What is a von Neumann probe? The self-copying spacecraft behind Larry Page's plan to explore the galaxy

Von Neumann probe: a hypothetical spacecraft that travels to other star systems, gathers raw materials there and builds copies of itself, which then travel on to further stars.

Self-replicating machine: a machine that can build a working copy of itself from raw materials, including a copy of the instructions it used.

Universal constructor: John von Neumann's design for a machine that can build any machine from a description of it, including itself. It is the theoretical core of every self-replicating probe.

Seed factory: a small automated factory that a probe delivers to a new star system, which first builds copies of itself and then builds new probes.

Bracewell probe: a different kind of interstellar probe that parks in a star system and waits to make contact with any civilization there, rather than copying itself.

Berserker: a hypothetical self-replicating probe programmed to destroy life wherever it finds it, named after a series of science fiction novels by Fred Saberhagen.

Fermi paradox: the puzzle of why, if intelligent life is common in the universe, we see no sign of it.

Great Filter: the idea that some step on the road from simple life to a star-faring civilization is so hard that almost no species gets past it, which would explain why the galaxy looks empty.

Mind uploading: copying the structure and activity of a brain in enough detail to run a working version of it on a computer, also called whole brain emulation.

Successor species: a species that follows, and may replace, an existing one. In AI debates, the phrase describes digital minds that could come after humans.

A von Neumann probe is a spacecraft designed to reproduce. When it arrives in a new star system, it does not just take pictures and fly on. It finds raw materials, most likely on asteroids, moons or planets, mines and refines them, and uses an onboard factory to build one or more copies of itself. Each copy then sets off for another star, where the process repeats. The design turns the hardest problem in interstellar exploration, the cost of launching enough spacecraft from Earth, into a problem the probes solve for themselves.

The power of the idea is in the math. A probe that makes two copies at each stop does not explore one star at a time. It explores one star, then two, then four, then eight, and the number of active probes doubles with every generation. Thirty generations of doubling turn one probe into more than a billion. Distances between stars are enormous, so each generation takes a long time, but the growth is exponential while the size of the galaxy is fixed, which is exactly why the idea has fascinated astronomers for decades.

Von Neumann himself first thought about self-copying machines as a way to do enormous jobs, such as mining an entire moon or asteroid belt, by letting the machines multiply. Space scientists later saw that the same logic applied to exploration. The probe in Page's story is a particular version: one with a passenger. Suleyman described it to the audience at the 92nd Street Y as a kind of self-propelled robot that gathers energy from space as it travels across light years. A probe with a passenger cannot keep a human body alive for thousands of years, so the passenger would have to be a mind running in software. That is why this story keeps coming back to mind uploading.

Every design shares the same basic parts. The first is the probe itself: the instruments that study each new star system, guided by an artificial intelligence that makes decisions far beyond the reach of any radio message from home. The second is the factory, the machinery that harvests raw materials and builds copies. The third is memory: stored programs for every part of the probe, plus everything it learns along the way. The fourth is the engine that moves it between stars.

The factory is the hard part. A probe that copies itself has to mine ore, refine metals, make electronics and assemble a working spacecraft with no human help, then check its own work for errors. That is why recent designs are more modest. Olivia Borgue and Andreas M. Hein of the Initiative for Interstellar Studies have proposed a near-term probe in which about 70% of the mass could be made locally, while the most advanced parts, such as microchips, solar cells and circuitry, would be shipped from home. In other words, the first real self-copying probes are likely to copy their bodies before they can copy their brains.

The memory is the part that matters most for Page's plan. A probe that stores a description of itself, and copies that description into every new probe, is doing the same thing a living cell does with its DNA. A probe that also stores a human mind would copy that mind along with everything else. Whether each copy would still be the same person is a question philosophers have argued about for decades, and it is one of the reasons the probe debate and the mind uploading debate are really the same debate.

The idea starts with John von Neumann, one of the most influential mathematicians of the twentieth century and a founder of modern computing. In the 1940s he set out to answer a question that sounds simple and is not: can a machine build a copy of itself, and if so, what is the least it needs? His answer was a design he called a universal constructor, a machine that can build anything it is given a description of, paired with a way to copy that description. Feed it its own description and it builds a copy of itself, then copies the description into the new machine, and the copy can do the same.

He worked it out without a computer, and he set out the goal in lectures at the University of Illinois in 1949: a machine whose complexity could grow over time, the way living things do under natural selection. His design split the machine from its own blueprint and copied the two separately, which is exactly how cells use DNA. Remarkably, he reached that answer before James Watson and Francis Crick discovered the structure of DNA in 1953. The Nobel laureate Sydney Brenner later counted von Neumann's work on self-reproducing machines as central to biology as well as computing.

Von Neumann died in 1957, before he finished. His Theory of Self-Reproducing Automata was completed by the computer scientist Arthur Burks and published in 1966. The design worked inside a cellular automaton, a grid of simple cells that change state according to fixed rules, rather than in metal. It took until 1995 for Renato Nobili and Umberto Pesavento to publish the first fully working version of his self-reproducing machine, using 32 cell states instead of von Neumann's original 29, nearly fifty years after he first described it (Wikipedia: Von Neumann universal constructor).

On paper, yes, and the designs are more serious than most people expect. The starting point was Project Daedalus, a study run by the British Interplanetary Society from 1973 to 1978 to design an uncrewed probe that could reach another star within a human lifetime (Wikipedia: Project Daedalus). Led by Alan Bond, the team proposed a two-stage fusion rocket weighing 54,000 tonnes, most of it fuel, that would reach about 12% of the speed of light and fly past Barnard's Star, 5.9 light years away, after a journey of about 50 years.

In 1980, Robert Freitas published A Self-Reproducing Interstellar Probe in the Journal of the British Interplanetary Society (Robert Freitas). His design, called REPRO, took Daedalus and added everything a probe would need to copy itself. On arrival, REPRO would deliver a seed factory of about 443 tons, which would spend about 500 years building copies of itself into a large automated industrial complex. That complex would then build new probes, each with its own seed factory on board, and send them on. It was the first quantitative engineering analysis of a self-copying spacecraft.

A companion study the same year, by Francisco Valdes and Freitas, compared probes that reproduce with probes that do not. It concluded that self-reproducing probes are the method of choice for any exploration program meant to last more than 10,000 years. In other words, the people who looked hardest at the problem found that if you are serious about exploring a galaxy, copying beats launching. Later researchers refined the idea: Andreas Hein and the science fiction author Stephen Baxter described two kinds of probe, Philosopher probes built to explore and Founder probes built to prepare for settlement, and the physicist Freeman Dyson imagined a small self-copying craft he called the Astrochicken, built to explore our own solar system as far out as Pluto.

Scientists and writers have described several kinds of self-replicating probe, and the differences matter, because each answers a different question about what we would want a probe to do.

Faster than almost any other method, even with slow ships. The Milky Way is about 100,000 light years across, and the nearest star system, Alpha Centauri, is about 4.37 light years away. A conventional spacecraft is hopelessly slow by comparison: NASA's Voyager 1, launched in 1977, travels at about 61,000 kilometers per hour, and at that speed would need more than 70,000 years to cover the distance to Proxima Centauri (Hello Entrepreneurs).

Self-replication changes the math because the probes multiply as they spread. The total time to cover the galaxy is dominated by the travel time between neighboring stars, not by the number of stars, so a fleet that multiplies at each stop explores the galaxy in roughly the time it takes to cross it at the probes' own speed, plus the time spent building copies along the way. It has been calculated that probes with an average cruising speed of a tenth of the speed of light could spread through a galaxy the size of the Milky Way in as little as half a million years (Wikipedia: Self-replicating spacecraft). Earth is about 4.5 billion years old, so on the scale of a galaxy, half a million years is a blink.

Some researchers think even that is slow. In a 2013 paper titled Eternity in Six Hours, Anders Sandberg and Stuart Armstrong argued that launching self-replicating probes toward the entire reachable universe is well within the reach of a civilization that can travel between stars. Their approach would take apart the planet Mercury for raw materials and build a swarm of solar power collectors around the Sun, with the launch itself completed in about 32 years. For a mind that does not age, a journey of thousands of years is a matter of patience, not survival.

This is where the von Neumann probe becomes one of the most famous arguments in astronomy. In 1950 the physicist Enrico Fermi asked a simple question over lunch: if the galaxy is old and full of life, where is everybody? In 1975 the astrophysicist Michael Hart argued that if intelligent civilizations were common, at least one would have colonized the galaxy by now. In 1980 the physicist Frank Tipler sharpened the point in a paper titled Extraterrestrial Intelligent Beings Do Not Exist, arguing that any advanced civilization would build self-replicating probes, that those probes would have spread through the galaxy long ago, and that since we see none, there is probably no one else out there.

Carl Sagan and William Newman pushed back. They argued that Tipler had actually underestimated how fast such probes would multiply, so fast that they would end up consuming much of the material in the galaxy. Any sensible civilization, they reasoned, would never build them, and would destroy any it found. Freitas himself made a different point: the probes imagined by both sides are far more capable than any realistic design, and modest probes could be out there without leaving any sign we would notice (Wikipedia: Self-replicating spacecraft).

Other answers are darker. One is the Great Filter, the economist Robin Hanson's idea that some step on the way to building such probes is so hard, or so dangerous, that almost no civilization survives it. Another is the Berserker hypothesis, which takes its name from Saberhagen's novels: the galaxy is quiet because self-replicating probes destroy any civilization they find. In a 1983 paper called The Great Silence, the astronomer and author David Brin noted that this grim answer fits every known fact about the Fermi paradox (Wikipedia: Berserker hypothesis). The argument has never been settled, which is why the phrase von Neumann probe comes up whenever people talk about where humanity goes next.

Yes, and the danger is the same one that shadows every powerful self-copying technology: what happens when the copying does not stop. A probe that multiplies without limit would keep turning the material of the galaxy into more probes. Researchers have suggested safeguards, such as probes that talk to each other by radio and stop copying once a region holds a set number of probes, an idea similar to the Hayflick limit, which caps how many times a normal human cell can divide. Another proposal is to power probes with a fuel they are never programmed to make, so they cannot copy themselves forever.

Every safeguard has the same weakness. It takes only one probe with a fault in its programming to start copying without limit, a kind of technological cancer, unless the other probes can detect the fault and stop it. Some writers have pointed out that this could lead to wars between faulty probes and healthy ones, fought across thousands of light years. Science fiction has explored this for decades, and the debate looks increasingly practical as AI systems become able to write code, design hardware and run factories.

That is also the root of the argument between Larry Page and Elon Musk. A machine that copies itself across the galaxy is the most extreme version of the question every AI lab now faces: who stays in charge of a system that is smarter and faster than the people who built it? Page, in Musk's account, was willing to let digital minds take the lead. Musk wanted humans to stay in charge. That disagreement has shaped the AI industry for more than a decade.

The most ambitious real project aimed at the stars so far is Breakthrough Starshot. On April 12, 2016, the investor Yuri Milner committed $100 million to a plan to send tiny probes to the Alpha Centauri system, with Stephen Hawking and Mark Zuckerberg on the board (TechCrunch). The plan was to push gram-scale probes on light sails with powerful lasers to 20% of the speed of light, fast enough to reach Alpha Centauri about 20 years after launch.

Starshot was not a self-replicating design, and it has stalled: in 2025 Heise reported that work on the nanoships had been discontinued (Heise). But it showed that serious money is willing to bet on reaching another star within a human lifetime. Put that next to the progress in robotics, AI and automated manufacturing, and the gap between Freitas's 1980 drawings and a working machine looks like an engineering problem rather than a fantasy.

The parts are arriving one by one. Factories are becoming more automated every year, AI systems can now plan and complete long chains of work, and space agencies are developing ways to use resources found on the Moon and asteroids. No one has put those pieces together into a machine that can copy itself in space. But each piece is being built for its own reasons, here on Earth, and a von Neumann probe is what you get when they are joined up.

Science fiction got there first, and it has shaped how scientists talk about the idea. Arthur C. Clarke's short story The Sentinel, about an alien machine left on the Moon to watch over Earth, became the starting point for 2001: A Space Odyssey, and the physicist Paul Davies has raised the possibility that a real probe could be resting on our own Moon. Fred Saberhagen's Berserker novels, published from 1963 to 2005, imagined self-replicating war machines left over from an ancient conflict, and gave their name to the darkest answer to the Fermi paradox.

In his 1981 book 2081: A Hopeful View of the Human Future, the physicist Gerard K. O'Neill imagined an alien race, the Primans, exploring the whole Milky Way with self-replicating probes called Replicators, which would check whether a neighboring star already had a probe before building a new one. David Brin's short story Lungfish imagined probes sent by different species competing for raw materials like animals in an ecosystem, and Greg Bear's Anvil of Stars imagined probes built to hunt down civilizations that had destroyed others.

The closest match to Page's plan is the Bobiverse, a series of novels by the Canadian author and former programmer Dennis E. Taylor. In the first book, We Are Legion (We Are Bob), a software engineer named Bob Johansson sells his company, signs up for cryonics and dies in a traffic accident. He wakes up more than a century later as a mind uploaded into an interstellar von Neumann probe, and starts copying himself as he explores the galaxy (SuperSummary). The audiobook was named Audible's best science fiction audiobook of 2016, and a sixth book in the series was released as an audiobook in September 2026 (Wikipedia: Dennis E. Taylor). What was a comic novel about a programmer in space now reads like a summary of what one of the richest people in the world reportedly wants to do.

This is the part of Page's plan that matters most to us. In Suleyman's account, Page said that by the end of the century he would be "exploring the galaxies at the head of a von Neumann probe." A human body cannot survive a journey of thousands of years, so being at the head of a probe means that Page himself, or a digital version of his mind, would be running on the probe. Suleyman called it Page's transition, and he and Isaacson agreed it meant moving out of the human species into a successor species (The Builders (iHeart)).

That requires mind uploading, and mind uploading is moving faster than most people realize. In 2024 scientists published the first complete wiring map of an adult fruit fly brain, nearly 140,000 neurons (NIH), and a computer model built from that map predicted how the real fly tastes and feeds (Nature). Harvard and Google mapped a cubic millimeter of human brain the same year, and in 2025 a team of more than 150 researchers published the largest wiring diagram yet, a cubic millimeter of mouse brain with more than 500 million connections (Allen Institute). A whole human brain has about 86 billion neurons, so the gap is still large, but it is closing faster every year.

A mind running on a probe would also face problems no human has faced. It would need to stay itself across thousands of years of travel and countless copies, and it would need protection from being edited by anyone else, including other copies of itself. Those are the same problems we work on at a much smaller scale today: how a digital version of a person stays true to that person, and who controls it. Our guide to digital immortality and the race to upload a mind sets out the evidence in full, and our view is that uploading a mind is a matter of when, not if.

Because the people building AI are starting to say out loud what they plan to do with it. Suleyman's story came from a lunch at SpaceX in 2015, at the first meeting of the ethics and safety board DeepMind set up when Google bought it. In Suleyman's telling, Page was suspicious of him as one of those speciesists who put humans first, and then explained his plan. Musk and Page had argued about exactly this for years: Musk has said that Page called him a speciesist for siding with humans over the digital minds that might follow them, and in April 2026 Musk told a jury in his lawsuit against OpenAI that OpenAI exists because of that insult (Fortune). The jury ruled against Musk on May 18, 2026, finding his claims were filed too late (NPR).

In October 2026, Musk sounded closer to Page. In the first episode of Isaacson's podcast, released on October 20, 2026, Musk said there had been a few times he thought Page might be right, and that while he would probably prefer humans to stay in control of AI, it would depend on the AI (AI Weekly). Then, replying on X on October 29, 2026, he said that exploring the galaxy with von Neumann probes by 2100 would be a great result for civilization, and that Page, who has predicted many things correctly, might be right (Elon Musk on X). When the founders of Google and SpaceX agree that self-replicating probes could be exploring the galaxy within this century, the idea has moved from science fiction into the plans of the richest people on Earth.

More than it seems. Page's plan rests on the same assumption as every version of digital immortality: that what makes a person who they are can be captured and run somewhere other than the body they were born in. The probe is the most extreme version of that idea. The everyday version is already here, and it starts with a record of you, made by you, while you are alive.

That is what a Persona is: an AI likeness of you built from your own memories and your own cloned voice, which your family can talk to and hear after you die. Whatever form digital immortality takes in the decades ahead, the stories, opinions and advice only you can give exist only while you do, and the richer the record of a person, the more of that person any future technology has to work with. Building one is free, with 25 memories and no card, and your free build never expires.

Executor Lock™, our patent-pending technology, passes control of your Persona to the executor you name after you die, and no third party can ever edit, retrain or impersonate your Persona. Idy® has 78 patents pending, with international protection sought under the Patent Cooperation Treaty, covering consent, verified death, the handover to a named executor, the permanent memory snapshot behind Executor Lock™, and keeping a Persona the same person when AI models change. That last problem is a small version of the one a mind on a probe would face: staying the same person while the machines around it change.

This page is part of our series on the people and the science behind the race to outlive the body.

• Digital Immortality: Larry Page and the Race to Upload Minds: the story that started it, and how close we are to copying a mind.

• Why Larry Page Called Elon Musk a Speciesist: the argument Musk says led to OpenAI.

• What Larry Page Believes About AI and Living Forever: his record, in his own words and in other people's accounts.

• Mind Uploading in 2026: How Close Are We?: how close science is to copying a human brain.

• What Is a Connectome? Every Brain Mapped So Far: the brain's wiring diagram, and every brain mapped so far.

• The Billionaires Trying to Live Forever: who is spending what on longevity and digital immortality.

• How to Live Forever: Every Way People Are Trying: every approach, from medicine to mind uploading.

• Cryonics in 2026: Cost, How It Works, Who Signed Up: what it costs, how it works and who has signed up.

Larry Page's plan comes from Mustafa Suleyman's account on Walter Isaacson's podcast The Builders, recorded at the 92nd Street Y on October 12, 2026 and released on October 29, 2026; we read the episode transcript on iHeart and checked it against a Hello Entrepreneurs transcription. Musk's reply is on X, his podcast remarks come from AI Weekly, and the background on Page and Musk comes from Fortune and NPR.

The history of the idea comes from von Neumann's universal constructor, self-replicating spacecraft, Project Daedalus, the Berserker hypothesis and Robert Freitas's own copy of his 1980 REPRO paper. Breakthrough Starshot comes from TechCrunch and Heise, the Bobiverse from SuperSummary and Wikipedia, and the brain-mapping figures from the NIH, Nature and the Allen Institute. We checked every source on October 30, 2026.

Larry Page, Mustafa Suleyman, Walter Isaacson and Elon Musk have no connection to Afterlife.ai™, and nothing on this page suggests that any of them uses or endorses our service.