The Sandberg and Bostrom whole brain emulation roadmap eighteen years on
In 2008, Anders Sandberg and Nick Bostrom published Whole Brain Emulation: A Roadmap (Technical Report #2008-3, Future of Humanity Institute, Oxford University), formalizing the scientific and computational milestones required to upload a biological brain into synthetic hardware. Rather than treating mind uploading as a single speculative leap, the Sandberg-Bostrom roadmap broke the engineering challenge down into three distinct, measurable phases: high-throughput scanning of biological tissue, automated reconstruction and segmentation of neural connectomes, and real-time execution of the reconstructed electrophysiology on computational hardware.
Eighteen years later in 2026, the scanning and segmentation phases have produced milestones that were theoretical targets when the roadmap was drafted. Connectomics consortia have mapped the complete synaptic wiring diagram of the fruit fly Drosophila melanogaster (FlyWire, 2024, resolving roughly 139,000 neurons and 54 million synapses) and reconstructed a one-cubic-millimeter sample of human temporal cortex (Harvard and Google, 2024, cataloging 57,000 cells and 150 million synapses across 1.4 petabytes of imaging data).
Yet these structural achievements have brought the central theoretical challenge of the roadmap into sharper focus: possessing an exact static wiring diagram does not automatically produce a functioning, conscious emulation.
| Emulation level | Description | Biophysical detail captured | Computational cost per brain |
|---|---|---|---|
| Level 1: Population model | Coarse brain regions, mean-field firing rates | Macro-scale anatomical pathways | $10^{11}$ FLOPS (standard desktop) |
| Level 2: Point neuron | Spiking network (LIF / Izhikevich) | Individual spike timings, basic thresholds | $10^{15}$ FLOPS (single GPU server) |
| Level 3: Single-compartment | Spiking with simple adaptation | Spike-frequency adaptation, basic conductances | $10^{16}$ FLOPS |
| Level 4: Multi-compartment | Spatially branched dendritic trees | Dendritic computation, active ion channels | $10^{18}$ FLOPS (exascale cluster) |
| Level 6: 3D electrodiffusion | Extracellular fields, ion concentration dynamics | Ephaptic coupling, local field potentials | $10^{22}$ FLOPS |
| Level 7: Molecular / Proteomic | Second-messenger cascades, gene expression | Synaptic plasticity, neuromodulation states | $10^{25}$ FLOPS |
The Scale versus Resolution Trade-Off
The foundational insight of the Sandberg-Bostrom roadmap is that whole brain emulation is not a binary capability, but an 11-level hierarchy of scale and resolution.
At Level 2, the brain is modeled as a network of point neurons interacting via discrete spike events. As demonstrated in the analysis of Eugene Izhikevich’s 2D dynamical neuron models, point-neuron approximations can capture regular spiking, bursting, and adaptation with modest computational overhead (roughly $10^{15}$ floating-point operations per second for a human cortex).
However, as one descends toward Level 4 (multi-compartment dendritic trees) and Level 6 (3D electrodiffusion and extracellular field dynamics), the computational requirements expand by seven orders of magnitude. Whether an emulation needs to operate at Level 2, Level 4, or Level 7 depends entirely on which level of biophysical organization is constitutive of conscious experience:
- Computational Functionalism: If subjective experience depends solely on the abstract informational and causal relationships among functional units, as Hans Moravec argued in his gradual neural replacement thought experiment, then Level 2 or Level 3 point-neuron spiking models are sufficient to preserve both cognitive function and phenomenality.
- Substrate-Sensitive and Biological Naturalism: If phenomenal consciousness requires specific continuous electrochemical dynamics, astrocytic modulations, or ephaptic field interactions, as Ned Block argues in his meat machines analysis, then an emulation operating at Level 2 will produce an unconscious behavioral simulator, an unwitting philosophical zombie, while the physical conditions for qualia remain uninstantiated.
The Emulation Gap: Connectomes versus Dynamical State
The primary lesson of recent connectomics research is the reality of the “Emulation Gap.” A connectome is a static structural map, an anatomical snapshot frozen in resin or chemical fixative.
A biological brain, however, is a non-equilibrium dynamical system. The functional state of a living neural circuit is determined not only by which axon connects to which dendrite, but by invisible physiological variables that static electron microscopy cannot resolve:
- Neuromodulatory Wash: Ambient concentrations of dopamine, serotonin, noradrenaline, and acetylcholine continuously reshape synaptic gains across whole brain regions, dynamically reconfiguring the effective connectivity of the underlying physical network without altering its anatomy.
- Internal Ion Channel Distributions: Individual neurons constantly alter the expression and phosphorylation of sodium, potassium, and calcium channel subtypes, shifting their firing thresholds and burst tendencies.
- Synaptic Weights and Plasticity History: The physical size of a dendritic spine provides only an approximation of its synaptic strength. The exact molecular composition of AMPA and NMDA receptor pools remains largely unmeasured in static serial-section imaging.
Without measuring or accurately inferring these dynamic physiological variables, initializing a simulation from a static connectome produces silent or uncoordinated networks rather than an active stream of consciousness.
How the Roadmap Informs The Consciousness AI Architecture
The Substrate Console on this site embodies the exact trade-off that Sandberg and Bostrom formalized.
Layer 1 of the architecture runs in the browser using anatomical region templates constructed from the Allen Institute, BrainGlobe, and Julich-Brain atlases. The console deliberately preserves proportional connectivity and synaptic pathways between distinct functional clusters (such as the striatum, globus pallidus, and subthalamic nucleus in the basal ganglia action selection circuit) while discarding full biological scale, simulating hundreds of point neurons per cluster rather than billions.
In the Sandberg-Bostrom taxonomy, this positions the Substrate Console as an exploratory Level 1 / Level 2 hybrid: an architecture designed to test how macroscopic causal topology and threshold dynamics govern signal propagation and recurrent loops, while honestly acknowledging that full-scale biological emulation remains a distinct, higher-resolution milestone.
For projects investigating emerging consciousness under functionalist emergentism, including The Consciousness AI (https://github.com/tlcdv/the_consciousness_ai), the 2008 roadmap remains authoritative on one central methodological principle: progress in artificial consciousness cannot rely on vague claims of computational equivalence. It requires explicitly identifying which level of biophysical resolution an architecture implements, and demonstrating that the chosen resolution preserves the causal dynamics that consciousness science demands. What the physical limits of electrode arrays imply for connecting biological tissue to synthetic emulations is evaluated in Paradromics Neuralink and what a real cortical interface would need.
Anders Sandberg is Senior Research Fellow at the Future of Humanity Institute, Oxford University. Nick Bostrom is Professor at Oxford University and author of Superintelligence (Oxford University Press, 2014, ISBN 978-0-19-967811-2). Whole Brain Emulation: A Roadmap is published as FHI Technical Report #2008-3.