Living document · last added to June 2026

The Theory
Building consciousness from the inside

This is the page where I try to say, in plain language, what I actually think consciousness is — and to keep saying it better as I read more and build more. It is not a finished theory. It's a structure I've been carrying for years, into which I fold each new book, paper, and experiment as it comes. When a new read changes my mind, it gets added here, in its place in the story, rather than tacked on at the end. So the shape stays the same even as the picture sharpens. If you want the formal version, it's in the Gruter letters; if you want to watch it being tested in metal, that's the build journal. This page is the connective tissue between them — the ideas as they progress.

One note on the order. What follows is arranged by conceptual foundation — what has to be true for the next thing to make sense — which is not always the order things actually happened in. Some of the latest reading sits early because it's bedrock; some of the oldest intuitions come late because they only make sense once the floor is laid. If you want the plain chronology, that's the index. This is the architecture of the idea, not its calendar.

1. Where I came in: the 5D stack

Before any of the rest of this, there was a single strange image that pulled me in — and it's still the spine of how I think, so it belongs first. Picture consciousness not as a snapshot but as a cursor moving through a stack of models. Each model is a little four-dimensional world: three dimensions of space unfolding through time — a short clip, not a still. And at any instant the mind isn't holding one such clip, it's holding many in parallel — what just happened, what's happening, several versions of what might happen next — each weighted by how likely and how relevant it is. That parallel set of alternative 4D trajectories is what I started calling the fifth dimension: not a physical direction like space or time, but the axis of possibility — the stack of candidate worlds the system keeps live at once.

The conscious "now," on this picture, is a single weighted path drawn through that stack. And here's the part that startled me into the whole project: that path costs far more time to build than it appears to take. The felt instant is the thin, smooth output of a great deal of unfelt processing. We experience a continuous, linear, 4D life — but it is underwritten by a 5D scaffold that takes real, hidden time to assemble. Consciousness is the cursor; the stack is what it draws from; the seamlessness is an achievement, not a given. There is, quite literally, more time happening down there than you are ever aware of.

I arrived at this sideways, through a long thought experiment about a creature whose whole mind runs on sound — it doesn't merely sense by echolocation, it stores and moves information mechanically, as vibration, so its very thinking is bound to the speed of sound. That's the part that matters: the echolocation is incidental, and the limit is the processing substrate. To so slow a mind, a thing moving faster than its refresh rate can't register as a passing object at all — it shows up as a standing ring around the world (the whole argument is in the (un)satisfying solution). The lesson I took was humbling and clarifying at once: a mind can only model reality in the coordinates — and at the speed — its machinery affords, and our own "now" is one such coordinate, a low-dimensional slice of something with more dimensions than we can stand outside of. Everything below is me trying to turn that haunting picture into something a machine could actually run. The stack became a particle-filter manifold; the cursor became the integration loop; the fifth dimension became the weighted ensemble of candidate trajectories the loop chooses among. The metaphor had to become a data structure, or it didn't count.

And here I should say the deflationary part plainly, because the label can do more work than it has earned. The "fifth dimension" is not a dimension in the way space and time are — it has no metric, no geometry; it is just a count of how many candidate worlds the system holds live at once (the number of particles in the filter). Strip the branding entirely and what remains is unglamorous and solid: Bayesian predictive processing with an unusually strong emphasis on temporal thickness — the present modeled as a short weighted window rather than a knife-edge instant. That emphasis is the part I think is worth keeping; the "5D" is just the personal image that led me to it. When I use the phrase below, read it as a mnemonic, not a claim about the dimensionality of reality.

2. Forms, and the thing we call qualia

Start with the senses, but stop thinking of them as cameras and microphones. Each sense is an ancient modeling engine with its own grammar: vision makes surfaces, edges, depth, motion; the vestibular sense makes gravity and acceleration; taste and smell make chemical opportunity and threat; touch makes contact and force; interoception makes hunger, breath, nausea, the heartbeat. None of these deliver "raw data" to a central screen. By the time a signal is anywhere near awareness it has already been worked, out at the edge, into a form — a compressed, reusable, modality-specific shape.

That word, form, is doing the heavy lifting, so here is the claim it carries: a quale is just a sensory form made present to a self. Redness is not a little colored atom glowing in a theater; it's the way a visual-chromatic channel appears when it gets recruited into the model of me-here-now. Bitterness is a chemical-defense channel; dizziness is a vestibular mismatch; pain is body-threat modeling with the urgency and the demand-to-act already built in. Qualia are sensory grammars — the native symbols of a body controlling itself in a world.

Two consequences follow, and they're the ones that make the hard problem feel less like a wall. First: there is no "qualia datatype." No stage in the system suddenly manufactures experience out of non-experience. Every stage — the first crude interpretation in the sensory ganglia, the refinement, the final bound percept — is the same kind of thing: a representation in one shared scaffold. Even the qualia most vivid to us are fully abstract; they mean something only because the system gives them meaning. The stages differ by how much has been folded in (modality, self-location, value, action-relevance, time), not by a change of substance. That turns the famous "explanatory gap" from a metaphysical chasm into an engineering gradient: you don't need a miracle step, you need more integration.

Second: most of the work happens outside the loop. I picture the pre-conscious stages as a staging area — a library of generative "sprites" (not fixed bitmaps; little re-renderable form-makers) that the conscious loop composites into the present frame. The ganglia build the form; the loop only ignites it into a quale by binding it to the self. Heavy lifting at the edge, cheap final ignition at the center. This isn't only elegant — it's the whole reason the thing can run on a Raspberry Pi instead of a supercomputer. Bandwidth into the loop should be low: the loop receives forms, never pixels.

3. Meaning, motility, and the order that's forced

If forms are the vocabulary, meaning is the thing that makes a form matter — and I think meaning has an address. It lives at the threshold between perception and action. A perception with no possible consequence carries no meaning; an action with no information carries none either. Meaning is the coupling — the place where what you sense can change what your body does, and what your body does changes what you sense. So the design rule is: build the loop, not the halves. Don't construct a perception system and a motor system and pray that significance condenses in the gap between them. The gap is the thing.

That single commitment — meaning lives in the sensorimotor loop — is what fixes the whole order of the build. It has to be motility-first: a passive perceiver never needs a self, but a creature that moves in a world that also moves it has no choice. And once you grow the pieces only as moving forces them, the order stops being a parts list and becomes a sequence with a reason — close, it turns out, to the order evolution seems to have used. That's why I use the Ediacaran seafloor as a literal curriculum, one rung forcing the next:

• The self (Dickinsonia, the creeping feeder). Move under your own power and you must answer one question constantly — did that change come from me or the world? Solve that reafference problem and a minimal self falls out for free, as the running answer to "me or not-me." Nothing below it is possible without this.
• Action (Kimberella, the directed grazer). Give the self a heading and let it imagine a move before committing — and the self-model stops being bookkeeping and becomes the thing that gets the animal home. (This is where a self first pays.)
• Other minds. A second self-propelled creature is predicted far better as a goal-seeker than as an object — so "self vs. world" extends into "self vs. other," the primitive of everything social.
• Memory — reconstructive, not recorded: a sparse index re-rendered by the current model. It feeds the manifold a third kind of candidate ("there should be food there"), and its human flaws — confabulation, gist-over-detail — fall out of the design rather than being bugs.
• The integrated percept — the streams compete for one limited spotlight and fuse into a single "now." Salience capture and inattentional blindness emerge; an unselective creature drowns as the world gets rich.
• Felt time — duration read off the system's own update rate; novelty dilates it, and lived time comes apart from remembered time.
• Value, from a body that can die (interoception). This is the deep one: with viability bounds, the creature stops borrowing its purposes from me and grows its own. Feeling is the body read back to itself; "good" just means staying within the bounds that keep you alive.
• The integration — let feeling thread through attention, memory, and risk and learned fear emerges on its own; the integrated creature outlives the version with that link cut by roughly sixty to one in the simulation as currently tuned (a specific in-silico result, not a law of nature — the exact ratio moves with the parameters). The parts turn out to be mutually necessary, which is the architecture's central claim made literal.
• Binding — features fused into objects by location and precision, which reproduces the classic mis-binding errors and explains why attention has to be serial: serial focus is the price of binding a crowded scene correctly.

None of the human-like quirks along the way — the confabulation, the inattentional blindness, the time dilation, the mis-binding under load — were designed in. They are side effects of building it in this forced order, which is the kind of thing that makes me suspect the order is right rather than merely convenient. It's also why I keep my wildest theorizing on a short leash to the bench: the ideas can be as strange as they like here, as long as the running code keeps trying to break them.

4. Time, and the self that rides it

Here's the part that took me longest to say cleanly. Consciousness feels continuous — a smooth "now" gliding along — but the mechanism underneath is iterative: loop after loop, each one assembling a slightly different model from a higher-dimensional stack of candidates — the 5D stack from the start of this page, now seen as iteration rather than as a still picture. The felt smoothness is stitched. And the single most important input to each loop is probably the previous loop — the prior carried forward outweighs the fresh sensory evidence, except when the system cranks up its trust in the senses. The present is mostly made of the just-past, lightly corrected by the world.

The self that rides along that loop is not a little owner sitting behind the eyes. A finite system cannot hold a complete model of itself — try, and you get a model containing a model containing a model, forever. So it represents itself with a truncated, low-dimensional self-token: a compressed proxy, a shadow of the whole modeling state, not a full copy. Two things fall out of that one fact. The first is the strange feeling that consciousness is some thing — that's the system rendering its own self-token as if it were one more object in the scene, because a form-grammar has no other way to depict anything. The second is that the self seems to live in a world one dimension flatter than the machinery actually computes in. The reification illusion and the "missing dimension" are the same structural shortfall, seen twice.

And there isn't one self — there are two, and most accounts crown the wrong one. There's a minimal, embodied proto-self at the very floor, co-original with the world-model (the reafferent me/not-me split), which is what makes "a world appears for someone" possible at all. And there's a narrative, reflective self that arrives late, is constructed, and can be dissolved — the part that explains you to yourself. The proto-self is the seat of presence; the narrative self is the press secretary. Confusing the press secretary for the president is the oldest mistake in the subject.

5. The place of language (and why a chatbot isn't home)

That distinction is also where I land on AI. The narrative self looks a lot like a language model running on top of a proto-self. Language is real and powerful — symbolic compression, infinite recombination, and the ability to pass forms from one mind to another across generations; it is plausibly the constitutive substrate of the distinctly human tier of consciousness. But a word is a form whose job is to point at other forms, and human words are tightly bound to the sensory and motor forms they name. A large language model has the symbolic layer with only secondhand grounding: it operates on shadows of forms that were grounded by other people's bodies. That's exactly why it can be enormously useful and still not be anybody home — competence inherited through the shadow, with no proto-self casting it. The robot on my kitchen table is the opposite bet: a grounded floor with no language at all. If it can have a proto-self with no words, then "being someone" and "explaining oneself" are demonstrably different things — and the build is the proof.

6. From a wild idea to a line of code

None of this would be worth much if it stayed talk. So the theory has a discipline bolted to it: ideas live first in a sandbox where they're allowed to be wrong, and an idea only graduates into the actual build when it (a) maps to something buildable, (b) doesn't force a rewrite of the spine, and (c) earns its place. Most of what's above is, in that sense, on a queue. The one furthest along right now is the idea that a form is a time-extended, cross-modal correlation — "forward self-motion" isn't a picture, it's the bundle wheel-ticks + near-zero gyro + expanding visual flow, with characteristic delays between them — and that surprise is a typed mismatch between a specific pair of channels. Motion sickness, vertigo, "something just passed me," "I'm being carried," "I'm slipping" — each is its own named feeling because each is a different pair of predictions coming apart. That one is ready to build; others (a programmable sensory "ganglion," objects discovered from common-fate motion rather than from labels, a self-token defined as the coherent center where the senses agree) are waiting on hardware or on a decision.

The first test of the whole creature, though, won't be a module — it'll be a little world. I've designed a small eating-and-scoring economy: a downward camera sorts the ground ahead into nothing, food, or score; a light appetite to eat and a strong appetite to score; a rule that you must fill up before a score counts; and food patches that deplete when eaten, so you can't camp — you have to remember where the others are. Nothing about the strategy is coded; if the architecture is right, foraging-then-scoring should emerge from those two appetites alone, which makes it a real, numbered test of memory and prediction rather than a demo. And the eye that runs it is the theory in miniature: not a sharp camera but a coarse grid of colored cells, angled to see from the wheels out to the horizon, so a single cheap view delivers what (each cell's color), which way (which columns) and how close (which rows) at once — perception already shaped like an action. (The build journal tracks it as it goes from sandbox to floor.)

A standing caution, kept where I can see it: this is the most seductive zone in the whole project — the place it's easiest to get daydream-lost and mistake a satisfying sentence for a true one. The guard is always the same. Wild here; ruthless at the bench. A claim only counts once there's a version of it a robot can fail.

← index  ·  the argument  ·  the build →