뉴톰테크NeutomTech

An Independently Theorized Global Workspace:
BOSIFWEM's "Image" (2001–2015) and Anthropic's J-space (2026)

Ji Young-min (Sookmook) — NeutomTech, Busan, Republic of Korea · neutomtech.com · [email protected]
Bridge note compiled with Claude (Anthropic) · July 7, 2026 — one day after the publication of "A global workspace in language models."

Summary. Anthropic's interpretability team reports that a global-workspace-like structure (the J-space) emerged spontaneously in Claude. This note records a point-by-point correspondence with BOSIFWEM, an independently developed psychological architecture documented in Korea between 2001 and 2015, whose "Image" element was specified — in English, on a 2015 diagram — as a "board" of all senses, feelings, concepts, and stories. Because BOSIFWEM embeds its workspace in a full eight-element ecosystem — and because the same author's multi-valued logic treats knowing itself as graded — it yields testable predictions about what should be found next in and around the J-space. They are offered in the spirit of convergence, not priority.

1What BOSIFWEM is, in one paragraph

BOSIFWEM (an acronym of Body, Operating Soul, Sense, Image, Feeling, Will, Ego, Memory) is a structural theory of mind developed by an independent researcher over a 14-year study (2001–2015; information-flow map © 2008; full architecture diagram dated 2015-11-15), drawing on long East-Asian traditions of disciplined introspective observation read as records of observation rather than objects of faith. It treats mind as an information-processing phenomenon arising from eight interacting elements, connected by ~30 typed information flows that come online in 9 developmental layers. A comprehensive manuscript is in preparation; overview pages (Korean/English) are on the website above.

2The correspondence

The J-space paper lists distinctive functional properties of the workspace it found in Claude. Each has a counterpart specified in BOSIFWEM's "Image" (I) years earlier:

J-space property (Anthropic, 2026)BOSIFWEM "Image" specification (2001–2023)
Claude can report what is in the J-space; other representations are less reportable.Image is the inner board where content becomes describable; the 2015 diagram lists a dedicated language representation channel alongside the other two.
Claude can modulate J-space contents on request — think about something silently.Of Image's three representation types, one is defined as "reason representation (by Will)" — contents deliberately placed on the board by the goal system.
Intermediate steps of silent multi-step reasoning light up in the J-space and causally mediate performance.The 2022 research note gives the equation "Image (sense + feeling + memory) + Will (program) = simulation" — thinking is running images on the board to predict.
J-space contents are flexibly reusable across tasks (France → capital, currency, continent).In BOSIFWEM's relation-oriented framing, a concept on the board is a node whose edges open: foregrounding it exposes its relation bundle for any downstream use.
The J-space is not involved in routine processing; blocking it preserves fluent interaction but removes higher-order cognition.A 2023 note states it in four words: "consciousness is surplus capacity" — fluency and grammar belong to automatic programs (Memory); the board exists for the higher-order remainder. BOSIFWEM likewise rejects a separate "unconscious": there are only automatic programs plus the board.

A terminological footnote: the classic metaphor behind global workspace theory is the blackboard. The 2015 BOSIFWEM diagram — drawn before its author had encountered GWT — labels Image, in English, as "board of all sense, feelings, concepts, stories." The same diagram marks consciousness as "momentarily disappear", anticipating the moment-by-moment, non-substantial view of conscious access.

3What the surrounding ecosystem predicts

GWT — and the J-space results — specify the workspace itself. BOSIFWEM additionally specifies who operates it (an OS-like scheduler, "Operating Soul"), what fills it (Sense and Memory), what tags it (Feeling), and what reads and writes it (Will and Ego). If the J-space is a genuine workspace of this kind, the ecosystem view makes at least three findable predictions:

P1 — Valence tagging. BOSIFWEM's Flow 11 ("emotional salience") stamps every board entry with a value tag before it can drive behavior. Prediction: J-space activations carry a detectable affect/valence component; suppressing that component selectively degrades motivation- and preference-dependent tasks while leaving neutral recall largely intact.
P2 — A gating controller (attention as resource allocation). Flows 12 and 14 describe Will re-tuning sensory thresholds and allocating processing to the board. Prediction: entry into the J-space is governed by an identifiable controller circuit with competition dynamics — steerable, and distinct from the workspace contents themselves; its dysfunction should look like distractibility or fixation rather than knowledge loss.
P3 — A privileged self-occupant. BOSIFWEM's Ego is a construct assembled on the board (narrative self-construction flow), not a homunculus behind it. Prediction: self-referential representations show elevated baseline residency in the J-space; ablating them selectively harms first-person consistency (persona stability, autobiographical coherence) while leaving third-person reasoning intact.
P4 (bonus) — Graded figure/ground. BOSIFWEM describes the board as ~100 layered representations alternating foreground/background (a gestalt structure). Prediction: J-space membership is graded rather than binary — a small bright foreground with a measurable penumbra that primes without being reportable.
Why P4 is load-bearing — a note on graded knowing. The last prediction is not a loose end. The same author has spent years on a multi-valued logic in which "unknown" is a first-class truth value rather than a missing one. In it, a belief moves among four strata — unevaluated, indeterminate, superposed, and settled — and maturity is measured not by how much reaches "settled" but by how freely a mind can travel among them. Read through that lens, J-space membership is exactly such a spectrum: the bright foreground is settled, the penumbra is superposed or indeterminate, and most of the network sits unevaluated. The prediction then sharpens: the penumbra should behave less like faint noise and more like a distinct epistemic state — one that can prime, be queried, and resolve — and a system able to represent its own "not yet known" should hold an advantage, in calibration and in knowing when to stop, over one that cannot.

4Why this note

Two mapmakers, starting from different continents and different methods — one from decades of disciplined introspection formalized into an architecture, one from the mathematics of a trained model's Jacobian — appear to have drawn the same landmark. That is often a sign the landmark is real, though the author would rather hold even this as provisional: a convergence to be tested, not a claim to be won.

There is a reason the correspondence runs deep rather than cosmetic. If consciousness is surplus capacity, the next question is what the surplus is for. The author's long-term program, ECONAS — Evolvable CONscious/CONscience Artificial Systems — proposes that the aim is not to maximize certainty but to move freely among the epistemic strata above, holding "I don't know" as a stable, first-class state rather than a failure to be corrected. On this view humility is meant to be built in as a computational rule, not added afterward as manners.

A word on where this comes from. The author works alone — a single researcher who keeps conceiving, building, and revising by himself, moving back and forth between the philosophy of mind and computer engineering, and still thinking it through rather than finished with it. The research program (JIDEL, a thought-description language; the Neutom Network; BOSIFWEM; and, long-term, ECONAS) is small, independent, deliberately open about what is proposal versus hypothesis, and would welcome contact.

Overview: neutomtech.com/bosifwem.html (EN) · the research manifesto · foundation: sookmook.org
Sources — Anthropic, "A global workspace in language models" (interpretability research post, July 6, 2026) · BOSIFWEM primary records: founding document (2015), information-flow map (© 2008), full architecture diagram (2015-11-15, English labels quoted verbatim), research notes (2022–2023), comprehensive manuscript in preparation (2026). Compiled July 7, 2026 by Claude at the request of the author. This note asserts convergence, not priority; all BOSIFWEM claims are documented in dated primary records available on request.