A new paper in Neuroscience of Consciousness puts a long-debated idea to an empirical test: whether the brain's level of consciousness can be quantified from imaging data. Using functional MRI, the authors constructed simple Markov models of large-scale brain networks and calculated Integrated Information Theory's metric, Phi-a formal measure intended to capture the quantity of conscious experience. Their central prediction was straightforward and falsifiable: Phi should decline when consciousness fades, and rise when it returns.
They examined two well-characterized routes into altered consciousness. In a propofol anesthesia dataset (17 healthy adults progressing from wakefulness to mild and deep sedation, then recovery), Phi fell during sedation and rebounded in recovery. The drop was evident at the global level and within theory-guided network collections, including a frontal set and a balanced set combining core executive-control and default-mode components. In a separate sleep dataset (17 adults recorded with simultaneous EEG-fMRI across wake, N1, N2, N3, and REM), the global trend was muted, but posterior systems carried the signal: Phi decreased as sleep deepened, then rose in REM relative to N3. Together, the patterns align with the study's hypothesis that Phi behaves like a quantitative readout of conscious level - but they also show that anesthesia and sleep alter Phi through different network pathways.
Methodologically, the study is unusually transparent. Resting-state activity was parcellated using Yeo's 7- and 17-network atlases. From these networks, the authors assembled five-element systems at three scales - global, intra-network, and inter-network - then binarized each element's time series, built transition probability matrices, and computed Phi with a publicly available IIT 4.0 implementation (PyPhi). The five-element cap reflects a hard computational limit in current IIT toolchains; the team addressed this by selecting elements systematically at the global level and by theory at local levels, testing frontal "workspace-like" combinations alongside posterior sets emphasized by other consciousness theories.
The results reveal a meaningful dissociation. Under anesthesia, Phi changes tracked sedation in broad, frontal/global configurations, consistent with independent evidence that propofol preferentially disrupts associative and executive networks. During natural sleep, shifts in Phi were more localized and posterior, with an occipital/parietal set showing the clearest sleep-stage dependence and a notable REM > N3 increase. That topography echoes classic findings of visual-system engagement during dreaming and suggests that anesthesia and sleep reduce consciousness via partly different large-scale dynamics.
The authors frame the work as a data-driven test of IIT's core expectation: when conscious level drops, integrated information should drop. Importantly, they show this at a macro-network scale, not only in microcircuits, dovetailing with theoretical arguments for causal emergence -the idea that system-level organization can carry meaningful causal power beyond its parts. They also situate the findings among other frameworks, including the global neuronal workspace (fronto-parietal broadcasting) and recurrent processing accounts (local sensory recurrency), noting that their posterior sleep results are compatible with the view that sensory regions contribute to conscious content when sufficiently integrated.
Just as important are the limits. Computing Phi scales exponentially with the number of elements, which forced the analysis to five-node systems and discrete time-series binarization. That choice, while pragmatic, means the brain's continuous dynamics are approximated, and subtle effects may be averaged away at coarse granularity. The authors call for algorithmic advances to handle larger systems and for complementary "continuous" formulations of integrated information that may capture richer dynamics. They also underscore that anesthesia and sleep are not interchangeable: despite shared reductions in Phi, the spatiotemporal signatures differ, pointing to distinct mechanisms of unconsciousness.
Still, as an empirical probe of a mathematically precise theory, the study lands a clear message: Phi can be estimated from human fMRI and behaves as a state-sensitive metric across two very different contexts-pharmacologic sedation and natural sleep. For clinicians, that does not translate to a bedside monitor yet. But for researchers, it's a concrete step toward connecting formal consciousness theory with measurable brain activity, and toward mapping where and at what scale integrated information shifts when consciousness dims and returns.
If future work extends these methods to larger network sets, more diverse populations, and continuous formulations of Phi, we may learn not only that consciousness has a measurable level, but how different brain architectures maintain it - and how various routes to unconsciousness perturb it in distinct, clinically relevant ways.
