Mashour and colleagues’ 2020 review synthesizes two decades of work on the global neuronal workspace (GNW) hypothesis, a framework that tries to explain how the brain turns a private flicker of neural activity into something you can hold in mind, talk about, and act on. The GNW idea is deceptively simple: the cortex is full of specialized “local processors” (vision, language, action, valuation), and conscious access happens when one local representation is amplified and broadcast through a highly interconnected network—especially fronto-parietal hubs—so that many systems can use it at once.

If you’ve ever driven home on autopilot and suddenly “come back online” when someone cuts you off, you already know the difference the GNW is trying to capture. Plenty of processing happens without consciousness. The question is: what changes when experience becomes available for deliberate control, flexible decision-making, and verbal report?

This is where neurofeedback and biofeedback become particularly interesting. Neurofeedback is a form of learning in which a person receives real-time feedback about brain activity (most commonly via EEG) and gradually learns to shift it; biofeedback uses the same principle with bodily signals such as heart rate variability, breathing, muscle tension, or skin conductance. Both approaches are, at their core, training systems for state regulation—helping the brain-body system move from less adaptive patterns toward more adaptive ones.

The GNW model offers a useful lens for understanding why “state” matters so much in training. It suggests that effective self-regulation is not only about changing a single rhythm or region, but about whether the brain can reliably sustain, integrate, and “broadcast” information across networks when it needs to—especially under stress, fatigue, or overwhelm.

Methods

This article is a narrative review rather than a single experiment, but it is methodologically rich in a different way: it organizes converging evidence across multiple paradigms that contrast (1) conscious versus nonconscious contents (what is experienced) and (2) conscious versus nonconscious states (whether experience is possible at all). The review pulls together computational modeling, invasive recordings in animals, human EEG/MEG, and human neuroimaging—then asks whether the data fit the central GNW predictions.

A key methodological backbone of GNW research is the use of experimental designs that hold the stimulus constant while manipulating access. Classic examples include visual masking and threshold detection tasks, where the same stimulus is sometimes reported and sometimes missed. In GNW terms, misses are not “no processing”; rather, they reflect processing that remains largely feedforward and local, failing to recruit the long-range recurrent loops required for ignition.

The review also highlights “signature” measures that repeatedly appear when access occurs. Across many studies, early sensory responses (often within the first ~200 ms) can be present even when the stimulus is not consciously reported, suggesting that initial feedforward processing is not the defining feature of consciousness. Instead, GNW emphasizes later dynamics: sustained activity, widespread cortical recruitment, and late event-related components such as the P3-like wave that often accompanies reportable detection.

Importantly, the authors extend these ideas beyond moment-to-moment perceptual access into levels of consciousness—the capacity for any content to become conscious. Here, the methods shift to clinical and physiological perturbations: general anesthesia, sleep, and disorders of consciousness after brain injury. These domains are valuable because they can dissociate early sensory processing from global access. For example, during sleep, MEG studies suggest early sensory processing can be relatively preserved while later ignition and the late P3 response diminish.

Finally, the review methodically positions GNW relative to other influential theories (e.g., integrated information theory, recurrent processing theory, and higher-order thought theories), not as a philosophical debate, but as an empirical challenge: can we design experiments and causal interventions that separate local recurrent processing from global broadcasting, and posterior “hot zones” from fronto-parietal hubs?

Results

Across the reviewed evidence, the GNW hypothesis is supported by a recurring pattern: conscious access is associated with a non-linear, all-or-none transition in large-scale brain dynamics—what the authors describe as ignition. Computational simulations predict two regimes for the same input. In one, activity remains subthreshold and fades as it ascends the hierarchy. In the other, recurrent excitation and feedback trigger self-amplification, recruiting a distributed set of regions and sustaining the representation long enough to be used by multiple systems.

Electrophysiological studies in nonhuman primates provide a compelling testbed for these predictions. When animals report detection of weak stimuli, activity is not simply “a bit bigger” than on missed trials; it tends to show a sharper emergence and persistence, including engagement of higher-order areas such as prefrontal cortex. The review also notes that false alarms—reports without a stimulus—can show patterns resembling ignition, underscoring that conscious access is not a passive mirror of the outside world but a brain-generated selection-and-broadcast process.

In humans, EEG/MEG research supports a distinction between early sensory processing and later global signatures. Early components can appear for both seen and unseen stimuli, while later components—including the late P3-like response—more reliably track reportable access. The authors also discuss an important and healthy controversy: some data suggest that certain late components (including P3) may reflect post-perceptual processes related to decision-making or report rather than awareness per se, fueling the rise of “no-report” paradigms. GNW research has responded by emphasizing that ignition can, in principle, be measured without explicit report—by looking for sustained, widespread broadcasting dynamics and their causal perturbations.

When the question shifts from contents to states, the pattern becomes even clearer. Under general anesthesia, a major therapeutic effect can be framed as disruption of conscious access: information becomes less available to working memory and other cognitive systems. Despite diverse molecular targets across anesthetic agents, the review highlights converging evidence for a common proximate effect: disruption of the reverberant networks (especially fronto-parietal connectivity) that support global broadcasting. Functional disconnection or metabolic depression of fronto-parietal networks appears across major anesthetic classes.

Sleep provides a natural experiment with graded changes in responsiveness. MEG work discussed in the review suggests that during lighter sleep stages, the first ~200 ms of sensory processing can remain present (though weakened), but later ignition and the late P3 response to salient oddball stimuli drop off—aligning with the GNW prediction that the loss of conscious access is tied to failure of global ignition rather than complete shutdown of early sensory pathways.

The review also addresses dreaming and lucidity. Dreaming during REM sleep can involve higher-frequency activity that includes frontal and prefrontal regions, even in the absence of immediate report. Lucid dreaming—where awareness and some control emerge within the dream—is associated with increased gamma activity and coherence in dorsolateral prefrontal cortex, and causal manipulations (e.g., oscillatory entrainment) have been reported to increase lucidity. These findings are framed as supportive of a role for prefrontal participation in the degree of conscious access, even when the content is internally generated.

Discussion

One reason GNW remains clinically and practically useful is that it treats consciousness as an operational capability: the ability to flexibly access, sustain, and share information across brain systems. In daily life, this capability is what lets a person notice a subtle internal shift (“my body is getting tense”), keep it online long enough to interpret it (“this is anxiety building”), and then recruit strategies (“slow my breathing, widen my attention, change my self-talk”)—all while continuing to function.

From a self-regulation lens, GNW suggests two broad failure modes. The first is when information never reaches the workspace: early processing happens, but it remains fleeting and local. This can look like missed cues, poor interoceptive tracking, or a sense of acting before thinking. The second is when the workspace is unstable: ignition occurs, but it is noisy, hijacked by threat, or hard to sustain. This can look like rumination, attentional capture, or the feeling that the mind is “stuck on broadcast” with the wrong channel.

Biofeedback and neurofeedback map neatly onto these dynamics because they are training environments that (a) increase the salience of internal signals, (b) scaffold sustained attention and working memory for state tracking, and (c) provide reinforcement for more stable state transitions. If conscious access depends on recurrent amplification and large-scale integration, then training that gently improves arousal regulation, attentional stability, and network coordination should, in principle, support the very conditions under which adaptive ignition is more likely.

The anesthesia and sleep evidence in this review adds a particularly practical insight: early sensory processing can be present even when the system cannot sustain global access. In other words, the brain can still “hear” without being able to use what it hears in an integrated way. Translating that to clinical work, it cautions against assuming that presence of a response (a reflex, a startle, a transient EEG change) equals meaningful access. It also explains why some people can describe moments of being “there but not there” under fatigue, medication, dissociation, or shutdown: local processors keep running, but the workspace has trouble igniting or maintaining broadcast.

The ongoing debate about late EEG components (like P3) is also a gift rather than a problem. It reminds us to separate three related but distinct phenomena: awareness, attention, and report. Neurofeedback training often depends on all three. A person can attend without being aware of a cue, be aware without being able to report it clearly, or report something without truly sustaining it. GNW encourages designing training that supports the entire chain: stabilizing arousal, supporting selective attention, and improving the capacity to hold and manipulate information (working memory) without overload.

Finally, the comparison with other theories (IIT, recurrent processing theory, higher-order theories) is a reminder that the field is converging on shared features—especially the importance of recurrent processing—while still debating scope. GNW’s distinctive claim is that conscious access involves a broader architecture of long-range loops that include fronto-parietal hubs enabling global routing. For neurofeedback practice, this matters because it nudges protocol thinking beyond a single “hot zone” and toward network-level goals: integration, stability, and context-appropriate broadcasting.

Brendan’s perspective

There’s a moment in many neurofeedback sessions that I’ve come to think of as “the lightswitch.” It’s not dramatic. No fireworks. It’s usually a tiny shift: the client stops chasing the screen, their face softens, their breath drops a notch, and they say something like, “Oh… I get it.”

From the outside, it looks like learning. From the inside, it often feels like the nervous system found a new gear.

When I read Mashour, Roelfsema, Changeux, and Dehaene’s review of the global neuronal workspace, that clinical moment clicks into a bigger model. GNW isn’t a neurofeedback theory, but it describes a problem neurofeedback wrestles with daily: how do we help the brain move from scattered, local processing to integrated, usable access—especially when stress, fatigue, trauma, ADHD, anxiety, or sleep disruption are constantly pushing the system away from integration?

In GNW terms, neurofeedback is often trying to do one (or more) of three things:

1. make internal signals more detectable (so they can enter the workspace),

2. stabilize the brain’s ability to sustain a state once it appears (so ignition can be maintained), and

3. reduce the noise and interference that cause the workspace to be hijacked (so broadcasting is context-appropriate rather than threat-driven).

That framing changes how I think about protocols. Not because it tells us a single “best” target, but because it reminds us that the target should serve a functional goal: improving the capacity for stable, integrated access.

A practical example: clients who describe feeling “spaced out,” foggy, or disconnected often show EEG patterns consistent with under-arousal or unstable arousal regulation—sometimes elevated slow activity, sometimes excessive variability, sometimes a kind of rhythmic rigidity that doesn’t flex with demand. In GNW language, that can look like a system that struggles to reach or sustain ignition. The clinical aim isn’t simply “reduce theta” or “increase beta.” The aim is to help the brain build a reliable bridge between state regulation and cognitive availability.

One starting point I often consider is sensorimotor rhythm (SMR) training. A classic SMR range is roughly 12–15 Hz, frequently trained around central sites such as C3, Cz, or C4. Clinically, SMR training can support sleep stability, motor inhibition, and a calmer, more organized baseline—conditions that can make global access easier when the day demands it. If a person’s workspace is constantly disrupted by hyperarousal or impulsive state shifts, creating a steadier baseline can be the difference between “the brain knows what to do” and “the brain can actually do it when it counts.”

For clients whose main issue is anxious overdrive—racing thoughts, somatic tension, attentional capture—alpha-based approaches can be useful, particularly when they are framed as training flexibility rather than chasing a single number. Alpha (often ~8–12 Hz) is a complex rhythm, but in practice it often relates to relaxed alertness, sensory gating, and the ability to widen attention without collapsing into drowsiness. Posterior alpha training (e.g., POz) can help some people downshift visual vigilance and soften hypermonitoring. More frontally aligned alpha and beta strategies (carefully selected based on assessment) can sometimes support a calmer executive stance: less reactive broadcasting, more thoughtful broadcasting.

Then there are clients who don’t lack light (in our analogy... their switch is maybe stuck "on"), they have too much of it in the wrong direction. Think of rumination, perseveration, intrusive imagery, or the sense that the mind is stuck in a loop. In GNW terms, the broadcast system may be over-amplifying threat-salient representations, repeatedly re-igniting the same content. Here, protocols that build inhibitory control and improve switching can be more relevant than protocols that simply “increase activation.” This is where individualization matters most. Two people can present with the same symptom (“I can’t stop thinking”), but one needs downshifting of arousal and threat circuitry, while the other needs improved executive stability and working memory control.

So how does GNW help, concretely? It encourages three design principles in neurofeedback.

First, prioritize state stability before state sophistication. If ignition is a threshold phenomenon, then a jittery, exhausted, or hypervigilant brain will struggle to cross thresholds cleanly. This is why combining neurofeedback with biofeedback can be so powerful. Teaching a person to regulate breathing, increase heart rate variability, or reduce electrodermal reactivity can create the physiological conditions for cleaner cortical learning. When the body stops screaming “danger,” the workspace has bandwidth.

Second, train context-appropriate broadcasting. The goal is not a permanently “high integration” brain. The goal is a brain that can flex: focused when focus is needed, open when openness is needed, quiet when quiet is needed. Clinically, I think of this as training transitions. A session might involve alternating between a focus task and a recovery task, or pairing neurofeedback with brief cognitive challenges that require working memory transformation—because the review highlights that when working memory contents must be transformed, active, decodable states re-emerge with signatures resembling conscious access.

Third, respect the report problem. GNW research wrestles with the fact that some neural signatures may partly reflect reporting rather than awareness itself. Neurofeedback has a parallel problem: the feedback interface can become the task, and the client can learn to game it in ways that don’t generalize. In practice, I counter this by anchoring the training to felt sense and functional outcomes. Instead of “make the bar go up,” the aim becomes “find the internal shift that makes your attention steadier,” or “notice what changes in your chest when the tone stabilizes,” or “see if your thinking becomes quieter but clearer.” Those are workspace-relevant skills.

If I were to translate GNW into a simple clinical metaphor, it would be this: the brain is a newsroom. Many journalists (local processors) are constantly gathering information. Conscious access happens when one story makes it to the editor’s desk, gets verified, and is sent out on the wire to every department. Neurofeedback is not writing the story for the newsroom. It’s improving the newsroom’s ability to choose wisely, verify efficiently, and distribute information without panic.

That metaphor also keeps us honest about the limits of research versus practice. GNW is built largely on controlled paradigms: masking, detection thresholds, reportability. Real lives are messier. People don’t just fail to detect a stimulus; they fail to detect their own rising arousal, their own fatigue, their own narrowing attention—until the system tips. That’s why protocols must be individualized, and why a qEEG (when used thoughtfully) is less a “diagnosis machine” and more a map of constraints and tendencies.

For example, someone with pronounced fronto-central fast activity and high autonomic arousal might benefit from protocols that support downshifting and inhibitory control (often starting centrally with SMR and adding relaxation-oriented alpha work as tolerated), paired with breath training and pacing. Someone with under-arousal and slow-heavy patterns might need careful activation and stabilization—sometimes beginning with sleep regulation and gentle attentional training rather than pushing intensity. Someone with strong posterior rhythms but weak executive stability might benefit from approaches that support frontal-midline stability and task engagement, but only once arousal is under control.

The most exciting implication of the Mashour et al. review for neurofeedback is that it reframes “consciousness” as a trainable capacity for access and integration. We are not training consciousness directly. We are training the ingredients that make conscious self-regulation more likely: stable arousal, recurrent engagement, sustained attention, and flexible broadcasting.

And when that ignition moment happens in session—when the client suddenly feels the shift and can reproduce it—it’s hard not to think: this is the workspace learning to light up on purpose.

Conclusion

Mashour and colleagues’ review of the global neuronal workspace hypothesis offers a powerful, testable story about conscious access: local processing becomes conscious when recurrent loops push activity past a threshold, triggering widespread ignition and global broadcasting. Across paradigms that contrast seen versus unseen stimuli, and across altered states such as sleep and anesthesia, a consistent theme emerges: early sensory responses may persist, but late, sustained, widely distributed dynamics are what seem to track access.

For biofeedback and neurofeedback, GNW provides a useful lens rather than a rigid recipe. It encourages thinking in terms of network integration, state stability, and the capacity to sustain and flexibly use information—skills that are central to self-regulation. Whether the goal is calmer attention, better sleep, improved inhibition, or reduced reactivity, training tends to work best when it supports the conditions for adaptive “broadcasting” rather than chasing isolated signals.

If there is a take-home message from GNW for clinical practice, it is this: helping the nervous system regulate is not just about quieting noise—it is about making the right information globally available at the right time, so the person can steer their life with more clarity and choice.

References

Mashour, G. A., Roelfsema, P., Changeux, J.-P., & Dehaene, S. (2020). Conscious processing and the global neuronal workspace hypothesis. Neuron, 105(5), 776–798. https://doi.org/10.1016/j.neuron.2020.01.026