Dehaene–Changeux model¶

A neural-network implementation of the global neuronal workspace account in which local processors and long-range workspace neurons model conscious access, decision-making, and reportable cognitive integration.

Version: v1 · 2026-09-28 · History
Domain-specific #: 8897
Domain group: Natural Sciences
Origin domain: Neuroscience
Subdomains: Computational Neuroscience, Global Workspace Theory → Neuroscience

Core Idea¶

The Dehaene–Changeux model is a computational-neuroscience implementation of global workspace theory. Specialized processors interact with long-range workspace neurons; when competition and recurrent amplification cross a threshold, a representation is modeled as igniting and becoming globally available.

The model has generated predictions for paradigms including inattentional blindness and planning tasks. It is a framework for neural correlates and access, not a complete solution to subjective experience. Architecture, parameterization, task mapping, and empirical validation must remain separate claims.

How would you explain it like I'm…

The Brain's Big Bulletin Board

Your brain has lots of little helpers, each good at one job, like seeing colors or hearing words. There is also a big shared message board that connects them. In this computer model, when one helper's message gets strong enough, it suddenly lights up the whole board so every helper can use it. Scientists use that sudden light-up to study when we notice something.

When Brain Messages Ignite

Scientists have an idea called the global workspace: the brain has many specialized parts, plus a shared 'workspace' that connects them over long distances. The Dehaene–Changeux model is a computer model that turns that idea into simulated brain cells. Different pieces of information compete, and when one gets boosted past a tipping point, the workspace 'ignites' and that information becomes available everywhere in the brain. Scientists use the model to make predictions, for example about why we sometimes fail to notice things right in front of us. It helps explain when information reaches the whole brain, but it does not claim to fully explain what it feels like to experience something.

Global Workspace Ignition Model

The Dehaene–Changeux model is a computational neuroscience version of global workspace theory. It contains specialized processors plus "workspace" neurons with long-range connections that link them. Representations compete, and recurrent (looping) activity can amplify one; if that amplification crosses a threshold, the representation "ignites" and becomes globally available to the rest of the system. The model has been used to make predictions about tasks such as inattentional blindness, where people miss an unexpected object, and planning tasks. It is a framework for studying neural correlates of conscious access, not a complete theory of subjective experience. Its architecture, its parameter choices, how it is matched to a task, and whether experiments support it are separate questions.

The Dehaene–Changeux model implements global workspace theory as a computational neural network. Specialized, largely modular processors handle particular contents, while a population of workspace neurons with long-range, reciprocal connections links them. Representations compete for access to the workspace; recurrent amplification between processors and workspace can drive one representation past a threshold, at which point the model shows ignition: a sudden, sustained, self-reinforcing activation that makes that content globally available to other processors. This nonlinear ignition is the model's proposed mechanism for conscious access. The model has generated predictions for paradigms including inattentional blindness and planning tasks. It should be read as a framework for neural correlates of access rather than a solution to the problem of subjective experience. Evaluating it requires keeping separate the architecture, its parameterization, the mapping from model variables to a task, and empirical validation.

Structural Signature¶

Sig role-phrases:

specialized processors. Perform local or domain-specific representations. Constitutive architecture. If altered: A unitary undifferentiated network misses workspace competition.
workspace neurons. Provide long-range integrative connectivity. Identity-bearing population. If altered: Local recurrence alone is not global workspace broadcast.
competitive ignition. Selects one coalition into sustained global activation. Constitutive transition. If altered: Weak transient activation need not imply conscious access.
global broadcast. Makes selected information available to multiple cognitive systems. Constitutive functional claim. If altered: A report channel alone is not system-wide availability.
behavioral paradigm mapping. Connects model states to tasks such as inattentional blindness or planning. Necessary validation frame. If altered: Simulation resemblance is not proof of consciousness.

What It Is Not¶

Global workspace theory. Is the broad cognitive theory or DCM neural implementation meant?
Attention. Is selection alone or conscious-access broadcast claimed?
Recurrent processing. Is long-range workspace recruitment present?
Artificial neural network. Is a general model being mistaken for this architecture?

Scope of Application¶

Use DCM for source-specific computational and experimental analysis of global neuronal workspace mechanisms.

Conscious access. Models ignition and report availability.
Attention. Studies competition and masking.
Inattentional blindness. Links weak versus global activation.
Decision and planning. Models workspace coordination.
Computational neuroscience. Tests neural-network predictions.

Clarity¶

Global activity is not enough. The model's claim is a structured transition from local processing to sustained long-range availability under competition.

Manages Complexity¶

A model compresses anatomy and behavior into populations and parameters, enabling predictions while leaving alternative explanations, subjective phenomenology, and biological detail outside its direct proof.

Abstract Reasoning¶

Specify specialized and workspace populations.
Define competition, recurrence, and ignition criterion.
Trace broadcast to multiple modeled systems.
Map internal variables to a behavioral or neural paradigm.
Treat agreement as support for a mechanism, not proof of consciousness itself.

Knowledge Transfer¶

Competition–ignition–broadcast structure transfers to other workspace models. DCM's neural populations, parameter choices, and consciousness interpretation do not; a broadcast software architecture is only an analogy. The nearest stopping boundary is explicit: Baars's global workspace theory is closest: it is the broader cognitive architecture that DCM implements in a specific neuronal computational form. The inclusion test remains: A case qualifies when it instantiates the Dehaene–Changeux global-workspace architecture and maps modeled ignition/broadcast to declared neural or behavioral evidence. The structure no longer applies when the case exits when long-range workspace and ignition are absent or when simulation output is treated as direct evidence of subjective consciousness.

Examples¶

Canonical¶

A simulated stimulus activates a local processor but only a sufficiently amplified winner recruits long-range workspace neurons and becomes available to modeled report and decision systems.

Mapped back: specialized processors → local stimulus network; workspace neurons → long-range population; competitive ignition → thresholded recurrent winner; global broadcast → report and decision access; behavioral paradigm mapping → visibility task.

Applied / In Practice¶

An inattentional-blindness simulation reproduces reduced workspace ignition for an unattended stimulus, but the paper treats this as a testable neural prediction rather than direct access to experience.

Mapped back: specialized processors → sensory representation; workspace neurons → global network; competitive ignition → suppressed under inattention; global broadcast → limited availability; behavioral paradigm mapping → inattentional blindness.

Structural Tensions¶

T1: mechanistic precision vs. phenomenological reach. A neural computation can predict access while leaving subjective experience underdetermined. Diagnostic: Which consciousness claim is actually tested?

T2: global integration vs. specialized processing. Broadcast depends on local representations it does not replace. Diagnostic: Where does each computation occur?

Structural–Framed Character¶

Description turns on specialized processors, workspace neurons, competitive ignition, global broadcast, behavioral paradigm mapping. Skeletal core. Specialized processes compete for entry into a shared long-range coordination layer. Domain-bound accent. Neurons, consciousness, attention, ignition, report, and cognitive tasks define DCM. Transfer remains bounded because Why not prime. Global broadcast is portable; this is a named consciousness model. The negative boundary is concrete: Any consciousness theory, neural network, attention model, central executive, recurrent activity, or global workspace metaphor is not the DCM. DCM is mixed: network dynamics are formal, while mapping them to consciousness and report is theoretically framed. Its character: a neuronal global-workspace model of competitive ignition and broadcast.

Structural Core vs. Domain Accent¶

Skeletal core. Specialized processes compete for entry into a shared long-range coordination layer.

Domain-bound accent. Neurons, consciousness, attention, ignition, report, and cognitive tasks define DCM.

Why not prime. Global broadcast is portable; this is a named consciousness model.

This entry is a kind of Representation.

Workspace. Shared access coordinates otherwise specialized processing.
Broadcast. Selected information becomes widely available.
No strict parent is asserted.

Relationships to Other Abstractions¶

Current abstraction Dehaene–Changeux model Domain-specific

Parents (1) — more general patterns this builds on

Dehaene–Changeux model is a kind of Representation Prime

Dehaene–Changeux model is a strict kind of Representation: its frozen identity entails the parent's defining structure while adding domain-specific restrictions.

Hierarchy path (1) — routes to 1 parentless root

Dehaene–Changeux model → Representation → Abstraction

Neighborhood in Abstraction Space¶

Dehaene–Changeux model sits in a moderately populated region (51^st percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Cognitive & Behavioral Theories (16 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Global workspace theory. Tell: Is the broad cognitive theory or DCM neural implementation meant?
Attention. Tell: Is selection alone or conscious-access broadcast claimed?
Recurrent processing. Tell: Is long-range workspace recruitment present?
Artificial neural network. Tell: Is a general model being mistaken for this architecture?

References¶

Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Dehaene%E2%80%93Changeux_model (revision 1334885413).
Preserved source candidate: http://cogprints.org/5275/1/BrainDynamics.pdf
Preserved source candidate: http://www.coma.ulg.ac.be/papers/vs/murphy_propofol.pdf
Preserved source candidate: https://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0014224
Preserved source candidate: http://www.tbiomed.com/content/4/1/10
Preserved source candidate: https://www.templetonworldcharity.org/projects-database/0389
Preserved source candidate: https://www.science.org/doi/10.1126/science.abj3259
Preserved source candidate: https://osf.io/mbcfy/
Preserved source candidate: https://www.science.org/doi/10.1126/science.adj4498

The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.