It’s September 3, 2026 (yesterday), and a team from HHMI Janelia, together with collaborators from Cambridge and Google Research, published in Cell the complete wiring diagram of the central nervous system of a male fruit fly: brain and ventral nerve cord, 166,000 neurons, 125 million synapses [4]. It’s the largest brain map ever built, and it weighs less than a grain of rice.
Connectome
A connectome is the complete map of connections of a nervous system: which neuron talks to which, in what direction, and through how many contacts. It’s the electrical wiring diagram, wire by wire.
It’s obtained by slicing tissue into layers a few tens of nanometers thick, photographing each layer with an electron microscope, and letting an artificial neural network trace each neuron through the stacked block. Afterward, people have to manually correct the algorithm’s errors — false mergers, false splits — one by one, for years.
The problem is that the brain is much more than a simple network of neurons connected by axons and dendrites; there are neurons that communicate by releasing monoamines and neuropeptides into the surrounding medium, which diffuse and affect cells they never physically touch [12]. This means a lot of information is still missing.
Mapping connectomes
Adult fly. The FlyWire consortium published the complete brain of a female in 2024: 139,255 neurons and about 50 million synapses [3]. And now the male, full body [4]. Having both sexes allows something new: comparing two complete connectomes of the same species and starting to see how much variability there is between individuals.
The human cubic millimeter, known as H01, was published in 2024 by Alexander Shapson-Coe and Jeff Lichtman’s lab at Harvard, together with Viren Jain’s team at Google [5]. It was extracted from a living person during epilepsy surgery. A fragment that was headed for a biohazard bin became the most detailed map that exists of the human cortex. It contains 57,000 cells, 230 millimeters of blood vessels, and about 150 million synapses, in 1.4 petabytes of images.
That cubic millimeter held surprises: unusual axons showed up making as many as fifty synapses onto a single target neuron, when the norm is one or two. Our anatomical knowledge of the human brain still has gaps at the level of synaptic detail.
Logistical problems
One human cubic millimeter is 1.4 petabytes. A human brain is a bit over one million cubic millimeters. The standard estimate in the field, signed by a group of leaders in the area in 2020, is that a complete mouse brain at synaptic resolution would come out to somewhere around an exabyte, and a human one around a zettabyte [8]. Roughly the annual traffic of the entire internet — for a single skull.
The female fly’s connectome required about 33 person-years of manual correction, spread across more than a hundred labs and a community of volunteers working from home [7]. That works out to about twenty minutes per neuron. Now multiply twenty minutes by 86 billion neurons [36].
The connectome is just a map
Here’s the part that interests me most: even if the complete human connectome appeared tomorrow, we wouldn’t have a mind. We’d have a blueprint.
If you know the diagram but not the ion channel densities, you have hundreds of thousands of brains compatible with your data and no way to know which one was yours.
Daniel Witvliet and his team reconstructed the complete brains of eight genetically identical worms, raised under identical conditions. Around 43% of the cell-to-cell connections aren’t conserved between individuals [10]. Clones — and nearly half the wiring differs. The idea that the C. elegans connectome is a printed circuit was dismantled by its own research community.
Either way, even with the map in hand, when it comes to simulating the brain — with that amount of information, trying to load it into working memory and having software capable of emulating sensory inputs and reacting through the synapses — that’s still science fiction for humans. There’s a long way to go.
But the connectome works
Philip Shiu, in Kristin Scott’s lab at Berkeley, took the fly connectome and built the dumbest possible model: leaky integrate-and-fire neurons, with each synapse labeled only as excitatory or inhibitory based on the predicted neurotransmitter. No ion channels, no neuromodulation, no electrical junctions. He activated the sugar-taste neurons on the computer, and the model predicted which neurons would fire and which ones were needed for the fly to extend its proboscis. Then they went to the lab, activated and silenced those specific neurons with optogenetics in live flies, and the predictions held up [13].
In parallel and independently, Janne Lappalainen and Srinivas Turaga’s group built networks whose connectivity was fixed by the connectome of the fly’s visual system, left the dynamic parameters free, and optimized them for motion detection. The emergent responses matched experimental measurements published across 26 studies that hadn’t been used for training [14].
Two different methods, two different labs, experimental validation in both cases.
Memory
Memories have a physical, localizable, manipulable substrate. In 2012, Susumu Tonegawa’s lab at MIT genetically tagged the dentate gyrus neurons active during fear learning, then reactivated them with light in a neutral context. The mouse froze as if it were reliving the experience [15]. That’s been replicated and extended in many independent labs.
The substrate is continuously recycled. Alessio Attardo, James Fitzgerald, and Mark Schnitzer measured in vivo, using two-photon microendoscopy, the half-life of dendritic spines in the CA1 region of the mouse hippocampus: between one and two weeks, with essentially complete turnover within about a month [16]. The specific synapses holding up a memory today won’t be the same ones two months from now.
David Glanzman, at UCLA, pharmacologically erased a long-term memory in Aplysia, reversed the associated synaptic growth, and then managed to reinstate it with minimal training that wouldn’t have been enough in a naive animal [18]. His reading is that storage doesn’t live in the synapse but in something intracellular, possibly epigenetic. His group went further and transferred RNA from trained slugs into untrained slugs, transferring the sensitization along with it [19]. This is a real, documented disagreement between serious labs.
Nobody knows what the minimum set of variables would be that you’d need to capture to preserve a memory. That’s the real bottleneck. The petabytes are a budget problem, but the real problem is that we still don’t know for certain how memory is encoded in the brain.
And feelings?
Worse still, and for a structural reason. Emotions depend on the global neuromodulatory state, and that state is exactly what a connectome doesn’t capture. Dopamine, serotonin, noradrenaline, and acetylcholine don’t travel down a wire from A to B: they bathe entire regions and reconfigure how everything else is processed. The experimental demonstration of this comes from the same Marder tradition: an identical circuit produces radically different outputs depending on the chemical cocktail it’s immersed in.
On top of that you have to add the body. A simulated brain with no signals from the gut, the heart, and the endocrine system wouldn’t be in some neutral state; it would be in a state no human being has ever been in. We don’t know what that would produce, and on top of that, the part about emulating external senses is still missing.
The global map
In the United States, the NIH’s BRAIN CONNECTS program is distributing $150 million across eleven projects, and it’s worth underlining the actual goal: developing the tools to map a complete mouse brain, not finishing the map [40].
In Japan, the Brain/MINDS program has been working with marmosets since 2014 — a deliberate bet on a small primate instead of the mouse — and has developed tissue-clearing techniques that allow whole neurons to be seen without slicing [22]. In China, the China Brain Project, launched in 2016, combines macaque connectomics, disease models, and brain-inspired computing; its director, Mu-ming Poo, has proposed mapping the cell-type-specific connectome of key macaque brain regions as an intermediate step toward humans [23].
The EPFL’s Blue Brain Project, led by Henry Markram, reconstructed and simulated a rat cortical microcolumn of about 31,000 neurons on a supercomputer in 2015 [24]. It closed at the end of 2024 after nearly twenty years, roughly 300 publications, and a petabyte of released data and tools [25]. It left behind valuable infrastructure and did not leave behind the whole-brain simulation that was announced at the outset. That’s a result, and it deserves to be counted as one.
Small projects
OpenWorm, launched in 2011, is an international open-source project sustained largely by volunteers, whose goal is to simulate a complete C. elegans cell by cell. It has a soft-body physics engine, a neuronal model framework, and a web visualizer, all public on GitHub [26]. And after more than a decade, it still doesn’t reproduce the worm’s behavioral repertoire. We’ve had that animal’s connectome since 1986, we have its complete cell lineage, and we still can’t make it walk convincingly — mostly because electrophysiological data for most of its ion channels is still missing. Anyone who promises you they can emulate a human brain first has to explain why the worm is putting up such a fight.
In December 2024, Lei Ma’s team at Peking University, with the Beijing Academy of Artificial Intelligence, published the BAAIWorm model in Nature Computational Science: a biophysically detailed neural network of the worm’s 302 neurons, coupled in closed loop to a soft body of 3,341 tetrahedra with 96 muscles, inside a real-time 3D fluid environment [27]. The digital critter zigzags toward food because its sensory neurons detect the gradient. Padraig Gleeson, of the OpenWorm team and a reviewer of the paper, called it a major advance. It’s still a worm, and it’s still the most complete digital organism that exists.
The Carboncopies Foundation, founded by neuroscientist Randal Koene in 2012, is a nonprofit that promotes and coordinates research toward whole-brain emulation, and has proposed sensible things like generating synthetic brain data with known ground truth so researchers can evaluate whether an emulation pipeline actually works [28]. It’s not an experimental lab; it’s a roadmap-and-outreach organization.
The 2045 Initiative, funded by Russian businessman Dmitry Itskov since 2011, laid out a four-phase ladder toward transferring a consciousness to a non-biological avatar by 2045 [29]. Its peer-reviewed scientific output is practically nonexistent, and its public activity has been dormant for years.
The Terasem Movement Foundation has offered, since 2006, free to anyone, the chance to create a “mindfile” on LifeNaut: you upload your memories, your photos, your answers to personality questionnaires, and the foundation stores them on the hypothesis that some future technology will be able to reconstruct you from that data. They even offer to beam your data into deep space in case something happens to Earth [30]. This isn’t connectomics and doesn’t claim to be. It’s a biographical archive with a metaphysical promise attached, and it needs to be kept separate from the other projects here, since it has no scientific method behind it.
Preserve today to digitize tomorrow
Another option is to preserve tissue with its ultrastructure intact until the technology catches up.
The reference technique is aldehyde-stabilized cryopreservation, published by Robert McIntyre and Gregory Fahy in Cryobiology in 2015. The brain is perfused with glutaraldehyde, which fixes the structure within minutes, and then with cryoprotectant to vitrify it without ice formation [31]. The Brain Preservation Foundation, which independently evaluated the samples with electron microscopy, awarded its small-mammal prize in 2016 for a rabbit, and its large-mammal prize in 2018 for a pig. Preservation of synaptic connectivity in a complete large brain has, in fact, been demonstrated and independently verified.
If the digital copy wakes up, who wakes up? Because it isn’t you
Electron microscopy scanning is destructive: the brain gets sliced apart. So what’s left is a corpse and a program. Derek Parfit devoted a large part of Reasons and Persons to arguing that the question “is it the same person?” is badly framed — that what matters is psychological continuity, and that it comes in degrees [34].
My own guess is that the first “uploaded mind” to convince anyone won’t be human, and it won’t be a worm either: it’ll be a larval zebrafish. It’s the only vertebrate with a brain small and transparent enough to be mapped in its entirety.
If Glanzman is right even partially [18], geometry of the synapses won’t be enough: you’d need the internal molecular state of every cell. That multiplies the data volume by several orders of magnitude and makes obsolete any cost estimate circulating today — including the ones cited above.
References
[1] Dorkenwald, S. et al., FlyWire Consortium (2024). Neuronal wiring diagram of an adult brain. Nature 634(8032), 124–138. doi:10.1038/s41586-024-07558-y — Reliable. Methodology: transmission electron microscopy of serial sections of a complete adult female brain, automated segmentation with convolutional networks, and manual correction distributed across more than a hundred labs and external volunteers.
[2] HHMI Janelia, Cambridge Connectomics Group and Google Research (2026). Sexual dimorphism in the complete connectome of the Drosophila male central nervous system. Cell. doi:10.1016/j.cell.2026.08.015 — Recent. Methodology: same pipeline, extended to the complete ventral nerve cord; 166,000 neurons and 125 million synapses, annotated and verified by Janelia experts. Published September 3, 2026, alongside three papers already applying it to visual systems, taste, and social behavior.
[3] Shapson-Coe, A. et al. (2024). A petavoxel fragment of human cerebral cortex reconstructed at nanoscale resolution. Science 384(6696), eadk4858. doi:10.1126/science.adk4858 — Reliable. Methodology: one cubic millimeter of human temporal cortex removed during epilepsy surgery, fixed, stained with heavy metals, sectioned and imaged by electron microscopy; 1.4 petabytes automatically reconstructed. The authors themselves note the volume has not been fully hand-reviewed.
[4] Nature Reviews Neuroscience (2024). Fly connectome over the wire (research review on the FlyWire connectome). nature.com/articles/s41583-024-00879-x — Reliable. Source of the 33 person-years manual-correction figure, estimated by the consortium itself from the platform’s edit history.
[5] Abbott, L.F. et al. (2020). The mind of a mouse. Cell 182(6), 1372–1376. doi:10.1016/j.cell.2020.08.010 — Theoretical framework. Position paper signed by a group of leaders in the field, with estimates of data volume, cost, and timeline for whole-brain connectomes.
[6] Witvliet, D. et al. (2021). Connectomes across development reveal principles of brain maturation. Nature 596(7871), 257–261. doi:10.1038/s41586-021-03778-8 — Reliable. Methodology: eight isogenic C. elegans individuals of different ages, fully reconstructed by serial-section electron microscopy, with direct comparison among them and against the 1986 connectome.
[7] Bentley, B. et al. (2016). The multilayer connectome of Caenorhabditis elegans. PLoS Computational Biology 12(12), e1005283. doi:10.1371/journal.pcbi.1005283 — Reliable. Methodology: reconstruction of the monoamine networks and part of the neuropeptide networks from biosynthetic- and receptor-gene expression data, topologically compared against the synaptic and electrical-junction networks.
[8] Shiu, P.K. et al. (2024). A Drosophila computational brain model reveals sensorimotor processing. Nature 634(8032), 210–219. doi:10.1038/s41586-024-07763-9 — Reliable. Methodology: leaky integrate-and-fire model built using only connectivity and predicted neurotransmitter, later validated with optogenetic activation and silencing in live flies. The model deliberately excluded electrical junctions and neuromodulation, and still predicted behavior.
[9] Lappalainen, J.K. et al. (2024). Connectome-constrained networks predict neural activity across the fly visual system. Nature 634(8036), 1132–1140. doi:10.1038/s41586-024-07939-3 — Reliable. Methodology: mechanistic network with connectivity fixed by the connectome for 64 cell types in the optic lobe and unknown dynamic parameters, optimized via deep-learning techniques to detect visual motion; predictions were checked against measurements from 26 prior studies.
[10] Liu, X. et al. (2012). Optogenetic stimulation of a hippocampal engram activates fear memory recall. Nature 484(7394), 381–385. doi:10.1038/nature11028 — Reliable. Methodology: activity-dependent genetic tagging of dentate gyrus neurons active during fear conditioning, followed by reactivation with light in a different context. Replicated and extended in multiple independent labs.
[11] Attardo, A., Fitzgerald, J.E. and Schnitzer, M.J. (2015). Impermanence of dendritic spines in live adult CA1 hippocampus. Nature 523(7562), 592–596. doi:10.1038/nature14467 — Reliable. Methodology: two-photon microendoscopy at intervals in CA1 of live mice over weeks, with kinetic modeling of turnover and resolution controls via STED microscopy to rule out nearby spines being counted as one.
[12] Chen, S., Cai, D., Pearce, K., Sun, P.Y-W., Roberts, A.C. and Glanzman, D.L. (2014). Reinstatement of long-term memory following erasure of its behavioral and synaptic expression in Aplysia. eLife 3, e03896. doi:10.7554/eLife.03896 — With reservations. Methodology: co-cultures of Aplysia sensory and motor neurons sensitized with spaced serotonin pulses, followed by reconsolidation blockade and PKM inhibition, with morphological counts of presynaptic varicosities and behavioral experiments in intact animals.
[13] Bédécarrats, A., Chen, S., Pearce, K., Cai, D. and Glanzman, D.L. (2018). RNA from trained Aplysia can induce an epigenetic engram for long-term sensitization in untrained Aplysia. eNeuro 5(3), ENEURO.0038-18.2018. doi:10.1523/ENEURO.0038-18.2018 — With reservations. Methodology: RNA extraction from the nervous system of sensitized animals and injection into untrained animals, measuring the withdrawal reflex with controls using RNA from untrained animals.
[14] Okano, H., Sasaki, E., Yamamori, T. et al. (2016). Brain/MINDS: a Japanese national brain project for marmoset neuroscience. Neuron 92(3), 582–590. doi:10.1016/j.neuron.2016.10.018 — Reliable. Methodology: description of the Japanese national program, combining structural and functional MRI, tissue clearing, projection tracing, and genetically modified marmosets to model disease. Data is publicly accessible.
[15] Poo, M., Du, J., Ip, N.Y., Xiong, Z.-Q., Xu, B. and Tan, T. (2016). China Brain Project: basic neuroscience, brain diseases, and brain-inspired computing. Neuron 92(3), 591–596. doi:10.1016/j.neuron.2016.10.050 — Reliable. Program paper signed by its directors; describes the three-map strategy (cell types, connectivity, activity) centered on the macaque.
[16] Markram, H., Muller, E., Ramaswamy, S. et al. (2015). Reconstruction and simulation of neocortical microcircuitry. Cell 163(2), 456–492. doi:10.1016/j.cell.2015.09.029 — Reliable as far as the published reconstruction goes. Methodology: digital reconstruction of a young rat somatosensory cortical microcolumn of about 31,000 neurons from morphological and electrophysiological data, simulated on a supercomputer using the NEURON engine.
[17] EPFL Blue Brain Project (2024). Project closure after nearly twenty years. epfl.ch/research/domains/bluebrain — Reliable as institutional documentation. The project concluded at the end of 2024 with about 300 publications, a petabyte of data, and an open platform, without reaching the whole-brain simulation laid out in its original proposal.
[18] Sarma, G.P. et al. (2018). OpenWorm: overview and recent advances in integrative biological simulation of Caenorhabditis elegans. Philosophical Transactions of the Royal Society B 373(1758), 20170382. doi:10.1098/rstb.2017.0382 — Reliable. Methodology: description of the simulation architecture (c302 for the neural network, Sibernetic for the body and fluid, Geppetto for visualization) and its limitations. The authors themselves acknowledge that the level of detail achieved remains insufficient for biological research. All code is open source.
[19] Zhao, M., Ma, L. et al. (2024). An integrative data-driven model simulating C. elegans brain, body and environment interactions. Nature Computational Science. doi:10.1038/s43588-024-00738-w — Recent. Methodology: multicompartment neural network of the 302 neurons with synapses and electrical junctions modeled in detail, validated against patch-clamp recordings from five representative neurons, coupled in closed loop to a finite-element soft body with 96 muscles in a real-time 3D fluid environment. Published December 2024, featured on the journal’s cover; open-source code. (Note: confirm the exact author order in the DOI before publishing.)
[20] Carboncopies Foundation. carboncopies.org — Theoretical framework. Nonprofit organization founded by Randal Koene dedicated to roadmapping and outreach for whole-brain emulation. Not an experimental lab; its output consists of workshops, reviews, and methodological proposals such as the Brain Emulation Challenge.
[21] 2045 Initiative. 2045.com — Problematic. Organization founded in 2011 by Dmitry Itskov with a phased plan toward transferring a consciousness to a non-biological avatar by 2045. Included for its media relevance: it has produced no peer-reviewed literature supporting its goals, and its public activity has been stalled for years.
[22] Terasem Movement Foundation / LifeNaut. terasemmovementfoundation.com/lifenaut — Problematic. Free biographical “mindfile” creation service since 2006, with the explicit hypothesis that future technology will be able to reconstruct the person from them. No experimental result supports that inference. Included because it shares vocabulary with connectomics without sharing its method.
[23] McIntyre, R.L. and Fahy, G.M. (2015). Aldehyde-stabilized cryopreservation. Cryobiology 71(3), 448–458. doi:10.1016/j.cryobiol.2015.09.003 — Reliable. Methodology: glutaraldehyde perfusion followed by cryoprotectant and storage at cryogenic temperature, with ultrastructure evaluation by electron microscopy after rewarming. Independently verified by the Brain Preservation Foundation in complete rabbit and pig brains.
[24] Parfit, D. (1984). Reasons and Persons. Oxford University Press — Theoretical framework. Philosophical analysis of personal identity. Contains no empirical data and doesn’t claim to.
[25] Azevedo, F.A.C. et al. (2009). Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain. Journal of Comparative Neurology 513(5), 532–541. doi:10.1002/cne.21974 — Reliable. Methodology: isotropic fractionator, which dissolves tissue into a homogeneous suspension of cell nuclei and counts them, applied to complete human brains from four donors.
[26] AA.VV. (2025). State of Brain Emulation Report 2025. arXiv:2510.15745 — Recent. State-of-the-art review, not peer-reviewed. Useful for mapping projects, funding, and comparisons between whole-organism models; should not be used as a source of primary experimental results