Global Workspace Theory says a mental state is conscious when its content becomes widely available across the system — broadcast out of a narrow workspace to the many processes that can use it. That is a claim about access. It is not a claim about experience, and its architects have rarely pretended otherwise.
In Bernard Baars's 1988 formulation, and in the neuronal version developed later by Stanislas Dehaene and Jean-Pierre Changeux, the theory predicts which information becomes reportable, which is lost, and roughly when the switch happens. Those are real predictions, and in 2025 they were tested hard. What the theory does not address is why the broadcast is accompanied by anything at all — why there is someone in the room to receive the bulletin.
Naming that boundary is not an attack. A theory that explains access and says so plainly is in better shape than one that quietly widens its claim. The interesting work starts exactly where the workspace stops.
What is global workspace theory?
Start with what is established. The Stanford Encyclopedia of Philosophy entry on consciousness (section 6.5, consulted 14 September 2026), describes global workspace theories as distinguishing conscious states partly through "enhanced wide spread access to the state's content." That sentence is doing careful work. The mark of consciousness, on this family of views, is not some extra glow attached to a mental state. It is the state's availability — the fact that its content can now be used by memory, by speech, by planning, by attention, by whatever else is listening.
The intuition behind it is ordinary. Most of what your nervous system does stays local: your visual system computes edges without any of it reaching the part of you that can talk. Occasionally a piece of content breaks out of its local process and becomes available everywhere at once — you can report it, remember it, act on it. Global Workspace Theory takes that transition to be the phenomenon worth explaining, and offers an architecture for it. If you want the wider map rather than this one theory in depth, we set out how the major theories of consciousness compare separately.
How does the "theater" metaphor actually work?
The usual exposition runs like a stage. A vast number of unconscious specialist processes work in parallel, each competent and each blind to the others. Attention acts as a spotlight, and whatever the spotlight lands on is lit for the whole house — made available system-wide rather than staying in its own corner. Capacity is the point: the stage is small, the wings are enormous, and competition for the stage is what makes consciousness serial and slow while the underlying machinery is parallel and fast.
Taken as engineering, this is a strong idea and it earns its keep. The steelman deserves stating at full strength: a theory that explains why conscious contents are few, ordered in time, reportable, and usable by any downstream process at once has explained something substantial about the mind, and it has done so in a way that generates experiments rather than slogans.
But metaphors carry passengers. A theater implies an audience — someone seated in the dark for whom the lit stage appears. Nothing in the architecture supplies that audience. Broadcast is a relation between subsystems; every process that receives the signal is another mechanism, not a spectator. Read strictly, the workspace has no seats. Read loosely, it seems to explain far more than it does, and the loose reading is the one that circulates.
What is the difference between GWT and global neuronal workspace theory?
The distinction is standard in the field rather than a quotable finding, so take it as consensus framing and not as a citation. Baars's global workspace is a psychological architecture: a functional story about capacity, competition and availability, stated at the level of cognitive processes. The global neuronal workspace, developed by Dehaene and Changeux, is the attempt to cash that architecture out in neurons — long-range connections, mostly frontoparietal, that sustain and amplify a signal once it wins. Its signature claim is ignition: a sudden, nonlinear, widely distributed burst of activity that marks the moment content enters the workspace.
The difference matters for a practical reason. A functional architecture is hard to refute directly, because many neural arrangements could implement it. A neuronal claim about when and where ignition occurs is falsifiable in a specific experiment — which is why the neuronal version is the one that ends up in the lab, and the one that can lose.
What did the IIT–GNWT adversarial collaboration find?
Here the established layer gets sharp. The Cogitate Consortium ran a preregistered adversarial test — predictions from both camps locked in before data collection — pitting the global neuronal workspace against integrated information theory. The results were published in Nature on 30 April 2025 (volume 642, issue 8066, pages 133–142). The paper's own summary: "These results align with some predictions of IIT and GNWT, while substantially challenging key tenets of both theories."
No winner was declared, and reporting it as a defeat for either side misreads it. Each theory was caught by a prediction it had made in advance. The table below sets the two failures side by side, since that symmetry is the finding.
| Theory | Prediction that failed | What the data showed |
|---|---|---|
| Integrated information theory | Sustained connectivity within posterior cortex during conscious perception | The predicted sustained posterior connectivity was not found |
| Global neuronal workspace | Ignition at the offset of the stimulus, as well as its onset | Ignition at stimulus offset was largely absent |
Both theories survived in part; both lost a tenet they had staked in public. That is what a healthy adversarial collaboration looks like, and it is more informative than a verdict would have been. For how the rival in this contest defines and quantifies its central quantity, see how integrated information theory measures consciousness.
If the honest scorecard interests you more than the headline, the free guide works the same way — arguments with their failure conditions attached.
What does GWT not explain?
Two layers have to be kept apart here, and the second is still open.
What is established: broadcast tracks reportability well, and the neuronal version makes timing predictions precise enough to fail. What remains live debate: whether ignition is the signature of consciousness or a signature of reporting it — a system asked to describe its states will show widespread availability of exactly the content it is describing, and separating the two experimentally is genuinely unsettled. The Cogitate null result on offset ignition is one reason that debate has not closed.
Underneath both sits the boundary that no amount of further data moves. Suppose every prediction had been confirmed. You would know which contents become globally available, through which pathways, on what timescale. You would not know why availability feels like anything. A system could broadcast content to every subsystem it has, report on it fluently, and — as far as the architecture goes — be no one at all. Access is a third-person fact about traffic. Experience is a first-person fact about there being a point of view. The workspace explains the traffic.
Global Workspace Theory is a physicalist theory: brains produce consciousness by organizing information. Holopsychism runs the explanation in the other direction, and this is a proposal, not a result. The framework treats consciousness as pure potential rather than as something the brain manufactures, with awareness as selection — the act by which one coherent possibility becomes actual — and the brain as receiver, filter and translator, tuning a signal rather than generating it. On that reading, the workspace is a description of the filter's bandwidth: what a nervous system of this resolution can bring into coherence at once. Resolution here is a spectrum, not a hierarchy.
That proposal is not supported by the Cogitate data and owes an account of its own failure conditions before it can ask for assent. It explains less than it hopes: nothing about why this filter yields this point of view, and no equivalent of an ignition prediction to lose. That debt is the whole reason for the question this whole site is framing.
The useful conclusion is methodological. A theory of access that knows it is a theory of access can be tested, revised and broken — as 2025 showed. The damage comes from the upgrade in translation, where "explains which contents become available" becomes "explains consciousness." Hold the line and both halves stay honest: a working science of access, and an unembarrassed question about the one who receives it.
FAQ
Is Global Workspace Theory proven?
No, and its neuronal version took a direct hit recently. The Cogitate Consortium's preregistered test, published in Nature on 30 April 2025, found that results "align with some predictions of IIT and GNWT, while substantially challenging key tenets of both theories" — including the global neuronal workspace's prediction of ignition at the end of a stimulus, which was largely absent.
Does GWT solve the hard problem of consciousness?
It does not set out to. The theory characterizes conscious states partly by widespread access to their content, which is a claim about what information becomes available system-wide. Why that availability is accompanied by experience is a separate question the architecture leaves open.
Are GWT and GNWT the same theory?
They are the same idea at two levels of description. Baars's global workspace is a psychological architecture of capacity and competition; the global neuronal workspace of Dehaene and Changeux is its neural implementation, with long-range frontoparietal ignition as the mechanism. The neuronal version is the testable one.
Did integrated information theory lose the adversarial test?
Neither theory won. IIT's predicted sustained connectivity within posterior cortex was not found, and the global neuronal workspace's offset ignition was largely absent. The paper challenges key tenets of both and names no victor.
Is Holopsychism a version of Global Workspace Theory?
No. The workspace account is physicalist — the brain generates conscious access by organizing information. Holopsychism proposes the opposite direction, with the brain as receiver and filter rather than source. It is a tentative framework, not an extension of GWT and not supported by GWT's evidence.
The guide takes the same boundary seriously and marks every proposal as a proposal.