Whole Brain Emulation

📋Page Status

Page Type:ContentStyle Guide →Standard knowledge base article

Quality:48 (Adequate)⚠️

Importance:24 (Peripheral)

Last edited:2026-01-28 (4 days ago)

Words:3.5k

Structure:

📊 21📈 1🔗 3📚 44•6%Score: 13/15

LLM Summary:Comprehensive analysis of whole brain emulation finding <1% probability of arriving before AI-based TAI, with scanning speed (100,000x too slow for human brains) as the primary bottleneck despite resolution requirements being met. Documents technical requirements (10^18-10^25 FLOPS depending on detail level), current progress (fruit fly complete at 140K neurons vs 86B human), and concludes WBE is peripheral to AI prioritization given AI's faster trajectory.

Issues (1):

QualityRated 48 but structure suggests 87 (underrated by 39 points)

Overview

Whole Brain Emulation (WBE), sometimes called “mind uploading,” refers to creating a functional copy of a biological brain in a computational substrate. Unlike AI systems trained on data, a brain emulation would replicate the actual neural structure of a specific brain. The approach assumes that “total understanding of the brain is not needed, just understanding of the component parts and their functional interactions,” as stated in the foundational Sandberg-Bostrom Roadmap.

This was once considered a leading path to artificial general intelligence. The 2008 Whole Brain Emulation: A Roadmap by Anders Sandberg and Nick Bostrom at FHI provided detailed technical analysis across scanning, data processing, and simulation requirements. The report estimated that depending on the level of biological detail required, computational demands range from 10^18 to 10^25 FLOPS. However, progress has been much slower than AI, and most researchers now expect AI to reach transformative capabilities first. A 2011 AI workshop estimated an 85% probability that neuromorphic AI would arrive before brain emulation.

Estimated probability of being the dominant path to transformative intelligence: less than 1%

Technical Requirements

The WBE pipeline involves three major phases, each with distinct technical challenges and bottlenecks. The following diagram illustrates the complete pathway from biological brain to running emulation:

Loading diagram...

Brain Scanning Approaches Comparison

Different scanning technologies offer tradeoffs between resolution, speed, and preservation requirements. The table below compares major approaches being developed for connectomics:

Technology	Resolution	Speed	Preservation Method	Current Status	Key Limitation
Serial Section EM (ssEM)	4-10nm	≈1mm^3/year	Aldehyde fixation	Gold standard for connectomics	Extremely slow
Focused Ion Beam SEM (FIB-SEM)	4-8nm isotropic	≈0.01mm^3/year	Chemical fixation	High resolution, very slow	Not scalable
Multi-beam SEM	4nm	≈10mm^3/year	Aldehyde fixation	Allen Institute mouse cortex	Still 100x too slow
SmartEM (ML-guided)	4-10nm	7x faster than standard	Chemical fixation	Published 2024	Early stage
X-ray Holographic Nanotomography	20-50nm	Non-destructive	Cryogenic	Research phase	Insufficient resolution
Expansion Microscopy	≈60nm effective	Faster than EM	Chemical expansion	LICONN 2025	May miss fine detail

Nature Methods selected EM-based connectomics as Method of the Year 2025, noting that “connectomics has so far outpaced Moore’s law predictions about technological progress.”

Key Technical Challenges

The Sandberg-Bostrom Roadmap identified specific quantitative requirements for each stage. Current progress and remaining gaps are summarized below:

Challenge	Current Capability (2025)	WBE Requirement	Gap Factor	Key Bottleneck
Scanning resolution	4-10nm with multi-beam SEM	5-10nm for synapse detail	≈1x (achieved)	No longer limiting
Scanning speed	≈10mm^3/year (mouse visual cortex)	≈1,200,000mm^3 for human brain	100,000x	Primary bottleneck
Data volume	1.4 petabytes (1mm^3 mouse cortex)	≈1-2 exabytes for human brain	1,000x storage	Manageable with scale
Preservation	ASC preserves connectome (pig brain, 2018)	Preserve synaptic weights	Uncertain	May need more than structure
Segmentation	AI-automated for EM (fruit fly complete)	Human-scale automation	600,000x neurons	Scaling challenge
Neural modeling	31,000 neurons + 36M synapses (Blue Brain)	86 billion neurons	2,700,000x	Computational scaling
Compute (run-time)	1.7 exaFLOPS (El Capitan, 2024)	10^18-10^25 FLOPS depending on detail	1x to 10^7x	May already be sufficient

The resolution requirement has essentially been met, but scanning speed remains the dominant bottleneck. At current rates, scanning a complete human brain would take approximately 100,000 years. Even with optimistic 100x improvements per decade, whole-brain scanning would require multiple additional decades of technological development.

Computational Requirements by Simulation Detail

The compute required for WBE varies enormously depending on the level of biological fidelity. The Sandberg-Bostrom Roadmap and subsequent analyses provide the following estimates:

Simulation Level	FLOPS Required	Current Hardware Status	Year Affordable (≈$1M)	Key Trade-off
Functional/behavioral	≈10^15 (1 PFLOPS)	Available since 2008	Already	May miss critical dynamics
Neural network level	10^18-10^19 (1-10 EFLOPS)	El Capitan: 1.7 EFLOPS (2024)	2024-2030	Standard target estimate
Detailed compartmental	≈10^22 FLOPS	Not yet available	2040-2050	Includes dendritic computation
Molecular/metabolome	10^25-10^29 FLOPS	Far future	2087+ (per Sandberg)	May be unnecessary

Henry Markram (Blue Brain Project founder) estimated 10^18 FLOPS for detailed simulation, though his 2018 estimate for “real-time molecular simulation” was dramatically higher at ~4x10^29 FLOPS. The key uncertainty is which level of detail is actually necessary to preserve cognition and identity.

Power comparison: An exascale computer consumes 20-30 megawatts. The human brain consumes approximately 20 watts - a factor of one million more efficient. This suggests that even if WBE becomes computationally feasible, energy costs may constrain how many emulations can run simultaneously.

Key Properties

Property	Rating	Assessment
White-box Access	LOW	Brain structure visible but not interpretable
Trainability	N/A	Copied from biological learning, not trained
Predictability	LOW	Human-like cognition is inherently unpredictable
Modularity	LOW	Brains are highly interconnected
Formal Verifiability	LOW	Too complex, poorly understood

Current Progress

The history of connectomics demonstrates steady progress but also reveals how far WBE remains from human-scale implementation. Each milestone has exposed new challenges previously underappreciated.

Major Achievements Timeline

Milestone	Year	Details	Significance
C. elegans connectome	1986	302 neurons, 7,000 synapses mapped via serial EM	First complete connectome of any organism
OpenWorm project launched	2011	Open-source C. elegans simulation	Demonstrated gap between structure and function
Blue Brain cortical column	2015	31,000 neurons, 36 million synapses simulated	Most detailed mammalian circuit simulation
ASC wins Brain Preservation Prize	2018	Pig brain preserved with verifiable connectome	Showed long-term storage is feasible
Fruit fly connectome (brain)	2023	≈140,000 neurons fully mapped	Largest complete brain connectome
Fruit fly complete CNS	2024	Brain + ventral nerve cord in both sexes	First complete adult insect nervous system
Mouse visual cortex mapped	2025	1mm^3, 500 million connections	Largest mammalian connectome volume
Blue Brain Project concluded	2024	Delivered open-source mouse brain models	Transitioned to Open Brain Institute

Scale Comparison: Connectome Progress

Organism	Neurons	Synapses	Connectome Status	Functional Simulation	Gap to Human
C. elegans	302	≈7,000	Complete (1986)	Partial (BAAIWorm 2024)	285 million x
Fruit fly	140,000	≈50 million	Complete CNS (2024)	Functional analysis ongoing	614,000 x
Mouse	70 million	≈100 billion	1mm^3 complete (0.08%)	No	1,200 x
Marmoset	600 million	≈1 trillion	Not started	No	140 x
Human	86 billion	≈150 trillion	Not started	No	-

The C. elegans Lesson

Despite having the complete C. elegans connectome since 1986 - nearly 40 years ago - functional simulation remains incomplete. As OpenWorm researchers note: “Although we have the complete structural connectome, we do not know the synaptic weights at each of the known synapses. We do not even know whether the synapses are inhibitory or excitatory.”

The OpenWorm project focused on anatomical data from dead worms, but the connectome alone doesn’t specify the relative importance of connections or their dynamic properties. This suggests that structural mapping alone may be insufficient - WBE may require capturing dynamic state information that current preservation methods don’t retain.

Brain Preservation Technologies

A critical prerequisite for WBE is preserving brain structure at sufficient resolution. Recent advances in preservation technology have made significant progress, though key questions remain about what information must be preserved.

Preservation Methods Comparison

Method	Mechanism	Resolution Preserved	Scalability	Key Advantage	Key Limitation
Aldehyde Fixation	Chemical crosslinking	Synaptic (nm-scale)	High	Standard, well-understood	Requires perfusion; destructive
Vitrification	Glass-state freezing	Cellular to synaptic	Moderate	No ice crystal damage	Dramatic brain shrinkage
ASC (Aldehyde-Stabilized Cryopreservation)	Fixation + vitrification	Synaptic verified	High	Won Brain Preservation Prize	Irreversible
Plastination	Polymer replacement	Variable	High	Stable at room temperature	May alter fine structure
Cryonics (current practice)	Vitrification attempt	Uncertain	Limited	Preserves biological viability hope	Not verified to preserve connectome

The Brain Preservation Foundation awarded its Small Mammal Prize in 2018 to 21st Century Medicine for demonstrating that ASC preserves synaptic ultrastructure in a whole pig brain. Their evaluation found “preservation was uniformly excellent: processes were easily traceable and synapses were crisp.”

What Needs to be Preserved?

A fundamental uncertainty is whether the connectome (structural wiring) alone is sufficient, or whether additional information is required:

Information Type	Likely Necessary?	Currently Preservable?	Notes
Connectome (neuron connectivity)	Yes	Yes (ASC verified)	Necessary but possibly not sufficient
Synapse weights	Likely yes	Uncertain	Not directly observable in dead tissue
Synaptic protein composition	Possibly	Partially	The “synaptome” may encode memory
Neuromodulator state	Possibly	No	Dopamine, serotonin levels lost at death
Epigenetic markers	Possibly	Partially	May modulate memory storage
Glial cell states	Unknown	Partially	Astrocytes may contribute to computation

As one review notes: “Many neuroscientists would agree that preserving the connectome alone may not be sufficient to preserve memory. Aspects of what is called the synaptome and perhaps the epigenome may modulate human memory storage as well.”

Safety Implications

Potential Advantages

Advantage	Explanation
Human values by default	Emulation of human has human values (in theory)
Understood entity type	We have experience with humans
Gradual development	Progress is incremental and visible
Legal/ethical frameworks	Could extend human rights frameworks

Risks and Concerns

Anders Sandberg’s analysis of WBE ethics notes that “emulations can be instantiated several times, stopped, deleted, restored from backups and so on. This confuses many ethical systems.”

Risk	Severity	Explanation	Mitigation Difficulty
Identity discontinuity	HIGH	Is the copy the same person? Creating an emulation may be “equivalent to assisted suicide with an unknown probability of success”	Philosophical - cannot be resolved technically
Speed-up risk	HIGH	Emulations could run 1000x+ faster than biological time, making oversight impossible	Requires hard speed limits
Copy proliferation	HIGH	Could create millions of copies; who owns them? Can they vote?	Requires entirely new legal frameworks
Suffering at scale	HIGH	Digital minds might experience suffering; could create vast quantities of suffering entities	Unclear if detectable
Deletion ethics	HIGH	Is deleting an em murder? What about backup restoration?	No current ethical framework applies
Modification risk	MEDIUM	Easier to modify than biological brains; could remove values, add compulsions	Technical access controls possible
Value drift	MEDIUM	Emulations may diverge from human values, especially if running faster	May be unavoidable
Malicious use	MEDIUM	Creating copies of unwilling individuals; torture; forced labor	Requires strong legal protections

Stuart Russell’s warning: Computer scientist Stuart Russell, in Human Compatible, calls creating a WBE-based superintelligence “so obviously a bad idea” due to the control problem - human-derived motivations may not remain stable under self-improvement, and emulations may inherit humanity’s “darker motivations.”

Comparison with AI Safety

Aspect	WBE	AI (Transformers)
Value alignment	Starts aligned (human brain)	Must be trained/aligned
Interpretability	As opaque as human minds	Opaque but different
Speed of development	Slow, predictable	Fast, unpredictable
Controllability	Similar to humans	Unknown
Existential risk	Unclear	Actively debated

Potential Impacts

Impact	Assessment
Labor market	Could create unlimited skilled labor (copies of experts)
Economic growth	Potentially explosive if emulations work faster
Inequality	Who gets emulated? Who controls emulations?
Mortality	Potential path to “immortality” for the wealthy

Robin Hanson’s “Age of Em”

Economist Robin Hanson’s 2016 book The Age of Em provides the most detailed analysis of a world dominated by brain emulations. Drawing on economics, physics, and computer science, Hanson explores the social and economic implications:

Prediction	Mechanism	Implications
Economic doubling every 1-2 weeks	Copying ems is as easy as copying software	Growth rates unimaginable by current standards
Wages at subsistence (compute costs)	Perfect labor markets; unlimited copy supply	Ems work for the cost of running them
Variable-speed operation	Faster ems cost more to run	Speed stratification by wealth/importance
Em “clans”	Copies of successful individuals dominate	A few thousand original humans might provide all labor
Subjective centuries in years	Ems experience time faster than wall-clock	1-2 years of “human time” = ≈1000 years of em experience
Humans become like retirees	Can’t compete economically, but not eliminated	Human wealth fraction falls, but absolute wealth rises

Hanson estimates this scenario could occur within roughly a century if WBE becomes feasible. However, he notes this analysis is conditional on WBE arriving before other transformative AI - a condition that seems increasingly unlikely.

Research Landscape

Key Organizations (2025)

Organization	Focus	Status	Key Output
Allen Institute for Brain Science	Brain mapping and connectomics	Active	Mouse visual cortex connectome (2025)
Open Brain Institute	Successor to Blue Brain Project	Launched March 2025	Open-source brain models
EBRAINS	European brain research infrastructure	Active	Hosts Blue Brain-derived models
Carboncopies	WBE advocacy and coordination	Active	Roadmap updates, community building
Brain Preservation Foundation	Preservation technology validation	Active	Prize competitions, evaluations
Princeton/Janelia Connectomics	Fly and mouse connectomes	Active	Fruit fly complete connectome
Google Connectomics	AI for neural reconstruction	Active	Flood-filling networks, SmartEM

Key Publications and Milestones

Publication/Milestone	Year	Contribution
Whole Brain Emulation: A Roadmap	2008	Foundational technical analysis
The Age of Em	2016	Comprehensive economic/social analysis
Blue Brain neocortical microcircuit	2015	First detailed mammalian simulation
ASC Brain Preservation Prize	2018	Verified preservation quality
Drosophila connectome	2024	First complete adult brain connectome
EM-based connectomics (Method of Year)	2025	Nature Methods recognition
Mouse visual cortex connectome	2025	Largest mammalian connectome
BAAIWorm C. elegans model	2024	Most complete worm simulation

Timeline Analysis

Expert predictions for WBE timelines vary significantly, reflecting deep uncertainty about both technical requirements and the pace of progress.

Published Timeline Estimates

Source	Estimate	Methodology	Notes
Sandberg (2013)	50% by 2064	Expert judgment	Conservative estimate
Kurzweil (2005)	By 2045	Technology extrapolation	Optimistic; relies on continued exponential progress
2024 Technology Trend Analysis	Mouse ~2034, Marmoset ≈2044, Human >2044	Trend extrapolation	Based on supercomputer, connectomics, and activity measurement progress
Henry Markram (2009)	Human brain by 2019	Blue Brain extrapolation	Did not occur - illustrates prediction difficulty
80,000 Hours	This century likely, but after AI	Expert synthesis	Conditional on AI not transforming trajectory first

Milestone Probability Estimates

Milestone	Optimistic (20%)	Median (50%)	Pessimistic (80%)	Key Dependencies
Complete mouse connectome	2028	2032	2040	Scanning speed improvements
Functional mouse brain emulation	2032	2038	2050	Modeling + compute
Human brain scanning feasible	2040	2055	2080	Major scanning breakthrough
Human brain emulation	2045	2065	Never	Multiple breakthroughs needed

Why WBE Will Likely Not Arrive Before AI

Factor	Impact	Confidence
AI progress rate	Deep learning capabilities doubling every 6-18 months	High
WBE progress rate	Connectomics roughly doubling every 3-5 years	High
Current gap	AI approaching human-level; WBE at insect level	High
Investment differential	>$100B/year in AI vs. ≈$1B/year in connectomics	High
Parallel vs. serial	AI research parallelizes; brain scanning is sequential	Medium

A 2011 workshop of AI researchers estimated an 85% probability that neuromorphic AI would arrive before brain emulation. Given subsequent AI progress (GPT-4, etc.), this probability has likely increased.

The “AI Moots It” Scenario

The most likely outcome is that transformative AI arrives before WBE becomes feasible, fundamentally changing the trajectory. This could manifest in several ways:

AI accelerates WBE - Superhuman AI dramatically speeds scanning, reconstruction, and modeling
AI makes WBE unnecessary - If AI achieves similar capabilities, motivation for WBE decreases
AI poses new risks - Resources shift to AI safety rather than WBE development
AI enables alternatives - Brain-computer interfaces or neural enhancement may become more attractive

Relevance to AI Safety

Why Study WBE Despite Low Probability?

Backup path - If AI proves unalignably dangerous, WBE is an alternative
Informs AI safety - Understanding biological intelligence helps understand artificial
Hybrid systems - WBE insights may inform brain-computer interfaces
Comparative analysis - Different path illuminates unique AI risks

Key Differences from AI Safety Concerns

AI Safety Concern	WBE Equivalent
Goal misgeneralization	Human values may be context-dependent
Mesa-optimization	Humans already have inner optimizers
Deceptive alignment	Humans can be deceptive
Capability overhang	Speed-up creates sudden capability jump

Key Uncertainties

The feasibility and timeline of WBE depends on resolving several fundamental uncertainties, both technical and philosophical:

Technical Cruxes

Uncertainty	Optimistic View	Pessimistic View	Resolution Method
Required simulation detail	Neuron-level (10^18 FLOPS) sufficient	Molecular-level (10^25+ FLOPS) required	C. elegans functional validation
Connectome sufficiency	Structure encodes everything	Dynamic state (neuromodulators, etc.) essential	Preservation + revival experiments
Scanning speed ceiling	100x improvement per decade possible	Physical limits near current rates	Engineering progress
Preservation completeness	Current methods preserve enough	Critical information lost at death	Brain Preservation Foundation validation

Philosophical Cruxes

Question	Implications if Yes	Implications if No
Is the copy “you”?	WBE is a path to continuity/immortality	WBE creates a copy, not continuation; original dies
Can digital minds be conscious?	Moral status clear; WBE creates persons	Uncertain moral status; may be “philosophical zombies”
Does continuity require gradual transition?	Destructive scanning acceptable	Must develop non-destructive methods first
Are human values stable under self-modification?	WBE inherits human alignment	WBE may diverge rapidly from human values

Key Crux: The Information Preservation Problem

The most consequential uncertainty may be what information must be preserved. The C. elegans case is instructive: despite having the complete structural connectome for 40 years, functional simulation remains incomplete because:

Synaptic weights are not directly observable in fixed tissue
Inhibitory vs. excitatory nature of synapses must be inferred
Dynamic properties (learning rates, plasticity) are not encoded in structure
Neuromodulator states are lost at death

If these gaps prevent C. elegans emulation, human WBE faces even greater challenges given the brain’s 285-million-fold greater complexity.

Brain-Computer Interfaces - Related technology path
Biological/Organoid Computing - Alternative biological approach
Dense Transformers - The likely faster path