Mind Uploading scenario 2050
EN 日本語
Dated scenario / 2026 → 2050

From a worm to a human.
A timeline to the year
mind uploading is finished.

Mind uploading — whole brain emulation (WBE) — means measuring the structure and state of a brain and reproducing its information processing on a computer. Following the format of AI 2027, this page starts from where research actually stands worldwide and asks what could happen in each year from 2026 to completion in 2050, keeping demonstrated fact and projection clearly apart.

Format: dated scenario Scope: published research worldwide Span: 2026–2050 Dates derived from: four laws 44 references

This is the companion volume to the integrated guide, and it is a thought experiment. For the factual record see Chapter 3 · Current status; for a comparison of forecasts see Chapter 4. The timeline below is a conditional scenario in which every optimistic assumption holds, and it is deliberately faster than the consensus forecast (median around 2070).

302 → 8.6×10¹⁰
Neurons to copy (nematode → human)
6
Phases
2050
Completion (conditional)
44
Real studies cited

Abstract

Abstract

Whole brain emulation is the engineering problem of "scan and emulate", systematised by Sandberg and Bostrom in 20081. In the twenty years since, research worldwide has pushed structural mapping from the nematode (302 neurons)2 to the whole adult fly brain (about 140,000 neurons)7 and 1 mm³ of human cortex9, while readout has reached speech BCIs operating at conversational rates2123.

This review extrapolates that curve and draws 2026 → 2050 as a six-phase dated scenario. Every date is derived from four extrapolation laws, with the arithmetic and the failure conditions fully disclosed: automating the map → the first mammalian whole brain → running it → the jump to humans → reconstructing an individual → completion and the fork. Each phase builds on the last. Completion in 2050 is not a promise but a conditional scenario that holds only if three rate limiters are passed — degeneracy (uniqueness of reconstruction)16, whole-brain scale, and validating identity — and it ends by splitting into "continuation" and "the copy problem"31.

1. The Exponential

Brain mapping is already climbing exponentially.

Plotted logarithmically, the largest brain reconstructed at synaptic resolution spans about nine orders of magnitude from nematode to human. The solid line is demonstrated; the dashed line is this scenario's projection. Completion can be read as the point where the curve reaches the ceiling of a whole human brain.

Largest brain reconstructed at synaptic resolution A logarithmic plot from the nematode's 302 neurons to a whole human brain of 8.6×10¹⁰ neurons, showing results demonstrated up to 2025 and projected from 2026 onward. projected → ← demonstrated 10² 10⁴ 10⁶ 10⁸ 10¹⁰ 1990 2005 2025 2040 2050 Neurons reconstructed Nematode 302 (1986) Fly central brain 25K (2020) Whole fly 140K · mouse 1mm³ (2024–25) Whole mouse 7×10⁷ (~2030) Whole human 8.6×10¹⁰ (~2045) done
Demonstrated (to 2025) Projected (2026 onward) Whole human scale

How to read it: WBE is not a single invention but a dependency pipeline — fix → acquire → reconstruct → run → validate. The curve above tracks only the acquisition axis. Copying structure does nothing on its own without the dynamical state and the validation to go with it. The timeline that follows asks, year by year, which axis advances and what is left over.

2. The Derivation

Why these dates?

The years in this scenario were not chosen by feel. Each phase's duration comes from four extrapolation laws, and 2050 is what the sum produces. The arithmetic is fully disclosed here so that readers can judge for themselves which parts are solid and which are a bet.

Law 1 — Scale

Mapping scale: the largest complete connectome, in neurons

There are only four measured points, but the slope is unambiguous.

1986: 302 (nematode)2 → 2020: 25K (fly central brain)6 → 2024: 139K (whole fly)7 → 2025: 167K (male fly CNS)32 Long run (1986→2024): 2.66 OOM ÷ 38 yr = 0.070 OOM/yr (14.3 yr per order) Recent (2020→2024): 0.75 OOM ÷ 4 yr = 0.187 OOM/yr (5.4 yr per order)

Applying each rate to the next target gives:

Fly 139K → mouse 7×10⁷ : 2.70 OOM → 38.6 yr at the long-run rate (2063); 14.4 yr at the recent rate (2038) Mouse → human 8.6×10¹⁰ : 3.09 OOM → 16.5 yr at the recent rate

The thing to notice: placing the whole mouse brain at 2030 demands roughly 2.4× the recent rate. But "mouse 2030 → human 2045" is 3.09 orders in 15 years, or 0.21 OOM/yr — essentially the recent rate itself. In other words, almost the entire bet in this scenario is concentrated in phase 2, the mouse.

Law 2 — Data

Data volume: converting from H01 (1 mm³ of human cortex = 1.4 petabytes)

Imaging volume scales with tissue volume. Since reconstructing 1 mm³ of human cortex produced 1.4 PB9, the requirement is simple multiplication.

Whole fly brain ≈ 0.08 mm³ → about 110 terabytes Whole mouse brain ≈ 500 mm³ → about 0.7 exabytes (≈6,000× the fly) Whole human brain ≈ 1.2×10⁶ mm³ → about 1.7 zettabytes (≈15,000,000× the fly)

Estimates assuming today's fastest EM (multibeam SEM) put more than ten years on imaging and reconstructing a whole mouse brain alone33. This scenario's "start in 2027, finish imaging in 2028, publish in 2030" rests on the assumption of a roughly fivefold acceleration through instrument parallelism and throughput gains. The 10 mm³ pilots now underway (about 2% of a mouse brain)34 will be the first indicator of whether that holds.

Law 3 — Compute

Compute: exascale is already here; the next four orders take about 19 years

The 2008 roadmap estimated a spiking-level human brain at about 10¹⁸ FLOPS and an electrophysiological-level one at about 10²² FLOPS1.

TOP500 leader: 2008 1 petaFLOPS → 2022 1 exaFLOPS35 = 3 OOM ÷ 14 yr = 0.21 OOM/yr Spiking level (10¹⁸): reached in 2022 Electrophysiological (10²²): 4 OOM ÷ 0.21 = about 19 yr → around 2041–2045

The thing to notice: compute is not the rate limiter in this scenario. The capacity to run a whole human brain at high resolution arrives naturally at the same time as the map. The real wall is not arithmetic but the memory and bandwidth to hold and update 10¹⁴ synaptic states, and the pipeline to process zettabyte-scale data.

Law 4 — Automation

Proofreading: the one place that needs a two-order breakthrough, not an extrapolation

The output of automatic segmentation is not yet a wiring diagram. For the whole fly brain, automatic processing was followed by millions of community edits — proofreading estimated at tens of person-years736.

Fly, 139K neurons: proofreading on the order of tens of person-years Naive scaling to mouse, 7×10⁷ (500×): on the order of 10,000 person-years → not viable

Phase 1 (2026–28), "automating the map", is therefore different in kind from the others. It is not an extrapolation but an assumption that proofreading cost falls by two orders of magnitude. Designs that remove the need for proofreading altogether — E11 Bio's protein barcoding (PRISM)37 — are what would carry that assumption. If it fails, the whole scenario slides toward Law 1's long-run rate, and the mouse arrives in 2063.

Phase by phase: the derivation and the failure conditions

PhaseYearsHow the date is derivedAssumptions (which, if broken, delay it)Confidence
1. Automating the map2026–28 Requires no new discovery: this is standardising pipelines already demonstrated by FlyWire and MICrONS710. Duration is set by adoption speed Law 4 (two-order reduction in proofreading cost); continuity of funding High
2. The first mammalian whole brain2029–32 Law 1 extrapolated at 2.4× the recent rate. Assumes the 10 mm³ pilot completes in three years and scaling to 50× that takes three more Fivefold imaging throughput, fully automated proofreading, a decade of continuous funding. Losing any one pushes this to around 2038 Most fragile
3. Running it2033–37 In the fly, "model38 → embodied39" took only two years. The mouse is 560× larger, but by Law 3 compute is not limiting, so five years is allocated to the science of modelling Establishing methods to infer dynamics — synaptic weights, neuromodulation — from structure. The only phase requiring new science, so the estimate carries maximum uncertainty Science-dependent
4. The jump to humans2038–43 Law 1's recent rate applied unchanged (about 16 years mouse to human). Assumes work begins in parallel rather than waiting for the mouse, hence a 2039 start. Consistent with Law 3's electrophysiological level arriving 2041–45 Supply of preserved brains (depends on clinical-grade preservation beginning in phase 2); ethical and legal frameworks Conditional
5. Reconstructing an individual2044–48 Five years allocated to consensus formation rather than technology: designing a validation battery, replicating it across institutions, and having a standards body adopt it Agreement on validation criteria themselves. This axis effectively starts from zero today Not started
6. Completion and the fork2049–50 Simply where the sum lands. The round number 2050 was not chosen and worked backwards from; it is what the phase durations add up to — Endpoint

How to reconcile 2050 with outside forecasts. The forecasting community's median is around 207040, and the 2025 review from inside the field puts it at 30–40 years at minimum (2055–65 onward)41. This scenario is 15 to 20 years faster. Where does the difference come from? As the table shows, almost entirely from phases 2 and 3 — the speed of mapping a whole mouse brain, and whether the science of inferring dynamics from structure resolves in five years. Estimate those two conservatively and this timeline translates directly to 2065–70. This page is therefore not a forecast but an instrument for showing that two assumptions move the future by twenty years.

3. The Scenario

2026 → 2050, in six phases.

The year panel on the right follows your scroll and shows which phase you are reading. Click to jump.

Where we are · 2026 · demonstrated

Before extrapolating, the foundation. All of the following has actually been achieved.

  • Complete nematode wiring diagrams exist for both sexes and across development234. In Drosophila, both the larva5 and the whole adult brain (about 140,000 neurons, 50 million synapses) are complete678.
  • In mammals, 1 mm³ of human cortex has been reconstructed at petavoxel scale9, and mouse visual cortex has been mapped structurally and functionally together (MICrONS)10. Whole-brain cell-type atlases exist1112.
  • Measurement includes Neuropixels recording thousands of sites at once1819, BCIs decoding handwriting and speech at speed in paralysed patients20212223, and restored walking via a brain–spine bridge24.
  • Preservation by aldehyde-stabilized cryopreservation retains synaptic structure in whole pig brains17. The basis for reconstruction is cortical microcircuit simulation15.
Phase 12026–2028projected

Automating the map

AI-driven electron microscopy reconstruction drops the cost of wiring diagrams by orders of magnitude.

  1. 2026

    AI removes the reconstruction bottleneck

    Flood-filling and transformer-based automatic segmentation slash the human cost of petabyte-scale EM imagery. The pipelines demonstrated on FlyWire, H01 and MICrONS7910 become standard, and reconstruction throughput stops being the main limiter on acquisition.

  2. 2027

    A whole-mouse-brain project starts

    Public bodies such as the BRAIN Initiative and private consortia adopt a whole mouse connectome of about 70 million neurons as an explicit intermediate goal and begin whole-brain EM imaging. The 10 mm³ BRAIN CONNECTS pilots begun in 202334 provide the foundation for a roughly three-order scale-up from 1 mm³.

  3. 2027

    Speech BCIs reach conversational speed

    Intracortical speech neuroprostheses212223 reach speed and stability adequate for everyday conversation and are replicated by multiple teams. Reading state from neural activity becomes a clinical standard.

  4. 2028

    Whole-mouse imaging completes

    Whole-brain EM acquisition finishes and the automatic reconstruction phase begins. Preparations advance for overlaying molecular and cell-type atlases1112 and expansion microscopy13.

Why 2026–2028

Derivation: this phase alone is not an extrapolation. It standardises pipelines already running at FlyWire7 and MICrONS10, so the duration is set by adoption rather than discovery. Three years is what adoption takes.

Failure condition: Law 4 — failing to cut proofreading cost by two orders of magnitude. If that does not move, every later phase slides toward the long-run rate.

Phase 22029–2032projected

The first mammalian whole brain

From worm to mouse. On the structural axis, a mammalian brain closes for the first time.

  1. 2030

    Whole mouse connectome, first release

    A draft wiring diagram of about 70 million neurons and 10¹¹ synapses is published, amounting to roughly 0.7 exabytes (Law 2). With the jump from nematode to fly to mouse, the mammalian structural axis reaches a resting point3334.

  2. 2031

    A multimodal mouse brain

    The wiring diagram is overlaid with whole-brain cell types from transcriptomics11 and Neuropixels-class functional recording19 — the first mammalian dataset in which structure, molecules and activity share one coordinate frame.

  3. 2031

    Foundation models of dynamics

    Large models predicting activity from structure — a "virtual mouse brain" — begin reproducing function in sensorimotor circuits. The foundation-model digital twins demonstrated in visual cortex in 202542 and Blue Brain's microcircuit reconstructions15 scale, data-driven, toward the whole brain.

  4. 2032

    Clinical-grade brain preservation

    Procedures derived from ASC17 are standardised, and cases appear in which whole human brains are preserved at synaptic resolution under end-of-life and research protocols. Securing "originals" for future acquisition begins.

Why 2029–2032 · the most fragile joint in this scenario

Derivation: Law 1's recent rate (0.187 OOM/yr) applied unchanged puts the whole mouse brain at 2038. Placing it in 2030 requires about 2.4× acceleration. The assumed breakdown: the 10 mm³ pilot34 completes in three years, and scaling 50× takes three more — the figures you get when fivefold throughput and automated proofreading land together.

Failure condition: lose any one of throughput, automation, or a decade of continuous funding and this reverts to around 2038, shifting every later phase by eight years. That is why Chapter 3's watchlist tracks this stretch.

Phase 32033–2037projected

Running it: functional emulation

The stage of making the map move. The body loop and the uniqueness of reconstruction become the main battleground.

  1. 2034

    Neuromorphic hardware runs mouse scale in real time

    Spiking networks abstracted from Hodgkin–Huxley membrane currents14 run at mouse scale in real time and at low power on SpiNNaker-class25, Loihi-class26 and NorthPole-class27 hardware alongside exascale compute.

  2. 2035

    Closing the body loop

    Regional and then whole-mouse functional emulations reproduce specific behaviours inside sensorimotor loops with virtual or robotic bodies24 — the mammalian counterpart of the embodied fly emulation demonstrated in 202639. Since a brain does not function in isolation, embodiment becomes a condition of fidelity.

  3. 2036

    A standard answer to degeneracy

    Against degeneracy — different parameters producing identical activity16 — perturbation response plus held-out data validation becomes the standard of proof. Many apparent matches are rejected, and reconstruction becomes falsifiable.

  4. 2037

    An agreed mouse emulation

    A mouse brain emulation passing a suite of behavioural and perturbation held-out tests reaches consensus. This is the benchmark culture begun by ZAPBench43 institutionalised as mammalian pass criteria. The question sharpens at the same time: does matching behaviour imply matching internal states?

Why 2033–2037

Derivation: in the fly, going from a connectome-based model38 to autonomous embodied behaviour39 took only two years. The mouse is 560× larger in neurons, but by Law 3 compute is not limiting. The five-year allocation is therefore time for the science of modelling — structure to synaptic weights and neuromodulation — not for equipment.

Failure condition: this is the only phase requiring new science, and its estimate is the most uncertain. If dynamical state turns out not to be recoverable from static structure, five years will not do it, and the work restarts from inventing the necessary measurements.

Phase 42038–2043projected

The jump to humans

Three more orders of magnitude from mouse to human, acquiring whole-brain structure from preserved brains.

  1. 2039

    Human whole-brain structural acquisition

    Exascale structural connectomes begin to emerge from preserved donor brains, scaling H01's 1 mm³ fragment9 to a whole organ. Mapping of roughly 86 billion neurons and 100 trillion synapses begins.

  2. 2040

    State inference from structure matures

    Methods mapping EM morphology to functional synaptic weights and neuromodulatory state reach a level sufficient to initialise dynamics. Combining perturbation-based elimination of degeneracy16 with digital-twin learning42 narrows the gap left by structure alone.

  3. 2041–42

    Partial human emulation

    Subsystems such as sensory cortex and hippocampal memory circuits are validated against pre-mortem recordings and behavioural data — material drawn from the intracortical BCI records that had become clinically standard2123. Not a whole brain, but human circuitry runs for the first time.

  4. 2043

    The economics of emulation reaches policy

    The questions Hanson raised in The Age of Em31 about an emulation economy become concrete, and drafting begins on governance of copying rights, labour and personhood.

Why 2038–2043

Derivation: mouse to human is 3.09 orders. Law 1's recent rate gives about 16.5 years, and no acceleration is assumed here — this is the most straightforward extrapolation in the scenario. Structural acquisition is placed at 2039 on the assumption of parallel starts rather than waiting for the mouse. It also aligns naturally with Law 3's electrophysiological level (10²² FLOPS) arriving 2041–45.

Failure condition: supply, not technology. Human brains preserved at synaptic resolution must exist by this point, which depends on clinical-grade preservation having arrived in phase 2 (2032). Without ethical and legal frameworks, this is where it stops.

Phase 52044–2048projected

Reconstructing an individual

Running the entire pipeline, end to end, for one person.

  1. 2045

    The pipeline connects

    Fix → acquire → reconstruct → run → embody → validate is joined end to end for a specific individual for the first time. The entry point is a brain fixed a decade or more earlier under the ASC-family protocols established in the 2020s1744.

  2. 2046

    Memory and personality reproduced

    Candidate whole-brain human emulations begin reproducing donor-specific memories and personality markers under blinded testing.

  3. 2047

    An identity validation battery is deployed

    A battery operationally testing the continuity of identity enters use. Because theories of consciousness remain unsettled30, the battery is designed to avoid metaphysics and ask only about agreement in behaviour, memory and perturbation response. Legal status, consent and the copy problem become live policy questions at the same time.

  4. 2048

    "Completion" defined operationally

    A standards body defines completion as passing the validation battery, explicitly decoupled from metaphysical claims.

Why 2044–2048

Derivation: these five years are not technical time but time for consensus — designing a validation battery, replicating it across institutions, and having a standards body adopt it, benchmarked against how long medical-device and clinical-trial standards take to establish.

Failure condition: the validation axis effectively starts from zero as of 202643, so five years is a generous allocation. Without agreement on pass criteria, the technology advances while nobody is in a position to say it is finished.

Phase 62049–2050projected

Completion, and the fork

Engineering arrives. The philosophical question splits in two.

  1. 2049

    The battery is passed

    A whole-brain human emulation passes the agreed battery of behavioural, perturbation and memory tests, reaching completion in the engineering sense. The years taken were the sum of Laws 1 through 4, not a target set in advance.

  2. 2050

    The fork — behaviour can be verified, subjectivity cannot be measured

    Behavioural identity is confirmed. But no physical measurement separates a perfect reproduction of behaviour from subjective continuity282930. Here the future splits into two routes (below).

Why 2049–2050

Derivation: not worked backwards from a round number. Adding the phase durations — 3 + 4 + 5 + 6 + 5 years — from 2026 lands the endpoint here. That the year reads "2050" is an outcome, not an objective.

Against outside forecasts: the 15-to-20-year gap against the Metaculus median of about 207040 and the field review's 2055–6541 comes entirely from the assumptions in phases 2 and 3. Take those conservatively and this timeline translates directly to 2065–70.

4. The Fork

Past completion, the future splits.

Even if the technology succeeds, the question of identity does not close on technical grounds. Just as AI 2027 forks into "race" and "slowdown", here two routes emerge from how the validation is interpreted.

Functional and psychological continuity adopted

Uploading is treated as the person continuing

Society accepts the functional and psychological continuity demonstrated by the validation battery as the criterion of identity. Emulations acquire legal personhood and are treated as continuations of the original. Gradual uploading — replacing neurons a few at a time, in the manner of the Ship of Theseus — is preferred over destructive copying as the procedure that preserves continuity.

The debate moves to when and under what conditions a person continues, and technical and validation work keeps advancing.

5. Why Conditional

Why 2050 is conditional.

Three rate limiters that extrapolation alone does not clear. Until they are solved, the scenario cannot be pulled forward.

1. The identifiability barrier. If multiple internal states explain the same observations no matter how much you measure, reconstruction never closes uniquely16. The key is not "measure more" but methodology that reduces the candidate states through perturbation and held-out data.
2. Scale and destructiveness. A whole human brain is about a million times 1 mm³ of cortex9. That means scaling imaging, reconstruction and compute by several orders of magnitude, against the physical limits of non-destructive measurement in living tissue. Destructive acquisition ties the problem directly to preservation and ethics.
3. Identity and ethics. No measurement separates perfect behaviour from subjective continuity2830. The copy problem, consent, social status, conditions for shutdown — the questions that technical success alone does not close remain at the end31.

References

Cited research (44).

Representative results on each axis, drawn from published research worldwide. Milestones from 2026 onward are projections from these trends; the citations establish the demonstrated foundation. Each entry links to the original or to a stable bibliographic page.

  1. Sandberg, A., & Bostrom, N. (2008). Whole Brain Emulation: A Roadmap. Tech. Report #2008-3, Future of Humanity Institute, Univ. of Oxford.
  2. White, J. G., et al. (1986). The structure of the nervous system of C. elegans. Phil. Trans. R. Soc. B, 314, 1–340.
  3. Cook, S. J., et al. (2019). Whole-animal connectomes of both C. elegans sexes. Nature, 571, 63–71.
  4. Witvliet, D., et al. (2021). Connectomes across development reveal principles of brain maturation. Nature, 596, 257–261.
  5. Winding, M., et al. (2023). The connectome of an insect brain (Drosophila larva). Science, 379, eadd9330.
  6. Scheffer, L. K., et al. (2020). A connectome and analysis of the adult Drosophila central brain (hemibrain). eLife, 9, e57443.
  7. Dorkenwald, S., et al. (FlyWire Consortium) (2024). Neuronal wiring diagram of an adult brain. Nature, 634, 124–138.
  8. Schlegel, P., et al. (2024). Whole-brain annotation and multi-connectome cell typing of Drosophila. Nature, 634, 139–152.
  9. Shapson-Coe, A., et al. (2024). A petavoxel fragment of human cerebral cortex reconstructed at nanoscale resolution. Science, 384, eadk4858.
  10. The MICrONS Consortium (2025). Functional connectomics spanning multiple areas of mouse visual cortex. Nature, 640.
  11. Yao, Z., et al. (2023). A high-resolution transcriptomic and spatial atlas of cell types in the whole mouse brain. Nature, 624, 317–332.
  12. Siletti, K., et al. (2023). Transcriptomic diversity of cell types across the adult human brain. Science, 382, eadd7046.
  13. Chen, F., Tillberg, P. W., & Boyden, E. S. (2015). Expansion microscopy. Science, 347, 543–548.
  14. Hodgkin, A. L., & Huxley, A. F. (1952). A quantitative description of membrane current… J. Physiol., 117, 500–544.
  15. Markram, H., et al. (2015). Reconstruction and simulation of neocortical microcircuitry. Cell, 163, 456–492.
  16. Prinz, A. A., Bucher, D., & Marder, E. (2004). Similar network activity from disparate circuit parameters. Nat. Neurosci., 7, 1345–1352.
  17. McIntyre, R. L., & Fahy, G. M. (2015). Aldehyde-stabilized cryopreservation. Cryobiology, 71, 448–458.
  18. Jun, J. J., et al. (2017). Fully integrated silicon probes for high-density recording of neural activity. Nature, 551, 232–236.
  19. Steinmetz, N. A., et al. (2021). Neuropixels 2.0: a miniaturized high-density probe. Science, 372, eabf4588.
  20. Willett, F. R., et al. (2021). High-performance brain-to-text communication via handwriting. Nature, 593, 249–254.
  21. Willett, F. R., et al. (2023). A high-performance speech neuroprosthesis. Nature, 620, 1031–1036.
  22. Metzger, S. L., et al. (2023). A high-performance neuroprosthesis for speech decoding and avatar control. Nature, 620, 1037–1046.
  23. Card, N. S., et al. (2024). An accurate and rapidly calibrating speech neuroprosthesis. N. Engl. J. Med., 391, 609–618.
  24. Lorach, H., et al. (2023). Walking naturally after spinal cord injury using a brain–spine interface. Nature, 618, 126–133.
  25. Furber, S. B., et al. (2014). The SpiNNaker project. Proc. IEEE, 102, 652–665.
  26. Davies, M., et al. (2018). Loihi: a neuromorphic manycore processor with on-chip learning. IEEE Micro, 38, 82–99.
  27. Modha, D. S., et al. (2023). Neural inference at the frontier of energy, space, and time (NorthPole). Science, 382, 329–335.
  28. Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience, 5, 42.
  29. Dehaene, S., & Changeux, J.-P. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70, 200–227.
  30. Cogitate Consortium; Ferrante, O., et al. (2025). Adversarial testing of global neuronal workspace and integrated information theories. Nature.
  31. Hanson, R. (2016). The Age of Em: Work, Love, and Life when Robots Rule the Earth. Oxford University Press.
  32. Janelia FlyEM (2025). Male CNS Connectome (brain + ventral nerve cord, 166,000 neurons).
  33. BioTechniques (2024). Building a comprehensive mouse brain connectome. (includes duration and data-volume estimates)
  34. Google Research / Harvard / Allen Institute (2023–). BRAIN CONNECTS: 10 mm³ mouse hippocampus and cortico-basal ganglia connectome projects.
  35. TOP500 Supercomputer Sites. Leading system performance over time (2008 Roadrunner 1 PFLOPS → 2022 Frontier 1 EFLOPS → 2024 El Capitan 1.74 EFLOPS).
  36. Notable progress has been made in whole brain emulation (2025). LessWrong. (secondary analysis of FlyWire proofreading effort and the automation bottleneck; not peer reviewed)
  37. E11 Bio (2024–25). PRISM: self-proofreading connectomics via protein barcoding and expansion microscopy.
  38. Shiu, P. K., et al. (2024). A Drosophila computational brain model reveals sensorimotor processing. Nature, 634, 210–219.
  39. Eon Systems (7 March 2026). How the Eon team produced a virtual embodied fly. (the first embodied whole-brain emulation)
  40. Metaculus. Date of first human whole brain emulation. (community forecast; median has tracked around 2070)
  41. Zanichelli, N., Schons, M., Freeman, I., Shiu, P. K., & Arkhipov, A. (2025). State of Brain Emulation Report 2025. arXiv:2510.15745.
  42. Wang, E. Y., et al. (2025). Foundation model of neural activity predicts response to new stimulus types. Nature, 640. (digital twin)
  43. Google Research, Janelia & Harvard (2025). ZAPBench: whole-brain activity prediction benchmark in larval zebrafish.
  44. Nectome (2026). Ultrastructural preservation of a whole large mammal brain with a protocol compatible with human physician-assisted death. (preprint)