Adversarial testing of global neuronal workspace and integrated information theories of consciousness

Ever wondered how the brain creates consciousness? This major study tests IIT vs GNWT theories of conscious perception using neuroimaging. We dive into the adversarial collaboration using iEEG, MEG & fMRI to pinpoint where awareness arises – posterior or prefrontal cortex? Discover the surprising results challenging leading theories and revealing clues about the neural basis of experience.

Frequently Asked Questions (FAQ)

What were the primary theories of consciousness being tested? GNWT vs. IIT, contrasting a prefrontal “ignition” with posterior integrated information.
What were the key predictions about neural activity?
- IIT: maximal decoding, sustained activity, and prolonged connectivity in posterior areas.
- GNWT: brief PFC ignition for decoding and late-phase PFC↔category-area bursts.
How was conscious content manipulated and measured? Suprathreshold faces, objects, letters, and false fonts shown in varied orientations/durations while iEEG, MEG, and fMRI recorded responses.
Where did decoding primarily occur? Category decoding in both posterior cortex and PFC; orientation decoding mainly in posterior cortex.
Did adding prefrontal data improve decoding? No; including PFC signals did not boost and sometimes reduced decoding accuracy.
What were the findings on maintenance of conscious content? Posterior sites showed sustained duration‑tracking activity; PFC responses were only transient at onset.
How was interareal connectivity assessed, and what were the results? Examined synchrony via iEEG, MEG, and fMRI gPPI; found some posterior and brief PFC-linked patterns but no definitive support for either theory’s precise predictions.
What were the main conclusions about the two theories? Results favor IIT’s posterior-focus on content representation and maintenance; key GNWT predictions about PFC involvement were not supported.

Frequently Asked Questions (FAQ - long form)

What were the primary theories of consciousness being tested in this study? The study focused on testing two prominent theories of consciousness: the Global Neuronal Workspace Theory (GNWT) and the Integrated Information Theory (IIT). These theories offer contrasting predictions about the brain regions and neural dynamics involved in conscious experience. GNWT emphasises the role of a widely distributed network, particularly the prefrontal cortex (PFC), in making information globally available for conscious processing. IIT, on the other hand, posits that consciousness arises from integrated information within specific brain regions, primarily in the posterior cortex.

What were the key predictions made by the two theories regarding neural activity during conscious perception? The study tested three main predictions. Firstly, concerning the decoding of conscious content, IIT predicted that conscious content would be maximally decodable in posterior brain areas, while GNWT highlighted a necessary role for the PFC. Secondly, regarding the maintenance of conscious content, IIT posited sustained activity in the posterior cortex, whereas GNWT predicted brief, transient “ignition” in the PFC at the beginning and end of stimulus presentation, with content stored in a non-conscious “silent” state in between. Thirdly, regarding interareal connectivity, IIT predicted sustained connectivity within the posterior cortex (between high-level and low-level visual areas), while GNWT predicted brief, late-phase connectivity and information sharing between the PFC and category-specific areas.

How was conscious content manipulated and measured in the experiment? To investigate the multifaceted nature of conscious experience, the researchers manipulated several attributes of visual stimuli. Participants were presented with suprathreshold stimuli across four categories (faces, objects, letters, and false fonts), each with unique identities and presented in three orientations (front, left, right). Stimuli were shown for varying durations (0.5s, 1.0s, and 1.5s). Participants were instructed to detect specific targets from predefined categories, making these categories task-relevant, while the other categories were task-irrelevant. Conscious perception of these stimuli was confirmed in a separate memory test. Neural activity related to conscious content was measured using intracranial electroencephalography (iEEG), magnetoencephalography (MEG), and functional magnetic resonance imaging (fMRI).

Where did the decoding of conscious content (like stimulus category and orientation) primarily occur according to the study’s findings? The study found that stimulus category (faces vs. objects) could be decoded in both posterior and prefrontal brain regions across different neuroimaging modalities (iEEG, MEG, and fMRI). This finding was broadly consistent with predictions from both IIT and GNWT. However, when decoding stimulus orientation (left, right, or front views), which was always task-irrelevant, the results were more divergent. Decoding of face orientation was primarily achieved in posterior brain regions, with little to no significant decoding observed in the prefrontal cortex across iEEG and fMRI. MEG showed weaker, yet above-chance, decoding in prefrontal regions, but signal leakage from posterior areas could not be ruled out.

Did adding prefrontal cortex activity improve the decoding of conscious content compared to using only posterior areas? No, the study found strong evidence against the idea that including activity from prefrontal regions improved the decoding accuracy of stimulus category or orientation. When comparing decoders trained solely on posterior regions to those trained on both posterior and prefrontal regions, adding the prefrontal data did not enhance decoding performance and, in some cases, even reduced it. This result was supported by Bayesian analyses across iEEG and MEG data and challenges the idea that the PFC adds unique information for conscious content decoding beyond what is present in posterior areas.

What were the findings regarding the maintenance of conscious content over time in different brain regions? The study investigated how neural activity tracked the duration of stimulus presentation. In posterior brain regions, a subset of electrodes showed sustained activity that correlated with stimulus duration, consistent with the maintenance prediction of IIT. This sustained activity was observed for both non-selective and category-selective responses, particularly in early visual and ventral temporal areas. In contrast, in prefrontal regions, while many electrodes showed transient responses at stimulus onset, none exhibited the sustained or onset-and-offset temporal profile predicted by GNWT. Bayesian analysis provided strong evidence against the GNWT-predicted temporal pattern in the PFC.

How did the study investigate interareal connectivity related to consciousness, and what were the results? Interareal connectivity was assessed by examining synchrony patterns between different brain regions, specifically within the posterior cortex (high-level to low-level visual areas) and between the PFC and category-specific areas. IIT predicted sustained gamma-band connectivity within the posterior cortex, while GNWT predicted brief, late-phase synchrony between the PFC and category-specific areas. Using iEEG and MEG, the study found some evidence of content-selective synchrony. iEEG dynamic functional connectivity (DFC) analysis revealed content-selective synchrony for objects in V1/V2 and for both faces and objects in the PFC. MEG DFC showed low-frequency (below 25 Hz) content-selective synchrony for faces in both V1/V2 and PFC. The fMRI generalized psychophysiological interaction (gPPI) analysis also indicated content-selective connectivity between the fusiform face area (FFA) and areas including V1/V2, right intraparietal sulcus (IPS), and right inferior frontal gyrus (IFG). While some connectivity patterns were observed in predicted regions, the results didn’t definitively and exclusively support the precise temporal and regional predictions of either theory for all tested conditions.

What were the main conclusions drawn from the study’s empirical findings regarding the two theories of consciousness? The study’s findings presented a complex picture, offering some support for aspects of both theories while also challenging key predictions. The decoding of conscious content was observed in both posterior and prefrontal regions, aligning partly with both theories. However, the strong evidence for orientation decoding primarily in posterior areas and the lack of improved decoding when including PFC data were more consistent with IIT’s emphasis on posterior processing for conscious content. The sustained neural activity tracking stimulus duration in posterior regions supported IIT’s prediction of content maintenance in these areas, while the absence of the predicted sustained or onset-offset activity in the PFC challenged a core prediction of GNWT regarding prefrontal involvement in maintenance. The connectivity analyses showed some predicted patterns but were not entirely conclusive in exclusively supporting one theory over the other, highlighting the complexity of interareal communication during conscious perception. Overall, the results suggest that while the PFC may be involved in processes related to tasks and goals, the posterior cortex appears to play a more central role in the representation and maintenance of rich, multidimensional conscious content, aligning more closely with certain aspects of IIT.

BRIEFING DOCUMENT: Adversarial Testing of Consciousness Theories

Date: October 26, 2024 Source: Excerpts from “Adversarial testing of global neuronal workspace and integrated information theories of consciousness.pdf” Authors: Cogitate Consortium, et al. Subject: Empirical testing of Global Neuronal Workspace Theory (GNWT) and Integrated Information Theory (IIT) of consciousness.

1. Executive Summary:

This study employs an adversarial approach using multiple neuroimaging modalities (iEEG, MEG, and fMRI) to empirically test key predictions of two prominent theories of consciousness: Global Neuronal Workspace Theory (GNWT) and Integrated Information Theory (IIT). The research focuses on three main predictions related to conscious visual experience: decoding of conscious content, maintenance of conscious content, and interareal connectivity supporting consciousness. The findings provide mixed support for both theories, highlighting the complexity of neural mechanisms underlying conscious experience and suggesting limitations in the current theoretical frameworks. Notably, the study found strong evidence against a purely prefrontal locus for conscious content decoding and maintenance as predicted by some interpretations of GNWT, while also finding limited support for sustained posterior activity and specific connectivity patterns predicted by IIT.

2. Background:

Understanding the neural basis of consciousness is a major challenge in neuroscience. GNWT and IIT are two influential theories proposing different neural correlates of consciousness.

Global Neuronal Workspace Theory (GNWT): Proposes that conscious content arises when information is broadcast widely across the brain, particularly involving a “workspace” in the prefrontal cortex (PFC) and parietal areas. Predictions often involve transient “ignition” of activity, particularly in PFC, and widespread information sharing.
Integrated Information Theory (IIT): Postulates that consciousness is related to the capacity of a system to integrate information, measured by the value Phi (Φ). While specific neural correlates are complex, IIT generally emphasizes the role of posterior cortical areas (like visual cortex) as the primary substrates of conscious experience due to their high capacity for information integration. Predictions often involve sustained activity in posterior regions and specific patterns of connectivity within these areas.

This study aims to directly compare these theories by generating opposing predictions for specific neural measures and testing them empirically.

3. Research Design and Methodology:

Adversarial Testing: The study is designed around empirically testable, often opposing, predictions derived from GNWT and IIT. (See Figure 1a).
Multimodal Neuroimaging: Data were collected using:
- Intracranial Electroencephalography (iEEG) from epilepsy patients (n=29-31 for main analyses). This offers high spatial and temporal resolution.
- Magnetoencephalography (MEG) from healthy participants (n=65 for main analyses). This provides high temporal resolution and whole-brain coverage.
- Functional Magnetic Resonance Imaging (fMRI) from healthy participants (n=73 for main analyses). This offers high spatial resolution.
Task Design: Participants viewed visual stimuli (faces, objects, letters, false fonts) of varying durations (0.5s, 1.0s, 1.5s). Stimuli were categorized as either task-relevant (targets or from target categories) or task-irrelevant (from other categories). Participants were instructed to detect predefined targets. Crucially, participants were confirmed to be conscious of both task-relevant and task-irrelevant stimuli in a separate experiment (see Section 3 in Supplementary Information). (See Figure 1c, d).
Regions of Interest (ROIs): Analyses focused on pre-defined posterior (aligned with IIT predictions) and prefrontal (aligned with GNWT predictions) brain regions.
Analysis Techniques: Various techniques were employed, including:
- Decoding: Using linear Support Vector Machine (SVM) classifiers to determine if stimulus information (category, orientation) could be predicted from neural activity patterns. Cross-task decoding (training on one task condition and testing on another) was used to assess generalization.
- Linear Mixed Models (LMMs): To analyze neural activity (high gamma, alpha, ERPs) over time and across stimulus durations.
- Representational Similarity Analysis (RSA): To examine how neural representations of stimuli evolve over time.
- Interareal Connectivity Analysis: Using Pairwise Phase Consistency (PPC), Dynamic Functional Connectivity (DFC), and Generalized Psychophysiological Interaction (gPPI) to assess synchrony and functional connections between brain regions.

4. Key Findings and Themes:

The study tested three main predictions:

Prediction 1: Decoding of Conscious Content

IIT Prediction: Maximal decoding of conscious content in posterior brain areas.
GNWT Prediction: Necessary role for the PFC in decoding conscious content.
Findings:
- Decoding of stimulus category (faces-objects) was successful in both posterior and prefrontal ROIs across iEEG, MEG, and fMRI data.
- However, the temporal profile of decoding differed significantly:
  - In posterior ROIs, category decoding was sustained for the approximate duration of the stimulus (more than 95% accuracy in iEEG for face-object decoding). (See Figure 2b, top row).
  - In prefrontal ROIs, category decoding was brief (approximately 0.2–0.4s in iEEG for face-object decoding) and restricted to around stimulus onset/offset. (See Figure 2b, bottom row).
- Decoding of stimulus orientation (a task-irrelevant attribute) was observed primarily in posterior brain areas across all modalities (iEEG, fMRI, MEG). Prefrontal ROIs showed little to no reliable orientation decoding.
  
  “decoding of face orientation (left, right or front views) was achieved in posterior but not in prefrontal ROIs, both with iEEG (Fig. 2f,h, approximately 95% with pseudotrial aggregation; Extended Data Fig. 3a) and the fMRI searchlight approach (Fig. 2g, approximately 45%).”
- Bayesian testing strongly supported the null hypothesis of no face orientation decoding in prefrontal regions for iEEG and fMRI.
- Adding prefrontal ROI activity to posterior ROI activity did not improve, and in some cases reduced, decoding accuracy for both category and orientation across iEEG and MEG.
  
  “Across critical time-resolved methods (iEEG and MEG) and various PFC ROI definitions, adding prefrontal ROIs did not improve—and in some cases reduced—category and orientation decoding…”
- Bayesian testing provided strong evidence against increased decoding accuracy when including PFC ROIs.

Prediction 2: Maintenance of Conscious Content

IIT Prediction: Conscious content is actively maintained in the posterior cortex, reflected in sustained neural activity tracking stimulus duration.
GNWT Prediction: Brief content-specific “ignition” (approximately 0.3–0.5s) in the PFC at stimulus onset and offset, with content stored in a non-conscious silent state between these events.
Findings:
- In posterior ROIs, a small proportion of electrodes (25 out of 657) showed sustained high gamma activity that tracked stimulus duration, consistent with IIT. (See Figure 3b).
- However, this sustained activity was sparse and only 15% of face-selective electrodes showed this pattern. The majority of face-selective electrodes showed transient onset activations.
- In prefrontal ROIs, none of the 655 electrodes measured the temporal profile predicted by GNWT (onset and offset activation). (See Figure 3c).
- Bayesian analysis provided strong evidence for models other than the GNWT-predicted pattern in PFC.
- Cross-temporal RSA analysis in posterior ROIs showed sustained categorical representation (face-object separability), which matched the IIT model better than the GNWT model for 1.5s duration trials. (See Figure 3d).
- No significant sustained representations were found in the prefrontal ROI RSA. (Extended Data Fig. 4).

Prediction 3: Interareal Connectivity Supporting Consciousness

IIT Prediction: Sustained gamma-band connectivity within the posterior cortex (e.g., between high-level visual areas and V1/V2) throughout conscious experience.
GNWT Prediction: Brief, late-phase metastable connectivity (more than 0.25s) with information sharing between the PFC and category-specific areas, manifested in long-range gamma-band or beta-band synchronization.
Findings:
- Using Dynamic Functional Connectivity (DFC) with iEEG, significant content-selective synchrony (higher connectivity for preferred category) was observed:
  - For object-selective electrodes in V1/V2 (consistent with IIT). (See Figure 4a, top row).
  - For both face- and object-selective electrodes in the PFC ROI (consistent with GNWT’s emphasis on PFC). (See Figure 4a, bottom row).
- MEG DFC showed significant content-selective synchrony below 25Hz (lower frequencies than typically associated with gamma-band, often linked to feedforward processing or local circuits) for face-selective areas with both V1/V2 and PFC. (See Figure 4b).
- fMRI gPPI analysis showed significant content-selective connectivity between face-selective areas (FFA seed) and V1/V2, right intraparietal sulcus (IPS), and right inferior frontal gyrus (IFG). (See Figure 4c).
- Overall, neither theory’s specific connectivity predictions were fully and consistently supported across all modalities and regions. For example, IIT predicted sustained gamma-band connectivity within posterior areas, which was not consistently found. GNWT predicted phasic connectivity between category-selective areas and PFC, and while some connectivity was found, its temporal and frequency characteristics varied.

5. Limitations and Caveats:

Electrode Coverage (iEEG): While coverage was noted as “exceptional,” it is inherently variable across patients, necessitating pooling electrodes into “super participants” for some analyses, which might average out individual differences.
Signal Leakage (MEG): The study acknowledges the possibility of signal leakage from posterior to prefrontal regions in MEG data, which could potentially influence the interpretation of prefrontal activity.
Statistical Approaches: The study employed various statistical methods (permutation tests, cluster-based correction, Bayesian analysis), and some results were sensitive to specific parameter choices (e.g., MEG alpha band activity).
Complexity of Conscious Experience: The simplified nature of the visual stimuli compared to real-world experience (Fig 1b example of Mona Lisa) is an inherent limitation in laboratory settings.
Theory Interpretation: The predictions tested are specific interpretations of GNWT and IIT, and the theories are complex and have evolving formulations.

6. Conclusion and Implications:

The study provides valuable empirical data directly comparing key predictions of GNWT and IIT. The findings suggest that neither theory, in their currently tested formulations, fully accounts for the neural data related to conscious visual experience.

The sustained representation of conscious content and the robust decoding of stimulus orientation in posterior brain areas, coupled with the lack of sustained activity and weak/absent orientation decoding in PFC, provide stronger support for the idea that posterior brain regions play a crucial role in representing the content of conscious perception, more aligned with aspects of IIT.
The absence of the predicted onset/offset ignition pattern in prefrontal regions challenges a core temporal prediction of some interpretations of GNWT.
Connectivity findings were complex, showing some support for interareal interactions involving both posterior and prefrontal areas, but not consistently matching the specific frequency bands and temporal dynamics predicted by either theory.

The results underscore the need for more refined and potentially integrated theoretical frameworks of consciousness that can account for the distributed nature of conscious processing, the differential roles of brain regions (posterior vs. prefrontal), and the temporal dynamics of neural activity and connectivity. Future research should continue to employ adversarial collaborations and multimodal approaches to probe the neural basis of consciousness with increasing precision.

Resources & Further Watching

Read the research paper: Adversarial testing of global neuronal workspace and integrated information theories of consciousness (Note: Please verify authors if possible, as ‘xxx’ was used).
Watch Next (Playlist): Psychology

💡 Please don’t forget to like, comment, share, and subscribe!

Youtube Hashtags

#Consciousness #Neuroscience #BrainScience #IIT #GNWT #CognitiveScience #Mind #Awareness #Perception #NeuralCorrelates #PosteriorCortex #PFC

Youtube Keywords

consciousness neuro science brain science iit gnwt neuro imaging fmri meg ieeg brain activity scientific study research adversarial collaboration brain decoding brain connectivity psychology cognitive science mind awareness perception subjective experience neural correlates prefrontal cortex pfc posterior cortex cognivite neuro science philosophy of mind human brain visual perception neuro philosophy

Stay Curious. Stay Informed.

Join the ResearchLounge community to get regular updates on the latest breakthroughs in science and technology, delivered clearly and concisely. Subscribe to our channels and never miss an insight.

Help us grow by sharing our content with colleagues, students, and fellow knowledge-seekers!

Your engagement fuels discovery!

Donate Now