Integrated Guide · Chapter 4

Forecasts — saying "when" with the conditions attached.

Timing is the most misunderstood topic in this field. This chapter first lays out the major forecasts with their sources and reasoning, then translates them into a dependency structure: which intermediate goals are needed, in what order. The aim is to read the future through conditions rather than through years. For a year-by-year narrative, see the companion scenario, 2026 → 2050.

10 forecasts comparedFour extrapolation laws20 references

4.1 Forecasts

The forecasts: who, when, and on what basis.

Forecaster (year)Human WBEType of reasoningAssessment from 2026
Moravec (1988)12030s–2040sExtrapolation of computeThe compute extrapolation was close to right; scanning and modelling were badly underestimated
Kurzweil (2005/2024)22030s scanning, ~2045Aggregate exponential curvesHeld to 2045 in the 2024 sequel. The 2030s scanning claim diverges sharply from current reality
Sandberg & Bostrom (2008)3No date given (conditionally mid-century or later)Technology-tree decompositionStill the methodological standard, and cited partly for its refusal to promise a date
Hanson (2016)4"Sometime in the next century"Economic inevitabilityDeclines to bet on timing and analyses consequences instead. The standard text on the resulting society
Watanabe (2019–)5Around 2040 (a 20-year goal)Gradual migration via BCI, avoiding scanning entirelyFramed by its author as a target. The prerequisites — interface bandwidth, a natural law of consciousness — are unmet
Metaculus community6Median ~2070Aggregated forecasting (hundreds of participants)Stable in the 2060–2080 range throughout the 2020s
Digital minds expert survey (2025)765% by 2100 (digital minds broadly)Survey of 67 expertsA broader definition than WBE. Median probability that it is possible in principle: 90%
State of Brain Emulation Report (2025)830–40 years at minimum (≈2055–65 onward)Inventory of progress across three capabilitiesThe most systematic estimate from inside the field
Eon Systems (2026)9No date; human scale stated as the goalCompany roadmapThe fly demonstration is real, but public statements carry an investment context
This site's companion scenario2050 (conditional)A thought experiment in which every optimistic condition holdsNot a forecast — a way of showing what would have to be true for the fastest path
How to read this

The forecasts fall into three groups: optimistic insiders (2040s–50s), aggregated and external assessments (2060s–80s), and sceptical neuroscientists (22nd century or never). The width of the distribution is itself an accurate reflection of how much uncertainty remains, by orders of magnitude, in modelling and validation.

4.2 Dependencies

Dependencies: what waits on what.

The path to human WBE decomposes into roughly this ordered set of intermediate goals. Every date is an estimate, and the width of each range is the uncertainty.

#MilestoneStatus and estimateIf it slips…
M1Whole-brain activity and structure in the same animal (zebrafish)Release imminent (2026–27) Activity acquired, connectome in reconstruction10All research on inferring dynamics from structure stalls
M2A mature, validated fly emulation including learning and neuromodulationUnderway (to ~2030) Extending the 2026 embodied demonstrationNo answer on whether structure suffices, so investment in larger species cannot be justified
M3Whole mouse connectome (≈7×10⁷ neurons)Early-to-mid 2030s, funding and automation permitting 10 mm³ pilots underway11Everything downstream moves back. The single largest bottleneck
M4A validated mouse emulation5–10 years after M3 Conditional on passing behaviour and perturbation benchmarksNo scientific basis for extrapolating to humans
M5Human whole-brain scanning (preserved brain, 1–2 zettabytes)2040s onward A further ~1,000× beyond the mousePreservation advances alone, accumulating an inventory of unreadable brains
M6Running and validating a human WBE2050s at the earliest; median ~2070— (the endpoint, though identity and consciousness questions remain → Chapter 5)

Where those estimates come from — four extrapolation laws

The years above are not guesses. They follow from four empirical trends. The full derivation, and its application to each phase, is in the companion scenario's "why this date" section.

LawSlope from measurementWhat follows
1. Scale
largest complete connectome, in neurons
Long run (1986→2024): 0.070 OOM/yr
Recent (2020→2024): 0.187 OOM/yr
Fly to mouse is 2.70 orders of magnitude: 14 years (2038) at the recent rate, 39 years (2063) at the long-run rate. Mouse to human is 3.09 orders, about 16.5 years at the recent rate
2. Data volume
H01: 1 mm³ human cortex = 1.4 PB17
Proportional to volume Whole mouse brain ≈ 0.7 exabytes; whole human brain ≈ 1.7 zettabytes. At current EM rates, imaging a whole mouse brain alone is estimated at over ten years18
3. Compute
TOP500 leader over time20
0.21 OOM/yr
(2008 petascale → 2022 exascale)
Spiking level (10¹⁸ FLOPS) is already reached. Electrophysiological level (10²²) is four orders up, arriving around 2041–45. Compute is not the binding constraint
4. Proofreading
manual editing after automatic reconstruction
Tens of person-years for the fly's 139,000 neurons19 Naive extrapolation to the mouse gives on the order of 10,000 person-years, which does not work. The only law requiring a breakthrough rather than an extrapolation, and the real limiter on M3
What the derivation tells you
  • Compute is no longer the problem. By law 3, the capacity to run a human brain at high resolution arrives naturally around the time the map would be finished. The "we don't have enough computing power" framing of the 2010s is out of date.
  • Uncertainty concentrates in M3 and M4. By law 1, believing the recent rate rather than the long-run rate changes the mouse connectome alone by 25 years — 2038 versus 2063. That one choice dominates everything downstream.
  • The spread of forecasts reflects exactly this. The optimists price in the recent rate plus the law-4 breakthrough; the aggregators price both conservatively. Forecasters are less in disagreement than putting different coefficients into the same equation.

4.3 Factors

Accelerants and brakes.

Accelerant

AI automation

Segmentation, proofreading, synapse detection and dynamical inference are all squarely in AI's strengths. It was the single largest factor enabling the whole fly brain, and progress in general AI transfers directly into WBE speed.

Accelerant

Private funding and competition

With Eon, E11 and Google in the field, the frontier can now move independently of public budget swings. Growth in the BCI industry pulls readout technology toward volume manufacturing.

Accelerant

Preservation as an option

If preserving first becomes institutionalised, the time constraint relaxes and the field can afford to wait for scanning to mature — while the ethical disputes expand in parallel.

Brake

Scientific unknowns in modelling

If synaptic weights and neuromodulatory state cannot be recovered from static structure, a finished map still will not run. Currently the largest single uncertainty (Chapter 2.3).

Brake

Funding and politics

The US BRAIN budget halved between 2023 and 2025 before recovering in 202612. Decade-scale data projects are structurally fragile against that volatility.

Brake

Unprepared ethics and regulation

Welfare of animal emulations, medicalisation of human preservation, and the suffering risk of failed emulations (Chapter 5). Proceeding without resolving these invites a backlash capable of halting the research itself.

4.4 WBE × AGI

The relationship to AGI: competing, and accelerating each other.

Progress in AI through the 2020s changed what WBE is for. Three relationships are worth separating.

The inverse risk

As the roadmap itself noted, the by-products of WBE research — brain-inspired learning algorithms, neuromorphic substrates — may produce powerful "brain-like AI" before any complete emulation3. Treating WBE as a safety route only holds if the management of those intermediate products is included in the plan.

4.5 After

The society that would follow.

Continues in the companion volume

How these questions surface year by year is dramatised in the second half of the companion scenario (phases 4–6). That scenario ends by splitting into two routes — "continuation" and "the copy problem" — and the philosophical content of that split is the subject of the next chapter.

4.6 Summary

Chapter summary.

Key points
  • Forecasts span the 2040s (optimistic insiders), around 2070 (aggregated forecasting) and later (sceptics). The width of that distribution accurately reflects the uncertainty in modelling and validation, and there is no reason to trust any single year.
  • Read conditions rather than dates: the largest single bottleneck is the whole mouse connectome (M3), followed by a validated mouse emulation (M4). How those two go essentially fixes the human timeline.
  • The relationship to AGI is, in practice, acceleration rather than competition — alongside a live argument for WBE as a safer path to advanced intelligence, and the countervailing risk posed by its intermediate products.

References

Chapter 4 references (20).

  1. Moravec, H. (1988). Mind Children. Harvard Univ. Press.
  2. Kurzweil, R. (2005). The Singularity Is Near; (2024). The Singularity Is Nearer. Viking.
  3. Sandberg, A., & Bostrom, N. (2008). Whole Brain Emulation: A Roadmap. FHI.
  4. Hanson, R. (2016). The Age of Em. Oxford Univ. Press.
  5. NeurotechJP. Interview with Masataka Watanabe on consciousness uploading and MinD in a Device.
  6. Metaculus. Date of first human whole brain emulation (community forecast).
  7. Caviola, L., & Saad, B. (2025). Futures with digital minds: expert forecasts in 2025. arXiv:2508.00536.
  8. Zanichelli, N., et al. (2025). State of Brain Emulation Report 2025. arXiv:2510.15745.
  9. Eon Systems. Company site (fly demonstration and stated human-scale goal).
  10. Google Research (2025). ZAPBench (zebrafish whole brain; connectome in progress).
  11. Google Research. The 10 mm³ mouse hippocampus project (BRAIN CONNECTS).
  12. The Transmitter (2025–26). BRAIN Initiative budget cuts and recovery.
  13. Karnofsky, H. (2021). How digital people could change the world. Cold Takes.
  14. Mineault, P., et al. (2024). NeuroAI for AI safety. arXiv:2411.18526.
  15. Foresight Institute. Whole Brain Emulation Workshop (2023–), arguing for WBE in an AI-safety frame.
  16. Kokotajlo, D., et al. (2025). AI 2027. (the format the companion scenario follows)
  17. Shapson-Coe, A., et al. (2024). A petavoxel fragment of human cerebral cortex reconstructed at nanoscale resolution. Science, 384, eadk4858. (source of 1 mm³ = 1.4 PB)
  18. BioTechniques (2024). Building a comprehensive mouse brain connectome (duration estimates).
  19. Notable progress has been made in whole brain emulation (2025). LessWrong. (secondary analysis of FlyWire proofreading effort; not peer reviewed)
  20. TOP500 Supercomputer Sites. Leading system performance over time.