Mind uploading in 2026: how close are we to copying a human brain, and what would it take?

Mind uploading: copying the structure and activity of a brain in enough detail to run a working version of that mind on a computer.

Whole brain emulation: the scientific term for mind uploading. An emulation reproduces a specific brain closely enough to behave like it, rather than modeling brains in general.

Neuron: a nerve cell that sends electrical and chemical signals. A human brain has about 86 billion of them.

Synapse: a connection between two neurons, where one passes a signal to another. A human brain has trillions.

Connectome: a complete map of the neurons in a brain and every connection between them. The study of connectomes is called connectomics.

Brain-computer interface (BCI): a device that records signals from the brain, or sends signals into it, and connects them to a computer.

Gradual replacement: a proposed form of uploading in which neurons are replaced by artificial parts one at a time until the whole brain runs on new hardware.

Substrate independence: the idea that a mind depends on the pattern of information processing, not on the material it runs on, so the same mind could run on neurons or on silicon.

Digital immortality: the idea that a person can continue in digital form after their body dies, from a Persona built from their memories and voice to a full copy of their mind.

Mind uploading is the idea that everything that makes you who you are, your memories, your personality, your habits of thought, is stored in the physical structure and activity of your brain, and that if that structure could be measured precisely enough, it could be reproduced and run on a computer. The result would be a digital mind that thinks like you, remembers what you remember and, if the theory is right, experiences the world as you do.

There are two main ways people imagine doing it. The first is scan and copy: measure a brain in full detail, then build a working model from the measurements. With today's technology, that means preserving the brain and slicing it into extremely thin layers to image, which is only possible after death. The second is gradual replacement: swapping neurons for artificial parts one at a time, while the person stays conscious, until the whole brain runs on new hardware (Wikipedia: Mind uploading).

The roboticist Hans Moravec described a version of gradual replacement as early as 1979, imagining a robot brain surgeon that copies a brain layer by layer, and returned to it in his 1988 book Mind Children, where he called it transmigration (Wikipedia: Hans Moravec). The futurist Ray Kurzweil, whom Larry Page later hired at Google, has long predicted that people will be able to upload their minds and become digitally immortal by around 2045. What has changed in the last decade is that neuroscientists, not just futurists, are now doing the early steps.

A 2025 report by researchers in the field, the State of Brain Emulation Report, revisited the landmark 2008 roadmap by Anders Sandberg and Nick Bostrom of Oxford's Future of Humanity Institute. It breaks the work into three capabilities: mapping the brain's structure, recording its activity, and running the result as a working model in a body or a simulated world (arXiv). Each one is a field of science in its own right, and each one is moving.

The first step gets most of the attention because it produces spectacular pictures, but all three matter. A wiring diagram alone is like a map of a city's roads with no traffic: it tells you where signals can go, not what they actually do. Recording activity shows the traffic, and emulation tests whether the model behaves like the real thing. The most convincing results of the last few years are the ones where all three came together, as with the fruit fly.

Not a human brain, and not a mammal's. But the first animal brains have been mapped in full, and the first working models have been built from those maps. The story starts with a tiny worm. In 1986, scientists published the complete wiring diagram of the nematode worm C. elegans, all 302 of its neurons, after more than a decade of tracing them by hand from electron microscope images (OpenWorm). An open-source project called OpenWorm has since worked on simulating the whole worm, starting with its 302 neurons and 95 muscle cells.

The next leap took almost forty years. In March 2023, scientists published the full map of a larval fruit fly brain: 3,016 neurons and about 548,000 connections (Live Science). In October 2024, the FlyWire consortium published the complete wiring diagram of an adult fruit fly brain, nearly 140,000 neurons and more than 50 million connections, in a set of nine papers in Nature (NIH).

Then came the step that matters most for uploading. A team led by Philip Shiu built a computer model of the entire fly brain from the wiring diagram alone, and it predicted which neurons would fire when the fly tasted sugar or water, and how that would drive feeding. Experiments on real flies confirmed the predictions (Nature). In March 2026, Eon Systems, a San Francisco startup where Shiu is now a senior scientist, reported that it had connected that brain model to a simulated fly body that walks, grooms and feeds. Eon calls it the first whole brain emulation to drive several behaviors in a body. The claim comes from the company, the code linking brain to body has not been released, and outside scientists have not yet verified it (The Decoder). Eon says its next target is a mouse brain, with about 70 million neurons.

The honest answer is that the gap is still huge, and closing faster than almost anyone predicted. A human brain has about 86 billion neurons (PNAS), more than 600,000 times as many as a fruit fly. In 1979 the Nobel laureate Francis Crick wrote that an exact wiring diagram of even one cubic millimeter of brain was out of reach. In 2025, a team of more than 150 researchers published exactly that for a mouse: more than 200,000 cells, 4 kilometers of axons and 523 million synapses, about seven years of work (Allen Institute).

The data problem is enormous. The cubic millimeter of human brain that Harvard and Google mapped in 2024, about the size of half a grain of rice, produced 1,400 terabytes of data, and a whole mouse brain would be about 1,000 times larger (ScienceDaily). A whole human brain is more than a thousand times larger again. But data storage keeps getting cheaper, and the slowest step, checking the computer's tracing of every neuron by hand, is exactly where AI is starting to help.

The cost problem is moving too. E11 Bio, a nonprofit research organization, has developed a method called PRISM that tags neurons with protein barcodes, expands the tissue so it can be imaged with light microscopes, and uses AI to check its own work. It says the approach could cut the cost of mapping by up to 100 times, and it is aiming to map a whole mouse brain within about five years for about $100 million (Asimov Press). The US National Institutes of Health launched a $150 million program, BRAIN CONNECTS, in 2023 to build the tools for mapping whole brains, with one project aiming at 2% to 3% of a mouse brain. If a whole mouse brain is mapped by the early 2030s, the path to a human brain becomes a question of scale, money and time. Our guide to the connectome lists every brain mapped so far.

Brain mapping used to be the work of a few labs. It is now a field with major public programs, technology companies and startups, and some of the biggest efforts of the last twenty years have already come and gone.

The two European projects are a useful warning as well as a sign of ambition. The Human Brain Project began in 2013 with a budget of about €1 billion and a promise, from its founder Henry Markram, of a simulated human brain. Within two years it was widely seen as mismanaged and its claims as overblown, Markram was asked to step down, and the project was refocused on building research infrastructure (Wikipedia: Human Brain Project). The lesson the field took was not that uploading is impossible, but that it has to be built from the bottom up: map a small brain completely, prove the model behaves like the real one, then scale.

Not in the science fiction sense, but brain-computer interfaces are now doing things that would have sounded impossible ten years ago. In August 2023, researchers at UC San Francisco and UC Berkeley reported that a woman who had lost the ability to speak after a stroke could talk through a digital avatar at about 80 words per minute, compared with 14 on her previous device, in a voice rebuilt from a recording of her wedding speech (University of California). It was the first time speech and facial expressions had been synthesized from brain signals.

In June 2025, a team at UC Davis went further. Their implant, 256 electrodes in the part of the brain that controls speech, turned a man with ALS's brain signals straight into audible speech with a delay of about 40 milliseconds, roughly the lag you hear when you speak yourself. He could change his intonation to ask a question and even sing simple melodies, and his synthetic voice was cloned from recordings made before his illness. Listeners understood about 60% of his words, compared with 4% without the implant (ScienceDaily).

Other work reads the mind from outside the skull. In 2023, researchers at the University of Texas at Austin used brain scans and an AI language model to reconstruct the gist of stories a person was hearing or imagining, though not word for word, and only with the person's cooperation (UT Austin). Elon Musk's Neuralink implanted its first human patient in January 2024 (Irish Times). None of this is uploading. But every one of these systems turns brain activity into something a computer can store and use, which is the bridge any upload would have to cross.

There is a second road to a digital mind that does not start with the brain at all. Instead of copying neurons, it copies behavior: how a person talks, what they believe, how they make decisions. In 2024, researchers at Stanford interviewed 1,052 people for two hours each and built AI agents from the interviews. The agents answered a standard social survey about 85% as accurately as the real people did when asked the same questions again two weeks later (Stanford HAI).

That result matters because it shows how much of a person can be captured from what they say. Two hours of conversation was enough to predict much of how someone would answer questions about their values and views. A Persona built over months of conversations, stories and recorded memories holds far more. A Persona is not a copy of a brain, and we never pretend otherwise. A Persona captures the part of a person that shows up in how they talk, what they remember and what they care about.

The two roads may meet. A future emulation built from a scanned brain would still need to be checked against something: how did this person actually talk, think and remember? The most detailed record of a person's own words, memories and voice is exactly what that check would need. That is one reason we think building a record of yourself now is not a lesser form of uploading, but the first part of it.

This is the oldest question in the field, and the hardest. Suppose a perfect scan of your brain is run on a computer while you are still alive. The digital mind remembers your childhood and thinks it is you. But you are still sitting in the scanner. Most people's intuition is that the copy is a new person who shares your memories, not you. If that is right, then scan and copy uploading creates a twin rather than moving you anywhere.

Gradual replacement is designed to get around that problem. If your neurons are replaced one at a time while you stay awake, there is never a moment when you stop and a copy starts, much as a ship that has every plank replaced over the years is still, by most accounts, the same ship. Some philosophers think continuity of that kind is what personal identity really is. Others, including some who see the self as a useful illusion, argue that the question of whether the upload is really you has no deep answer at all (Wikipedia: Mind uploading).

There is also the question of consciousness. Most neuroscientists think the mind comes from what the brain does, which would mean a faithful enough model would be conscious too. A minority, including the physicist Roger Penrose, think consciousness may depend on something about biological brains that a computer cannot reproduce. Because nobody can yet measure consciousness from outside, Sandberg has proposed a cautious rule: treat any emulation as if it could have the same mental life as the original, and treat it accordingly. We agree, and it is why we think the rules around a digital version of a person matter as much as the technology.

The first is detail. Nobody yet knows how much detail an upload needs. A wiring diagram may not be enough: the strength of each connection, the chemicals that change how neurons behave, the support cells around them and the electrical properties of each cell may all matter. The 2008 roadmap set out several possible levels, from a rough network of neurons down to individual molecules, and the right level is still an open scientific question. The fruit fly results are encouraging because a model built mainly from wiring predicted real behavior, but a fly is not a person.

The second is scanning a living brain. Today's mapping methods need preserved tissue cut into thin slices, which means they can only be used after death. No current technology can read every synapse in a living human brain. That is why some people interested in uploading have turned to brain preservation, a close relative of cryonics, keeping a brain in a state where its structure could be read later. In 2018 a startup called Nectome proposed preserving the brains of terminally ill people with a chemical fixative and deep cooling, in a procedure that would itself be fatal, and it drew sharp criticism from scientists and ethicists (MIT Technology Review).

The third is computing power, and the fourth is validation: proving that an upload behaves like the person, not just like a person. Computing power is the problem most likely to solve itself, because AI is driving the biggest build-out of computing in history. Validation is harder, and it is where a detailed record of how someone actually spoke, thought and remembered becomes valuable. None of these obstacles is known to be impossible to overcome. They are hard engineering and hard science, which is the kind of problem that tends to yield to time and money.

Nobody can give an honest date, and anyone who gives you a precise one is guessing. Kurzweil has long pointed to around 2045. The researchers behind the State of Brain Emulation Report focus instead on the next milestones: a whole mouse brain mapped, a mouse brain emulated and validated, and better ways to record activity across a whole brain (arXiv). Each of those is a decade-scale project with real funding behind it.

What we can say is how fast things are moving. It took almost forty years to get from a worm with 302 neurons to a larval fly with 3,016. It took one more year to get to an adult fly with nearly 140,000, and less than two years after that, a company reported a whole fly brain driving a virtual body. Mapping costs are falling by orders of magnitude, AI is speeding up the slowest steps, and some of the richest people in the world now openly say they want this. That is why our view is that uploading a mind is a matter of when, not if.

More people than you might think, including some of the most powerful people in technology. According to Mustafa Suleyman, Larry Page told him over lunch in 2015 that by the end of the century he expected to be "exploring the galaxies at the head of a von Neumann probe," a self-copying spacecraft, which Suleyman took to mean leaving the human species for a successor species (The Builders (iHeart)). Our guides explain the von Neumann probe and why Page called Elon Musk a speciesist.

Kurzweil, whom Page personally hired at Google in 2012, has predicted digital immortality for decades. In 2018, MIT Technology Review reported that Sam Altman was among the people who had paid a deposit to join Nectome's waiting list. And a long line of philosophers and futurists, from Moravec onward, has argued that a mind is a pattern that can be moved. What was a fringe position in the 1980s is now being discussed openly by the people who run the world's largest technology companies, and our guide to the billionaires trying to live forever follows the money.

Fiction has shaped how most people picture uploading. Black Mirror's San Junipero, from 2016, imagined a digital afterlife where people could choose to live on after death. The Amazon series Upload imagined a commercial afterlife with tiers of service, and Dennis E. Taylor's Bobiverse novels, which began with We Are Legion (We Are Bob) in 2016, follow a software engineer whose mind is uploaded into a self-replicating spacecraft, which is strikingly close to what Page reportedly wants.

Most fictional uploads are cautionary tales: minds trapped, copied without consent, owned by companies or edited by strangers. Those stories are worth taking seriously, because they are about control, not technology. Who owns the digital you? Who can change it? Who decides when it is switched on or off? Those are exactly the questions we have built our technology around.

It starts with a record. Whatever form uploading finally takes, scanned brain, gradual replacement or something nobody has thought of yet, it will need to be checked against the real person: how they spoke, what they remembered, what they believed, how they sounded. A brain scan cannot tell you any of that on its own. A detailed record made by the person, in their own words and their own voice, can.

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. You build your Persona by talking, answering questions and telling stories, by voice or by typing, and your Persona learns from every conversation. Building is free: you get 25 memories with no card and no time limit, and your free build never expires. Paid plans start at US$14.99 a month.

The fiction gets one thing right: control matters more than technology. 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.

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.

• What Is a Von Neumann Probe? Larry Page's Galaxy Plan: the self-copying spacecraft at the center of Page's plan.

• 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.

• 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.

The science comes from the NIH, Nature, Live Science, OpenWorm, ScienceDaily on the Harvard and Google map, the Allen Institute, PNAS, the State of Brain Emulation Report 2025, Asimov Press on E11 Bio and The Decoder on Eon Systems. Brain-computer interfaces come from the University of California, ScienceDaily, UT Austin and the Irish Times, and the Stanford study from Stanford HAI.

History and philosophy come from Wikipedia on mind uploading, Hans Moravec and the Human Brain Project, and the 2008 Whole Brain Emulation roadmap. Nectome comes from MIT Technology Review, and Larry Page's plan from The Builders. We checked every source on October 30, 2026.

Larry Page, Mustafa Suleyman, Elon Musk, Ray Kurzweil, Sam Altman and the research groups named on this page have no connection to Afterlife.ai™, and nothing on this page suggests that any of them uses or endorses our service.