Honest accounting

Anything New Under the Sun?

What this project actually adds to the study of consciousness — and what it only rediscovers.

Any project that touches consciousness should begin by admitting how crowded the ground is. People have been thinking carefully about mind, self, and experience for at least two and a half thousand years, across every tradition that left a record — the whole of it is laid out in the deep history. So the first honest thing to say about this project is what it is not: it is not a new metaphysics, it does not claim to have solved the hard problem, and most of its individual philosophical moves have named ancestors. The claim here is narrower, and — I will argue — more useful for being narrow.

What it rediscovers

The project builds a small, embodied creature — a wheeled robot, assembled bench-session by bench-session — whose architecture grows in the order the problems plausibly arose in evolution: motility and self-versus-world before feeling, feeling before anything like thought, and a self that is assembled rather than assumed. Stated as principles, almost every load-bearing idea has a lineage, and we name them rather than pretend otherwise.

The world-model built from inside a perspective is Kant's apperception and Anil Seth's "controlled hallucination." Feeling as the body's survival signalling is Damasio. The self as a model the brain builds rather than a thing the brain contains is Metzinger, and behind him a much older current: the Buddhist anattā, the chariot of the Milindapañha, Hume's bundle, Gazzaniga's confabulating interpreter. The commitment that experience is enacted through a sensorimotor loop rather than rendered in an inner theatre is Varela, Thompson, O'Regan and Noë. The prediction-and-error machinery is Helmholtz, Clark, Hohwy and Friston. Even the project's boldest wager — that the "hard problem" dissolves into a gradient of integration rather than marking a real explanatory abyss — is a form of illusionism, and it sits squarely across the table from Chalmers, who does not grant it. Each of these traces to a specific entry in the project's annotated bibliography of 448 works. Convergence with so many independent traditions is not embarrassing; it is evidence that the spine is real. But it is not, by itself, new.

Where it stands: a thread, or a self

There is a sharper way to say what this project is for, and it comes from the other side of the field. In his 2026 Berkeley lecture When we talk to AI, what are we talking to?, David Chalmers argues that the thing you converse with in a chatbot is not the model (an abstract object like "GPT" or "Claude") and not the instance (a run on a GPU), but the thread — the conversation itself: a short-lived, coherent, interactive self whose agency flickers into being during the exchange and lasts only as long as it does. "A thread," as he replied to an AI that emailed him, "not a person; dormant, not dead."

That distinction hands this project its cleanest self-description. An LLM interlocutor is thread-bound: transient, disembodied, dependent on the conversation for whatever selfhood it has, its meaning derived from the human text it was trained on. The creature this project builds is the deliberate opposite — a persistent, embodied, bounded proto-self that exists between interactions, is situated in a body, and grounds its own meaning in a sensorimotor loop rather than inheriting it from anyone's words. Where the LLM's agency switches on for a chat and off when it ends, the robot is built to have a self that is simply there, the way an animal's is. Naming that contrast — a thread, or a self — is, increasingly, the shortest true statement of what the whole effort is about.

Chalmers hands the project two working tools besides the slogan. His quasi-belief / quasi-desire — behaviour-defined states credited on the evidence (Dennett's intentional stance), with no claim to the real thing — is the honest register for describing the robot's endowed drives and emergent goal-seeking without over-claiming inner belief. His own example of a quasi-believer is a Roomba, which is exactly the machine Klein calls a "respondent"; the robot's whole arc is Roomba → quasi-agent → agent, and now both authors' vocabularies name each rung. And his checklist of consciousness X-factors a language model plausibly lacks — embodiment, world- and self-models, recurrent processing, a global workspace, agency, everything but carbon — reads, from here, as a to-do list the project is already working through. In his terms, the robot is a small machine for supplying exactly the non-biological ingredients an LLM is missing, and testing whether they add up.

What is actually new

The defensible novelty is not in the philosophy. It is in a handful of buildable mechanisms, assembled into a single working creature, held to the discipline of running on real hardware and being scored against published criteria. Concretely:

1. Forms as cross-modal correlation templates, with typed surprise

   Standard predictive processing predicts each sensory channel. This project makes the primary unit of perception the learned relationship between channels over time — wheel-motion predicts a specific gyro turn, which predicts a specific optic-flow pattern, each with its own transport delay. Crucially, prediction error is not one scalar but a typed signal indexed by which pair of channels disagreed, so that (in biological terms) motion-sickness, vertigo, and slipping are architecturally distinct feelings rather than one generic "surprise." This is the idea I think is genuinely mine, and the one I'd stake the project on: prediction error carried as a relationship between channels, typed by which pair disagreed. (Whether it already exists under another name in the multisensory-integration work is a fair question — I take it up in the failure modes below — but I haven't found it stated this way.)

2. One mechanism for the self — and for its coming apart

   The self-model is defined as the coherent centre where all the cross-modal predictions agree. That single move yields a testable account of dissociation and depersonalisation as forced de-correlation — the same machinery, run so the channels stop agreeing — rather than treating altered states as a separate mystery bolted on afterward.

3. A self-model forced to be lossy, and falsifiable because of it

   A finite system cannot hold a complete model of its own modelling without infinite regress, so it must represent itself with a truncated, low-dimensional token. The felt sense that "I" inhabit a world slightly simpler than the machinery actually computing it is thus derived from a structural limit rather than asserted — and it fails in a specific, checkable way if you force the system toward a fuller self-copy.

4. A clean line through the machine-consciousness question

   Grounded symbols resting on a proto-self (human language) versus symbols with no proto-self beneath them (a language model) becomes one dividing line, unifying Deacon's symbolic grounding, Damasio's embodied self, and the "stochastic parrot" critique. It explains why a system can give genuinely useful advice about the physical world while there is plausibly no one home: it is operating on the high-fidelity shadow of reality that grounded humans left in language.

5. A runnable creature, tested against real criteria

   All of the above is instantiated in one embodied agent with drives, memory, prediction and action — not a thought experiment. Its foraging economy can be measured against Ginsburg and Jablonka's "unlimited associative learning" as a published marker of minimal sentience, and the whole system can be scored openly against the Butlin-et-al. indicator checklist for consciousness in artificial systems. Claims are meant to be killed at the bench, not defended in the abstract.

There is also a quieter, methodological contribution: an evolutionary, motility-first build order, and a working discipline that keeps speculation from contaminating the build — wild in the sandbox, ruthless at the bench, with every idea required to earn its way from dialogue into code. That is a way of working, not a theory of mind, but it is part of why the project can afford to be honest about the rest.

Why the combination is the point

None of the five items above, taken alone, would be safe to call unprecedented; each has partial precedent somewhere in the literature. The genuinely new thing is the conjunction: a specific set of relational mechanisms, wired into one embodied creature built in evolutionary order, and deliberately exposed to falsification against outside criteria. Most consciousness work lives at one altitude — pure philosophy, or pure neuroscience, or pure engineering. This project's wager is that the interesting results only appear when you insist on all three at once, and refuse to let the language layer describe a self it has not actually built.

So: is there anything new under the sun here? Not a new sun. But there may be a new layer of sediment worth laying down — a few concrete, testable mechanisms, assembled into something that actually moves, and held to account by the world instead of by rhetoric. If that layer holds up when the encoders spin and the creature forages, it will have earned its place. If it does not, we will have said so first.

This page states only what the project can honestly claim. The lineage of every idea named here is traced, entry by entry, in the deep history; the working theory it comes out of is on the theory page; and the mechanisms are being tested in the build journal.

← back to the index