Whole Brain Emulation

The Basic Pipeline

Whole brain emulation requires three fundamental capabilities working together: scanning, interpretation, and simulation. Each presents distinct scientific and engineering challenges, and all three must be solved for WBE to become possible.

Scanning means capturing the physical structure of a brain at whatever resolution is necessary to preserve its functional properties. At minimum, this means identifying every neuron, every synapse, and the properties of each synapse (its strength, type, and location on the target neuron). It may also require capturing the distribution of ion channels across each neuron, the concentration of neuromodulators, and other molecular details whose functional importance is not yet fully understood. Current scanning approaches rely on electron microscopy of fixed tissue, which means the process is destructive, the brain must be removed, preserved, sliced, and imaged. Non-destructive scanning at synaptic resolution remains far beyond current technology.

Interpretation means converting the raw scan data into a computational model. This requires identifying individual neurons in the imaging data (segmentation), tracing their connections (reconstruction), classifying synapse types, and estimating functional parameters like synaptic weights. Machine learning has dramatically accelerated this process, but human proofreading is still required for accuracy, and the error rates of automated methods can compound when applied to large volumes of tissue.

Simulation means running the resulting computational model on hardware that can execute it at biologically relevant speeds. The computational requirements depend on the level of biological detail in the model. A simulation that models every ion channel in every neuron of a human brain would require computational resources far beyond anything currently available. A simulation using simplified neuron models might be feasible with current or near-future supercomputers, but it is uncertain whether simplified models preserve enough biological detail for faithful emulation.

Current State of Progress

Meaningful progress toward whole brain emulation is currently confined to small organisms where comprehensive connectomic datasets have become available. The nematode C. elegans, with its 302 neurons and complete connectome, has been the primary testbed. The OpenWorm project has built multi-scale, closed-loop simulations that reproduce basic behaviors like locomotion and chemotaxis by integrating neural dynamics, body mechanics, and environmental interaction.

However, despite having the complete wiring diagram of C. elegans for nearly 40 years, there is still no validated model that reproduces all of the worm behavioral repertoire. This sobering fact illustrates a crucial point: even with a complete connectome, WBE requires understanding how the structural connections translate into functional dynamics, and this understanding remains incomplete even for the simplest nervous systems.

The 2024 completion of the Drosophila connectome (roughly 140,000 neurons) represents a major step forward, but functional emulation of the fly brain remains a distant goal. While researchers can now model specific circuits within the fly brain, such as the circuits underlying visual motion detection or olfactory processing, integrating these into a coherent whole-brain emulation is far beyond current capabilities.

For mammalian brains, the challenges are orders of magnitude greater. The largest densely reconstructed volume of mouse cortex covers just one cubic millimeter and contains approximately 120,000 neurons. The complete mouse brain contains roughly 70 million neurons, and the human brain contains 86 billion. Neither the scanning pipeline nor the computational models are close to ready for emulation at these scales.

Computational Requirements

The computational resources required for WBE depend critically on the level of biological detail that must be simulated. Estimates range across many orders of magnitude depending on assumptions about which details are functionally important.

At the simplest level, modeling each neuron as a point integrate-and-fire unit, a human brain emulation would require simulating 86 billion neurons and roughly 100 trillion synaptic connections in real time. This is within reach of current supercomputing technology, at least in terms of raw computational capacity, though the memory requirements for storing synaptic weights are substantial (on the order of hundreds of terabytes) and the communication bandwidth requirements are challenging.

At the most detailed level, modeling every ion channel and dendritic compartment, the computational requirements increase by roughly a factor of 10,000 to 100,000 compared to point neuron models. A biophysically detailed simulation of the human brain would require computational resources comparable to those used to train the largest current AI models, and it would need to run these computations continuously in real time rather than in a batch training mode.

The energy comparison is striking. The biological brain operates on roughly 20 watts. Current estimates suggest that a biophysically detailed brain emulation would require megawatts of power, roughly 100,000 times less efficient than biology. Even neuromorphic hardware, which is far more efficient than conventional computers for neural simulation, is still orders of magnitude less efficient than biological neural tissue.

What Level of Detail Is Needed

The fundamental open question in whole brain emulation is: what level of biological detail must be captured for a simulation to be functionally equivalent to the original brain? This is sometimes called the "resolution question," and the answer determines whether WBE is a near-term possibility or a challenge for future centuries.

The optimistic view holds that the essential computations of the brain are carried out by the patterns of spiking activity in neural networks, and that these patterns are determined primarily by the connectivity and synaptic weights of the network. If this view is correct, then a connectome plus synapse-level parameters might be sufficient, and WBE could become feasible as connectomics technology continues to improve.

The pessimistic view holds that essential computations occur at the molecular level, in the specific ion channel distributions of individual neurons, in the dynamics of intracellular signaling cascades, in the behavior of individual dendritic spines, and possibly in the quantum mechanical properties of biological molecules. If this view is correct, then WBE would require scanning and modeling at a resolution far below the synaptic level, pushing the timeline out by decades or more.

The evidence is currently inconclusive. Some experimental results suggest that neural circuit function is remarkably robust to perturbations of individual synapses and channels, supporting the optimistic view. Others show that subtle molecular-level changes can dramatically alter circuit behavior, supporting the pessimistic view. Resolving this question is one of the most important tasks in neuroscience, not just for WBE but for our fundamental understanding of how brains compute.

Timeline Estimates

Projections for when WBE might become possible vary enormously depending on assumptions about the required level of biological detail, the pace of technological progress, and the amount of resources devoted to the effort.

For simple organisms, functional emulation of C. elegans behavior is likely achievable within the current decade, as connectomic data and neuron models continue to improve. A functional emulation of the Drosophila brain might be possible in the 2030s, assuming continued progress in both connectomics and simulation technology.

For mammals, the timeline is far more uncertain. Mouse whole-brain simulation at the cellular level (using simplified neuron models) is projected to be feasible in the 2030s by some researchers, while whole-brain emulation that includes detailed biophysics is unlikely before the 2040s. For the human brain, most serious estimates place even a simplified WBE several decades away, with biophysically detailed emulation remaining far beyond the foreseeable technological horizon.

These estimates carry substantial uncertainty. The history of both neuroscience and computing technology shows that progress often comes in unexpected bursts, driven by technological breakthroughs that were difficult to predict in advance. The development of automated connectomics, the scaling of neuromorphic hardware, or advances in our theoretical understanding of neural computation could accelerate the timeline significantly.

Ethical and Philosophical Implications

If whole brain emulation becomes possible, it would raise profound ethical and philosophical questions. Would a brain emulation be conscious? Would it have the same moral status as the biological person it was copied from? Who would own or control a brain emulation? Could brain emulations be copied, modified, or deleted, and what would the ethical implications of those actions be?

These questions are not merely hypothetical. They connect to ongoing debates in philosophy of mind about the nature of consciousness, personal identity, and moral status. The functionalist view, which holds that mental states are defined by their functional roles rather than their physical substrate, implies that a perfect emulation would be conscious and would be the same person as the original. Biological naturalist views, which tie consciousness to specific biological processes, would deny this.

The practical implications extend to economics and society. Brain emulations could potentially be run at different speeds (faster or slower than biological real time), copied to create multiple instances, or modified to enhance cognitive capabilities. These possibilities would have transformative effects on labor markets, intellectual property, and the meaning of individual identity.