Gal Vishne, Edden M. Gerber, Robert T. Knight and Leon Y. Deouell (2023), Cell Reports 42(7):112752 — DOI-10.1016/j.celrep.2023.112752
Key takeaways
- When a clearly visible image stays on, local activity in visual cortex does not: response amplitude and single-electrode category selectivity fall by about 80% within 900 ms of onset (81.8% and 77.4% from peak). The content does not fall with it. Linear decoders on the pattern across electrodes separate face from watch in ventral temporal cortex at an AUC of 99.8% averaged over 100–900 ms. A decoder trained at onset still works at the end and the reverse, and information ends about 230–450 ms after the image goes off, whatever its duration.
- In prefrontal and parietal cortex category and exemplar information is real but transient, mostly between about 150 and 600 ms, later than in visual cortex, not generalising across time and not tracking duration. No report was required on these trials, and the targets were a different category.
- The paradigm had been adopted by the COGITATE adversarial collaboration, which pre-registered predictions: IIT a stable, duration-long representation in a posterior hot zone, the global neuronal workspace a transient prefrontal ignition. Both held here, so the authors conclude the two predictions “are, in fact, not adversarial”.
- Their interpretation is that a stable “experience subspace” in the sensory population code fits continuous experience, and the frontoparietal burst fits discrete updating. Which one corresponds to experience depends on whether experience is continuous or discrete, a question the data do not settle.
Standing in the record
Filed on 2026-10-03 at the owner’s request, in a batch of four works on large-scale neural dynamics with Verzhbinsky et al. (LIT-633), Xu et al. (LIT-638) and Muller et al. (LIT-647). No anthology topic holds the neural correlates of human visual experience, and the paper carries no instruction for machine-learning practice.
It is the record’s first empirical test of the neural correlates of consciousness, and it lands on a question the record already holds from the philosophical side. Dennett and Kinsbourne’s Time and the Observer (LIT-431, read in NOTE-367) is this paper’s second reference. Colour phi is the puzzle both start from: Dennett and Kinsbourne argue that whether a revision of content came before or after experience has no fact at that scale, while Vishne et al. set “discrete perception” against “continuous” as rival readings of postdiction and then find a neural signature for each in a different place. The paper takes no position on multiple drafts, and tests nothing Dennett and Kinsbourne claim. It does show that “when” questions about experience have measurable neural correlates that differ by region, which is the empirical ground their argument was about.
On the theories it tests: the record holds IIT 3.0 (LIT-445) and the indicator report of Butlin et al. (LIT-056), whose GWT indicators come from the global neuronal workspace. The paper tests predictions made for this paradigm by those theories’ proponents (Tononi et al. 2016; Mashour et al. 2020), not the formal content of either, so no relation is declared. Its finding that the two theories predicted compatible things here is a caution for THEORY-023’s reading that theory-derived indicators decide what counts as evidence: in this paradigm, two theories that disagree about consciousness did not disagree about the data.
No relation is declared. The record holds no other NCC study for it to extend, rival or be compared against.