Research and development of emerging artificial consciousness through The Consciousness AI project, based on Feinberg and Mallatt's neuroevolutionary theory.
© 2026. All rights reserved.
Across a patch of cortex, brain activity moves as waves. In 2026 three research groups tied those waves to perception, working memory and the coordination of distant brain regions. This console runs small models of each finding in your browser. It also runs the oscillators inside this project’s own agent, with the same equations as the code, including a defect we found while building this page.
Nothing on the canvas is recorded brain data. Each chapter is a model, and each model states what it reproduces by construction and what it shows on its own.
This graphic needs a canvas. The chapter table below carries the same information.
| Chapter | Model | Built in | Emerges | Source |
|---|---|---|---|---|
| 1 Our agent | The five-oscillator AKOrN binding layer, same equations as the code | The code's update rule, with or without the rotation term | A fixed point on main. Continuing phase motion with the fix | Miyato et al. 2025; the research repo |
| 2 Sheet | 64 by 64 Kuramoto sheet, Gaussian coupling within 2.5 cells, optional conduction delay and coarse field term | Local coupling rules, pacemaker regions, initial phases | Plane waves, target waves, spirals, sparse spiking | Muller et al. 2026; Davis et al. 2021; Xu et al. 2023 |
| 3 Gate | The sheet with a plane wave and a threshold stimulus | A 15 percent phase-dependent gain | Nothing new. The bars display the built-in effect | Davis et al. 2020 |
| 4 Ripples | Six sites, Poisson ripple onsets, 90 Hz bursts, 40 Hz multi-unit activity | A 30 percent co-firing rise during co-ripples | Nothing new. The ratio converges to the built-in value | Verzhbinsky, Halgren et al. 2026 |
A neural traveling wave is a region of raised excitability that moves across the cortical sheet. It is carried by spikes along the horizontal fibres inside one cortical area. Those fibres are unmyelinated and slow, about 0.1 to 0.6 metres per second, so a signal from a neighbour a few millimetres away arrives tens of milliseconds late. Muller, Busch, Davis and Reynolds review the evidence in Neuron (2026). Their main points follow, in their order.
The review then proposes, and labels as speculative, that waves could carry recent input across the map and support short-term prediction of what comes next. The evidence for that role comes from network models such as Benigno et al. 2023. No recording has shown it yet.
The Kuramoto model is the standard model of coupled oscillators. Each unit has a phase and a natural frequency, and each one turns toward the phases of the units it is coupled to. With all-to-all coupling, as in the ring of the Thermodynamic Console, the only thing that can happen is more or less synchrony. A wave needs a place to travel, so the units must sit on a map and couple to their neighbours. Chapter 2 does that. The plane, target and spiral patterns all come from the same local rule.
Artificial Kuramoto Oscillatory Neurons (AKOrN, Miyato, Löwe, Geiger and Welling, ICLR 2025) put this model inside deep networks. The Muller review cites AKOrN as one of the machine learning models that show waves in their hidden layers, beside the Neural Wave Machines of Keller and Welling and the wave segmentation network of Liboni and colleagues.
In August 2026 Verzhbinsky, Halgren and colleagues reported in Nature Neuroscience on 35 patients with electrodes in the hippocampus, amygdala, ventromedial prefrontal cortex, anterior cingulate and pre-supplementary motor area. Ripple rates rose in every region during a working memory task. When two regions rippled at the same time, their neurons fired together about 30 percent more often, with no fall-off up to 220 mm, and more so at higher memory load. Chapter 4 shows that pattern. The regions are the recorded ones. The study did not record the whole brain.
Xu, Long, Feng and Gong found rotating spiral patterns in fMRI of 100 people, both at rest and during tasks. The direction and position of the spirals told the tasks apart. The data is the slow fMRI signal, with no electrical recording, so the spirals are a whole-brain pattern over seconds. The spiral in Chapter 2 is the same kind of pattern on a small sheet and a fast clock.
Earl Miller and colleagues propose that the brain computes in part through the electric fields that waves create, which then act back on neurons (Analog Cognition and Consciousness, J. Neurosci. 2026; Pinotsis and Miller, Cerebral Cortex 2026). Muller and colleagues disagree in Box 1 of their review. Measured wave speeds match axon conduction, field effects in slice work are below a millivolt, and a field is too coarse to reach single feature columns. Both sides cite real measurements. This page takes no side. The field slider in Chapter 2 lets you add a weak coarse field and see what it does to a wave that local coupling already carries. The site covers the measured side of ephaptic coupling in this post on Anastassiou’s work.
The Consciousness AI agent binds its five workspace modules with an AKOrN layer. While building this console we ported its update rule line by line and found that the natural frequency term was always zero. The code multiplied each oscillator by the diagonal of a skew-symmetric matrix, and that diagonal is zero by definition. So the oscillators could only settle into one fixed state. The project’s own results had already recorded the symptom. On the delayed match task, sync_R took one value, 0.450000, over 13,600 settled steps, and across three seeds it carried no information about the stimulus. The fix is behind a default-off flag so that every earlier result stays reproducible. Chapter 1 runs both versions.
The agent’s oscillators are still not spatial. There is one per module and no map, so no wave can travel in them yet. Whether waves already appear in the agent’s visual map, and whether adding them would help, are open questions that the project will test before building anything.
No study on this page shows that. The measured results are that wave phase changes what an animal detects, that co-ripples link firing across distant regions during working memory, and that spiral patterns differ by task. Each is a link between waves and a function that consciousness research cares about. None shows that waves are sufficient for experience, and the site’s consciousness fingerprint page sets out what is and is not known about oscillations and identity.
