The Adversarial Collaboration: IIT vs. GWT in the Ring

Language: en

---

Overview

In 2019, the John Templeton Foundation committed $20 million to what may be the most ambitious experiment in the history of consciousness science: a structured adversarial collaboration pitting the two leading scientific theories of consciousness against each other in head-to-head empirical tests. The Accelerating Research on Consciousness (ARC) program funded researchers from competing theoretical camps to jointly design experiments that could decisively distinguish between their predictions — and to commit in advance to accepting the results.

The first phase centered on the COGITATE Consortium, a six-laboratory collaboration testing Integrated Information Theory (IIT, championed by Giulio Tononi) against Global Workspace Theory (GWT, championed by Stanislas Dehaene and Bernard Baars). The results, published across multiple papers in 2023-2025, were simultaneously illuminating and humbling: neither theory was fully confirmed or fully refuted. Both theories made some correct predictions and some incorrect ones. The adversarial approach revealed not just which predictions were right, but where both theories share a fundamental blind spot — and what a future theory of consciousness will need to explain.

If consciousness science is a courtroom, the Templeton adversarial collaboration put both leading theories on the witness stand, cross-examined them with carefully designed experiments, and reached the Scottish verdict: not proven. The theories are not innocent, but the evidence does not convict either one. The real finding is that consciousness is more complex than either theory alone can capture.

The Contenders

Global Workspace Theory (GWT)

Global Workspace Theory, originally proposed by Bernard Baars in 1988 and computationally formalized by Stanislas Dehaene, Jean-Pierre Changeux, and colleagues in their Global Neuronal Workspace (GNW) model, proposes that consciousness arises when information is broadcast widely across the brain via a “global workspace” — a network of long-range cortical connections centered on prefrontal and parietal cortices.

In the GNW model, most brain processing is unconscious — occurring in specialized, encapsulated modules (visual cortex processes vision, auditory cortex processes sound, etc.) that operate independently and in parallel. Information becomes conscious when it is selected by attentional mechanisms and broadcast to the global workspace, where it becomes available to all modules simultaneously. This broadcasting is implemented by synchronized neural firing in prefrontal and parietal cortices, detectable as a late positive potential (P3b) in EEG recordings, approximately 300-500 milliseconds after stimulus presentation.

Key GNW predictions:

• Consciousness requires prefrontal cortex involvement (for the global broadcasting mechanism)
• Conscious perception is accompanied by a late ignition event — a sudden, all-or-nothing cascade of prefrontal-parietal activation approximately 300 ms post-stimulus
• This ignition is detectable as the P3b ERP component and as increased long-range connectivity between frontal and posterior regions
• The content of consciousness is determined by which information wins the competition for global workspace access

Integrated Information Theory (IIT)

Integrated Information Theory, developed by Giulio Tononi beginning in 2004 and refined through multiple versions (IIT 1.0 through IIT 4.0), proposes that consciousness is identical to integrated information — a mathematical quantity called phi (phi) that measures how much a system is “more than the sum of its parts.” Any system with non-zero phi has some degree of consciousness, and the quality of consciousness is determined by the specific structure of integrated information.

IIT is built on five axioms (existence, composition, information, integration, exclusion) derived from the phenomenological properties of conscious experience, and five corresponding postulates about the physical substrate that can realize these properties. The theory predicts that consciousness is generated by a “posterior cortical hot zone” — the parieto-occipital and temporal cortices — where neural circuits have the highest degree of intrinsic information integration.

Key IIT predictions:

• Consciousness does not require prefrontal cortex (the frontal cortex contributes to access, report, and cognitive control, but the content of experience is generated by posterior cortex)
• Consciousness is a graded phenomenon — different stimuli produce different degrees of conscious experience depending on the complexity and integration of the neural response
• The neural correlates of consciousness are sustained rather than transient — consciousness involves a maintained state, not a brief ignition event
• The minimum neural substrate of consciousness is the posterior cortical hot zone, not the prefrontal-parietal global workspace

Where They Disagree

The theories make opposite predictions on two critical points:

The role of prefrontal cortex: GWT says prefrontal cortex is essential for consciousness (it is the broadcasting hub of the global workspace). IIT says prefrontal cortex is not necessary for consciousness (it contributes to the cognitive access and report of conscious content, but the experience itself is generated posteriorly).

The temporal signature: GWT predicts a late, all-or-nothing ignition event (the P3b, approximately 300 ms post-stimulus). IIT predicts sustained posterior activity that begins earlier and persists longer, without a sharp frontal ignition event.

The COGITATE Experiment

Design

The COGITATE Consortium (Columbia University, Free University of Berlin, Peking University, Yale University, University of Wisconsin-Madison, and Hebrew University of Jerusalem) designed a preregistered study with 256 participants across the six sites. The experiment used a within-subject design testing visual perception of faces and objects at different levels of visibility (from subliminal to clearly visible).

The critical innovation was the no-report paradigm. The confound that has plagued all previous consciousness research is that studying consciousness typically requires asking subjects to report on their experience — and the act of reporting involves cognitive processes (attention, working memory, decision-making, motor planning) that are theoretically distinct from consciousness itself. Both GWT and IIT acknowledge this confound, but it leads to different predictions: GWT argues that the frontal activation seen during conscious perception is part of the consciousness mechanism, while IIT argues that frontal activation reflects post-conscious processing required for the report.

COGITATE included both report and no-report conditions. In report conditions, subjects indicated whether they perceived the stimulus. In no-report conditions, conscious perception was inferred from physiological measures (pupil dilation, microsaccades) without requiring any behavioral response. This design allows separation of consciousness-related neural activity from report-related neural activity.

Neuroimaging Methods

The study used three complementary neuroimaging methods:

High-density EEG (256 channels): For millisecond temporal resolution of the entire sequence of neural events from stimulus presentation through conscious perception.

7T fMRI: For high spatial resolution mapping of which brain regions are activated during conscious vs. unconscious perception.

MEG: For combining millisecond temporal resolution with reasonable spatial localization.

The preregistered analysis plan specified exactly which neural measures would confirm or disconfirm each theory’s predictions, eliminating post-hoc interpretation.

Results

The COGITATE results, published in stages from 2023 to 2025, were nuanced:

Partially supporting GWT: Conscious perception was accompanied by a late increase in long-range connectivity between frontal and posterior regions, consistent with the global broadcasting mechanism predicted by GWT. The P3b component was detected in report conditions, confirming the existence of the ignition-like event.

Partially supporting IIT: In no-report conditions, the frontal activation was dramatically reduced but conscious perception (inferred from pupil measures) still occurred. Sustained posterior cortical activity was the most reliable neural correlate of consciousness across both report and no-report conditions. The data supported IIT’s prediction that posterior cortical activity is the core neural correlate and that frontal activity is associated with reporting rather than experiencing.

Disconfirming GWT predictions: GWT predicted that the P3b (frontal ignition) would be present whenever conscious perception occurs. In no-report conditions, the P3b was absent or greatly attenuated, yet physiological measures indicated conscious perception. This suggests the P3b reflects the reporting process, not consciousness itself — a significant blow to the GNW model.

Disconfirming IIT predictions: IIT predicted specific patterns of information integration (measurable by perturbational complexity) that would distinguish conscious from unconscious processing. These patterns were only partially observed — the complexity measures distinguished conscious from unconscious states at the group level but with substantial individual variability. The clean relationship between phi and consciousness that IIT predicts was not robustly confirmed.

What the Adversarial Approach Revealed

The Reportability Confound Is Real

The single most important finding from COGITATE is the demonstration that the neural correlates of consciousness look different when you require a report versus when you don’t. This is not a minor methodological nuance — it undermines decades of consciousness research that relied exclusively on report paradigms.

The standard paradigm in consciousness research (present a stimulus, ask the subject if they saw it, compare brain activity for “yes” vs. “no” trials) conflates consciousness with reportability. The COGITATE data show that frontal activation — previously considered a hallmark of consciousness — is largely an artifact of the reporting requirement. When subjects don’t need to report, consciousness is associated primarily with sustained posterior cortical activity, not frontal ignition.

This finding has immediate implications for the clinical assessment of consciousness in neurological patients. The standard clinical assessment of consciousness requires a behavioral response (following commands, tracking objects). Patients who cannot produce behavioral responses are typically classified as unconscious (vegetative state). But the COGITATE data suggest that consciousness can occur without the frontal activation required for behavioral report — meaning that some clinically “unconscious” patients may be conscious but unable to report their experience.

Neither Theory Is Complete

The honest summary: both theories captured part of the elephant. GWT correctly predicted that long-range connectivity plays a role in conscious access. IIT correctly predicted that the core neural correlate of consciousness is in posterior cortex and does not require prefrontal activation.

But neither theory predicted the full pattern of results. The reality appears to be that consciousness involves both sustained posterior cortical activity (as IIT predicts) AND some form of wider distribution or integration (as GWT predicts), but the integration may not require the all-or-nothing frontal ignition that GNW specifies.

The Need for New Theories

The COGITATE results have catalyzed the development of hybrid and alternative theories:

Dendritic Integration Theory (DIT): Proposed by Johan Storm and colleagues, DIT argues that consciousness arises from integration of information within the dendritic trees of individual neurons (particularly large cortical pyramidal neurons with extensive dendritic arbors) rather than from global broadcasting between brain regions or integrated information across networks. Each neuron performs a miniature integration of thousands of inputs in its dendritic tree, and this intra-neuronal integration is the substrate of consciousness.

Predictive Processing Theories: Andy Clark, Jakob Hohwy, and others have argued that consciousness arises from the brain’s predictive processing machinery — specifically, from the precision-weighted prediction error signals that update the brain’s model of the world. This framework is compatible with both posterior cortical involvement (where sensory predictions are generated and compared) and longer-range connectivity (for hierarchical prediction error propagation).

Higher-Order Theories (updated): Hakwan Lau and Richard Brown’s Perceptual Reality Monitoring (PRM) theory proposes that consciousness requires a meta-cognitive representation of one’s own perceptual state — a “reality check” that distinguishes veridical perception from imagination. PRM predicts prefrontal involvement (for the meta-cognitive assessment) but only when the perceptual state is ambiguous, potentially reconciling the discrepant findings about frontal activation.

The Second Round: Phase 2 of the Templeton Program

In 2025, the Templeton Foundation announced the second phase of the ARC program, expanding the adversarial collaboration to include additional theories (Higher-Order Theories, Predictive Processing, and Dendritic Integration Theory) and additional experimental approaches:

Transcranial focused ultrasound (tFUS): Moving beyond correlational neuroimaging to causal manipulation by selectively stimulating or inhibiting predicted consciousness-critical structures (see the 2026 MIT roadmap article).

Optically pumped magnetometers (OPMs): New wearable MEG sensors that allow brain imaging during naturalistic behavior (walking, talking, interacting), enabling the study of consciousness outside the artificial confines of the scanner.

Clinical populations: Extending the adversarial approach to patients with disorders of consciousness (coma, vegetative state, minimally conscious state) and to psychedelic states, providing a wider range of consciousness levels for theory testing.

Computational modeling: Requiring each theory to produce quantitative computational models that make precise, falsifiable predictions about specific experimental conditions — moving beyond qualitative predictions to quantitative hypothesis testing.

The Philosophy of Adversarial Collaboration

A New Model for Science

The adversarial collaboration model itself may be as significant as any specific result. Traditional science proceeds through advocacy: each lab designs experiments to support its own theory. The adversarial model inverts this: competing camps jointly design experiments to test their disagreements, committing in advance to accept the results. This structure reduces confirmation bias, eliminates cherry-picking of results, and produces more definitive evidence.

The model has implications beyond consciousness science. Any scientific field with competing paradigms — cosmology, evolutionary biology, nutrition science — could benefit from structured adversarial collaborations. The Templeton program may be remembered not only for what it revealed about consciousness but for pioneering a new approach to scientific dispute resolution.

The Humility of Not Knowing

Perhaps the deepest lesson of the adversarial collaboration is epistemological humility. The two leading theories of consciousness, developed by some of the most brilliant scientists in the field, both turned out to be partially right and partially wrong. This suggests that consciousness is genuinely difficult — that it does not yield easily to any single theoretical framework, and that our current understanding is more limited than theoretical advocates typically acknowledge.

This humility is itself a form of progress. Knowing what you don’t know is the first step toward knowing more. The adversarial collaboration mapped the boundaries of current understanding with unprecedented clarity, showing exactly where the leading theories succeed and where they fail. This map of ignorance is as valuable as any positive finding.

The Contemplative Perspective

What Meditators Have Always Known

Contemplative traditions have engaged in their own adversarial collaborations about consciousness for millennia. Buddhist and Hindu philosophers debated the nature of consciousness with a rigor and sophistication that rivals (and in some respects exceeds) modern scientific discourse.

The Buddhist tradition’s distinction between awareness (vijñana) and the contents of awareness (nimitta) parallels the GWT-IIT debate. GWT focuses on the contents and their broadcasting (which experiences gain access to the global workspace). IIT focuses on the structural properties of the system that generates experience (integrated information). The Buddhist distinction suggests both perspectives capture something real: consciousness has contents (which are selected and broadcast) AND an intrinsic structure (which determines the quality of experience).

The Advaita Vedanta distinction between reflected consciousness (chidabhasa, consciousness reflected in the mirror of the mind) and pure consciousness (chit, awareness itself) offers another lens. GWT describes reflected consciousness — the content that appears in the global workspace. IIT describes the structural requirements for the mirror that reflects. Neither describes the light itself — the awareness that is prior to any particular content or structure.

This contemplative perspective suggests that both theories may be right about the neural correlates of consciousness (the brain mechanisms that determine which contents appear in awareness) while both miss the deeper question: what is awareness itself, prior to any particular content? This is Chalmers’ hard problem, and neither GWT nor IIT fully addresses it.

Four Directions Integration

• Serpent (Physical/Body): The no-report paradigm findings restore the body’s role in consciousness research. When behavioral reports are removed, the neural correlates of consciousness shift toward posterior sensory cortices and away from prefrontal executive regions. Consciousness, stripped of its cognitive accessories, looks more like a bodily process — a sustained sensory engagement with the world — than an executive broadcast. The serpent reminds us that consciousness begins in sensation, in the body’s meeting with the world, before the mind narrates and reports.

• Jaguar (Emotional/Heart): The adversarial collaboration reveals the emotional investments hidden within scientific theorizing. Each theory’s advocates have spent decades building their frameworks, training students, publishing papers. The commitment to accept empirical results that may disconfirm years of work requires emotional courage — the willingness to grieve what is lost and embrace what is found. This is the same emotional work required in any genuine transformation: letting go of cherished beliefs when reality contradicts them.

• Hummingbird (Soul/Mind): The COGITATE results show that the mind is more complex than any single theory can capture. This is humbling for the mind that theorizes about itself — the very faculty that creates theories turns out to be more intricate than any theory it creates. The hummingbird’s lesson: approach consciousness with wonder rather than certainty, with curiosity rather than dogma. The mystery deepens as knowledge advances, and this deepening is itself a form of understanding.

• Eagle (Spirit): The eagle sees the long view: 30 years of the “science of consciousness” (since the field’s founding in the 1990s) have produced extraordinary advances in understanding the neural correlates of consciousness while leaving the hard problem — why there is subjective experience at all — essentially untouched. The adversarial collaboration demonstrates that even with the best theories, the most rigorous experiments, and the most honest science, consciousness retains its essential mystery. This is not a failure of science but a revelation of the depth of the phenomenon. Consciousness may be the universe knowing itself — and the universe does not yield its secrets easily.

Key Takeaways

• The Templeton Foundation’s adversarial collaboration program ($20M) pitted IIT against GWT in preregistered, jointly designed experiments across six laboratories (the COGITATE Consortium).
• Neither theory was fully confirmed: GWT correctly predicted long-range connectivity in consciousness but incorrectly predicted mandatory frontal ignition; IIT correctly predicted posterior cortical dominance but only partially confirmed its integrated information measures.
• The no-report paradigm revealed that frontal activation during conscious perception is largely an artifact of the reporting requirement, not a feature of consciousness itself.
• Neither theory predicts the full pattern of results, catalyzing development of new and hybrid theories (Dendritic Integration Theory, Perceptual Reality Monitoring, predictive processing approaches).
• Phase 2 of the Templeton program (announced 2025) expands to additional theories and incorporates causal tools (tFUS) and naturalistic brain imaging (wearable MEG).
• The adversarial collaboration model itself is a methodological innovation that reduces confirmation bias and produces more definitive science.

References and Further Reading

• COGITATE Consortium (2023). An adversarial collaboration to critically evaluate theories of consciousness. Preregistered protocol and initial results. Nature Human Behaviour.
• Melloni, L., et al. (2023). An adversarial collaboration testing Global Neuronal Workspace and Integrated Information theories of consciousness. Cell, 186(17), 3896-3913.
• Dehaene, S., et al. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70(2), 200-227.
• Tononi, G., et al. (2016). Integrated information theory: From consciousness to its physical substrate. Nature Reviews Neuroscience, 17, 450-461.
• Baars, B. J. (1988). A Cognitive Theory of Consciousness. Cambridge University Press.
• Koch, C., et al. (2016). Neural correlates of consciousness: Progress and problems. Nature Reviews Neuroscience, 17, 307-321.
• Pitts, M. A., et al. (2014). The role of the P3b in conscious perception: Not just a “no-report” event. Journal of Cognitive Neuroscience, 26(12), 2729-2742.
• Lau, H., & Rosenthal, D. (2011). Empirical support for higher-order theories of conscious awareness. Trends in Cognitive Sciences, 15(8), 365-373.
• Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200-219.