Mind uploading remains firmly in the realm of science fiction, but the underlying research programs that would be necessary to make it real have advanced substantially in the past decade. No one has uploaded a mind, and no one is close. What has changed is that several of the prerequisite technologies, particularly brain mapping, neural recording, and computational modeling, have hit milestones that were considered far-off just a few years ago. The gap between where we are and where we’d need to be is still enormous, but it’s now possible to describe it in concrete terms rather than hand-waving.
What Mind Uploading Actually Requires
The idea sounds simple in principle: scan a brain at sufficient resolution, build a computational model of everything that scan reveals, and run it on hardware powerful enough to reproduce the brain’s behavior in real time. In practice, that breaks down into at least three separate grand challenges. You need to map every neuron and every connection between them. You need to understand how those connections produce thought, memory, and behavior, not just catalog them. And you need computing power that can simulate all of it at once. A 2025 reassessment of the field, building on a foundational 2008 roadmap by Sandberg and Bostrom, organizes these challenges into three pillars: recording brain function, mapping brain structure, and computational emulation.
Mapping Brains at the Cellular Level
The field of connectomics, which aims to produce complete wiring diagrams of nervous systems, has had its biggest wins in small organisms. In 2023, researchers published the first complete synaptic-resolution connectome of an entire insect brain, that of a fruit fly larva. The map covered 3,016 neurons and 548,000 synaptic connections, and it revealed structural features that nobody had predicted: pervasive multisensory integration, highly recurrent circuits, and architectural motifs that resemble those used in modern deep learning systems.1PubMed Central. The connectome of an insect brain That was a larva. By 2024, a team completed the first whole-brain connectome of an adult fruit fly, a far more complex animal with over 130,000 neurons and millions of synaptic connections.2Nature. Network statistics of the whole-brain connectome of Drosophila
These are genuinely impressive achievements. They also illustrate the scale of the problem. A fruit fly brain has roughly 130,000 neurons. A human brain has about 86 billion, connected by an estimated 100 trillion synapses. That is not a difference you close by waiting for incremental progress; it’s a difference of roughly six orders of magnitude in neuron count alone, and the wiring is far more tangled. Current approaches to human brain connectivity rely on neuroimaging techniques like diffusion MRI and tractography, which work at resolutions of millimeters, not the nanometer-scale resolution that electron microscopy achieves in insect brains. Bridging that gap requires new methods, and researchers have flagged that machine learning and simulation will likely be needed to fill in where experimental data can’t yet reach.3Science. Scale matters: The nested human connectome
Recording What the Brain Is Doing
Mapping the wiring is only half the problem. You also need to know what signals are flowing through those wires, and current recording technology faces hard trade-offs between coverage and precision. Functional MRI offers whole-brain coverage but resolves activity at the scale of millimeters, far too coarse to capture what individual neurons are doing. EEG captures activity across the whole scalp but with poor spatial resolution, because the signal from any one electrode blends contributions from vast numbers of neurons.4PubMed Central. Exploring the Frontiers of Neuroimaging: A Review of Recent Advances in Understanding Brain Functioning and Disorders
Implanted electrode arrays get much closer to the action. Recent work has pushed the channel counts of these devices sharply upward. One team developed a minimally invasive cortical array with 1,024 channels, including nearly a thousand recording electrodes at 50-micrometer scale.5Nature Biomedical Engineering. Minimally invasive implantation of scalable high-density cortical microelectrode arrays for multimodal neural decoding and stimulation Another group built a wireless brain-computer interface that packs 65,536 electrodes onto a single flexible chip just 50 micrometers thick, with the ability to record from 1,024 channels simultaneously.6Nature Electronics. A wireless subdural-contained brain–computer interface with 65,536 electrodes and 1,024 channels These are remarkable engineering feats. But even 65,000 electrodes cover only a small patch of cortex. Scaling electrode-based recording to the whole brain, with billions of neurons firing in concert, remains an unsolved problem by many orders of magnitude.
Decoding Thoughts from Brain Scans
One area that has generated a lot of public excitement is neural decoding, the ability to reconstruct what a person is seeing or thinking from their brain activity. Using fMRI data and generative AI models, researchers have shown they can take brain scans recorded while a person looks at an image and then produce a new image that matches the original on a semantic level. The approach works by training a model to map fMRI signals onto the kind of feature space used by image-recognition neural networks, then feeding those features into an image-generation model.7arXiv. Semantic Brain Decoding: from fMRI to conceptually similar image reconstruction of visual stimuli More recent work continues to refine these pipelines, using optimized AI architectures to analyze which brain regions contribute most to visual and semantic processing.8PubMed. Optimized AI-based neural decoding from BOLD fMRI signal for analyzing visual and semantic ROIs in the human visual system
This is striking work, but it’s worth being clear about what it does and doesn’t demonstrate. These systems reconstruct the gist of a visual experience, not a pixel-perfect replay. The generated images look thematically similar to what the person saw, capturing things like “a dog on grass” or “a building,” not the exact dog or exact building. And the approach works only for visual perception, one of the brain’s most well-studied and spatially organized functions. Reconstructing an abstract thought, a memory, or a sense of self from brain scans is a fundamentally different and far harder problem. Neural decoding shows that brain activity encodes information in patterns we can partially read. It does not show that we can capture anything close to the full content of a mind.
The Computing Problem
Even if you could perfectly map and record a human brain, simulating it would demand computing resources that dwarf anything currently available. Researchers studying the computational costs of brain tissue simulations have found that even simplified neuron models, when scaled to brain-sized networks, quickly become bottlenecked by memory bandwidth rather than raw processing speed. The choice of how synapses are modeled turns out to matter more than the level of anatomical detail in the neuron, which means shortcuts in biological realism don’t necessarily buy you much in computational savings.9PubMed Central. Understanding Computational Costs of Cellular-Level Brain Tissue Simulations Through Analytical Performance Models
One team has built a platform called Digital Twin Brain that can simulate spiking neural networks at roughly whole-human-brain scale using personalized structural data from brain scans. Their scaling experiments revealed that human brain simulation is not primarily a number-crunching problem; it’s a communication and memory-access problem. The sheer volume of data that needs to shuttle between simulated neurons, reflecting the brain’s sparse but heterogeneous wiring, overwhelms conventional computing architectures.10arXiv. Digital Twin Brain: a simulation and assimilation platform for whole human brain This suggests that even if Moore’s Law continued at its historical pace (which it hasn’t for clock speeds), simply throwing more conventional processors at the problem wouldn’t be enough. New computing architectures, possibly including neuromorphic chips that process information using spikes rather than traditional clock cycles, may be needed. One such chip, called Speck, consumes just 0.42 milliwatts at rest by processing signals asynchronously, the way biological neurons do.11Nature Communications. Spike-based dynamic computing with asynchronous sensing-computing neuromorphic chip Neuromorphic hardware is promising for certain tasks, but scaling it to emulate an entire brain is still speculative.
Preserving a Brain for Later Scanning
One practical question that comes up in any mind uploading discussion is whether you could preserve a brain well enough today that future technology might be able to scan and emulate it. This is the bet that cryonics organizations are making. The science here is mixed. A technique called aldehyde-stabilized cryopreservation (ASC) has shown that rabbit brains can be preserved with ultrastructural detail that is indistinguishable from fresh tissue under electron microscopy. Cell membranes remained intact, nuclear envelopes were clearly defined, mitochondria looked normal, and myelin sheaths were tight.12Cryobiology. Aldehyde-stabilized cryopreservation
The catch is that ASC relies on chemical fixation before freezing, a process that kills the tissue. It preserves structure beautifully but makes any biological revival impossible. And even with the best preservation techniques, brain tissue remains vulnerable to damage during cryopreservation, particularly from ice crystal formation as water transitions from liquid to solid.13PubMed Central. Cryopreservation of brain cell structure: a review Whether the structural information preserved by these methods would actually be sufficient for a future upload, whether it captures everything that matters about how the brain computes, is an open question that hinges on assumptions about which biological details carry the information that constitutes “you.”
Organoid Intelligence and Biological Computing
An unexpected thread of research that’s tangentially relevant to mind uploading is the growing field of organoid intelligence. Brain organoids are three-dimensional clusters of human brain cells grown in a lab. Unlike flat cell cultures, these organoids self-assemble into structures that partially mimic real brain architecture: they develop layered cell patterns, myelinated axons, spontaneous electrical activity, and complex oscillatory behavior. Researchers have proposed using them as a form of biological computing, leveraging the brain cells’ inherent ability to learn, store information, and process inputs.14Frontiers in Science. Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish
Organoid intelligence isn’t mind uploading. Nobody is proposing that you could transfer your consciousness into a clump of lab-grown neurons. But it highlights something important about the computational challenge: biological tissue is extraordinarily efficient at doing the kind of processing that brains do. If silicon-based hardware struggles to emulate a brain because of energy and memory bottlenecks, hybrid biological-digital systems might offer an alternative pathway. It’s early-stage and speculative, but it represents a different angle of approach than the brute-force “scan everything, simulate everything” model that most mind uploading discussions assume.
The Identity Problem That Technology Can’t Solve
Even if every technical challenge were overcome tomorrow, mind uploading would still face a problem that no amount of engineering can resolve: the question of whether the upload is actually you. If a perfect computational copy of your brain is running on a server somewhere, is that copy conscious? Does it have your subjective experience? Or is it a very convincing duplicate that believes it’s you while “you” are gone?
These aren’t just philosophical parlor games. They have practical implications for whether anyone would actually want to be uploaded. The assumption underlying most mind uploading proposals is a principle called substrate independence: the idea that a mind doesn’t depend on any particular physical material, only on the pattern of information processing. If the pattern is right, the thinking goes, consciousness follows regardless of whether it’s running on neurons or transistors. But this assumption has been challenged on energetic grounds. Real-world information processing depends on energy, and energy depends on material substrates. Attention to these energy requirements undermines the straightforward claim that a mind can simply be moved from one platform to another.15Philosophy of Science. Energy Requirements Undermine Substrate Independence and Mind-Body Functionalism If the physical substrate matters to how information is processed, then a perfect structural copy might not produce the same mind.
There’s no experimental way to test this right now because we don’t have a working theory of consciousness that everyone agrees on. Until we understand what generates subjective experience, we can’t know whether a digital copy would have it. This isn’t a problem waiting for better scanners or faster computers. It’s a gap in basic science.
Where Public Attitudes Stand
Public opinion on related questions, while not directly about mind uploading, gives some sense of how society might receive the technology if it ever arrives. Surveys show that most people are skeptical that current AI systems are conscious, with only about one in five believing they are. But views on future AI systems are more divided, with a much larger share accepting or leaning toward the possibility that future systems could be conscious. At the same time, large majorities believe that sentient AI should be treated with respect while also believing AI should remain subservient to humans, a tension that would only intensify if uploaded minds entered the picture.16Alignment Forum. Project Ideas: Sentience and Rights of Digital Minds These conflicting intuitions suggest that the ethical and legal frameworks for mind uploading, questions like who owns an uploaded mind, whether it has rights, whether it can be copied or deleted, are nowhere near ready, and the technology is nowhere near forcing the issue.
How the Field Measures Its Own Progress
The most useful framing for understanding where mind uploading research stands comes from the 2025 State of Brain Emulation Report, which revisited the field’s progress since a landmark 2008 roadmap. The report doesn’t claim that brain emulation is imminent. Instead, it identifies ongoing challenges and outlines strategic priorities, treating the problem as a multi-decade engineering program that requires coordinated advances across neuroscience, computing, and preservation technology.17arXiv. State of Brain Emulation Report 2025
The honest summary is that each of the three pillars, recording, mapping, and emulation, has seen genuine progress since 2008, but none is anywhere close to the level needed for a human mind upload. Connectomics has gone from mapping a worm’s 302 neurons to mapping an adult fly’s 130,000, a huge leap that still leaves the human brain’s 86 billion neurons essentially untouched at the resolution that matters. Recording technology can now capture thousands of channels simultaneously but needs to scale by many orders of magnitude. Computational modeling has grown more sophisticated and more aware of its own bottlenecks, but whole-brain real-time simulation on today’s hardware remains out of reach.
The trajectory matters, though. Each of these areas is advancing faster than linear trends would have predicted a decade ago, driven partly by AI-assisted analysis, partly by better hardware, and partly by the sheer momentum of investment in neurotechnology for medical applications like brain-computer interfaces. Mind uploading research doesn’t exist as a single funded program. It’s the sum of many adjacent fields, each solving its own problems, whose solutions might someday converge. Whether they actually will, and whether the result would be something anyone would recognize as mind uploading, depends on answers to questions that haven’t been asked yet, much less answered.