Dehaene and Changeux’s Global Neuronal Workspace Theory
Global Neuronal Workspace Theory proposes that conscious awareness involves selective ignition and recurrent broadcasting within distributed networks of cortical pyramidal neurons with long-range axons. Originating in Dehaene, Kerszberg, and Changeux’s 1998 model, and developed with Naccache and others, it specifies interactions among specialized processors, cortical and thalamic circuitry, inhibition, spontaneous activity, reward, and vigilance. Access occurs through ignition, an abrupt, self-sustained, brain-scale activation with measurable signatures. Conceived as a neurobiological model, the theory specifies mechanisms from receptor molecules and inhibitory interneurons to reward systems, long-distance connectivity, and development. Changeux and Farisco emphasize that molecular regulation, biological development, embodiment, and valence are integral to its original multilevel interpretation. Later computational interpretations differ over substrate dependence and artificial consciousness. Experiments support aspects of distributed amplification but challenge particular markers and anatomical predictions. Whether the mechanism explains phenomenal experience itself, rather than the accessibility and consequences of information, remains disputed.

Stanislas Dehaene
Cognitive Neuroscientist
Stanislas Dehaene (b. 1965, Roubaix) is a French cognitive neuroscientist and professor of Experimental Cognitive Psychology at the Collège de France, where he has held the chair since 2005. Trained in mathematics and with a doctorate in experimental psychology, he directs the Cognitive Neuroimaging Unit at NeuroSpin, investigating reading, numerical cognition, reasoning, learning, and neural signatures of conscious access. With Jean-Pierre Changeux, Michel Kerszberg, Lionel Naccache, and other collaborators, he developed and tested the Global Neuronal Workspace Theory of conscious processing. His research combines behavioral experiments, computational modeling, and neuroimaging. He received the Brain Prize (2014), and his books include The Number Sense, Reading in the Brain, and Consciousness and the Brain.

Jean-Pierre Changeux
Professor of Neurobiology
Jean-Pierre Changeux (b. 1936, Domont) is a French neurobiologist whose research connects molecular mechanisms, neuronal development, cognition, and consciousness. He held the chair of Cellular Communications at the Collège de France and directed the Molecular Neurobiology Unit at the Institut Pasteur. His work on allosteric proteins and nicotinic acetylcholine receptor, the first neurotransmitter receptor to be purified, helped establish molecular approaches to nervous-system function. He also developed theories of synaptic selection, proposing the theory of epigenesis by selective stabilization of synapses. With Stanislas Dehaene and colleagues, he proposed and advanced the Global Neuronal Workspace Theory. He received the Wolf and Balzan Prizes. His philosophical writings examine the neuronal basis of experience, knowledge, art, and ethics within a biologically grounded scientific framework.
Key Takeaways
Core Claim
A content becomes conscious when long-range cortical pyramidal neurons amplify and sustain it, making it globally available to specialized nonconscious processors.
How It Works
Ignition, an abrupt, self-sustaining, brain-scale activation, selects one content and inhibits rivals, creating a serial bottleneck in conscious access.
Measurable Signatures
Late amplification, the P3b wave, high-frequency bursts, and long-distance synchrony mark conscious access, emerging about 200–300 milliseconds after a stimulus.
Distinguishing Idea
Conceived as neurobiology rather than abstract architecture: receptors, inhibitory interneurons, reward and vigilance systems, and development are built into the model.
Phenomenal Consciousness
Proponents identify global availability with experience; critics argue that the theory explains access and report, not why access feels like anything.
Challenges & Tensions
The COGITATE adversarial test supported some predictions but challenged offset ignition and prefrontal coding; functionalist and multilevel readings still divide the theory.
Relation to GWT
Shares Baars’s high-level workspace idea but specifies a cortical, molecular, and developmental mechanism absent from the psychological theory.
Dehaene and Changeux’s Global Neuronal Workspace Theory
The Global Neuronal Workspace Theory (GNWT), introduced by cognitive neuroscientist Stanislas Dehaene, computational neuroscientist Michel Kerszberg, and neurobiologist Jean-Pierre Changeux in 1998, and developed with Lionel Naccache and others, holds that a content becomes conscious when it is amplified and sustained by a distributed network of cortical pyramidal neurons with long-range axons and thereby made available to the brain’s many specialized, nonconscious processors. Entry into this workspace takes the form of “ignition,” an abrupt, self-sustaining, brain-scale activation that the theory treats as the measurable signature of conscious access, and its proponents hold that this global availability is what subjects experience as a conscious state. Although it adopts Baars’s psychological idea of a global workspace, GNWT was framed from the start as a neurobiological model, specified from receptor molecules and inhibitory interneurons to reward and vigilance systems, long-distance connectivity, and cognition. Changeux and Farisco accordingly present it as a multilevel theory of conscious processing in living organisms that is not functionalist and should not be conflated with Baars’s Global Workspace (Dynamics) Theory, which has its own entry (Dehaene et al., 1998; Dehaene & Naccache, 2001; Mashour et al., 2020; Changeux & Farisco, 2026).
In other words, GNWT proposes that a representation becomes conscious through nonlinear amplification and recurrent exchange within a distributed neuronal workspace. Its original 1998 formulation specifies long-range cortical neurons, interactions with specialized processors, spontaneous activity, and reward and vigilance modulation, rather than merely an abstract broadcasting function (Dehaene et al., 1998). Changeux and Michele Farisco’s contemporary interpretation emphasizes the molecular, cellular, circuit, developmental, and embodied organization through which human subjective experience becomes possible. GNWT identifies conscious access with experienced awareness, but its phenomenal claims remain contested, and later computational interpretations differ over how indispensable the original biological commitments are (Changeux & Farisco, 2026).
From Baars’s Workspace to Neuronal Architecture
GNWT grew out of a decade of attempts by Changeux and Dehaene to model cognitive functions from the bottom up, beginning with networks that learn temporal sequences by selection, whose elementary units were clusters of neurons and synapses containing allosteric receptors, and continuing with models of prefrontal function in delayed-response and rule-switching tasks. In these models, variable “pre-representations” arise spontaneously and are selected by reward signals (Dehaene et al., 1987; Dehaene & Changeux, 1989).
Dehaene et al.’s 1998 paper applied this approach to effortful cognitive tasks. It distinguishes two computational spaces. The first is a set of specialized, modular processors for perception, movement, long-term memory, evaluation, and attention, which operate automatically and nonconsciously. The second is a single global workspace composed of distributed, heavily interconnected neurons with long-range excitatory axons, mobilized when the specialized processors do not suffice. Workspace neurons selectively amplify or suppress the contributions of processor neurons through descending connections, and they receive vigilance and reward inputs, presumably through specific neuromodulatory pathways.
Anatomically, the authors proposed that workspace neurons do not occupy a distinct set of “cardinal” areas but are distributed across the cortex in variable proportions, a region’s contribution depending on its fraction of pyramidal neurons in cortical layers II and III, which give rise to long-range cortico-cortical and callosal connections and are especially abundant in dorsolateral prefrontal and inferior parietal cortex (Dehaene et al., 1998). Later formulations add cingulate and temporal association regions, the associated thalamocortical loops, and layer V pyramidal cells (Dehaene & Changeux, 2011; Mashour et al., 2020).
Reciprocal thalamic connections contribute to stability. The workspace is not one cortical location or a separate conscious observer. The anatomical contrast emphasized by Changeux and Farisco (2026) is between Baars’s early diffuse ascending reticular–thalamic proposal and a largely horizontal network of long-range cortical connections; it is more specific than the contrast between a psychological description and a neural implementation.
Its connections can amplify some processor inputs and suppress others, establishing a temporary coalition. The network can maintain a representation beyond immediate stimulation, reconfigure when a candidate fails, and coordinate processors that would otherwise operate more independently. The original Stroop simulations modeled effortful task performance, acquisition, and responses to error. Successfully simulating these functions did not establish that the simulated network had phenomenal experience.
Changeux’s own description highlights competitive and cooperative exchange: “The conscious compartment, the global workspace, processes information from non-conscious processors and in turn broadcasts that information from one system of processors to all the other processors—but in a certain way, a ‘theater’ kind of ‘stock exchange’ of processors, where these specialized processors interact, coordinate, and control each other in competition or cooperation.” (Changeux, 2025, personal communication).
He specifies that GNWT “privileges a subset of interconnected cortical pyramidal cells with long-range excitatory axons, particularly dense in prefrontal, temporoparietal, and cingulate regions, that, together with the relevant thalamocortical loops, forms a horizontal ‘neuronal workspace’ interconnecting multiple, specialized, automatic, and nonconscious processors. The conscious content of a given sensory input (seeing, hearing, etc.) is assumed to be encoded by the sustained and amplified activity (termed ‘ignition’) of a subset of neurons from the GNW, the rest being inhibited, which ultimately leads to a report of conscious experience” (Changeux, 2025, personal communication).
Here, broad broadcasting does not require identical delivery to every neuron, and inhibition concerns competing populations rather than brain-wide silence. The reference to report describes a possible behavioral endpoint; it should not make overt report a necessary condition of every phenomenal experience.
Crucially, the architecture already included spontaneous generation of candidate activity patterns, vigilance signals, and evaluation by reward. A proposed molecular implementation involved reward-dependent changes in allosteric receptor states affecting the stability of active neuronal configurations. This was a mechanistic hypothesis, not merely a generic connectionist label (Dehaene et al., 1998; Changeux & Farisco, 2026).
Ignition and Its Signatures
The transition to conscious access is nonlinear. Ignition is the sudden, coherent, and exclusive activation of the subset of workspace neurons coding the current content, with the rest of the workspace inhibited; it is all-or-none at the level of the network, sustained by recurrent long-distance loops, and it imposes a serial bottleneck, since only one content can ignite at a time (Mashour et al., 2020). Simulations of thalamocortical networks reproduced this behavior and predicted that ignition can also occur spontaneously, so that ongoing activity can block access to an external stimulus, as in inattentional blindness (Dehaene & Changeux, 2005). The theory yields a taxonomy: subliminal stimuli are too weak to ignite; preconscious stimuli are strong enough but lack attentional amplification; conscious stimuli have both (Dehaene et al., 2006).
Experimentally, the program contrasts identical stimuli that are sometimes perceived and sometimes not, using masking, the attentional blink, and inattention. Activity for seen and unseen stimuli diverges some 200–300 milliseconds after onset, when seen stimuli evoke a surge of amplified, propagating activity. Dehaene summarizes four signatures of conscious access: late amplification extending into parietal and prefrontal circuits; the P3b wave of the event-related potential; a late burst of high-frequency oscillations; and long-distance synchronization across cortical regions (Dehaene & Changeux, 2011; Dehaene, 2014). The framework has been extended to states of consciousness, where anesthesia and disorders of consciousness are interpreted as loss of the capacity for ignition and long-range communication (Mashour et al., 2020).
Conscious Access and Phenomenal Consciousness
According to Dehaene and Naccache, the hypothesis of a global neuronal workspace should synthesize three empirical observations that, they say, any theory of consciousness should incorporate: “(1) a considerable amount of processing is possible without consciousness, (2) attention is a prerequisite of consciousness, and (3) consciousness is required for some specific cognitive tasks, including those that require durable information maintenance, novel combinations of operations, or the spontaneous generation of intentional behavior” (Dehaene & Naccache, 2001).
Their formulation explicitly links neuronal availability with subjective experience: “This framework postulates that, at any given time, many modular cerebral networks are active in parallel and process information in an unconscious manner. An information becomes conscious, however, if the neural population that represents it is mobilized by top-down attentional amplification into a brain-scale state of coherent activity that involves many neurons distributed throughout the brain. The long-distance connectivity of these ‘workspace neurons’ can, when they are active for a minimal duration, make the information available to a variety of processes including perceptual categorization, long-term memorization, evaluation, and intentional action. We postulate that this global availability of information through the workspace is what we subjectively experience as a conscious state” (Dehaene & Naccache, 2001).
These are substantive theoretical commitments, including the historical claim about attention, rather than definitions that settle the phenomenal question. Later work distinguishes attentional operations and revises proposed markers of conscious access.
Phenomenal consciousness concerns what experience feels like; access consciousness concerns the availability of information for use in reasoning and in guiding speech and action (Block, 1995). The two are not mutually exclusive: an experience can have phenomenal character and be accessible. The disputed question, set forth more fully below, is whether phenomenality can occur without workspace access, as Block’s distinction allows. GNWT generally challenges that separation (Mashour et al., 2020), whereas critics argue that its experiments primarily explain access and its consequences (Boly et al., 2017).
Actual report is distinguishable from access. Workspace information may be available for use without being verbally described or expressed through an immediate response. Conversely, unconscious processing can affect behavior without becoming globally available. A theory of conscious access is therefore not automatically a theory requiring overt reports, although report-based experiments risk confounding experience with the operations needed to communicate it.
Different experiences correspond to differently organized representations, not merely different amounts of global activation. Seeing a color and hearing a tone recruit different content-selective populations and relationships with other processors. GNWT proposes that workspace participation makes the relevant content consciously experienced; the particular content depends on which representation is amplified. This separates the question of why a representation is conscious from the question of why it has this specific qualitative character. Explaining the first does not automatically settle the second (Dehaene & Naccache, 2001; Farisco et al., 2024).
The theory’s explicit target is conscious access, and its authors differ in how they relate access to phenomenal consciousness. Dehaene regards Block’s distinction between the two as misleading and holds that access is the scientifically tractable definition; the impression of a rich experience exceeding what can be reported is attributed to preconscious contents that could be accessed but are not (Dehaene et al., 2006; Dehaene, 2014). Naccache (2018) argues in detail that a strict access account can explain what the notion of phenomenal consciousness was introduced to capture. In a related analysis, Dehaene, Lau, and Kouider (2017) decompose consciousness into global availability of information and self-monitoring, both described as computations, and assess machines against them.
They state their position plainly: “Our stance is based on a simple hypothesis: What we call 'consciousness' results from specific types of information-processing computations, physically realized by the hardware of the brain. It differs from other theories in being resolutely computational; we surmise that mere information-theoretic quantities do not suffice to define consciousness unless one also considers the nature and depth of the information being processed” (Dehaene et al., 2017, p. 491).
Changeux and Farisco prefer the term “conscious processing” and treat subjective experience as one of its components, with conditions that are biological and not only computational. Evaluation and reward circuits are held to be necessary for conscious processing to be valenced, and hence for subjective experience to be accessible; the individual’s experience is shaped by a long postnatal epigenesis in which synaptic connections are selected through interaction with the physical, social, and cultural environment; and the biochemical diversity of the brain controls the states and contents of consciousness. On this reading, current artificial systems that broadcast information among modules may approximate access consciousness while lacking phenomenal experience (Farisco et al., 2024; Changeux & Farisco, 2026).
A Multilevel Theory, Not a Functionalist One
Changeux and Farisco (2026) argue that GNWT has been wrongly assimilated to computational functionalism. A computational description of the levels of conscious processing does not, they hold, license treating that processing as independent of its physical basis. Living organisms have a deeper multilevel organization than inanimate objects, with entangled connections between levels from molecules to social relations, and a model of consciousness must include Marr’s implementation level along with the computational and algorithmic ones. For them, the 1998 formulation, with explicit implementation from molecules through circuits and global modulation to cognition, is the authoritative statement of the theory.
They single out three features as necessary for conscious processing and as marking the difference from a purely functional account. The first is bottom-up multilevel organization, including the molecular level of allosteric receptors and neuromodulators. The second is inhibitory interneurons, needed to maintain a self-sustained representation in the workspace in the absence of sensory input (Volzhenin et al., 2022). The third is evaluation and reward circuitry. To these, they add development: a minimal consciousness is proposed for the preterm fetus at 25–30 weeks, followed by recursive consciousness in the newborn, explicit self-consciousness by the end of the second year, and reflective consciousness with theory of mind at three to five years, with an electrophysiological homologue of adult ignition recorded in infants (Lagercrantz & Changeux, 2009). The theory’s pharmacological realism is presented as a practical benefit, since it directs attention to the neurotransmitter pathways and receptors, notably nicotinic receptors, whose alteration affects conscious processing in neuropsychiatric disease (Koukouli & Changeux, 2020); computers, they observe, have no pharmacology. The position does not exclude other forms of conscious processing in principle, including artificial ones, but it denies that reproducing the cognitive level as per computational functionalism suffices (Changeux, 2017; Farisco et al., 2024).
Farisco notes that this point bears on the theory’s own history: “while the original formulation of the GNWT is clearly different than GWT (see 1998 paper), successive developments have somehow diluted these differences, eventually supporting functionalist interpretations of GNWT that are not really compatible with its original multilevel definition, which is deeply connected to the biological and molecular organization of the conscious organism” (Farisco, personal communication, 2026). Presentations of the theory in computational terms are found chiefly in Dehaene’s later work on machine consciousness (Dehaene et al., 2017); the multilevel, anti-functionalist reading is Changeux’s.
GNWT and GWT: Commonalities and Differences
Dehaene and Changeux’s Global Neuronal Workspace Theory and Baars’s Global Workspace (Dynamics) Theory share a high-level characterization: conscious processing is access of information to a central workspace of limited capacity, for which contents compete and from which the winning content is made available to many specialized nonconscious processors. Mashour et al. (2020) credit the central thesis to Baars and describe his workspace as a psychological construct.
The differences that GNWT’s authors stress are anatomical and architectural. Baars’s proposed substrate was an extended reticular-thalamic activating system projecting diffusely to the cortex; GNWT specifies a network of cortical pyramidal neurons with long-distance cortico-cortical and subcortical connections, a neuronal granularity absent from Baars’s account (Changeux & Farisco, 2026). Baars’s instantiation does not separate the level of consciousness, governed by the reticular formation, from its content; GNWT assigns content to the ignited workspace assembly and level to the ascending systems that enable ignition (Mashour et al., 2020). GNWT adds mechanisms with no counterpart in the psychological theory: nonlinear ignition, top-down mobilization and suppression of processors, inhibitory interneurons, and reward and vigilance inputs. And the two differ in status: GWT is a cognitive architecture that can be implemented in software, whereas GNWT, on the Changeux–Farisco reading, makes sense only on the basis of the biological and molecular organization of the conscious organism. From the other side, Baars counts GNWT within a workspace “family” of theories and has moved his own account toward cortico-thalamic dynamics without a privileged prefrontal role.
Adversarial Collaboration: Empirical Tests and Criticisms
Debate has centered on prefrontal cortex. Critics argue that lesion, stimulation, and imaging evidence locates the correlates of conscious content in posterior cortex, and that prefrontal activation and late signatures such as the P3b reflect report and other post-perceptual processes that disappear in no-report designs (Boly et al., 2017). Defenders reply that null findings in prefrontal cortex are not decisive and that content-specific prefrontal activity can be found with sufficiently sensitive methods (Odegaard et al., 2017).
The well-known Templeton World Charity Foundation-supported, preregistered COGITATE “adversarial collaboration,” involving a “theory-neutral” consortium alongside theory proponents, tested specified GNWT and Integrated Information Theory (IIT) predictions in 256 participants using fMRI, MEG, and intracranial recordings. It reported results that agreed with some predictions of each theory and challenged central tenets of both. It found some predicted content-related activity and frontal–visual synchronization, but challenged GNWT through the general absence of predicted ignition at stimulus offset and limited representation or decoding of some conscious dimensions in prefrontal cortex (Cogitate Consortium, 2025).
GNWT’s proponents, especially Naccache and colleagues (2025), emphasized positive findings and disputed the centrality of offset ignition unless participants consciously registered the offset. They argued that the data confirmed ignition after stimulus onset in the predicted window, irrespective of stimulus duration and task relevance, together with content-specific synchronization between frontal and early visual areas; that offset ignition is predicted only if the offset is itself consciously attended; and that weaker prefrontal decoding is expected where neurons show mixed selectivity (Naccache et al., 2025). Because the study used clearly visible stimuli, it did not directly test a conscious–unconscious transition. These qualifications constrain interpretation, but the preregistered failures remain evidence requiring explanation; they should neither be dismissed nor generalized into a refutation of every workspace model.
State-level observations concerning anesthesia and disorders of consciousness provide additional constraints on the capacity for recurrent global interaction. Such clinical applications assess a system’s potential for access, rather than directly measuring everything a noncommunicating patient experiences. Preserved or absent responsiveness must not be equated automatically with preserved or absent phenomenality (Dehaene & Changeux, 2011).
Kuhn wondered whether the adversarial collaboration really addressed the fundamental essence of either GNWT or IIT: “[I]t seems to me that the ‘neurogeographic’ debate between the ‘front of the head’ folks — global [neuronal] workspace (GNWT) and higher-order theorists — and the ‘back of the head’ folks — integrated information (IIT) and recurrent processing theorists — while essential for a complete neurobiological correlation of consciousness (Block, 2023), is of only mild interest for an ontological explanation of consciousness. If GNWT suddenly shifted to the back of the head, and IIT to the front, would the ‘trading-places’ inversion make much ontological difference? I’d be surprised if the basic principles of GNWT, assigning consciousness to competition and brain-circuit broadcasting, and of IIT, as a system’s intrinsic cause-and-effect structure (the maximally irreducible whole it forms), would need to change at all" (Kuhn, 2026).
Philosophical criticism parallels that directed at GWT: that global availability explains access and report but not why availability is accompanied by experience, and that phenomenal content may exceed what the workspace holds.
Divergent Interpretations and Artificial Consciousness
Farisco cautions that later GNWT formulations have sometimes diluted its original anatomical and architectural differences from GWT and encouraged functionalist readings incompatible with its intended multilevel biological definition. He emphasizes: “The crucial point is the level of biological realism, which was one of the main inspirations of GNWT, introduced with the explicit goal of modeling conscious processing taking place in biological organisms” (Farisco, personal communication, 2026). This concern is developed explicitly in Changeux and Farisco (2026). It identifies an interpretive disagreement within the broader workspace literature, not a consensus that every subsequent formulation is historically illegitimate.
Dehaene, Lau, and Kouider (2017), for example, distinguish global availability (C1) and self-monitoring (C2) from unconscious computation (C0), asking how these capacities might be engineered. In public commentary on language models, Dehaene and Naccache (2026) again present a computational interpretation, while stressing limitations involving recurrent autonomy, embodiment, selfhood, and enduring episodic memory. Their commentary is not a peer-reviewed demonstration of machine consciousness.
The biological position also does not entail that artificial feeling is impossible. Farisco, Evers, and Changeux (2024) distinguish replicating human consciousness from realizing alternative forms; Changeux and Farisco (2026) preserve that possibility. Their argument is that reproducing information-sharing functions is insufficient evidence of reproducing the organization responsible for human subjectivity. Biological dependencies provide constraints and research leads, not a logical proof that a different substrate could never feel.
Assessment
GNWT is among the most fully specified and experimentally productive theories of consciousness: it names cell types and connections, has been simulated at several scales, and has generated clinical measures of consciousness. Its distinctive wager is that conscious access, operationalized by report and ignition, is the proper target of explanation, and that an account of access is, or grounds, an account of experience. At least two questions remain open. One is empirical: how far conscious contents depend on prefrontal participation and late global activity as opposed to posterior and earlier processes. The other concerns the theory’s identity. Read functionally, GNWT describes an architecture that might be realized in other substrates; read as Changeux and Farisco insist it was conceived, it is a multilevel biological theory in which molecules, inhibitory circuits, valuation, and development are constitutive. The second reading separates it sharply from Baars’s GWT and aligns it with biologically grounded views of consciousness; the first keeps it within the workspace family. Both readings have support in the theory’s own literature.
GNWT should therefore be presented as having a shared neuronal core and explicitly distinguished interpretations of its explanatory scope. GWT supplies a related cognitive architecture; GNWT specifies a biological mechanism and, in its multilevel interpretation, a program linking molecular regulation, circuit dynamics, development, and lived significance. Its unresolved phenomenal question is whether this organization constitutes experience, rather than merely enabling global access, and how its concrete dynamics explain the particular qualities of what is felt.
References
Block, N. (1995). On a confusion about a function of consciousness. Behavioral and Brain Sciences, 18(2), 227–247.
Block, N. (2023). The Border Between Seeing and Thinking. Oxford University Press.
Boly, M., Massimini, M., Tsuchiya, N., Postle, B. R., Koch, C., & Tononi, G. (2017). Are the neural correlates of consciousness in the front or in the back of the cerebral cortex? Clinical and neuroimaging evidence. Journal of Neuroscience, 37(40), 9603–9613.
Changeux, J.-P. (2017). Climbing brain levels of organisation from genes to consciousness. Trends in Cognitive Sciences, 21, 168–181.
Changeux, J.-P. (2025). Personal communication to R. L. Kuhn.
Changeux, J.-P., & Farisco, M. (2026). The Global Neuronal Workspace as a multilevel model of conscious processing. Trends in Cognitive Sciences. https://doi.org/10.1016/j.tics.2026.03.004
Cogitate Consortium. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature. https://doi.org/10.1038/s41586-025-08888-1
Dehaene, S. (2014). Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts. Viking.
Dehaene, S., & Changeux, J.-P. (1989). A simple model of prefrontal cortex function in delayed-response tasks. Journal of Cognitive Neuroscience, 1, 244–261.
Dehaene, S., & Changeux, J.-P. (2005). Ongoing spontaneous activity controls access to consciousness: A neuronal model for inattentional blindness. PLoS Biology, 3, e141.
Dehaene, S., & Changeux, J.-P. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70, 200–227.
Dehaene, S., Changeux, J.-P., & Nadal, J.-P. (1987). Neural networks that learn temporal sequences by selection. Proceedings of the National Academy of Sciences, 84, 2727–2731.
Dehaene, S., Changeux, J.-P., Naccache, L., Sackur, J., & Sergent, C. (2006). Conscious, preconscious, and subliminal processing: A testable taxonomy. Trends in Cognitive Sciences, 10(5), 204–211.
Dehaene, S., Kerszberg, M., & Changeux, J.-P. (1998). A neuronal model of a global workspace in effortful cognitive tasks. Proceedings of the National Academy of Sciences, 95(24), 14529–14534.
Dehaene, S., Lau, H., & Kouider, S. (2017). What is consciousness, and could machines have it? Science, 358(6362), 486–492.
Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: Basic evidence and a workspace framework. Cognition, 79, 1–37.
Dehaene, S., & Naccache, L. (2026). Does Claude possess a conscious global workspace? Public commentary accompanying research on global-workspace-like organization in language models; not a peer-reviewed article. https://assets.claude.com/cc4be2488d65e54a6ed06492f8968398ddc18ebe.pdf
Farisco, M. (2026). Personal communication to R. L. Kuhn.
Farisco, M., Evers, K., & Changeux, J.-P. (2024). Is artificial consciousness achievable? Lessons from the human brain. Neural Networks, 180, 106714.
Koukouli, F., & Changeux, J.-P. (2020). Do nicotinic receptors modulate high-order cognitive processing? Trends in Neurosciences, 43, 550–564.
Kuhn, R. L. (2026). Theories on the Landscape: An Orthogonal Approach. Journal of Consciousness Studies, 33, No. 1–2, 2026, pp. 40–56. DOI: 10.53765/20512201.33.1.040. https://www.ingentaconnect.com/content/imp/jcs/2026/00000033/f0020001/art00003
Lagercrantz, H., & Changeux, J.-P. (2009). The emergence of human consciousness: From fetal to neonatal life. Pediatric Research, 65, 255–260.
Mashour, G. A., Roelfsema, P., Changeux, J.-P., & Dehaene, S. (2020). Conscious processing and the global neuronal workspace hypothesis. Neuron, 105(5), 776–798.
Naccache, L. (2018). Why and how access consciousness can account for phenomenal consciousness. Philosophical Transactions of the Royal Society B, 373, 20170357.
Naccache, L., Sergent, C., Dehaene, S., Wang, X.-J., Farisco, M., & Changeux, J.-P. (2025). GNW theoretical framework and the “adversarial testing of global neuronal workspace and integrated information theories of consciousness.” Neuroscience of Consciousness, 2025(1), niaf037.
Odegaard, B., Knight, R. T., & Lau, H. (2017). Should a few null findings falsify prefrontal theories of conscious perception? Journal of Neuroscience, 37(40), 9593–9602.
Volzhenin, K., Changeux, J.-P., & Dumas, G. (2022). Multilevel development of cognitive abilities in an artificial neural network. Proceedings of the National Academy of Sciences, 119, e2201304119.