If you could upload your mind, the first problem would be size: mapping a human brain at the resolution of a 2024 Harvard and Google scan would produce about 1.7 zettabytes of images, roughly a million times the largest brain dataset made so far. The second problem is identity: even a perfect, running copy of your connectome would be a new person who remembers being you, and philosophers still disagree on whether that counts as you surviving. Today the most complete wiring maps are of a fruit fly, about 139,000 to 166,000 neurons.
- 1.4 petabytes: one cubic millimetre of human cortex imaged at nanometre resolution (H01, Science, 2024).
- A 1,200 cm³ brain is 1.2 million such cubes, so about 1.7 zettabytes raw.
- The same brain's synapses, at 4.7 bits each, compress to roughly 100 TB: the copy is small, the scan is not.
- Largest complete connectomes: 139,255 neurons (female fly brain, 2024) and 166,000+ (male fly CNS, 2026).
- Philosophy's catch: on Parfit's view, what matters is psychological continuity, and a copy that branches off may not be you.
"Uploading" usually means three steps: preserve the brain, scan its structure in enough detail, and run that structure as software. Each step has had real progress since 2015. I will go through them with the numbers, then come to the question no microscope can answer.
What would you need to copy to upload a mind?
At minimum you would need the connectome: every neuron and every synapse, with each synapse's strength. Most neuroscientists think more is needed too, such as receptor types, neuromodulators and possibly glial cells, but the connectome is the part we can measure today.
- Neurons: about 86.1 billion in an adult male brain (Azevedo et al., 2009).
- Synapses: a cubic millimetre of human temporal cortex held about 150 million.
- Strength: synapse size tracks strength, and rat hippocampus showed at least 26 distinguishable sizes, or 4.7 bits each (Bartol et al., 2015).

The worm is the cautionary tale. John White and colleagues published the first complete set of synaptic connections of an animal in 1986, and 26 years passed before computers made another study of that kind possible. Even with the worm's map in hand, how its 302 neurons produce behaviour is still being worked out. A map is necessary, not sufficient.
In February 2018, a startup called Nectome, founded by MIT-trained computer scientist Robert McIntyre and Michael McCanna, preserved the brain of an elderly woman starting about 2.5 hours after her death. Its method, aldehyde-stabilized cryopreservation, had already won an $80,000 prize from the Brain Preservation Foundation for preserving a pig brain so well that individual synapses could be seen under an electron microscope. By March, 25 people had paid a refundable $10,000 deposit to join a waiting list for a future version done under anaesthesia while still alive, a procedure the company itself described as 100 percent fatal. Neuroscientists pushed back hard: no one could show that preserved tissue holds memories.
MIT Technology Review: A startup is pitching a mind-uploading service that is "100 percent fatal" (2018)Rabbit and pig brains were perfused with a glutaraldehyde fixative, then with rising concentrations of ethylene glycol up to 65% w/v, and vitrified at −135 °C. After rewarming, electron microscopy across many regions showed uniformly excellent preservation, with traceable neurites and crisp synapses. Limit: the fixative kills the tissue first, so it preserves structure, not function, and it proves nothing about memory.
Read the abstractHow big is the biggest brain map so far?
The biggest complete maps are of fruit flies; the biggest mammalian maps are single cubic millimetres. Each one is a landmark, and each one is tiny next to a human brain.
See the data as a table
| Synapses (million synapses) | |
|---|---|
| Female fly whole brain (FlyWire, 2024) | 50 |
| Male fly brain and nerve cord (2026) | 125 |
| Human cortex, 1 mm³ (H01, 2024) | 150 |
| Mouse visual cortex, 1 mm³ (MICrONS, 2025) | 523 |
Source: Dorkenwald et al. Nature 2024; Google Research and Janelia 2026; Shapson-Coe et al. Science 2024; NIH/Allen Institute via ScienceDaily 2025 · A human brain has an estimated 100,000,000+ million synapses, too large to show on this scale.
A cubic millimetre of human temporal cortex, removed during epilepsy surgery, was imaged by electron microscopy and reconstructed. It contained about 57,000 cells, about 230 mm of blood vessels and about 150 million synapses, and the dataset is 1.4 petabytes. Glia outnumbered neurons two to one, and among thousands of weak inputs some neurons received rare powerful inputs of up to 50 synapses from one axon. Limit: one sample, from a patient with epilepsy, and the reconstruction still needs proofreading.
Read the paper (free)A whole-brain wiring diagram of an adult female Drosophila melanogaster: 139,255 neurons and about 5 × 107 chemical synapses, reconstructed from electron microscopy, with cell types and predicted neurotransmitters. The team traced pathways from photoreceptors to motor outputs. Limit: one fly, and a wiring diagram without the synapse-by-synapse dynamics.
Read the paper (free)
How much data would a whole human brain scan produce?
About 1.7 zettabytes at H01 resolution, or roughly the same at MICrONS resolution. Here is the full working.
Assumptions
- Brain volume V: 1,130 cm³ (women) to 1,260 cm³ (men); use 1,200 cm³ = 1.2 × 106 mm³.
- Data per mm³ D: 1.4 PB (H01, human) or 1.6 PB (MICrONS, mouse).
- Synapse density ρ: 1.5 × 108 per mm³ from H01, applied to the whole brain. This is a simplification: white matter has far fewer, cerebellum is different.
- Bits per synapse b: 4.7.
- Raw images. \( 1.4\ \text{PB/mm}^3 \times 1.2\times10^{6}\ \text{mm}^3 = 1.68\times10^{6}\ \text{PB} = 1.7\ \text{ZB} \).
- Range. From 1,130 to 1,260 cm³ and 1.4 to 1.6 PB, the total is 1.6 to 2.0 ZB.
- Synapses. \( N = \rho V = 1.5\times10^{8} \times 1.2\times10^{6} = 1.8\times10^{14} \), an upper-end estimate.
- Compressed connectome. \( N \times b = 1.8\times10^{14} \times 4.7 = 8.5\times10^{14} \) bits, about 106 TB.
- Compression ratio. \( 1.68\times10^{21} / 1.06\times10^{14} \approx 1.6\times10^{7} \): the scan is sixteen million times larger than the map you would keep.
The good news hiding here: the thing you would actually run is small. A few hundred hard drives could hold 106 TB. The hard part is getting it, because you cannot know which pixels matter until you have segmented all of them.
And the time?
The MICrONS cubic millimetre took about seven years from recording to release, cut into more than 25,000 slices. At that pace, \( 1.2\times10^{6} \times 7 \approx 8.4 \) million years for one brain. To finish in a decade you need a speed-up of about 840,000 times, from faster microscopes, more of them in parallel and automated proofreading.
Use this to check what you just read: change the volume, the data density or the speed-up.
How long and how big is a whole-brain scan?
Could a computer run the copy?
Possibly, if the simple model is enough, and nobody knows that it is. A rough count of the work, using my assumptions rather than measured values:
- Synaptic events per second: \( 1.8\times10^{14} \) synapses times an average activation rate of 0.1 to 10 Hz gives about \( 10^{13} \) to \( 10^{15} \) events per second.
- Cost per event: assume about 10 arithmetic operations, giving \( 10^{14} \) to \( 10^{16} \) operations per second.
- Compare: an exascale supercomputer does \( 10^{18} \) operations per second by definition, 100 to 10,000 times more.
So a point-neuron simulation fits in today's largest machines. If you instead need molecular detail at every synapse, the cost rises by many orders of magnitude, and nobody can yet say which level is required. The fly connectome is the test bed: if a simulated fly behaves like a fly, the simple model gains credibility.
Calcium imaging recorded about 75,000 neurons in the visual cortex of an awake mouse watching natural and synthetic movies; the same tissue was then reconstructed by electron microscopy, with more than 200,000 cells and about 0.5 billion synapses. Linking activity to wiring revealed connectivity rules between excitatory neurons across areas. Limit: one mouse, one cubic millimetre, and proofreading covered only a subset of neurons.
Read the paper (free)
Would the uploaded mind be you?
That depends on which theory of personal identity you hold, and none has won. The Stanford Encyclopedia of Philosophy's entry, last revised in 2023, lays out the main options.
| View | What makes you you | Verdict on the upload |
|---|---|---|
| Psychological continuity | Memories, beliefs and character inherited causally | Survives, if the copy runs from your brain's state |
| Non-branching version | Continuity, as long as only one successor exists | Survives only if the original is destroyed; two copies means neither is you |
| Brute physical (animalism) | Being this living organism | Does not survive; the copy is someone else |
| Parfit: identity is not what matters | Connectedness and continuity, not identity | What you care about can survive, even if "you" strictly does not |
The sharpest test is fission. If a scan leaves your biological brain intact and starts a copy, there are now two beings continuous with you. They cannot both be you, so either the simple continuity view is wrong, or survival does not require identity, which is Derek Parfit's conclusion.
Indian philosophy adds a third angle. The Gita (2.22) compares death to discarding worn clothes; on that view the self is neither the body nor the pattern, so an upload would be another garment, not the wearer. Buddhist thought, by contrast, denies a fixed self altogether and would ask what exactly you hope to move. Both are explored in atman, anatta and the neuroscience of self and what gets reborn without a self.
The scan is a storage problem. Whether the copy is you is not a storage problem at all.
No law of physics forbids copying a brain's structure. The barriers are scale and knowledge: about 1.7 ZB of imaging at H01 resolution, an 840,000-fold speed-up to scan one brain in a decade, and no proof that a connectome plus 4.7-bit synapse strengths is enough to run a mind. The changes needed are much faster electron microscopy, automated proofreading and a simulated fly or mouse that behaves like the real one. Even then, what you would get is a successor that remembers being you; whether that is you depends on a philosophical choice no experiment can settle.
What's new
- Complete male fruit fly brain and nerve cord mapped: 166,000+ neurons, about 125 million synapses (Cell) Google Research, Sep 2026
- MICrONS: a cubic millimetre of mouse brain with 523 million synapses and 4 km of axons, 1.6 PB ScienceDaily, Apr 2025
- Single-neuron connectivity mapped across the whole mouse brain from 20,247 neuron reconstructions Nature Methods, 2025
What if next?
- What if humans didn't need sleep? An uploaded mind would not need to sleep. A biological one apparently does, and the reasons are surprising.
- What would the Gita say to an AI? If the upload is software, the Gita's advice on acting without clinging becomes a design spec.
- What if you could never forget anything? A perfect copy of your memory might be the last thing you want.
Questions people ask
Is mind uploading possible in our lifetime?
Not on current trends. The largest complete map is of a fruit fly brain, and one cubic millimetre of human cortex produced 1.4 PB. A whole human brain would be about 1.2 million such cubes, and nobody has yet shown that a map alone can be run as a working mind.
How much storage would a human brain upload need?
About 1.7 zettabytes of raw electron-microscope images at 2024 resolution. The extracted connectome would be far smaller, roughly 100 TB if each of about 1014 synapses is stored at 4.7 bits.
Would an uploaded mind be conscious?
Nobody knows. There is no accepted test for consciousness in a machine, and theories disagree on whether structure and function are enough or whether biology matters.
What is a connectome?
A complete map of the neurons in a nervous system and the synapses between them. Complete ones exist for the roundworm C. elegans (302 neurons) and the adult fruit fly.
Sources
- Shapson-Coe A. et al.: A petavoxel fragment of human cerebral cortex, Science (2024) (source for 1.4 PB, 57,000 cells and 150 million synapses per mm³)
- Dorkenwald S. et al.: Neuronal wiring diagram of an adult brain, Nature (2024) (source for 139,255 neurons and 5 × 10⁷ synapses)
- MICrONS Consortium: Functional connectomics spanning multiple areas of mouse visual cortex, Nature (2025)
- ScienceDaily: MICrONS release (2025) (source for 523 million synapses, 4 km, 1.6 PB, 25,000 slices and about seven years)
- Google Research: A connectomics milestone, mapping the complete male fruit fly brain (2026) (source for 166,000+ neurons and 125 million synapses)
- McIntyre R.L., Fahy G.M.: Aldehyde-stabilized cryopreservation, Cryobiology (2015)
- Regalado A.: A startup is pitching a mind-uploading service that is "100 percent fatal", MIT Technology Review (2018) (source for the Nectome story)
- Bartol T.M. et al.: Nanoconnectomic upper bound on the variability of synaptic plasticity, eLife (2015) (source for 4.7 bits per synapse)
- Azevedo F.A. et al.: Equal numbers of neuronal and nonneuronal cells, J Comp Neurol (2009) (source for 86.1 billion neurons)
- Emmons S.W.: The beginning of connectomics, a commentary on White et al. (1986), Phil Trans R Soc B (2015)
- Wikipedia: Human brain (accessed 2026) (source for 1,130 to 1,260 cm³)
- Olson E.T.: Personal Identity, Stanford Encyclopedia of Philosophy (rev. 2023) (source for the identity views, fission and Parfit)
More on the mind and machines on the Manthan world page and Applied AI.
