George A. Mashour, Pieter Roelfsema, Jean-Pierre Changeux and Stanislas Dehaene (2020), Neuron 105(5):776–798, a Review — DOI 10.1016/j.neuron.2020.01.026. Read in the PMC author manuscript, PMC8770991.
Key takeaways
- The theory, in its own words. Conscious access is “a non-linear network ignition associated with recurrent processing” that “amplifies and sustains a neural representation, allowing the corresponding information to be globally accessed by local processors” (abstract). Its substrate is a second computational space of “GNW neurons” with long-range axons, mainly large layer II/III pyramidal cells with a layer V contribution, forming a “high-efficiency cortical core” of prefrontal, parietal, cingulate and temporal hubs (Figure 1). The prefrontal cortex is a key hub but “not … the exclusive territory for conscious access”.
- Ignition is the signature. Across vision, audition and touch, seen and unseen trials share the first ~200 ms and then diverge suddenly, with late activity spreading to prefrontal and parietal cortex (the P300 or late positive component). Subliminal stimuli give “a slowing, decaying wave” instead. The authors concede that an earlier negativity (VAN, ~260 ms) survives when stimuli are task-irrelevant yet reported seen, and that whether the P3 tracks awareness or post-perceptual processing is “unresolved”.
- The best single result. In van Vugt et al. (2018), monkeys reported weak stimuli with a saccade. Both seen and missed stimuli drove V1 and V4, but only seen ones, and false alarms, ignited sustained prefrontal firing (Figure 2). The signal-detection threshold corresponds to the ignition threshold, and sensitivity to how well the signal propagates to prefrontal cortex, so “the SDT and GNW are not independent theories”.
- Level as well as content. Propofol, sevoflurane and ketamine, with different molecular targets, all functionally disconnect prefrontal from posterior parietal cortex and reduce the diversity of functional connectivity across GNW nodes (Figure 5). Central thalamic stimulation reverses anaesthesia in monkeys and helped one minimally conscious patient. Sleep abolishes late ignition while sparing early responses.
- Access, not phenomenality. “Unless otherwise specified, the term ‘consciousness’ in this review will be replaced by conscious access.” The GNW “brings into question the very distinction between phenomenal and access consciousness”. Of Lamme’s proposal that local sensory recurrence suffices for phenomenal consciousness, they write that “short of an experimental method for producing conscious experience in the absence of conscious access, this proposal remains untested”.
Standing in the record
Filed on 2026-10-03 at the owner’s request. It is the record’s first source paper for the global workspace; until now the record knew the theory only through those who argue with it or borrow from it. No anthology topic holds a theory of conscious access, and the review carries no instruction for machine-learning practice; its one mention of machine learning is an aside on training networks to form sequences of memory states.
It rivals Block (2007), LIT-646. Block argues that phenomenology overflows cognitive access, that the workspace is not part of the neural basis of phenomenology, and that Dehaene and colleagues’ “refrigerator light illusion” (mistaking potential access for actual experience) is wrong. This review takes the other side in its own text: it adopts “global availability of information … is what we subjectively experience as a conscious state” (Dehaene and Naccache 2001), replaces consciousness with conscious access throughout, and reads the Pitts et al. stimuli that are reported seen without a P3 as possibly “merely potentially visible but unattended”, which is the potential-access reading Block rejects. It does not discuss Sperling or overflow by name, and it does not cite the 2007 paper; the Block work it cites, for the distinction itself, is “Two neural correlates of consciousness” (2005), LIT-629. Neither paper tests the other.
It rivals Block (2019), LIT-637, on prefrontal involvement. The review cites Panagiotaropoulos et al. (2012), whose monkeys passively watched binocular rivalry while prefrontal neurons encoded the current percept, as “not confounded by working memory or effects related to reporting”. Block’s paper argues that this is exactly what a no-report paradigm cannot show, since a monkey not asked to report may still think about what it sees, and it offers inconspicuous rivalry, in which fMRI found no prefrontal difference, as the better test. The two papers draw opposite conclusions about the prefrontal cortex from the same class of evidence. Block’s paper appeared in November 2019, after this review was submitted, and the review does not answer it.
It rivals IIT 3.0, LIT-445. Section “The GNW and Other Theories of Consciousness” and Table 1 set the GNW against IIT (citing Oizumi et al. 2014, the paper LIT-445 holds): the GNW is representational and gives the prefrontal cortex a hub role in ignition; IIT is “closed and non-representational”, takes a posterior “hot zone” to be sufficient, and treats prefrontal cortex as “not essential for experience itself”. The review runs no test between them. It argues that the perturbational complexity index, which IIT inspired, is “fully compatible with the GNW”, and it names the COGITATE adversarial collaboration as the test under way.
Block’s other papers in the record bear on it without a declared relation:
- Block (1995), LIT-643, is the source of the phenomenal/access distinction this review puts in question. Its claim that Baars’s workspace is “a model of A-consciousness” is the review’s own framing, accepted: the review says outright that its subject is access.
- Block (2009), LIT-632, compares the workspace with higher-order and biological theories and finds that the workspace replaces the explanatory gap with a puzzle of its own. The review does not engage the gap at all; its comparison (Table 1) is by mechanism, prefrontal role, measure and disruption.
The empirical papers. Vishne et al. (LIT-644, THEORY-102) tested predictions that, by their account, came from the GNW’s proponents, including this review: a transient prefrontal ignition at onset. That prediction held. The review’s own text adds a nuance the test bears on: ignition is followed by “metastable” sustained activity in prefrontal and associative areas while a representation is maintained, and the van Vugt monkeys held it until the response. A burst confined to onset, with no sustained prefrontal content while a seen image stays on, fits the review’s statement that ignition is the transition into the attended working-memory state better than any claim that prefrontal activity carries the percept for its duration. That reading is mine. Maldarelli and Güntürkün’s crow review (LIT-650, THEORY-100) asks what the GNW would need in a bird; this review’s answer is long-range reciprocally connected hubs capable of ignition, with no requirement of lamination, which is the point that review presses. This review does not discuss birds.
Butlin et al., LIT-056, take four indicator properties from the global workspace, and their reading in NOTE-052 lists Dehaene, Changeux and Mashour among the theory’s sources. This review supplies what those indicators abstract from: ignition, its threshold, and its neural evidence. THEORY-023, on architectural indicators presupposing computational functionalism, can cite it as the workspace in its own terms: a theory that “views consciousness as an evolved neurocomputational system that enables the global sharing of representations”, and whose authors close by hoping it can be implemented “in an actual computational device”.