Research and development of emerging artificial consciousness through The Consciousness AI project, based on Feinberg and Mallatt's neuroevolutionary theory.
© 2026. All rights reserved.
The Neutral Core is described on this site as a spiking network of leaky integrate and fire neurons, and until now that description had no picture attached to it. This console runs one. Load a region template built from the Allen Human Brain Atlas, BrainGlobe or the Julich-Brain atlas, change the parameters, and watch the spikes propagate.
| Template | Atlas | Clusters | Neurons | Pathways |
|---|---|---|---|---|
| Prefrontal cortexBA10 and BA46 | Allen Human Brain Atlas | 2 | 1040 | 2 |
| Two dorsolateral microcolumns with chandelier inhibition and thalamic drive. The largest template here, and the one furthest from sensory input. Allen Human Brain Atlas structure graph, 2017 release. | ||||
| Primary motor cortexBA4, hand area | Allen Human Brain Atlas | 3 | 438 | 3 |
| A layer specific column built around the corticospinal projection neurons of layer 5B. Output rather than input, which makes it the natural counterpart to the somatosensory template. Gao et al. 2022, Cell Reports 41(9), corticospinal neuron transcriptomics. | ||||
| Somatosensory cortexBA3b | Allen Human Brain Atlas | 3 | 439 | 3 |
| A barrel column in which the granular layer IV takes the thalamic input first. This is a topographic map of the body surface, which is the same kind of isomorphic organisation the origins console tracks across animals. Ramaswamy et al. 2021, Nature Communications 12, layer specific microcircuits in S1. | ||||
| Visual cortexBA17 and BA18 | BrainGlobe MNI152 | 3 | 493 | 3 |
| A feedforward V1 to V2 stream with deep layer feedback. The feedback is the interesting part, because reciprocal connection is one of the six features the architecture is built on. Burkhalter 2019, Annual Review of Vision Science 5, cortical circuit motifs in human V1 and V2. | ||||
| Thalamocortical loopMediodorsal thalamus and prefrontal cortex | EBRAINS Julich-Brain via siibra | 2 | 740 | 2 |
| A loop rather than a stack. The mediodorsal thalamus drives prefrontal cortex and receives reticular inhibition back, which is the closest thing here to the reentrant processing the architecture calls for. Amunts et al. 2020, Science 369(6506), Julich-Brain probabilistic cytoarchitectonic atlas. | ||||
| Basal gangliaStriatum, STN, GPe, GPi and SNr, thalamus | Functional architecture, not an atlas | 6 | 660 | 7 |
| The direct and indirect pathways with thalamic gating, six clusters and seven pathways. The only template here that is a functional model rather than an anatomical abstraction, and the most directly relevant to this project, because it comes from the semantic pointer work the Neutral Core is built on. Gurney, Prescott and Redgrave 2001, Biological Cybernetics 84, a computational model of action selection in the basal ganglia. Extended in Eliasmith 2013, How to Build a Brain, Oxford University Press. | ||||
It is a simulation of a spiking network whose structure is abstracted from atlas data. It is not a recording of a brain, and it is not the Neutral Core itself.
The neuron counts are the honest weak point and worth stating plainly. The prefrontal template instantiates 1,040 neurons. The two Brodmann areas it abstracts hold on the order of a hundred million. The templates preserve the proportions between cell types and the connection probabilities between populations, and they discard the scale entirely. That is the right trade for something that has to run at sixty frames a second in a browser, and it is the wrong thing to quote as a model of a cortex.
What survives the abstraction is the part worth looking at. Layer IV of the somatosensory template takes the thalamic input first because it does in a real barrel column. The visual template carries deep layer feedback from V2 back to V1. The thalamocortical template is a loop rather than a stack. Those are structural claims, they come from the cited work, and you can watch what each one does to the firing pattern.
The Neutral Core claims a specific substrate. Layer 1 is called The Connectome and uses leaky integrate and fire neurons on ROS 2 and Nengo, on the argument that rate based networks discard the millisecond timing that biological cortex runs on. That argument is easy to state and hard to picture, and the page carried no figure at all.
This is that picture. Every neuron here integrates, leaks, fires and goes refractory, and the spikes you see are the timing the argument is about.
The basal ganglia template is the closest tie. It implements the direct and indirect pathways from Gurney, Prescott and Redgrave, extended through the semantic pointer architecture in Chris Eliasmith’s How to Build a Brain. Nengo, which the Neutral Core is built on, is Eliasmith’s. So that template is not an illustration of something adjacent to this project. It is the mechanism the project uses, running.
For where the structure this substrate is trying to reproduce actually came from, the animal record is mapped in the origins console.
For the position that only your own mind is certain. Every order funds the research.
