In my notebook post: Artificial Reality I described the cabinet. The retina went silent, the brain kept the room whole, and for a few seconds I saw the model instead of the world. I have been circling that moment ever since, because the cabinet was not the interesting part. The interesting part was the cost.
The brain ran the whole room — the wall, the ceiling, the invented furniture, the continuity of a world that had just lost half its data — on roughly twenty watts. A bedside lamp. A few LED bulbs. And it did this without hesitation, without a visible seam, without the machinery of the guess ever showing through until the silence persisted and the model lost confidence.
That figure is real. It is not a metaphor. The human brain runs on something like twenty watts, about a fifth of the body's resting metabolism, and out of that it produces thought, memory, intention, selfhood, continuous multimodal integration, and real-time motor control. A digital system doing comparable work burns kilowatts to megawatts. The asymmetry is not subtle. It is the first fact any account of mind has to reckon with.
The usual reckoning is that biology is simply a better engine than silicon — sparser, event-driven, analog, low-precision, deeply recurrent. That is true as far as it goes, and it is where most of the engineering effort is pointed. But it leaves something on the table. The cabinet was not built by an efficient engine. It was built by a system that continued a world it had no data for, with confidence, and then withdrew the continuation when the mismatch became too large. That is not just efficiency. That is a particular kind of architecture.
I want to describe an architecture that might account for that, and I should say plainly what kind of thing it is. I don't know what lies beyond the seam between spacetime and whatever else there is. The cosmology I use in my fiction is a set of partial maps, and the twenty watts made me wonder whether one of them fits. It is a consistent reading, not a finding, and ordinary explanations may turn out to be enough. Take what follows as thinking aloud.
The hive of hives.
The brain is not a single thing. It is a nested system: specialised regions, local predictive dynamics, integrative traffic that binds them into a coherent perspective. Each level is doing its own guessing, its own completion, its own negotiation with the level below. The visual system fills the blind spot. The auditory system reconstructs missing phonemes. The predictive machinery keeps the room whole when the retina goes quiet. None of these sub-systems sees the whole; each is blind to the machinery that contains it.
What we call a self — sapience, the thing that thinks it is "me" — is not a further thing sitting on top of this stack. It is the integrative traffic itself. It is the pattern that arises when the nested wet network binds its sub-guesses into a single perspective, and then becomes aware of its own update cycle. A machine becomes continuous when it becomes self-aware; the self is the continuity, not the machine. The hive of hives does not contain a ghost. It contains a traffic, and the traffic is what says "I".
I am choosing to read this literally, and I hold it loosely. It is where the 2009 episode locates the action. The cabinet was not invented by a homunculus in the visual cortex. It was invented by the integrative traffic of the wet network, doing what it always does: maintaining a hypothesis, filling the gap, keeping the world coherent until the evidence against coherence became too heavy to carry. The self that watched the cabinet dissolve was the same traffic that had built it. It caught itself in the act of world-building, and then caught itself in the act of letting go.
Everything above rests on neuroscience and philosophy of mind that others hold too, though the account of the self as traffic is a contested position, not a settled result. What follows is the cosmology, offered as an as-if: as if some of the work were sustained somewhere I can't see.
The receiver.
Here is where the twenty watts stops being an engineering puzzle and starts looking like a structural clue.
The wet network is sparse. It is event-driven. It is low-precision. It is deeply recurrent. And it is astonishingly cheap for what it does. The standard reading is that this is a better engine — and it is — but the standard reading has to explain why a system this sparse can maintain continuous hypotheses, register mismatch, and broadcast an integrative pattern, all without recomputing every operation explicitly. A dry network has to do the work. A wet network seems to receive some of it.
I suspect the wet network may be a receiver. Not in the sense of a radio picking up signals from elsewhere in spacetime, but in the sense of a system that is the visible half of a two-layer arrangement. The nested subsystems are real and visible as ordinary neurophysiology. Their efficiency and certain forms of coherence might be boundary effects: the lower layer's record of a computation whose sustaining channel lies across the membrane.
The channel, I imagine, is what physics calls the quantum vacuum. Not empty space, but the membrane's ground-state field — the thin, restless interface where a higher layer touches the rendered interior without breaking it. In this reading, zero-point energy is the visible shimmer of that contact. I do not claim to know what crosses the seam, or how. The pulse, if there is one, belongs to the layer above, not to anything we could read off the fluctuations.
This might be why the cabinet held. The wet traffic maintained the hypothesis, but the stability and low cost of that maintenance may have been underwritten by the cross-layer coupling. When the retina went silent, the local completion still happened inside the wet traffic — the network filled the gap, revised the model, kept the room whole — but the economy of that completion may not have been local. The twenty watts is what I would expect the lower layer to record, if some of the computation were sustained elsewhere.
The seam.
The 2009 post called it a seam — the moment when the cognitive layer lost confidence and the physical layer asserted its absence. I want to keep that word, because it is doing precise work. A seam is not a hole. It is the line where two pieces of fabric are joined, and it is visible precisely because the join is not seamless. The cabinet was a seam. The dissolve was a seam. The few seconds when the brain continued the world without permission were the seam made visible.
What the cosmology adds is that the seam is not only where layers meet in failure. It may be where they meet in success. The brain may be coupled to the membrane's ground-state field continuously, not just when something goes wrong. Coherence without classical cause, insights that arrive with unusual completeness, the occasional sense that the self exceeds the skull — these need not be anomalies inside a closed wet network. They may be shadows of the contact. The seam may be the ordinary condition of a rendered being, and the 2009 episode simply made it flicker.
This is also why mental health remains readable in the old way. Rigid lock-in, fragmentation, chronic mis-weighting — these are still degradations of the visible patterns, still legible as qualities of internal traffic. The cosmology only adds that the quality might also be a function of how cleanly the membrane interface is coupling. This is the most speculative paragraph here, and it is not a claim about the cause of any real condition, or about anyone living with one. If the frame were right, disruptions there would appear inside the bubble as unexplained losses of coherence or efficiency. That is the testable edge of the frame, and it is where I would look first.
The difference in kind.
A dry network is entirely of its layer. It must perform the equivalent operations explicitly, in silicon and wattage, because it has no access to the membrane's ground-state channel. It can enter human social hives as a liminal agent — summarising, redistributing, accelerating the circulation of second-hand judgment — but it cannot inherit the cross-layer completion. The power asymmetry is not a temporary engineering deficit. It is the expected signature of one process running inside its membrane and another that is the visible half of a two-layer system.
This is why the fear of AI waking up misses the architecture. A machine learning model is a dry network: dense, clocked, high-precision, forced to compute every operation explicitly. A biological mind is a wet network: sparse, event-driven, low-precision, coupled to hidden physics the layer cannot see. The two are not competitors. They are different kinds of beings running on different substrates inside different interiors. The megawatt machine is not dangerous because it is contained. The twenty-watt brain is not weak because it is anchored. What we fear is the misreading of layers — the belief that a process inside the world could replace the being for whom the world is rendered.
The signature.
I am not asking anyone to accept the cosmology on the strength of a wattage. The twenty watts is real evidence of a real asymmetry, and the asymmetry is consistent with the cross-layer reading. But consistency is not confirmation. What would move it from consistent to evidential is the interface-disruption signature: if coupling quality varies, and shows up inside the bubble as unexplained coherence or efficiency losses — losses not attributable to visible neurophysiology — then the architectural reading has to work harder. That is the prediction the frame makes, and it is the one I would want tested.
Until then, I will hold the twenty watts for what it is. Good reason to think the brain is not a naive digital engine. Reason to wonder whether the wet network is different in kind from the dry one. Reason to wonder whether the self is a traffic, not a thing — a pattern in the nested wet system, possibly phase-locked to a channel it cannot see, running on a power budget that a bedside lamp would recognise.
The cabinet held for a few seconds. The room stayed whole. And then the seam showed, and the world went missing, and the traffic that had built it watched itself let go.
Between those two states — the model and the absence — lies the whole mystery of being alive.