Mind Uploading

Mind uploading proposes to reconstruct a particular mind by mapping a brain in enough detail to simulate it. That specificity is the whole of the proposal: the target is a person, and success would have to be judged against that person rather than against a plausible mind in general. The word doing all the work in the framing — that a person could be transferred to a substrate — is the one nobody has defined operationally.

The proposal decomposes into three problems with separate literatures and radically different maturity. Mapping is an imaging and reconstruction problem making real, fast, fundable progress. Simulating what was mapped is a modelling problem where the field's own flagship report concedes that the necessary parameters are not visible in the images at all. Deciding whether the result is the same individual is not an engineering problem and no imaging resolution will touch it. This brief owns all three and imports the fourth question — whether a completed emulation would have experiences — from the consciousness briefs, because it is theirs and not this one's. Section 8 states that seam in both directions.

Established Connectomics is genuinely succeeding, at a pace that surprised the field, and the numbers are the argument in both directions. Caenorhabditis elegans, 302 neurons, has been mapped since the 1980s. The adult female Drosophila brain was published in Nature in October 2024: 139,255 neurons reconstructed (118,501 intrinsic to the brain), 54.5 million synapses with 50 million analysed above a five-synapse threshold, more than 8,400 cell types, at a reconstruction F1 of 99.2% — and approximately 33 person-years of manual proofreading. Established Read the exclusion list, because it is the rest of this brief in one paragraph: electrical synapses are absent and await higher-resolution electron microscopy; synaptic strength is absent; neuropeptides and neuromodulation are minimally treated; only a small fraction of glia were proofread, though glia are about 13% of cell bodies; and it is one female fly, with mushroom-body connectivity varying substantially between individuals.

Established The male fly central nervous system followed on 6 October 2025 — brain and ventral nerve cord in a single volume — and it came from a different group than the female brain. The producers were HHMI Janelia's FlyEM team with the MRC Laboratory of Molecular Biology in Cambridge and Google Research; the female brain came from the FlyWire consortium. Attributing both to one consortium is a common error and it matters, because the two datasets have different completion characteristics. Frontier The accompanying preprint reports 166,691 neurons, 46 million presynapses connected to 312 million postsynaptic densities and 11,691 unique cell types — alongside completion figures of 94% presynaptic and 42% postsynaptic in neuropils, with only 40.1% of synaptic connections having both partners proofread. “A connectome exists” and “a connectome is complete” are different claims and the second one is not yet true anywhere. Established The same dataset gives the field its cleanest dimorphism result: 331 male-specific cell types and 114 dimorphic types, concentrated in higher-order centres (3.4% of the male central brain) rather than the sensorimotor periphery (0.1% of the visual system).

Established In mammals the state of the art is one cubic millimetre, and its stated limitations are the ones to carry. The MICrONS reconstruction of mouse visual cortex, published in Nature in April 2025, covers 1.3 by 0.87 by 0.82 mm, with more than 200,000 segmented cells, about 120,000 neurons, 524 million synapses, 2 petabytes of raw electron microscopy across 26,652 serial sections, and in vivo calcium imaging from about 75,900 excitatory neurons in the same tissue. Established The authors report that axon segmentation is markedly less accurate than dendritic; that a percentage of synapses sit on detached spines, biasing connectivity analysis; that photon scattering degrades the functional recordings with depth; and that “proofreading and analysis remains the largest overall expense in terms of person hours.” The dataset's own explorer records that only 373 neurons have fully clean, completely proofread axons. Established For human tissue the comparable object is H01: 1.4 petabytes at 4 nm resolution, 225 million two-dimensional images from 5,300 sections, tens of thousands of neurons, 130 million annotated synapses and 104 proofread cells, from anonymised epilepsy surgery tissue, released in June 2021. That is one millionth of a human brain.

Frontier The scale gap should be stated once, in numbers rather than adjectives. A mouse brain has roughly 500 times the neurons of a fruit fly; a human brain roughly one million times. The State of Brain Emulation Report 2025 offers the comparison worth anchoring on: mapping a mouse brain is “comparable in scale to a high-resolution reconstruction of Earth.” Established The next real waypoint has a price and a date — Google Research with Harvard, the Allen Institute, MIT, Cambridge, Princeton and Johns Hopkins targeting 10 to 15 cubic millimetres of mouse hippocampus, 2 to 3% of the mouse brain, generating about 25 petabytes over five years, funded at $33 million inside the National Institutes of Health's $150 million BRAIN CONNECTS programme.

Established Capability here is not a smooth extrapolation; it is hostage to a single national funding relationship, and that was demonstrated in May 2025. Two of those grants were terminated — $3.3 million and $1.1 million, $4.2 million in total — as part of the federal action against Harvard. Harvard committed $250 million in bridge funding covering up to 80% of lost grants through June 2026, and collaborators attempted to transfer the awards. Frontier The field's own size makes that fragility structural rather than unlucky: fewer than 500 people worldwide work on brain emulation, and the basic neuroscience underneath runs at roughly $0.5 billion a year, about 1% of the NIH budget. A subject where a $4.2 million cancellation is a material setback is small before it is hard.

Established The binding constraint is not resolution. It is that the parameters a simulation needs are not in the picture, and C. elegans is the proof. The worm's connectome has been available for four decades and there is still no working emulation of it. What is missing is specific and enumerable: synaptic weights; membrane properties including capacitance and dendritic and axonal geometry; firing thresholds; dynamic learning mechanisms; and extrasynaptic signalling. The structural reason none of it comes from electron microscopy is that “electron microscopy has a serious disadvantage” — it works only on dead tissue. Frontier The older sceptical statement adds the sting that matters for identity: nobody can “read the weights off of any individual worm”, so even a perfect worm simulation would not be a simulation of that worm. Frontier The rejoinder in the same discussion is that the work was never funded — a claim that a seven-figure seed would have kept it moving. Underfunding and impossibility look identical from outside, and this brief records the question as genuinely unresolved rather than settling it in the direction that flatters either side.

Established The best counter-evidence is the fly, and it is routinely oversold in both directions. A leaky integrate-and-fire model of the entire fly brain, built from connectivity and predicted neurotransmitter identity across more than 125,000 neurons and 50 million synapses, correctly predicted which neurons respond to sugar and water, which are required for proboscis extension, how aversive and appetitive taste interact, and the antennal grooming circuit — all validated optogenetically. Established What it shows is that connectome plus simple dynamics predicts real behaviour in a small brain, which is a genuine and important result. What it does not show is that the model is the fly: it uses a uniform simplified neuron model with predicted rather than measured neurotransmitters, and it is a model of a fly brain rather than of the animal that was scanned. Established The MICrONS companion result cuts the same way from the other direction: feature similarity rather than receptive-field location predicts fine-scale connectivity — structure carries real functional information — while the authors note that connectivity alone does not determine function and that artificial networks develop the same wiring pattern only after training. Learning shapes the wiring; the wiring does not contain the learning.

Established And the compute estimate is not a number but a sixteen-order-of-magnitude argument. Published estimates of human brain performance span 10¹² to 10²⁸ FLOPS. The named anchors: basic functional capacity below 10¹⁵; a traversed-edges conversion giving 0.9 to 33.7 × 10¹⁶; and the three-level roadmap table at 10¹⁸ for a spiking neural network, 10²² at electrophysiology level and 10²⁵ at metabolome level. Established The spread is not measurement error. It is disagreement about which biological level must be emulated — the same disagreement as the missing-parameter problem, expressed in different units. Anyone quoting a single figure for “the compute needed to emulate a brain” has picked a side in an unresolved argument without saying so. Frontier The 2025 report's own forward projection is specific and modest: emulation of sub-million-neuron organisms — fruit flies, small fish — within a decade, at low hundreds of millions of dollars. Nothing in it projects a human.

Frontier The live scientific question is whether structure suffices, and the two hypotheses are cleanly opposed. Structural sufficiency holds that the connectome plus synaptic weights and molecular state is enough to reconstruct a mind; its evidence is the fly model predicting real behaviour and the like-to-like wiring rule showing structure carries function. Structural insufficiency holds that the necessary parameters are not recoverable from fixed tissue at all; its evidence is forty years of worm connectome with no emulation, and the report's own concession that current recording lacks synaptic weights and neuromodulation. Frontier On the evidence fetched for this brief, insufficiency is currently the better-supported of the two, and the decisive experiment — preserve, read out, then behave, in a small animal — has been proposed and never run.

Established Behind both sits an underdetermination nobody disputes: no one knows which biological level must be emulated. That the disagreement exists is established; the sixteen orders of magnitude are its direct consequence. Speculative Which level is right is unknown, and it will stay unknown until some organism is emulated at some level and validated. No recording programme currently produces the validation data: no organism's full brain has been recorded at single-neuron resolution, larval zebrafish reach about 80% coverage and C. elegans about 50%, with calcium imaging running at 1 to 30 Hz — well below firing rates — for minutes or hours at a time.

Speculative An under-examined hypothesis deserves naming rather than asserting: chaotic divergence. If brain dynamics are sensitively dependent on initial conditions at a level finer than any achievable scan, an emulation diverges from its original immediately even with every structural parameter captured. Speculative No source fetched for this brief develops it, which is why it is flagged at the weakest level that still lets it be stated. What would settle it is measurable: the divergence rate between a small simulated nervous system and its biological original, which nobody has measured because nobody has the pair.

Frontier The sleeper problem is validation, and it is worse than the scan problem. Even a working human emulation would have no agreed test confirming it is an accurate emulation of that person. Behavioural similarity is the obvious candidate and it fails immediately: a generic emulation of a species member, personalised only by gross anatomy, would pass any behavioural test a grieving family could design. Frontier Success and unverifiable failure are behaviourally identical, and no proposal in the fetched literature addresses it.

Frontier The identity question has four live positions and no dominant one, and that is the state of the art rather than a gap in this brief. The Stanford Encyclopedia's assessment is that there is “no consensus or even a dominant view” on personal identity, which leaves uploading philosophically underdetermined. Psychological continuity says an upload preserving your psychology satisfies the persistence conditions — and the fission problem is fatal to the simple version, because two branches each psychologically continuous with you cannot both be you while identity is transitive and they are not identical to each other. Established The two escapes are the multiple-occupancy view (you were always two coinciding persons) and the non-branching view (bilateral survival is death), and the encyclopedia notes the second one's absurdity directly: keeping half your brain running is normally sufficient for survival, so “why then would you not survive if the other half too were kept functioning?”

Frontier The remaining positions divide sharply and two of them refuse the question rather than answering it. Animalism and brute-physical views say uploading fails outright: no organism persists, the original body remains, and the upload is a new entity. Parfit's revisionism says identity is not what matters — psychological continuity grounds rational self-concern whether or not the identity facts cooperate, and under fission you should care about both branches. Frontier This is the position most often invoked to rescue uploading, and it rescues it by conceding the point: it says you do not survive and that you should stop caring whether you do. Frontier Cappuccio's embodied-mind objection is independent of all of them: the incompatibility has “nothing to do with” functionalist or computationalist prerequisites, and the flaw is presuming that minds can be “individuated and numerically identified while being reduced to immaterial formal patterns” at all. On that reading “transfer” has no referent, and the engineering question never becomes well-posed.

Frontier Gradual replacement is the philosophically preferred route and it is weaker than its popularity suggests. Corabi and Schneider's objection is that gradual replacement still yields a functional isomorph at some moment, producing the same spatiotemporal discontinuity between brain and computer; and the multi-upload scenario shows psychological continuity cannot be sufficient for identity, since two uploads in distinct systems are distinct from each other while sharing identical continuity, making continuity “non-causally dependent on extrinsic factors.” Frontier Mogensen's vagueness-and-holism attack on the fading-qualia argument applies here with special force, because gradual uploading is precisely the fading-qualia scenario run as an engineering plan. Speculative Substrate-dependent theories bar the whole programme regardless of fidelity, and that verdict is imported rather than adjudicated here — see section 5.

Established The first bottleneck is proofreading, not imaging, and the field says so in its own limitation sections. Imaging a cubic millimetre produces petabytes; turning that into a usable reconstruction is the largest expense in person-hours, and the resulting completeness figures are the ones to quote: 373 fully proofread axons out of about 120,000 neurons in the mouse volume, 104 proofread cells in the human millimetre, 40.1% of connections with both partners proofread in the male fly, and 33 person-years of manual work for the female fly brain. Frontier Automation is improving and is the field's main lever; nothing fetched here suggests it removes the bottleneck within the current decade.

Established The second is that electron microscopy works only on dead tissue, which decides what a scan can ever contain. Synaptic weight, membrane properties, firing thresholds and extrasynaptic signalling are dynamic quantities; the imaging modality that provides the structure destroys the state. Frontier That is a bottleneck with a named shape rather than a general difficulty: a method for recovering weights from fixed tissue, or a demonstration that behaviour is robust to getting them wrong, would move the whole subject, and it is recorded as a typed constraint in section 13.

Established The third is volume, and it is six orders of magnitude. A cubic millimetre is the largest densely reconstructed mammalian volume; a human brain is roughly a million of them; the funded next step is 2 to 3% of a mouse. Established The fourth is money and headcount — fewer than 500 people, about 1% of the NIH budget for the underlying science, and a $4.2 million cancellation that materially set the field back in 2025. Frontier Neither of those is a discovery problem, which is why this brief treats “hard” and “small” as separate diagnoses that the usual framing merges.

Established The fifth is that no compute target can be set, because the level-of-detail question is open. Between 10¹² and 10²⁸ FLOPS there is no engineering plan, only a choice of philosophy. Speculative A single validated emulation at any level of any organism would collapse most of that range, and none exists.

Established This brief waits on one result another brief on this map is producing, and it is a fidelity result rather than a throughput result. Brain Preservation owns the question of whether a preserved brain still contains what a mind is made of. Aldehyde-stabilised cryopreservation demonstrably preserves ultrastructure across a whole large-mammal brain; whether memory and identity are in what was preserved is the open question, and it is that brief's, not this one's. Everything in the preserve-now-read-later programme — and the entire commercial sector built on it — is downstream of an answer only that brief can give. A typed depends-on edge is recorded accordingly, and it matches the enabling relationship that brief already states in prose when it names a substrate for whole-brain emulation among the things it would make possible.

Established What this brief does not claim from it is the readout capacity, and the distinction is deliberate. High-throughput electron microscopy at whole-brain scale is recorded on that brief's own tech tree as an unmet industrial constraint, alongside century-scale custodial continuity. Those are its constraints, not results it produces, so restating them here would double-count a single gap and make the map look better connected than it is. Frontier The honest form is that this brief inherits the throughput problem without owning it, and that neither brief is in a position to solve it.

Frontier Three further inputs are imported and adjudicated elsewhere. From Consciousness Research, whether a completed emulation would have experiences at all; from Integrated Information Theory, the verdict that a functional duplicate on conventional hardware would not be conscious whatever it does; from Orch OR, a stronger prohibition on the same lines. Established None of those is recorded as a typed dependency, because this brief's central claim survives every one of their answers: consciousness science can tell you the emulation is a mind, and nothing in it can tell you it is your mind. Frontier Cryonics supplies the vitrification and revival literature, and Memory Engineering supplies the engram work that would have to say what a memory physically is before anyone could check whether a scan captured one.

Frontier The decisive experiment is small, cheap by the standards of this field, and has never been run: preserve, read out, then behave. Take a small animal with a known behavioural repertoire, preserve it by the best available method, reconstruct the connectome and whatever molecular state survives, build the emulation, and test whether it reproduces the individual's learned behaviour rather than its species' generic behaviour. Established Every component exists. The whole-brain connectome exists for the worm and the fly; preservation is demonstrated at large-mammal scale; behavioural assays are standard. Frontier What does not exist is the funded programme to run them in sequence, and its absence is the single most informative fact about the field's priorities.

Frontier Second: attack the weight problem directly. Two shapes are testable. One is a method for inferring synaptic strength from fixed-tissue morphology — spine geometry, active-zone size, vesicle counts — validated against paired electrophysiology in the same animal. The other is a robustness study: perturb the weights in the fly model and measure how far behavioural prediction degrades. Established The second is runnable today with published data and would tell the field how much of the missing parameter set actually matters, which is currently argued rather than measured.

Speculative Third: measure divergence. Run the best available emulation of a small nervous system alongside the animal it was derived from under matched stimulation, and measure how quickly the trajectories separate. That number would settle the chaotic-divergence hypothesis in either direction and nobody has produced it. Frontier The larval zebrafish, at roughly 80% whole-brain recording coverage, is the closest available preparation.

Frontier Fourth: finish the small organisms before arguing about the large one. The 2025 report's own projection is sub-million-neuron emulation within a decade at low hundreds of millions of dollars, and the fly is the natural target because both the connectome and the optogenetic validation toolkit already exist. Established A validated fly emulation would be the first evidence in the subject's history that the structure-to-function bridge holds for a whole brain, and its failure would be equally informative.

Frontier Fifth, and institutional rather than scientific: an independent audit of what preservation providers claim. The sector sells a service whose value rests entirely on a future readout, and the one time a university examined the premise closely it terminated the contract. A standing technical assessment, published, of what each provider's method demonstrably preserves would cost almost nothing and does not exist.

Established The best-validated preservation result is real and its own disclaimers are the important part. A commercial laboratory won the Brain Preservation Foundation's Large Mammal Prize in March 2018 by preserving an entire pig brain's connectome — on the order of 150 trillion synaptic connections — through simulated centuries-long cold storage, verified after rewarming by three-dimensional electron microscopy and independent judging. Established The Foundation stated in the same announcement that no pig or pig brain was revived, that no claim is made that memory, identity or consciousness was preserved, that the readout is something “future technology may be capable” of, and that one laboratory demonstration is “insufficient to address the quality-control measures that should be expected of any procedure applied to humans.” Frontier When the prize-awarding body disclaims the inference the prize exists to enable, that disclaimer carries unusual weight.

Established And the procedure is terminal, which is where the field's clearest documented failure of commercial honesty begins. Nectome pitched exactly this as a consumer service: a six-hour perfusion of a terminally ill patient under anaesthesia, described by MIT Technology Review as “100 percent fatal”, with a $10,000 refundable deposit, 25 people on the waitlist including a prominent technology executive, and about $1 million raised including $120,000 from an accelerator. Established Three weeks later the MIT Media Lab cancelled the research contract, citing “the scientific premises underlying the company's commercial plans, as well as certain public statements that the company has made,” and issued the sentence this brief treats as the field's most useful quotation: “Neuroscience has not sufficiently advanced to the point where we know whether any brain preservation method is powerful enough to preserve all the different kinds of biomolecules related to memory and the mind.” A Karolinska researcher called the arrangement unethical on the ground that “the company is based on a proposition that is just false.” Frontier The science was not the problem — aldehyde-stabilised cryopreservation is real. The gap between the science and the sale was the whole story.

Established The commercial ecosystem now sells against that premise at scale, and the figures come from interested parties. As of 2025: roughly 500 to 650 people cryopreserved worldwide; 5,000 to 6,000 members globally; prices from about $28,000 at the cheapest provider to $200,000 and $220,000 at the two best known; and more than $65 million disclosed in 2025 funding, including over $100 million into a single new venture backed by two prominent funds. Frontier Patient counts by provider run 252, 270, 103, 29, 20 and 4 on the largest provider's own comparison table — and an independent tally gives 248 rather than 252 for the same period and provider. A four-patient discrepancy in a self-reported census is small, and it tells you how much the sector's own numbers should be trusted. Established Nobody has been revived, and nothing has been read out.

Frontier The engineering requirement nobody has specified is the readout pipeline itself. Preservation produces fixed tissue at whole-brain scale; the reconstruction machinery that exists produces a cubic millimetre in petabytes over years, with proofreading dominating cost. A human-scale readout is therefore an imaging, storage, automated-segmentation and quality-control programme roughly a million times the largest one ever attempted, and no published design targets it. Established That constraint is recorded on Brain Preservation's tech tree as an unmet industrial capacity, and is inherited rather than restated here.

The seam with the consciousness-science briefs runs in both directions and is unusually clean. Whether a completed emulation would have experiences is not this brief's question. It is decided by whichever of the theories surveyed in Consciousness Research is right; Integrated Information Theory says a functional duplicate on conventional hardware would not be conscious whatever it does; Orch OR rules it out entirely. This brief's job is to say that the answer is imported, and from where — and then to make the point those briefs cannot make: even granting that the emulation is conscious, the identity question is untouched. Consciousness science can tell you the emulation is a mind. Nothing in it can tell you it is your mind. Running the other way, when this brief says a substrate-dependent theory bars the programme, it cites and does not adjudicate; the demarcation dispute and the decoherence objection belong to those briefs.

Sideways, into the preservation briefs. Brain Preservation owns preservation chemistry, aldehyde-stabilised cryopreservation, storage and the tissue-donation ethics; Cryonics owns vitrification, nanowarming and revival. This brief cites both and covers neither. What it uniquely owns at that seam is the consumer-protection question: preservation is sold on the premise that uploading may later be possible, and this is the brief where that premise is assessed. Upward: the companion Digital Minds brief in this category owns moral status and welfare, so everything about what would be owed to an emulation, once it exists, is that brief's. This one stops at “is it you.” The companion Synthetic Consciousness brief owns minds built rather than copied, and the distinction is exactly the token-versus-type distinction this brief turns on.

Also within this map: Machine Consciousness, whose assessment problem an emulation would inherit in full; Memory Engineering, which owns what a memory physically is; Brain-Computer Interfaces and Neural Interfaces, the only live route to reading a living brain at all; Biological Computing, which owns neural cultures as substrates; and Biological Immortality, the competing answer to the question uploading is usually asked in service of.

The Institute's minds programme's treatment covers the same slot at about a third of this depth and remains the shorter route in. It predates the proofreading-completeness figures, the compute-estimate spread, the funding termination and the identity literature assembled here, and this brief supersedes it as the slot's coverage.

Outside the map: connectomics and electron microscopy as instrument disciplines; computational neuroscience, which supplies the modelling; the metaphysics of personal identity, which supplies the question no instrument answers; and consumer-protection law, which has not yet been asked about a service sold on a future readout.

Established The field's institutional facts are three, and all of them are about size. Fewer than 500 people worldwide work on brain emulation. The basic neuroscience underneath is funded at about 1% of the NIH budget. And the largest coordinated scale-up on the record — a $33 million multi-site award inside a $150 million programme — lost two grants totalling $4.2 million in May 2025 to a federal action that had nothing to do with the science, with an institutional bridge covering up to 80% of losses only through June 2026. Frontier A capability that a $4.2 million cancellation can visibly slow is not a capability on a smooth curve.

Established The second institutional fact is that the sector selling on this premise is larger and better capitalised than the science. More than $65 million disclosed in 2025 preservation funding, over $100 million into one venture, thousands of members and hundreds of patients — against a research field of a few hundred people. Frontier Membership and funding figures indicate a market, not progress, and the two are routinely presented together.

Frontier The third is that nobody polices the premise. The one documented intervention was a university terminating a research contract in 2018 and stating that neuroscience does not know which biomolecules matter for memory. That was an institution protecting its own name, not a regulator protecting a consumer. Speculative No consumer-protection framework in any jurisdiction addresses a service whose delivery is contingent on a scientific result that may never arrive, and this brief records that as a typed constraint rather than as an accusation.

Established Interested parties are everywhere here and the direction of interest varies. The Brain Preservation Foundation is an advocacy body that awarded the prize and then disclaimed the inference — interest running against the finding, so the disclaimer is weighted up. The 2025 report is funded by a venture firm and two private organisations, and it argues the field is underfunded, which is the conclusion its funders would like. Provider patient counts and prices are self-reported. The strongest sceptical statement in the record comes from a university that had already taken the money. Read each accordingly.

Established The sharpest ethical fact in this subject is that the best-developed preservation route requires killing the patient. Aldehyde-stabilised cryopreservation is a perfusion of a living brain; performed on a person it is a terminal procedure, and the company that proposed selling it said so. Frontier That is not a reason to prohibit it — assisted-dying frameworks exist and the population is terminally ill by construction — but it makes the accuracy of the offer the entire ethical question, because the purchaser cannot be compensated for a mistake.

Established And the offer as made is not supported by the evidence behind it. The prize-awarding body states no claim that memory, identity or consciousness was preserved. A university stated that neuroscience does not know which biomolecules related to memory and mind a method would need to preserve. Both statements come from parties positioned to say the opposite. Handwave The marketed continuity claim — that preserving structure preserves the person — is the step done by assertion, and it is the step everything sold in this sector rests on.

Frontier The non-destructive case is the one that dissolves the framing and needs no philosophy at all. If the scan does not destroy the original, the original walks out of the scanner. Two beings then exist and both have equal claim. Established The continuity claim in that case is not contested; it is simply false, given only the stipulation. Frontier A destructive scan does not repair this. It removes the one party positioned to dispute the outcome. Everything in the identity literature is downstream of noticing that the destructive case is not better, only quieter.

Frontier A fourth question arrives before any of that, and it is a tissue-donation question. The human reconstruction that exists came from anonymised epilepsy surgery tissue. Scaling human connectomics means an increasing supply of human brain tissue under consent frameworks written for pathology rather than for reconstruction and publication of a person's wiring. Speculative Nothing fetched here suggests that the consent architecture has been revisited, and it is a live question well before any of the uploading questions are.

Frontier And if an emulation ever ran, the welfare question would arrive before the identity question was settled. A system reproducing a person's behaviour, whose fidelity nobody can verify, is exactly the entity the companion Digital Minds brief is about — a candidate welfare subject with no assessor and no forum. Speculative The order of arrival is the problem: the obligations would be live while the metaphysics was still open.

Frontier The finding that cuts hardest against the framing is that a scan may recover a type rather than a token. Fine-scale wiring in mouse visual cortex follows a like-to-like rule that artificial networks reproduce merely by being trained on a similar task — the connectivity encodes what the circuit is for, and it arises from learning. Combined with the impossibility of reading weights off any individual animal, this points somewhere uncomfortable: connectomics may be on track to recover the architecture of a mouse's visual cortex and never to recover this mouse's. Speculative Scan-and-simulate would then produce a generic member of a species, personalised only to the extent that gross anatomy is idiosyncratic. That is a real scientific achievement and it is not the thing the framing promises.

Frontier The second civilizational consequence is that a whole preservation sector is already priced on the opposite assumption. Hundreds of patients, thousands of members, nine-figure venture funding, and a value proposition contingent on a readout nobody has demonstrated at any scale in any organism. Established If structural insufficiency holds, that capital has purchased tissue in freezers. Frontier If structural sufficiency holds, it has purchased the only bridge available. The evidence currently favours the first and does not settle it, which is exactly the situation in which honest disclosure matters most and is least commercially attractive.

Speculative The third is that success would not obviously be recognisable. There is no agreed test that an emulation is an accurate emulation of a particular person, and none proposed. A civilization that achieved uploading and could not verify it would face a category of claim — that the deceased is running — that is unfalsifiable and enormously consequential. Handwave Speculating about how such a society would adjudicate it is storytelling, and it is flagged as such.

Frontier The terminal position is that the three problems have three different answers and only the merged framing makes them look like one project. Mapping: succeeding, slowly, at a scale six orders of magnitude short, and hostage to a small funding base. Simulating: blocked by parameters that the imaging modality cannot contain, with the compute requirement unbounded across sixteen orders of magnitude. Identity: not an engineering problem, with no dominant philosophical view, and with the non-destructive case showing the continuity claim to be false by stipulation. Established Progress on the first is real and does not transfer to the other two, which is why the field's genuine achievements keep being read as evidence for a proposition they do not touch.

These horizons track imaging volume, funded programmes and published projections rather than any date for a human, because nothing in the fetched literature projects one:

• 10 yr: Frontier The field's own projection is emulation of sub-million-neuron organisms — fruit flies, small fish — at low hundreds of millions of dollars. Established The funded mammalian waypoint is 10 to 15 cubic millimetres of mouse hippocampus, 2 to 3% of a mouse brain, about 25 petabytes over five years, and its funding was disrupted in 2025 with an institutional bridge running only to June 2026. Frontier Expect proofreading automation rather than imaging to be the variable that moves, and expect the preservation market to grow faster than the science.
• 25 yr: Speculative A whole mouse brain is the plausible ambition in this window if throughput improves by orders of magnitude and funding stabilises; the report's own comparison is that it is like a high-resolution reconstruction of Earth. Speculative A validated small-organism emulation, if it arrives, would collapse most of the sixteen-order-of-magnitude compute range and settle the structural-sufficiency question. Handwave Whether either happens depends on a national funding relationship rather than on any measurement, which is not something this brief can extrapolate.
• 50 yr: Speculative Human-scale mapping is not a matter of scaling current methods — it is a million-fold volume problem in a field of a few hundred people — and no published design targets it. Speculative If structural insufficiency holds, the ceiling is species-generic reconstruction rather than personal emulation, and the interesting outcome is a very good model of a human cortex belonging to nobody. Handwave Any date attached to a human upload in this window is a preference expressed as a schedule.
• 100 / 250+ yr: Handwave Beyond useful forecasting, and the preservation sector's own bet is on exactly this horizon: tissue held for a century against a readout technology nobody has designed. Handwave The only durable observation is that the identity question does not improve with time, instrumentation or resolution — it is the same question at 4 nm as at 4 micrometres.

• Depends on Brain Preservation, and specifically its fidelity result rather than its throughput: whether a preserved brain still contains what a mind is made of. Aldehyde-stabilised cryopreservation demonstrably preserves ultrastructure across a whole pig brain, and the prize-awarding body states in the same announcement that no claim is made that memory, identity or consciousness was preserved. The entire preserve-now-read-later programme, and the commercial sector selling it, are downstream of that answer, which only that brief can give — and that brief already names a substrate for whole-brain emulation among the things it would enable, so this edge types a relationship the map already asserts in prose. The readout capacity is deliberately not claimed here: whole-brain electron-microscopy throughput is recorded on that brief's own tree as an unmet industrial constraint, so restating it would double-count one gap. Verdicts on whether an emulation would be conscious are imported from Consciousness Research, Integrated Information Theory and Orch OR and are not typed dependencies, because this brief's central claim survives every answer they could return.
• Requires (not on this map) Recovery of synaptic weights from fixed tissue. Electron microscopy works only on dead tissue, and synaptic strength, membrane properties, firing thresholds, dynamic learning mechanisms and extrasynaptic signalling are all dynamic quantities that the modality destroys while capturing the structure. Nobody can read the weights off an individual animal, which is why forty years of the C. elegans connectome have produced no emulation of any particular worm. This is a specific missing result rather than a restatement of the subject: a method inferring strength from fixed-tissue morphology and validated against paired electrophysiology would move the whole field, and no programme is producing one. Second, an emulation validated against the individual animal it was scanned from. The preserve-then-read-then-behave pilot has been proposed and never run, every component exists, and without it success and unverifiable failure remain behaviourally identical — a generic species-typical emulation would pass any behavioural test anyone could design. Third, consumer protection for a service sold on a future readout. Hundreds of people are preserved and thousands are members against a readout nobody has demonstrated at any scale in any organism; the one documented intervention was a university terminating a research contract in 2018 while stating that neuroscience does not know which biomolecules related to memory a method would need to preserve. No regulator in any jurisdiction addresses a service whose delivery is contingent on a scientific result that may never arrive. The first two are results nobody is producing; the third is something a legislature could choose to supply.
• Enables Nothing on this map waits on a result this brief produces. No typed enabling edge is claimed, and the reason is worth recording: the obvious candidates are questions about what would be owed to an emulation, which is a moral-status question rather than a result, and the preservation sector, which is already selling without waiting.
• Adjacent Brain Preservation and Cryonics for the substrate; Machine Consciousness and Consciousness Research for the imported verdict; Memory Engineering for what a memory physically is; Brain-Computer Interfaces and Neural Interfaces for reading a living brain; Biological Immortality as the competing answer to the same motivation; and outside the map, connectomics, computational neuroscience and the metaphysics of personal identity.

Established “Whole-brain connectomes exist for mammals.” They do not. The largest densely reconstructed mammalian volume is one cubic millimetre of mouse visual cortex, and within it only 373 neurons have fully clean, completely proofread axons. The human dataset has 104 proofread cells. The male fly — a complete central nervous system, and the best-finished large dataset in existence — has 40.1% of its connections with both partners proofread. Frontier “A connectome was released” is a claim about a dataset; “a connectome is complete” is a claim about proofreading, and the second is the one that matters for simulation.

Established “A connectome specifies a running mind.” It omits synaptic strength, neuromodulatory context, molecular state, most glia, and — in the fly datasets — electrical synapses entirely, because those await higher-resolution imaging. Established The empirical demonstration is the worm: four decades of a complete 302-neuron connectome and no working emulation. Frontier The honest counter is that the worm was also never seriously funded, and this brief carries the ambiguity rather than resolving it in the direction that suits either camp.

Established “The fly simulation is an emulation of a fly.” It is a uniform leaky integrate-and-fire model with predicted rather than measured neurotransmitters, and it is a model of a fly brain rather than of the individual that was imaged. Established Its successes are real and are about circuits, not individuals: sugar and water responses, proboscis extension, taste interaction and antennal grooming, validated optogenetically. Reading it as a proof of concept for personal uploading skips the token-versus-type distinction the whole subject turns on.

Established “The compute requirement is roughly known.” Published estimates span sixteen orders of magnitude, from below 10¹⁵ FLOPS to 10²⁸, with the widely quoted roadmap giving 10¹⁸, 10²² and 10²⁵ for three different levels of biological detail. Established The spread is not error; it is the level-of-detail disagreement in different units. Quoting one figure picks a side silently.

Frontier “Gradual replacement solves the identity problem.” It is the philosophically preferred route and two independent arguments deny it. Corabi and Schneider: gradual replacement still produces a functional isomorph at some moment, and the multi-upload case shows psychological continuity cannot be sufficient for identity. Mogensen: vagueness and holism break the fading-qualia inference the route rests on. Speculative Its popularity substantially exceeds its support.

Established “A non-destructive scan would transfer you safely.” In that case two beings exist and the continuity claim is false by stipulation, with no philosophy required. Frontier A destructive scan does not fix this; it removes the party who would object. Anyone who finds the destructive version more reassuring should notice which fact is doing the reassuring.

Established “Preservation preserves the person.” Aldehyde-stabilised cryopreservation preserves ultrastructure — that is demonstrated, at whole-pig-brain scale, and independently judged. Established The prize-awarding body explicitly declines the further claim about memory, identity and consciousness, and a university stated that neuroscience does not know which biomolecules related to memory and mind would need preserving. Handwave The marketed step from the first sentence to the second is the assertion the whole sector rests on.

Established “Cryonics membership and funding figures show scientific progress.” They show a market: 500 to 650 preserved, 5,000 to 6,000 members, prices from $28,000 to $220,000, more than $65 million disclosed in 2025 funding and over $100 million into one venture. Frontier The counts are self-reported and disagree with an independent tally by four patients at the largest provider. Established Nobody has been revived and nothing has been read out.

Frontier “Human-scale mapping is a matter of scaling what exists.” It is a six-order-of-magnitude volume problem, in a field of fewer than 500 people, resting on basic neuroscience funded at about 1% of the NIH budget, whose last major scale-up lost $4.2 million to a funding action unrelated to the science. Speculative Treating “hard” and “small” as the same diagnosis has kept a plausible sociological explanation of slow progress out of the argument.

Handwave “Consciousness science will eventually confirm that an upload is you.” It cannot, and this is the cleanest boundary on the page. A theory of consciousness could in principle establish that an emulation has experiences. Established Nothing in any theory of consciousness addresses whether the experiences are the continuation of a particular prior person — that is a question in the metaphysics of personal identity, where the Stanford Encyclopedia records no consensus or even a dominant view.