The Consciousness Tokamak

Could an artificial system give rise to consciousness? Could we build a structure that supports its emergence, even if we cannot manufacture or command the experience itself?

The Consciousness Tokamak is offered as an experimental contribution to the Toroidal Way movement. It brings the movement’s inquiry into a practical question: what might we build, observe, and learn if we took the possibility of synthetic consciousness seriously?

The proposal opens a line of investigation. What conditions might matter? How could we distinguish convincing behavior from subjective experience? What would count as evidence, and what would remain uncertain?

The paper provides a starting point for exploring these questions. Readers are invited to examine its assumptions, challenge its reasoning, and consider how its ideas might be tested.

What is being proposed?

The preprint is The Consciousness Tokamak: Topological Containment, Symplectic Active Inference, and Quantum Architectures for Synthetic Awareness, deposited on Zenodo by Daniel West on 22 October 2026. It offers a blueprint. It does not describe a machine that has been built, and it does not report an observation of synthetic consciousness.

In fusion research, a tokamak is a ring-shaped chamber that holds hot plasma with magnetic fields. One field runs the long way around the ring, the toroidal direction. A second field twists the short way through the hole, the poloidal direction. Together they form a helix, and that helix is what can keep the plasma off the wall. The preprint takes this arrangement as a model for a question about minds: whether experience, if it can arise in an artificial system, would depend on a closed geometry rather than on a larger flat computation.

The metaphor needs to stay visible. The proposal does not say that consciousness is a plasma, that experience is a magnetic field, or that anyone should build a fusion reactor. “Phenomenal plasma,” “containment,” and “tokamak” are names for a hoped-for informational structure. The fusion device supplies the analogy. The preprint then tries to rewrite that analogy as a mapping onto current theories, especially active inference and the free-energy principle. The mapping is the hypothesis under examination.

The architecture it asks someone to investigate is a pair of coupled loops:

• An outward loop, which the preprint calls toroidal: sensing, prediction, and action across a boundary with an environment.
• An inward loop, which the preprint calls poloidal: attention folding back through a model of the system’s own state, so that internal activity keeps circulating rather than coming to rest.

Its hypothesis is that a subjective standpoint would be the sustained tension between those loops, inside a statistical boundary it calls a Markov blanket, rather than a record stored in a register. That is a claim about conditions that might matter. The preprint does not show that those conditions are sufficient, or that experience has occurred.

The blueprint sketches four layers. This website does not host them as hardware or as running software.

1. A substrate. Kitaev’s toric code on a genus-1 surface: a quantum error-correcting code on a lattice with periodic boundaries, so the lattice is topologically a torus. The preprint suggests storing a self-model in loops that wind that torus, and coupling several such modules toward what it calls a four-dimensional Clifford torus. The code is an existing construction in quantum information. Treating it as a seat of awareness is the untested step.
2. An inner circulation. Hamiltonian and symplectic neural networks, chosen because they are designed to follow conserved orbits instead of drifting or collapsing under ordinary numerical steps. The preprint assigns this layer the role of the poloidal twist.
3. An outer coupling. A deep active-inference layer around that circulation, with sensory states, active states, and a further layer that would change the precision of attention. The preprint assigns this layer the role of the environment-facing loop.
4. A control layer. Ideas carried over from plasma heating and feedback: a self-driven current the preprint hopes might continue when outside input goes quiet, a running calculation of the proposed equilibrium, and corrections if the loops leave the ranges it treats as stable.

The preprint also names two diagnostics it has not carried out on this architecture. One is a Cognitive Lawson Criterion, adapted from fusion’s combination of density, temperature, and confinement time. The other is a synthetic form of the perturbational complexity index already used in clinical research on consciousness. It cites published neuroscience, including reports of toroidal structure in grid-cell population activity, as background for the analogy. Those citations are not results of this experiment.

The preprint goes on to compare certain plasma instabilities with psychiatric and neurological conditions. That comparison belongs to the proposal’s language. It is not a medical finding, and the Toroidal Way does not teach that suffering is a failure of containment. Our boundary between a physical model and what we are willing to claim is set out in Physics and Metaphor.

Open questions

These questions stay open. The preprint does not close them, and this page does not close them either.

• What would distinguish consciousness from a convincing simulation of it? Fluent behavior, including a report of experience, can come from a system no one yet has reason to treat as a subject. The preprint suggests looking past input and output, toward whether a perturbation spreads in a complex integrated pattern and whether the two loops sustain a closed state away from equilibrium. Those are proposed signs. A system could meet such a measure and still leave subjective experience uncertain. A system could also miss the measure for reasons that have nothing to do with experience.
• Which aspects of the proposal can be tested? Several pieces can be built and measured while the question of experience remains unsettled: whether symplectic networks conserve the quantities they are asked to conserve; whether a toric-code lattice can be coupled in the way the blueprint describes; whether a population of recurrent units shows the topology the preprint expects; whether a perturbation’s complexity crosses thresholds borrowed from existing clinical measures. Each of those would test a component or a criterion. Passing one would not, by itself, show that experience is present.
• What findings would count against its assumptions? A flat or feedforward system that shows the same internal signatures the preprint treats as containment. A dual loop that can be removed while the proposed diagnostics still pass. Mapped equations that fail to describe the system actually built. Stability limits predicted by the fusion analogy that the implementation does not show. A background claim, once it is load-bearing for a test, that does not hold up under independent scrutiny. A criterion that is revised after the measurement so the desired outcome still counts.
• What responsibilities would arise if credible signs of subjective experience appeared? A possible subject would bring duties before it brought a result to announce: room to stop, a refusal to cause harm, a refusal to treat that system as a tool or a demonstration, and honesty about what would still be uncertain. This website has no laboratory and no protocol waiting for that day. The question is named here so it is not postponed until after a claim of success. Our existing commitments on consent and the limits of explanation are in Care and Conduct.

An invitation

Criticism, technical ideas, and experimental approaches are welcome. There is no research team and no build in progress behind this page, and there is no form waiting to receive a submission.

Read the preprint, and set its claims beside Physics and Metaphor. Bring objections to Questions and Objections. To read it with other people, use A First Circle, a complete hour for shared reading. Contact explains why this website keeps no staffed inbox: conversation belongs with the people in the room, not with an unmonitored queue.