Contribution
The paper names a circle in consciousness science and argues that it can be broken empirically. To decide whether unreportable representations are phenomenal, one needs the neural basis of phenomenology. To find that, one must decide whether to include the machinery of cognitive access, which seems to need the very answer sought. Block argues that inference to the best explanation, applied to psychology and neuroscience together, settles it, and that the evidence already says the access machinery is not part of the basis of phenomenology. It replaces “reportability” with “cognitive accessibility” and retires “phenomenal consciousness” for “phenomenology”.
Key insight
Capacity is a fingerprint of a system. If people demonstrably see more items than their access system can hold, the machinery of seeing is at least partly distinct from the machinery of access. If, further, the brain contains a mechanism that would produce that surplus only on the assumption that the access machinery is outside phenomenology, the fit between the two sciences is evidence for the assumption.
Assumptions
- Physicalism about phenomenology (§2): it is identical to its total neural basis. The paper’s question is not physicalism versus dualism.
- Core versus total neural basis (§2), and causal versus constitutive conditions. The question is whether access is constitutively necessary for phenomenology (§4).
- The global workspace model of cognitive access (§11), shared with the paper’s opponents.
- Subjects’ reports that they see all or almost all items are taken at face value, defeasibly (§8).
- Awareness is not cognitive access (§6): the sense in which one is aware of one’s experiences can be met by the deflationary or “same order” views, so it does not require the access underlying report.
Key results
- Reportability is the wrong notion (§4). A locked-in aphasic could have experience and no capacity to report. Owen et al.’s “vegetative” patient who imagines tennis on request shows an intentional act without report.
- Sperling (§9). 3 × 4 arrays for 50 ms: subjects report about 4 letters but, cued after offset, almost all of any row. The result is the same at 15 ms and 500 ms exposure.
- Landman et al. 2003 (§9). 8 rectangles for 500 ms. Capacity (Cowan’s K) is about 4 when the cue comes with the test array, almost 8 when it comes with the first, and 6–7 when it comes in the blank up to 1.5 s after offset. Detecting a change in size or orientation is no harder than detecting a change in size alone.
- Sligte et al. 2008 (§10). With up to 32 items, capacity near 32 for an early cue with black/white (rod-driven) stimuli and about 17 for isoluminant red/grey, both falling to about 16 for a late cue and to just over 4 once the new array appears. A light flash just after offset removes the black/white advantage; at about 1 s it has no effect, while a pattern mask at 1 s drops capacity to working-memory level. So the later persistence is cortical, not retinal.
- Engel 1970 (§10). Monocular persistence of random-dot stereograms lasts about 80 ms; the experience of depth persists about 300 ms, so persistence is phenomenal and above early vision.
- Working memory holds about four (§12). Luck and Vogel (four items with up to four features each); Hauser et al.’s monkeys choose the larger of two buckets up to four apple slices and choose at random at 3 versus 8; infants at about three (Feigenson et al.).
- Loftus’s grid task (§11). Subjects found the missing dot if and only if they rated the two half-grids as one integrated matrix, confirming their phenomenological reports without partial report.
- The brain (§§13–14). Delay-period firing in dorsolateral prefrontal cortex maintains working memory; prefrontal damage impairs delayed but not simultaneous performance. A V1–V5–V1 loop is the core basis of one kind of seen motion (Silvanto et al.). In Sergent and Dehaene’s attentional blink visibility is all or none, a frontal competition. Kouider et al. 2007 found strong occipito-temporal activation without frontal activation for supraliminal words under distraction.
- Predictions (§14). Representations of unaccessed items in Sperling-type tasks will involve recurrent loops; if G.K.’s fusiform activations for unseen faces are recurrent, that is experience he cannot know about.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Whether access is constitutive of phenomenology is an empirical question | moderate: an argument from the method of inference to the best explanation, with a measurement-error analogy | §8 |
| C2 | Phenomenology overflows cognitive access (narrow sense) | moderate: partial-report superiority in three paradigms plus subjects’ reports; the reading of those reports is the point in dispute | §§9–11 |
| C3 | The persistence that carries overflow is cortical, not retinal | moderate to strong: Engel, Treisman, Sligte et al. | §10 |
| C4 | The refrigerator-light (illusion) reading of overflow is not supported | moderate: six considerations, none decisive | §11 summary |
| C5 | Visual working memory holds about four items or fewer | strong: many independent paradigms | §12 |
| C6 | The neural basis of phenomenology does not include the workspace | weak to moderate: an inference to the best explanation from coalition models and Kouider et al. 2007; the neural details “are wrong, or at least highly incomplete” by the author’s own caution | §§13–14 |
Concepts
- methodological puzzle: we cannot find the basis of phenomenology without deciding whether it includes access, and the evidence seems to come through access.
- metaphysical correlationism: access is part of what phenomenology is.
- epistemic correlationism: access is part of our knowledge of phenomenology, so whether it is part of phenomenology is intractable.
- cognitive accessibility, narrow and broad: representations in the workspace (Dehaene and Naccache’s I3), versus those plus representations that could enter it (I2). Overflow is overflow of the narrow sense.
- overflow: the capacity of phenomenology exceeds that of the workspace.
- refrigerator light illusion: mistaking potential phenomenology, available whenever attended, for actual phenomenology.
- hyper-illusion: a mismatch between an appearance and the appearance of that appearance (§11).
- mesh: agreement between psychological and neural results that is a reason to accept the theory producing it.
Connections
- Block (2005), LIT-629. The two NCCs, the MT/V5 evidence and the coalition picture are carried over; the mesh argument is new.
- Block (1995), LIT-643. Block says this argument “is thus importantly different” from that paper’s, which claimed Sperling directly shows inaccessible phenomenal states (§9). Here only overflow is claimed, and items are accessible one at a time when cued.
- Dennett and Kinsbourne (LIT-431, NOTE-367) and “Quining Qualia” (LIT-433, NOTE-357). Block names Dennett (1988, 1991) for concluding from “empirically indistinguishable” views that there are no facts, and his “first person operationism” as remnant verificationism (§§7–8). The overflow argument is a direct case: whether the uncued letters were seen is the kind of question the Orwellian/Stalinesque argument says has no answer. The relation is declared on the LIT.
- Chalmers. Quoted from a 1998 paper as an epistemic correlationist. His 1997 claim that consciousness correlates with availability is answered: the broad sense of access “encompasses too much”, since unconscious representations can be amplified by attention too (Carrasco).
- Global workspace (Dehaene, Changeux, Naccache). The paper’s main opponents; Dehaene and Naccache’s I1/I2/I3 taxonomy and their refrigerator-light reply are quoted at length (§11).
Bearing on the record
- THEORY-100. The theory’s caveat about report paradigms is this
paper’s puzzle. The paper also bears on the theory’s
promote_when: Block (2019, LIT-637) argues that removing report does not remove the problem. - THEORY-102. The mesh argument’s picture, content held in recurrent sensory activity with frontal involvement only for access, is the picture that theory’s data fit. That theory concerns seen images, so it does not test overflow.
- Filed as THEORY-103: perceptual consciousness overflows cognitive access. Sources LIT-646 and LIT-640; rivals in the record LIT-431 and LIT-433.
- No ML instruction.
Limitations
- Overflow rests on reading subjects’ reports that they see all the items, which is what the illusion hypothesis denies. Block’s case against the illusion is cumulative, not decisive.
- Neural details are provisional by the author’s own caution (§13).
- The 32 commentaries were not read. They include Kouider et al.’s “partial awareness”, Naccache and Dehaene, Grush on generic phenomenology, and Sergent and Rees, and Block’s response; Block (2011, LIT-640) answers the later forms of their objections.
Open questions
- Are the representations of unaccessed items in partial-report tasks recurrent, as the paper predicts?
- Is G.K.’s fusiform activation for faces he denies seeing recurrent?
- Can the method be extended to representations inside Fodorian modules, the case that opens the paper?
Corrections
- none to a seeded skim (there was no seed)