, when Lockheed proved you could "hit a bullet with a bullet" and take out an incoming ballistic missile, something that had not been done before. That line of work became today's hit-to-kill systems: PAC-3, the Patriot Advanced Capability-3 that Lockheed builds, and THAAD. Hypersonic targets are not ballistic. Trajectories, speeds, and maneuverability change, so defensive capability has to take the next step. He leaves it there.

Fridman, speaking as an engineer who wants to inspire students, admits it is sad how much of this stays classified. Jackson's reply is calm. The public can already see enough, including students in grade school through college, to understand the hardness of the problems. Detailed F-35 performance numbers are sensitive, but you can still talk about what the jet is: supersonic supercruise, stealth, and a flying information node with data fusion and sensor fusion not seen before. Mission is what actually holds the workforce. When Jackson led GPS-related programs, an Air Force major whose call sign was Gimp briefed him after operational test in Iraq. On a Black Hawk bringing a sergeant and troops back from a deployed site, half-asleep, the sergeant lit up: GPS "brings me and my troops home every day." That, Jackson says, is the moment you know the work is useful.

Is a future coming in which fighter jets are so autonomous that a human almost never needs to be in the cockpit? In one sense, Jackson says, that world is already here: remotely piloted aircraft log hundreds of thousands of flight hours a year. The F-35 itself has huge layers of autonomy, so the pilot is more mission manager than stick-and-rudder operator, and in some ways it is the easiest aircraft in the world to fly. Carrier landings catch a tailhook on one of typically three wires. The joke, true or not, is that the F-35 is so consistent it always snags the second wire and wears that one out. Control systems have dialed out a lot of environmental variability. He does not believe cockpits empty of humans anytime soon. He does expect much more human-machine teaming at the tactical edge.

Skunk Works and the Air Force Research Lab ran a demo called Have Raider, using an F-16 as an autonomous wingman. They flew maneuvers and mission scenarios with that jet as a loyal, trusted wingman. Once you have up-leveled the pilot to mission manager, that person can control multiple other aircraft as extensions of their own. Safety is the other payoff. High-performance fighters pull nine Gs; a pressure suit is not always enough to keep blood in the brain, and G-LOC—G-induced loss of consciousness—is a disaster if you are low over terrain. Aeronautics built Auto GCAS, the Autonomous Ground Collision Avoidance System, into the F-16. It had already saved seven aircraft and eight pilots in a relatively short deployment. The Air Force wanted it on the F-35 immediately; testing was underway, and an air-to-air collision-avoidance piece had been folded in. It is a system of last resort. It does not try to decide whether the pilot is awake. If the jet has determined it is going into terrain, it gets out.

Sikorsky Innovations built a related idea called MATRIX. Jackson is picky about the words: not optional piloting, not unpiloted, but optimal piloting. It is an FAA-certified, largely deterministic system that can fly the same mission with two pilots, one, or none, with autonomy running in the background like a dial you can turn. In the North Sea it is meant to land on an oil rig in zero-zero weather. Miss the landing and you ferry everyone home, reschedule, and pay overtime for the crew and for the people still on the rig—real dollars. Militarily, the same aircraft can fly a crewed mission by day and switch to unmanned resupply at night.

Fridman's real fear is an AI arms race that slips human control. Jackson does not see that happening today. He first sketches why the appetite for assistive technology is so high: a return to great-power competition with China and Russia, new weapons and treaty breaches, artificial islands in the South China Sea, the annexation of Crimea and the war in Ukraine, North Korea, Iran, and a terrorism threat that has not gone away. Then he puts a clock on it. Photons hit the eye and a human moves on the order of a few tenths of a second. Computers run on nanoseconds. A nanosecond compared with a second is like a second compared with about thirty-two years. If autonomy can compress the OODA loop—observe, orient, decide, act—inside an adversary's loop, that is an incredibly powerful force on the battlefield.

The time-scale argument runs the other way too. Humans are useful from about a second out to about eight hours; after that they get tired and need a bathroom. Surveillance, guarding, moving material, logistics, and sustainment—the dull, dirty, and dangerous jobs—need activity past that limit. Those are the ranges where AI and autonomy are particularly well suited. Keeping it inside human control is the policy question, and the Pentagon has a directive for it: 3000.09. Jackson says it is well worth reading, only a couple of pages, and it is not a free-for-all. Humans must have agency over semi-autonomous and autonomous weapons. The systems must be tested, verified, and evaluated in realistic scenarios. Operators must be trained. Interfaces must show what state the system is in and what decisions it is making. Doctrine, tactics, techniques, and procedures have to exist. The logic, he says, is the same as for chemical and biological weapons: there are rules, and they are meant to be taken seriously. Lockheed complies 100 percent with customer policy.

Inside the company, AI is also an efficiency story: production, generative design for engineers, logistics, energy cost. Ethics still sits on the table—privacy, driving out bias, keeping humans in decisions, not delegating accountability to algorithms. It all comes back to three core values: do what is right, respect others, perform with excellence. Lockheed has a long-recognized ethics program, complete with the slightly surprising fact that there are awards for ethics programs, and the ethics team was working with engineering to turn those values into AI principles.

Born in the Soviet Union, Fridman asks why civilization has not destroyed itself in nuclear war, and whether deterrence still holds. Jackson notes that Lockheed is proud and privileged to play a part in multiple legs of U.S. nuclear and strategic deterrent systems, including Trident submarine-launched ballistic missiles. Is there still a possibility the human race could destroy itself? He says that possibility is real. He also thinks strategic deterrents have, in some sense, prevented the kind of world wars that marked the first half of the twentieth century. The world is more complicated now: a multipolar great-power landscape instead of a handful of ballistic-missile states. By a count a few years old, more than seventy nations had that kind of capability, with space capability in a similar range. Threats have proliferated in ways that were not expected. The United States, he says, is in the middle of a recapitalization of its strategic deterrent, which he ranks among the most important things the nation can do.

Deterrence is a bit of both offense and defense, a complicated game-theoretic program whose aim is to prevent anyone from using the weapons. The mechanism is that even if an adversary strikes first, you can strike back and do unacceptable harm, so they do not start. That only works if the systems will function when required and can adapt to a variety of scenarios, which is why hardware built over previous decades has to be updated for today and for the decades ahead. He wants one more category on the ledger. Weapons of mass destruction used to mean nuclear, chemical, and biological. Today there is a cyber threat. Attacks on critical infrastructure—electrical grids, water systems—are strategic for a nation's survival. His hope is that technical systems, perhaps enabled by autonomy, can help contain and fight threats that were not conceived when strategic deterrence was first designed.

Fridman quotes Eisenhower's 1961 farewell warning about a military-industrial complex that should not grow beyond what is needed. Jackson answers on three levels. First, Lockheed is not only a defense house: energy, space, and commercial aircraft matter, and he would not neglect them. Second, proportion. As an MIT graduate student he remembers talk of a peace dividend at the end of the Cold War. Military spending as a fraction of GDP, he says, is far below past peaks, while global threats look as if they warrant relevant defensive investment. Third, this is not quite a free market. You do not sell these systems to people on the street. That requires a close partnership between the customer and the people who design, build, and maintain the hardware, because the requirements are unique, safety is non-negotiable, and the systems have to work as intended every time. That long partnership, he says, is one of Lockheed's great strengths. Building the systems is very costly. There is very little room for mistake.

Fridman is happier back in space, and he asks the provocative question: what about SpaceX and Elon Musk? Jackson's answer is crisp. SpaceX is a competitive force in the industry. Does Lockheed like competition? Yes, and it thinks it is a strong competitor. Competition is what the United States is founded on, always coming up with a better way, and fresh ideas matter. The condition is a level playing field: you do not want different requirements for different players.

Then comes the line he tells people at places like MIT: this is the most exciting time to be in the space business in his entire life. Computing power, communications, and advanced and additive manufacturing have brought the barriers down. Innovation is happening at startups and inside Lockheed. Many people do not realize that Lockheed, working with Stanford and Stellar Solutions, built the first CubeSat launched from the United States, QuakeSat, just after 2000. Now the company is building what he calls smart sats and software-defined satellites that can restructure and remap their mission on orbit, giving almost unlimited flexibility over a satellite's lifetime.

The Wright brothers flew 116 years before this conversation. Now there are supersonic stealth aircraft and everything the two men have just walked through. Fridman asks where the edge of possibility sits if you take another hundred-year leap. "I don't know that there is an edge." The Lockheed brothers and Glenn L. Martin started in a church basement and an old service station. The company is unrecognizable because it has kept reinventing itself. Jackson's bet on the future comes in four words: faster, more connected, more autonomous, more complex. As CTO he has to decide which technologies to fund—AI and autonomy, quantum computing, which Lockheed has already invested in—and how to stay ahead of military threats nobody has thought of yet.

He expects humans to be out in the solar system, defending against threats that have not been imagined, and using the same capabilities for far greater knowledge of Earth: the depths of the oceans, the upper atmosphere, the Sun, and the edge of the solar system. Looking at the decade ahead, he is excited. "I don't think there's a better place" to be. The path he has described is not a pile of secret gadgets. It is a culture that keeps asking whether the test was good enough, whether the human is still in the loop, and whether the machine will still do the right thing when the envelope moves.