Contribution
The paper restates the overflow argument and answers the objections it met after 2007. Its new move is to make the sparse view pay for its posits: an unconscious iconic memory detailed enough to support the tasks, and a conscious “generic” representation. It argues that neither has independent support, while a third memory store, fragile VSTM, gives positive evidence for a cortical, high-capacity representation distinct from working memory.
Key insight
A sparse account of what subjects see does not remove the information they demonstrably use. It moves that information into an unconscious store. So the dispute is not rich experience versus no extra posit; it is rich experience versus a detailed unconscious image plus a conscious one with no specific content. Asked which hypothesis is better supported, rather than which is default, the rich view wins.
Assumptions
- Introspective judgments that one sees more than one can report are data, used in the aspects both sides accept (p. 5).
- Iconic memory isolates consciousness from the retina: the neural locus of the high-capacity memory is in areas that are candidates for the basis of conscious perception (p. 1).
- Working memory holds about four simple items (Box 1), with capacity falling for complex items (Alvarez and Cavanagh: about 4 for colours, under 2 for cubes).
Key results
- Accessibility versus access (p. 1). Overflow says necessarily most items are unaccessed, not that any is inaccessible.
- Kouider et al.’s modified Sperling task (pp. 2–4, Fig. 3). Low contrast plus a backward mask reduced recall to 1.47 of 4 per row (4.41 of 12; 3.9 of 12 with a pseudo-letter present), at the level usually taken to mean the icon is gone. Rotated or flipped letters in uncued rows were taken for real letters at a rate the authors put at 10–15%. Block’s reply is that the low error rate shows enough specific conscious information persisting even in a diminished icon. An isoluminant stimulus would block retinal persistence without diminishing the icon.
- Fragile VSTM (pp. 5–7, Fig. 4). With 32 items, a cue 10 ms after offset gives 30 correct for black/white and 20 for isoluminant red/grey; from 1 s on, the two are equal, and a flash before the cue affects only the early cue. With 8 items, capacity is 7 at 1 s, 6 at 2.5 s and 5 at 4 s, working memory 4. Sligte et al. 2009 locate the persisting representation in V4, not V1–V3. For coloured drawings, high-resolution representations number about 4 in fragile VSTM against slightly above 1 in working memory. Decreasing attention (three methods) and TMS to right dlPFC reduce working memory but barely fragile VSTM. Makovski and Jiang find a smaller but real advantage (30% for colours, 40% for odd shapes).
- Unconscious working memory (p. 7). Soto et al.: subjects who rated a masked grating invisible compared it with a later one at slightly above 55% after 5 s (d′ = .297 for presence), too weak to explain 5 rectangles held for 4–5 s.
- Box 2. Brockmole et al.: subjects hold an imaged 12-dot partial grid for 4–5 s at about 90% and superimpose it on a displayed one, as their introspective reports say they do.
- Two fallacies (p. 8). That a vote for overflow is a vote for inaccessible consciousness (Cohen and Dennett’s “dissociative” grouping); and that unconscious images are the default (Phillips).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Overflow does not entail that any item is inaccessible | strong (a logical point) | p. 1, lottery analogy |
| C2 | Sparse views require detailed unconscious iconic memory for which there is no independent evidence | moderate | pp. 4, 8 |
| C3 | Kouider et al.’s illusion results are compatible with rich phenomenology | moderate: rests on the diminished icon and the 10–15% error rate | pp. 2–4 |
| C4 | Fragile VSTM is a cortical store, distinct from working memory, with roughly double its capacity | moderate to strong: several Amsterdam experiments and a replication with a smaller effect | pp. 5–7 |
| C5 | Fragile VSTM is conscious | weak to moderate: inferred from the four-fold high-resolution advantage and the weakness of unconscious working memory; Block concedes it “could” be unconscious in principle | p. 7 |
| C6 | Conscious generic letter or rectangle representations are implausible | weak to moderate: appeals to crowding and peripheral vision reports | p. 5 |
Concepts
- overflow argument: the capacity of phenomenal consciousness exceeds that of cognitive access.
- generic illusion: subjects mistake a conscious representation of letter-likeness for specific letter shapes.
- fragment illusion: subjects mistake conscious letter fragments for whole specific letters (Kouider et al.).
- pure iconic memory: rod-fuelled retinal persistence of a few hundred milliseconds.
- fragile visual short-term memory: a cortical persisting representation, intermediate in capacity, lasting 4–5 s.
- visual working memory: a smaller, longer-lasting store that probably requires categorisation.
Connections
- Block (2007), LIT-646. This paper’s predecessor; the overflow argument and the refrigerator-light reply are carried forward, and the “iconic memory” of 2007 is split in two.
- Dennett. Cohen and Dennett (2011) is the principal target, not held. The record’s Dennett, LIT-431 and LIT-433, are related through the 2007 paper.
- Higher-order theory. Lau and Rosenthal (2011) and Brown (2011) are among the critics answered; neither is held.
Bearing on the record
- Filed as THEORY-103: perceptual consciousness overflows cognitive access, sourced from this paper and LIT-646.
- THEORY-102 reports a stable ventral temporal code for as long as an image stays on. Sligte et al.’s V4 locus for fragile VSTM points to the same territory, but for representations after offset. The link is mine.
- No ML instruction.
Limitations
- Second-hand evidence, as in all of Block’s Opinion pieces; the primary Amsterdam and Kouider papers are not in the record.
- That fragile VSTM is conscious is the contested step, and the paper’s case for it is plausibility, as Block says (“we will not get anywhere in consciousness research by accepting the less plausible account”).
- In-press copy. Printed page numbers could not be checked.
Open questions
- Neuroimaging could test whether attention moves from item to item during the long Amsterdam retention intervals, boosting each from generic to specific, which would revive the generic-illusion account (p. 8).
- Does fragile VSTM survive an isoluminant version of Kouider et al.’s pseudo-letter task?
Corrections
- none to a seeded skim (there was no seed)