Palmer Luckey: Autonomous Weapons Are Ancient and Why Anduril Won't Build Humanoids | EP #295
Is it worth your time?
Worth your time if you want to understand how autonomous weapon systems actually operate and why military robotics demands extreme specialization over general-purpose humanoids.
Luckey explains why building hardware for defense and firefighting requires local autonomy and specialized engineering rather than consumer tech trends. He details how his company disrupts traditional military contracting by developing physical products with private capital.
Autonomous weapons and command security. Luckey argues that autonomous weapons are not a new invention, pointing to decades-old systems like the Aegis system and thousands of years of historical precedents from landmines to traps. He warns that relying on central command for compute creates massive vulnerabilities because enemies can jam or attack communication links, making local autonomy an absolute necessity. He adds that specialized systems must process raw data like radar feeds locally rather than depending on slow network pipes.
Specialized robots versus humanoids. Luckey contends that general-purpose humanoid robots are poorly suited for military applications because specialization always yields better performance in extreme environments. He points out that accepting a five percent performance penalty for commonality is a non-starter when fighting adversaries, citing his robotic submarines designed to dive six thousand meters. He notes that mainstream foundation models are non-deterministic and unsuited for weapon guidance, requiring custom applications built since 2017.
Real threats from artificial intelligence. Luckey dismisses popular anxieties about artificial intelligence causing existential doom, arguing that newcomers have spent almost no time analyzing actual major threats. Instead, he warns that the real danger comes from evil people using moderately competent artificial intelligence to commit malicious acts. He argues that the technology's true value lies in massively reducing resource extraction and processing costs to bring economic efficiency to society.
Building hardware for defense. Luckey recounts that Anduril bypassed traditional defense consolidation by building hardware with its own money first, proving its products before government procurement officials stepped in. He urges wealthy tech figures to use their capital and credibility to solve massive physical problems like defense rather than cycling endlessly through cars and rockets. He advises young builders to get a job and construct projects in the open because organizational dynamics matter far more than writing code.
AI in Defense and Firefighting
The host welcomes Palmer Luckey as the king of autonomous weapon systems and congratulates him on winning the wildfire X Prize. Luckey explains that Anduril won $2.2 million, including a million from Lockheed, noting that Anduril works with Lockheed in the defense space by competing with their fighter jet division while partnering on some payloads.
Luckey and the host discuss how they have tried to get autonomous firefighting started for a long time, with Luckey mentioning the X Prize was announced back in 2019. He notes that one of Anduril's first two products was an autonomous firefighting tank built starting in 2017 and finished in 2019. He calls it insane that society lets fires become conflagrations before putting them out, and points out that while people claim the technology only recently made this possible, it has actually been possible for over a decade. He states that the real issue is that ignoring the problem has become politically impossible.
Fire Detection, Classification, and Response
Luckey describes how the testing site operated by detecting and classifying fires, while also rejecting false fires such as someone smoking a cigarette, a campfire, or a road flare. He emphasizes that the system must flag real fire characteristics to people and emergency services, and then match the fire type to the most cost-effective asset available in the area that can respond in time, adjusting for factors like whether the fire is in woods or on a brush hill. He contrasts this with military work where a delay of an hour compared to ten minutes might not change the outcome of a bomb, whereas a metastasizing fire requires rapid response to prevent it from growing too large.
The host teases a new upcoming $10 million X Prize funded by Luckey, which will be announced in a month.
Biological Intelligence and Animal Communication
Luckey explains that he is not interested in joining the race to push artificial intelligence forward with his money because foundation models are already covered by plenty of people. Instead, he believes humanity has not done enough to understand biological intelligence beyond humans. He announces a new prize focused on understanding how animals think, with the goal of communicating with wild animals in their own language ten times better than the current state of the art for the smartest species. He suggests that studying other species can offer lessons for artificial intelligence, particularly regarding power and time efficiency.
Moving to the official questions, the host asks about Luckey's p-doom. Luckey replies that he is not worried about artificial intelligence causing doom through killing, but is deeply concerned about evil people using moderately competent artificial intelligence to commit evil acts.
AI’s Impact on Costs and Society
Luckey expresses frustration with newcomers to the dooming hobby, comparing them to people who held naive views about the world living at the end of history until world events like the war in Ukraine forced them to change their minds. He argues that those predicting artificial intelligence as the ultimate threat have spent almost no time analyzing actual major threats or speaking with experts. Placing artificial intelligence doom low on his list, Luckey is an optimist who predicts the technology will massively reduce the costs of resource extraction, processing, and transformation, bringing economies of scale and efficiency to every part of society, much like automated agriculture and textiles.
The host then shifts the discussion to the technical challenge of deploying intelligence and weapons in the field, contrasting the high compute power of centralized headquarters against the latency of transmitting data back and forth.
Autonomous Weapons and Historical Examples
Luckey argues that the real concern is not latency or computation speed, but command and control. He explains that relying on a central location for compute requires maintaining a high-bandwidth communication link with potentially tens of thousands of active assets, which leaves systems vulnerable if an enemy jams, intercepts, or hacks the communications, or simply attacks the central control facility. He compares this to a scenario from a George Lucas documentary where the Trade Federation's battle droids all fell over after their central command and control was destroyed. He adds that underwater environments or situations where radio emissions expose a position make local autonomy a necessity.
Luckey disputes the idea that this represents an unprecedented opening of Pandora's box.
Scaling Laws and the Future of Autonomous Robots
Luckey points out that weapons making their own kill decisions have existed for decades, citing the Aegis system on destroyers which features a fully autonomous mode to shoot down threats within range without human confirmation, as well as SeaWiz and SeaRAM systems used against incoming missiles and speedboats. He notes that even during the Vietnam War, aircraft deployed missiles that flew over the horizon to identify specific radio emitters and strike surface-to-air missile sites independently, while human operators retained the decision to launch into a specific area based on probability. Expanding his argument, Luckey states that autonomous weapons have existed for thousands of years if defined as weapons executing programmed intent independently from the operator, ranging from pitfall traps and poison traps to landmines, sea mines, and torpedoes, as well as mythological golems defending cities. He concludes that advanced technology is simply making them better.
Specialized Robots vs. Humanoids
Luckey states that Anduril has been building AI into its systems since 2017, putting the company a decade ahead in this area. He explains that mainstream foundation models are non-deterministic, non-auditable, and unsuited for weapon guidance or radar classification. His team processes raw data like radar feeds and passive submarine sonar streams rather than text, creating highly specialized applications that share very little with consumer software.
The host notes that modern frontier models can now be dropped into robotic embodiments for strong results, and asks whether autonomous weapon systems follow something like Moore's law as underlying models improve.
Centralized Defense AI and Military Robotics
Addressing whether defense robotics will converge on a general-purpose form factor, Luckey doubts it on the military side. While generalist humanoid robots might dominate manufacturing, hospitality, and medicine for economies of scale, military competition requires massive overmatch rather than minor margins. He argues that accepting a five percent performance penalty for the sake of commonality is a non-starter when the goal is to fight unfair and utterly overwhelm adversaries.
Why Palmer Avoids Building Humanoids
Luckey gives the example of his robotic submarines designed to dive six thousand meters and sit on the ocean floor, alongside surface systems and specialized aerial vehicles, noting no single shape can serve all these extreme military roles. When the host asks why he does not build humanoids given his capability, Luckey explains that specialization always yields better results. He points out that trying to build a machine that can dive to the ocean floor and fly to the moon would only make it worse at both tasks.
Defense Contractors, Innovation, and Investment
Luckey explains that while consumer cars share platforms and standards, defense will likely remain a branch of exotic and highly specialized robotic systems. He notes that he opted out of building humanoids because other companies handle them well, though he believes humanoid robots will serve as an ideal interface to automate legacy military platforms like expensive radar systems without requiring humans to man them during combat. He adds that Anduril similarly avoids building small quadcopter attack drones or becoming a space launch provider, leaving those markets to other companies.
Solving Big Global Problems
Turning to the challenge of military centralization, Luckey recounts that early investors doubted the government's ability to buy products at scale, noting that eighty percent of major defense acquisition programs historically went to just five companies. Anduril bypassed this consolidation by operating as a product company that builds hardware with its own money first. Once government officials saw these products delivered on time and on schedule, subsequent program managers repeatedly chose to work with Anduril to make their jobs easier.
Billionaires, Innovation, and Expanding Participation
Reflecting on the audacity required to enter the defense industry, Luckey states that Anduril's initial pitch deck promised to save taxpayers hundreds of billions of dollars a year by displacing legacy systems that cost ten times more than they should. He recalls his early mentor John Carmack, the creator of 3D game engines and founder of Armadillo Aerospace, who pioneered vertical takeoff, vertical landing rockets long before SpaceX by applying commercial graphics hardware and game programming experience to aerospace challenges.
The Future of AI, Robotics, and Advice for Innovators
Luckey recounts working with John Carmack, who became the CTO of Oculus VR after Luckey managed to poach him from Id Software. Carmack started his own company to work on artificial general intelligence, reasoning that even if his chance of success was only one percent, the risk-adjusted outcome demanded that he do it. Luckey felt the exact same way about defense, believing that a one percent chance of helping solve a massive problem was more impactful than starting another gaming company, a space launch provider, or a car company. He notes that billionaires tend to cycle through the same few industries—cars and rockets—while far too few people enter defense or wear a metaphorical bat suit.
Addressing the private chat group of retired wealthy figures, Luckey recalls challenging them to use their capital and underutilized resources, such as their credibility and ability to gather teams and capital, to solve major world problems instead of just sitting on boats. He notes that many of them have stepped up recently, as nihilism has declined and more people feel the itch to tackle big challenges. He clarifies that he does not mean people should distract themselves by spreading thin across multiple fields, but rather that doing nothing is repulsive. He mentions that the group chat recently shut down after members posted the wrong memes too many times and that they need to restart a version 2.0.
Shifting to virtual reality, Luckey discusses Meta's new VR glasses. While they do not match his exact vision, they are much closer in several ways. He has long believed in decoupling compute and the battery from the front of the headset. During his time at Oculus before being fired, the prototype series code-named Santa Cruz placed the battery and compute on the back of the head to keep the front as light as possible, an ergonomic balance he achieved with the Oculus Rift CV1 in 2016. He expresses frustration that subsequent designers, including Apple, routinely mess up ergonomics by packing everything into a lump on the front. He praises the new Meta headset for its lightweight design, off-boarded compute and batteries, high-quality pass-through, and competitive price.
Offering advice to the audience of entrepreneurs and visionaries, Luckey draws on his experience starting an online community called ModRetro when he was fourteen or fifteen, which was an internet forum for modifying vintage game consoles and building handheld units. Running that forum taught him how to build and market an organization, rally people around a cause, and manage interpersonal conflicts between brilliant people who hate each other. Because running an organization requires managing human dynamics, he strongly believes that kids should get a job and build projects with other people in the open. Addressing whether artificial intelligence changes building, Luckey compares code creators mourning the end of hand-coding to artists who complained when the camera made capturing images accessible to everyone. He explains that founders do not get to work on the technology anyway, because the job of a founder is to hire people, manage legal and interpersonal issues, and make the final architecture and fundraising decisions. He asserts that artificial intelligence will not replace those responsibilities anytime soon, warning hobbyist coders who think they will spend all their time building that being a founder means other people take over the technical work.