Validation of Global Neuronal Workspace Theory (GNWT) predictions using iEEG and MEG data was inconclusive, as no prefrontal iEEG electrodes showed the GNWT-predicted combination of an onset and offset response (BF01 > 3 for all prefrontal electrodes).
The MEG results regarding GNWT predictions were sensitive to parameter choices, and signal leakage from posterior sites could not be ruled out.
The researchers assert that the combination of predictions, tested through highly powered, multimodal studies, places a high bar for either Integrated Information Theory or Global Neuronal Workspace Theory to pass, rendering failures more informative.
Conceptualization of the experimental framework for standardization across sites for the adversarial testing of global neuronal workspace and integrated information theory was conducted by A.L., K.B., R.H., L. Mudrik and L. Melloni.
The adversarial study of integrated information theory (IIT) and global neuronal workspace theory (GNWT) involved 256 human participants who viewed suprathreshold stimuli for variable durations while researchers measured neural activity using functional magnetic resonance imaging, magnetoencephalography, and intracranial electroencephalography.
The article 'Disorders of consciousness diagnosis, interventions, and prognostication for the intensivist: Report of the 2025 ISICEM roundtable' published in Intensive Care Medicine in 2026 cites the Nature article 'Adversarial testing of global neuronal workspace and integrated information theories of consciousness'.
The experimental results for Global Neuronal Workspace Theory (GNWT) consisted of a partly challenged prediction regarding an all-or-none threshold and amplification of information updating the content of consciousness in the prefrontal cortex (PFC), and a partly supported prediction given the inconclusive result for orientation regarding global broadcasting of information in the prefrontal cortex.
The preregistered phase-phase coupling (PPC) metric results for prediction 3, which is critical for both Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT), supported neither theory.
According to Global Neuronal Workspace Theory (GNWT), the theory would be challenged if transient prefrontal activation at stimulus onset and offset was not observed.
The study defined the winning model using the Bayesian Information Criterion (BIC) and computed Bayes factors based on the difference in BIC values to compare the Global Neuronal Workspace Theory (GNWT) model against either the null model or the time-window model.
The authors argue that the lack of ignition at stimulus offset in Global Neuronal Workspace Theory (GNWT) is surprising because the change in conscious experience at the onset of a blank fixation screen represents a clear update to the content of consciousness that the global workspace should have represented.
To test Global Neuronal Workspace Theory (GNWT) and Integrated Information Theory (IIT) decoding predictions, researchers decoded stimulus categories (faces versus objects and letters versus false fonts) separately for task-relevant and task-irrelevant conditions, and decoded orientation (front view versus left view versus right view) on combined data from both task conditions.
The article 'Fragmentation and multithreading of experience in the default-mode network' published in Nature Communications in 2025 cites the Nature article 'Adversarial testing of global neuronal workspace and integrated information theories of consciousness'.
Both Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) predict that conscious content should be evident in theory-relevant brain areas independently of other cognitive processes, such as report and task manipulations.
Interareal connectivity assessment for Integrated Information Theory (IIT) regions of interest involved 4 patients, while assessment for Global Neuronal Workspace Theory (GNWT) regions of interest involved 21 patients.
Regarding the maintenance of conscious percepts over time, Integrated Information Theory (IIT) predicts that conscious content is actively maintained by neural activity in the posterior ‘hot zone’ throughout the duration of a conscious experience, whereas Global Neuronal Workspace Theory (GNWT) predicts ignition events in the prefrontal cortex (PFC) at stimulus onset and offset, with activity-silent information maintenance in between.
Researchers evaluated theoretical predictions of Global Neuronal Workspace Theory (GNWT) and Integrated Information Theory (IIT) by subsampling observed cross-temporal representational matrices in four time windows: 0.3–0.5 s, 0.8–1.0 s, 1.3–1.5 s, and 1.8–2.0 s.
Stanislas Dehaene served as the proponent for the Global Neuronal Workspace Theory (GNWT) in the adversarial collaboration.
An exploratory decoding analysis identified only one electrode in the inferior frontal sulcus that showed the temporal pattern predicted by Global Neuronal Workspace Theory (GNWT), though the transient responses occurred earlier than expected at 0.15 seconds post-onset and post-offset.
Global Neuronal Workspace Theory (GNWT) posits that prefrontal cortex workspace neurons broadcast information but do not add information to the conscious experience.
Regarding interareal connectivity during conscious perception, Integrated Information Theory (IIT) predicts sustained short-range connectivity within the posterior cortex, linking low-level sensory areas like V1/V2 with high-level category-selective areas, whereas Global Neuronal Workspace Theory (GNWT) predicts long-range connectivity between high-level category-selective areas and the prefrontal cortex (PFC).
The Integrated Information Theory (IIT) model significantly correlated with the observed Representational Similarity Analysis (RSA) matrix, providing a better fit to the data than the Global Neuronal Workspace Theory (GNWT) model.
The research group tested three preregistered, peer-reviewed predictions of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) regarding how the brain enables conscious experience.
Exploratory dynamic functional connectivity (DFC) results in iEEG were consistent with Global Neuronal Workspace Theory (GNWT) predictions while challenging Integrated Information Theory (IIT) predictions, as connectivity with V1/V2 was not sustained.
Focusing on the presence of fully attended, consciously experienced stimuli is beneficial for testing the primary positive predictions of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) because failures to confirm predictions are harder to dismiss as weak signals.
Bayesian analysis (BF01 > 3) provided evidence for an intercept-only or time-varying amplitude model over the Global Neuronal Workspace Theory (GNWT) model for all electrodes in prefrontal cortex (PFC) regions of interest.
The study found no content-selective phase-phase coupling (PPC) between face-selective and object-selective electrodes and prefrontal cortex (PFC) electrodes in the relevant time window, which contradicts the prediction of Global Neuronal Workspace Theory (GNWT) (BF01 = 2.62–5.32).
Three control analyses—using feature selection, modified time-windows, and decoding analysis time-locked to stimulus offset—provided additional evidence for the absence of Global Neuronal Workspace Theory (GNWT)-predicted patterns in prefrontal cortex (PFC) regions of interest.
Prior to the study published in Nature, predictions from Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) were typically tested using only one data modality at a time, which allowed negative results to be attributed to the limitations of that specific modality.
Global Neuronal Workspace Theory (GNWT) predicts activation in the 0.3–0.5-s window, while Integrated Information Theory (IIT) predicts activation in the 0.3–0.5 s, 0.5–0.8 s, and 1.3–1.5 s time windows.
In prefrontal cortex (PFC) regions of interest, cross-temporal representational similarity analysis (RSA) revealed transient face–object categorical representation at stimulus onset, but not at stimulus offset, resulting in no significant correlation with the Global Neuronal Workspace Theory (GNWT) onset and offset model.
Global Neuronal Workspace Theory (GNWT) predicts brief, content-specific ignition in the prefrontal cortex within 0.3–0.5 seconds after stimulus onset, followed by a decay back to baseline where information is maintained in a latent state until another ignition marks the offset of the current percept and the onset of a new one.
To assess decoding accuracy within unique regions of interest (ROIs), the researchers trained separate classifiers using all electrodes in a given parcel, using time windows of 0.3–0.5 seconds for Global Neuronal Workspace Theory (GNWT) and 0.3–1.5 seconds for Integrated Information Theory (IIT).
The article 'There can be more to consciousness research than theory testing' published in Communications Psychology in 2025 cites the Nature article 'Adversarial testing of global neuronal workspace and integrated information theories of consciousness'.
The failure to confirm the Global Neuronal Workspace Theory (GNWT) prediction regarding the prefrontal cortex (PFC) challenges both GNWT and higher-order theories of consciousness that rely on the PFC to supply the content of visual consciousness.
The fusiform face area showed content-selective connectivity (face stimuli greater than object stimuli) with V1/V2, the inferior frontal gyrus, and the intraparietal sulcus, which is consistent with predictions from both Global Neuronal Workspace theory and Integrated Information Theory.
Researchers used an interaction term between stimulus category (faces, objects, letters, and false fonts) and the predictors of the Global Neuronal Workspace Theory (GNWT) and Integrated Information Theory (IIT) to account for brain regions showing selective responses to specific categories.
Global Neuronal Workspace Theory (GNWT) faces a challenge regarding the maintenance of a conscious percept over time, specifically due to the lack of ignition at stimulus offset observed in the study.
The researchers aimed to challenge and potentially falsify Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) by examining where their predictions differ, rather than attempting to discover the neural correlates of consciousness.
The leaders of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) mutually agreed upon the study's methodological approach before data collection and results disclosure, designating it as the most powerful and conclusive method for testing their theories.
The study focused on testing the competing biological implementations of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) by targeting brain regions where their predictions diverge most notably: the posterior cortex for IIT and the prefrontal cortex (PFC) for GNWT.
In posterior cortex regions of interest, cross-temporal representational similarity analysis (RSA) revealed sustained face–object categorical representation, which matched the Integrated Information Theory (IIT) model better than the Global Neuronal Workspace Theory (GNWT) model.
For MEG analysis, the researchers extracted reconstructed source-level data within predefined anatomical regions of interest (ROIs) based on the theories: GNWT ROIs included 'G_and_S_cingul-Ant', 'G_and_S_cingul-Mid-Ant', 'G_and_S_cingul-Mid-Post', 'G_front_middle', 'S_front_inf', and 'S_front_sup'; IIT ROIs included 'G_cuneus', 'G_oc-temp_lat-fusifor', 'G_oc-temp_med-Lingual', 'Pole_occipital', 'S_calcarine', and 'S_oc_sup_and_transversal'.
None of the 655 prefrontal cortex (PFC) electrodes measured the temporal profile of onset and offset predicted by Global Neuronal Workspace Theory (GNWT).
In the study 'Adversarial testing of global neuronal workspace and integrated information theory', MEG cross-task decoding of stimulus categories (letters versus falsefonts) shows significantly above-chance (50%) decoding when classifiers are trained on relevant stimuli and tested on irrelevant stimuli (purple) or vice versa (orange), specifically within the posterior and prefrontal regions of interest.
Integrated Information Theory (IIT) predicts that conscious content is maximal in posterior brain areas, whereas Global Neuronal Workspace Theory (GNWT) predicts a necessary role for the prefrontal cortex (PFC).
Significant connectivity was observed between the prefrontal cortex (PFC) and both face-selective and object-selective areas in the gamma frequency band within the time window predicted by Global Neuronal Workspace Theory (GNWT).
Global Neuronal Workspace Theory (GNWT) faces a challenge regarding the representation of conscious experience contents, as the study found no representation of identity in the prefrontal cortex (PFC) and only limited representation of orientation in MEG, despite these dimensions being part of the participants' conscious experience.
Integrated Information Theory (IIT) predicts a sustained representation of conscious content, while Global Neuronal Workspace Theory (GNWT) predicts a phasic (onset and offset) representation.
Extended Data Fig. 7 provides an overview of theoretical predictions, experimental outcomes, and interpretations, including preregistered predictions of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT).
Global Neuronal Workspace Theory (GNWT) predicts that neural activation (ignition) occurs following stimulus offset (0.3–0.5 s) independent of stimulus duration, with virtually no response in between.
iEEG decoding was performed on the high-gamma signal, averaged over non-overlapping windows of 0.02 seconds for electrodes located in the Global Neuronal Workspace (GNWT) and Integrated Information Theory (IIT) regions of interest (ROIs).
The study questions whether the prefrontal cortex (PFC) is involved in broadcasting all conscious content as predicted by Global Neuronal Workspace Theory (GNWT), or only a subset such as abstract concepts and categories, suggesting the role of the PFC in consciousness might need to be redefined.
Across modalities, face–object decoding occurred in both posterior and prefrontal regions of interest, which is consistent with predictions from both Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT).
The article 'A methodological guideline for consciousness assessment via neural electrophysiological activity' published in Military Medical Research in 2025 cites the Nature article 'Adversarial testing of global neuronal workspace and integrated information theories of consciousness'.
The experimental design of the study on global neuronal workspace and integrated information theory used task relevance to distinguish between the effects of task demands and conscious perception.
Global Neuronal Workspace Theory (GNWT) posits that conscious content can be decoded from prefrontal cortex (PFC) activity.
The article 'Make science more collegial: why the time for ‘adversarial collaboration’ has come' published in Nature in 2025 cites the Nature article 'Adversarial testing of global neuronal workspace and integrated information theories of consciousness'.