Theories of Consciousness Smackdown: IIT vs GNWT
Related Jim Rutt Show podcasts:
— Team IIT: EP105 Christof Koch on Consciousness
— Father of GWT: EP108 Bernard Baars on Consciousness
Consciousness science has long been rich in hypotheses but poor in experimental results, apart from accumulating numerous “correlates of consciousness” data, which have done little to distinguish among competing theories.
A new paper, Adversarial Testing of Global Neuronal Workspace and Integrated Information Theories of Consciousness (free pdf available), recently published in Nature, marks a significant step forward by directly testing two leading theories against each other.
The two leading theories of consciousness that were tested against each other are Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT). Both theories attempt to explain how brain activity gives rise to subjective experiences — our sense of awareness and conscious perception. Despite each theory having some empirical support, they’ve seldom been directly juxtaposed within the same experimental framework. This project, conducted by the international Cogitate Consortium, adopted a collaborative and adversarial approach, meaning proponents of each theory worked together alongside neutral teams to clearly define their theories’ predictions and set objective criteria for evaluating them.
Theoretical Background
- Integrated Information Theory (IIT) suggests that consciousness arises from complex interactions in networks of neurons, particularly focusing on posterior (rear) brain areas involved in sensory processing. IIT predicts that conscious experiences correspond to sustained neural interactions within these areas, and notably, does not consider the frontal cortex as necessary for consciousness.
- Global Neuronal Workspace Theory (GNWT), in contrast, proposes that consciousness results from the global broadcasting of information across widely distributed neural networks involving frontal brain regions. According to GNWT, when information enters consciousness, it becomes broadly accessible across the brain, particularly engaging the prefrontal cortex (PFC).
To rigorously test these theories, the consortium carefully designed and preregistered an experiment with precisely defined outcomes that could support or challenge each theory.
Experimental Design and Methodology
The experiment involved 256 human participants who viewed clearly visible visual stimuli — such as faces, objects, letters, and meaningless symbols (“false fonts”). The stimuli were shown for varying durations (0.5, 1.0, and 1.5 seconds) and orientations (front view, left, and right profiles). Participants performed tasks to identify specific targets while their brain activity was measured using three advanced neuroimaging techniques:
- Functional Magnetic Resonance Imaging (fMRI): measures changes in blood flow, capturing spatially precise brain activity.
- Magnetoencephalography (MEG): detects magnetic fields generated by neural activity, offering high temporal resolution.
- Intracranial Electroencephalography (iEEG): directly records electrical activity within the brain, providing both high temporal and spatial detail but is more invasive.
The study took extensive measures to ensure methodological rigor, including large participant numbers, multimodal imaging, standardization across different laboratories, and minimizing bias by involving neutral analysts separate from the theory proponents.
Key Predictions and Results
The researchers made specific, theory-driven predictions about how brain activity should behave according to IIT and GNWT. Three core aspects were tested:
- Representation of Conscious Content: Could brain areas predictably encode visual stimulus categories (faces, objects) or orientation independently of task requirements?
Result: Both IIT and GNWT partially succeeded. Conscious content could indeed be decoded effectively from posterior brain regions (aligned with IIT) and to a lesser extent from frontal regions (supporting GNWT). However, adding frontal cortex data did not consistently enhance decoding accuracy, challenging a key GNWT claim about frontal necessity for consciousness. - Duration and Timing of Neural Activity: Does brain activation closely follow stimulus presentation, and is it sustained throughout the conscious experience?
Result: IIT predictions were partly supported. Posterior brain regions demonstrated sustained activity corresponding closely to stimulus duration. GNWT’s prediction — that there should be a clear neural “ignition” or global activity peak particularly at stimulus offset — was notably not consistently observed, significantly challenging GNWT’s assumptions about temporal dynamics. - Interregional Synchronization: Do specific brain areas synchronize their activity when consciously processing stimuli?
Result: Surprisingly, neither theory’s predictions were fully supported. IIT expected sustained synchronization within posterior regions, which was not reliably observed. GNWT expected strong synchronization involving frontal regions upon conscious perception; results were mixed, revealing some limited frontal-posterior synchronization but not the strong, consistent pattern predicted.
Overall, the results strongly challenge central claims of both theories. IIT is primarily challenged by the lack of expected neural synchronization patterns within the posterior cortex, undermining its claim that specific types of network connectivity directly specify consciousness. GNWT, meanwhile, faces significant difficulty due to the absence of consistent frontal ignition events and weaker than expected frontal representations of conscious information.
Broader Implications and Scientific Contribution
Beyond testing IIT and GNWT specifically, this research is a model of rigorous, transparent, and collaborative science, aimed at objectively advancing theoretical understanding. The findings imply that neither IIT nor GNWT alone fully accounts for the complexity of consciousness, suggesting that hybrid or new theories may need to incorporate insights from both frameworks. Additionally, the researchers call for developing quantitative methods to integrate findings across multiple measurement techniques and experimental paradigms, emphasizing the necessity of precise, predictive models in cognitive neuroscience.