The Workspace Is Real. The Mystery Is Borrowed.

There is a standard move in the science of consciousness, and Taylor makes it as clearly as anyone writing for a general audience. The move is to identify consciousness with a mechanism of access: information becomes conscious when it wins a competition and gets distributed widely, “broadcast,” to the rest of the brain. Taylor’s vehicle is working memory, which he shows is close kin to the global workspace, and his essay is a small masterclass in why the pairing is tempting. Doorways erase our plans; phones hide unicycling clowns; fridges lie to us about their own lights. Every example lands.

I want to take the essay seriously enough to point at the one word it leans on and never cashes. It is in the title. Working memory may mysteriously give rise to consciousness. That adverb is not a flourish. It is the entire hard problem, standing in the doorway, and the essay walks past it.

What The Essay Gets Right

The framework I work from, RSP (Recursive Social Prediction, the model in Beyond the Bat), does not treat the global workspace as a rival. It treats it as a complementary theory — a correct account of one thing that is not the thing in dispute. So most of Taylor’s essay reads, from where I sit, as true.

The workspace is the machinery of access, and RSP needs it

Taylor is a self-described “huge fan of the global neuronal workspace theory, which suggests that consciousness arises as a result of information being ‘broadcast’ in a ‘global workspace’ in the brain.” He notes that “the global workspace and working memory are similar notions,” and that attention is the gate: “when we pay attention to information that is stored in working memory, that information gets boosted in volume and is broadcast across the brain.” None of this is in tension with RSP. Broadcasting and attentional gating are real, and they are part of the Level 3 machinery RSP assumes. If you want to know which contents are available to report, plan, and act on, the workspace story is a good answer. RSP simply denies that this answer is an answer to a different question.

His opening example is a Level 2 event boundary

Take the essay’s first image, the doorway effect: you walk into a room and forget why. Taylor’s reading is capacity — “the brain flushes away the old facts from working memory, to get ready for the fresh information that we might need in our new setting.” RSP borrows the example and offers a sharper mechanism, one cognitive psychology already prefers. Crossing a threshold is an event boundary: the brain closes one event model and opens another, the location-updating effect in Radvansky’s event-horizon account. The old contents are not evicted because the bucket is full. They are cleared because they stop predicting. Highly predictive of the room you were in, they are worthless in the room you just entered, so precision drains from them and they drop away. (The effect itself is real but conditional, weaker in some replications than the popular version implies; the mechanism it illustrates is the durable part.)

In RSP this is Level 2 doing its ordinary job. The spatial schema, which the framework places in the same hippocampal-parietal systems the doorway literature implicates, is a generative model of where you are and what that affords. Change the place and you swap the model, and the swap, not a shortage of slots, is what drops the keys. It is a clean piece of predictive machinery, and it is entirely a story about access: which contents are available in which context. Nothing in “the event model updated” touches why being in either room is felt. Hold that thought; it is the whole essay in miniature.

The refrigerator light illusion is already half of RSP’s move

The essay’s best passage is about a puzzle: a country scene feels overwhelmingly rich, but working memory holds almost nothing. Taylor’s resolution is that the richness is partly a trick.

“We are only ever conscious of a few little bits at a time, but the ease with which attention can make things conscious fosters the illusion that we’re conscious of a lot more.”

He calls it the refrigerator light illusion: whenever you check whether the light is on, it is on, because checking is what turns it on. This is a genuinely deflationary claim about felt richness, and it is close to something RSP says in its own voice. On RSP’s account there is no meaning-independent reservoir of rich experience sitting behind the report, waiting to be sampled. The felt plenitude is generated in the act of being modeled, not read off a full store. Taylor has walked right up to the idea that the richness of experience is a construction of the checking system. He uses it to defend the workspace’s small capacity. RSP uses the same insight one level deeper, and I will come back to why that matters.

There is a name for the shape of this illusion, and it is bigger than consciousness. The illusion of explanatory depth (Rozenblit and Keil, 2002) is the finding that people rate their grasp of ordinary things far above what they can actually explain — a zipper, a flush toilet, a bicycle — right until you ask them to explain, at which point the confidence collapses. Same engine as the fridge: a shallow thing available on demand, mistaken for a deep thing possessed. They are two species of one genus, generation-on-demand misread as a standing store, and the signature recurs well past these two cases: in choice blindness, in confabulated reasons, in the rich visual field that change-blindness experiments puncture.

This genus is what RSP predicts. If the self-model generates its reports when probed rather than reading them off an inner record, introspection should overstate what is already there, in domain after domain, which is what these separate literatures keep finding. RSP commits to exactly this: “introspection is not transparent access to the structure of experience” but “itself a cognitive process that can mislead us about its own workings,” the givenness of raw experience being “an artifact of how consciousness presents itself.” The framework already carries one member of the family in its comparison set, Chater’s introspection-as-illusion thesis, which it classes as complementary: it agrees on the confabulating self-model and parts only on whether to call the result an illusion. Taylor’s refrigerator light is that same claim shrunk to a single appliance. I develop the pattern on its own in a companion piece; here it is enough that his best deflationary move is a local instance of RSP’s global one.

The spectrum is graduated phenomenality by another name

Taylor’s own contribution is to reject the binary. “I suggest that there could be some information that is not ‘in’ or ‘out’ of working memory, but somewhere in between.” He argues working memory “comes on a spectrum,” with “a large grey area,” and asks the obvious follow-up: “whether there can be a grey area between something being conscious and not being conscious.” RSP answers yes, and has from the start. Consciousness on this framework is graduated, not switched: an invertebrate’s bare valence, a mammal’s richer self-and-world model, a human’s recursive self, are points on a slope, not members of a club. Taylor’s grey area is that slope, seen through the lens of working-memory encoding. His intuition is right. What it needs is a mechanism, and the mechanism is where our accounts part.

The Word Doing All The Work

Here is the fault line. After a tour of what working memory explains, the essay closes on a promissory note.

“These experiments and others continue to unearth the mechanics of working memory. There is still much more to find out … As these mysteries are slowly explained, perhaps we will be in a better place to tackle the arguably biggest conundrum in science: consciousness itself.”

Read that carefully. The mechanics of working memory are one project; “consciousness itself” is another, harder one, and the hope is that finishing the first will position us for the second. That is an honest statement, and it concedes the whole game. It means the workspace account, however complete, does not touch the question of why any broadcast is accompanied by experience. The title’s “mysteriously” is not a placeholder for a fact we will soon discover. It marks a structural limit of access theories: you can specify, to any level of detail, which information gets distributed and where, and you will not have said one word about why distribution is like anything from the inside. Broadcasting is content-neutral. A system could globally share information about a rippled patch, a phone number, a sunset, and route it flawlessly, and nothing in “it was broadcast” entails “it was felt.” Taylor knows this, which is why the word is in his title and the payoff is deferred to a future paragraph that never arrives.

What RSP Supplies: The Step After “Broadcast”

RSP is a theory of what consciousness is, not of which contents reach report, and its central claim is that the felt is not produced by broadcasting but is identical to a particular kind of self-modeling. The workspace makes information available. Being conscious of that information is the system recursively modeling itself as an experiencer of it — the strange loop RSP calls Level 3b. On this view the explanatory gap is not narrowed by more mechanism; it is dissolved, because there was never a second thing (the feeling) over and above the self-model that needs to be caused into existence. That is the step past “broadcast” that access theories, by their nature, cannot take.

This immediately does three things for Taylor’s own puzzles.

• It answers the capacity problem he leaves open. Taylor’s refrigerator light illusion says the richness is partly manufactured by attention. RSP says the felt quality itself is manufactured — by the self-model’s report, not stored in slots and then illuminated. There is no overflow to explain because there was never a full bucket of experience to overflow. Independent behavioral work on human throughput, which the paper builds a prediction around, puts the actual rate of conscious processing at roughly ten bits per second against an enormous sensory input. The mismatch Taylor finds romantic is exactly what RSP expects: a narrow self-model narrating a wide world and reporting the narration as plenitude.
• It supplies the sufficiency condition he cannot state. Taylor can say working memory plus attention is where consciousness shows up. He cannot say why that combination suffices rather than merely correlates. RSP names the missing ingredient: it is not enough that information is broadcast; the system must recursively predict itself as the thing undergoing the broadcast. Access without that recursion is a very capable unconscious router.
• It explains his “working memory without consciousness” case. Taylor is troubled by the 2011 Gabor-patch result — information apparently held in working memory while the subject is unaware of it — and reaches for his spectrum to soften it. RSP gives the dissociation a mechanism rather than a grey area: lower-level encoding, the Level 2 schemas and the Level 3a categorization beneath the self-model, can run without the recursive self-model (Level 3b) that would make it experienced. The same dissociation appears under anaesthesia: the paper reads the surviving hippocampal encoding of speech semantics, grammar, and even upcoming-word features as Level 2 schema processing continuing while the Level 3 self-model is offline. No one is home to experience it. Unconscious contents in working memory are not a paradox on this view. They are the ordinary case of access without a self that is present to it.

What The Essay Gives Back

The debt runs both ways. RSP is strong on what a subject is and thin on the empirical texture of the buffer that feeds it. Taylor’s essay is a tour of exactly that texture: the four-slot experiments and their flexible-resource rivals, chunking, the doorway studies, the courtyard clown, the Gabor patch. RSP’s claim that conscious throughput is narrow needs precisely this literature to have teeth, and Taylor lays it out with more care and charm than the appendices do. His refrigerator light illusion, in particular, is a cleaner popular statement of constructed richness than anything I have written, and I would rather cite it than paraphrase it. An account of what consciousness is still owes an account of the machine it rides on, and this essay is a good part of that account.

Where We Disagree

One disagreement does not dissolve into complementarity, and it is worth stating plainly. Taylor treats consciousness as the “biggest mystery facing both science and philosophy today” and bets that grinding out the mechanics will eventually let us “tackle” it. RSP’s wager is the opposite: the mystery is not a hard fact awaiting data but a category error awaiting dissolution, and no amount of workspace detail will close a gap that closes only when you stop treating the feeling as a separate effect. Under this reading Taylor is an optimist about a problem RSP thinks should be retired rather than solved.

Beneath that sits a sharper split. Taylor’s countryside scene, with its hills, birdsong, and cut grass all present “at once,” reads as though there is a rich phenomenal field that the workspace either does or does not deliver in full. RSP denies the field. There is no meaning-independent raw quale that the broadcast conveys intact; there is a self-model constructing and reporting states, all the way down. Taylor’s refrigerator light illusion flirts with this and then stops at the fridge door. RSP walks through it. If he took his own deflation one step further — from “the richness is exaggerated by attention” to “the felt quality is itself a construction of the modeling system” — the mystery in his title would not need solving. It would have been dissolved by the very move he already half-makes.