A central challenge in consciousness research is identifying the neural correlates of conscious perception independent of behavioral reports or decision-making processes. Previous studies, primarily focusing on visual modalities, have reported a qualitative “bifurcation” in neural activity around the perceptual threshold, where late sustained activity is associated with conscious reports. However, several limitations exist: the findings haven't generalized across sensory modalities (like audition); complex stimuli (words, faces) and stimulation procedures might introduce non-linearities; the act of reporting itself might confound the neural response; and analyses have been largely descriptive, potentially mistaking non-linear for bifurcation dynamics. This study aims to overcome these limitations by investigating auditory perception using a model-based approach that leverages inter-trial variability to distinguish between different types of brain dynamics, including linear, non-linear unimodal, and bifurcation dynamics. This approach allows the researchers to investigate whether bifurcation dynamics represent a general signature of the transition between non-conscious and conscious perception, irrespective of the need for a behavioral response.

Existing research on neural correlates of consciousness primarily contrasts neural activity for the same stimulus when consciously perceived versus not. This approach has revealed clear changes in neural activity associated with conscious perception, including increased activity in sensory regions, wider network involvement, increased global functional connectivity, and late sustained activity. However, these studies relied on behavioral reports, potentially confounding conscious processing with decision-making or motor preparation. Studies using attentional blink and visual masking paradigms identified sharp neural changes separating conscious from non-conscious processing on a trial-by-trial basis, specifically observing an all-or-none pattern in late evoked potentials associated with conscious reports. These findings point to a qualitative “bifurcation” in neural processes. However, these studies lack generality across sensory modalities, used complex stimuli, and were vulnerable to reporting-related confounds. This research aims to address these gaps.

The study employed EEG to record brain activity in 20 healthy adults during both active and passive listening sessions. In the active sessions, participants performed a simple auditory detection task, identifying and rating the audibility of vowels embedded in noise at various signal-to-noise ratios (SNRs) around the behavioral threshold. In the passive sessions, participants listened to the same sounds without a specific task, instead performing various unrelated visual, amodal, or mind-wandering tasks. The order of sessions was counterbalanced. A computational framework compared three models of brain dynamics: unimodal linear, unimodal non-linear, and bifurcation. The key distinction lies in inter-trial variability; only the bifurcation model predicts a non-monotonic profile with a peak in variability around the perceptual threshold. Data analysis included event-related potentials (ERPs), multivariate pattern analysis (MVPA) with temporal generalization, and Bayesian model comparison. MVPA was used to derive a continuous measure of stimulus-related neural activity (projected activity) for each trial, enabling the analysis of both mean and variability of neural activity across trials. The bifurcation model was quantitatively fitted to the trial-by-trial neural data and its predictive power was tested.

Behavioral data from the active sessions showed a non-monotonic profile of audibility variability, with a burst around the detection threshold and bimodal distributions of audibility ratings at threshold SNRs. ERP analysis revealed classic auditory potentials modulated by stimulus strength. Crucially, inter-trial variability in the EEG data displayed a non-monotonic profile after 300 ms, peaking around the threshold SNR, consistent with bifurcation dynamics. MVPA confirmed this pattern, showing a non-monotonic variability profile after 250 ms, correlating with individual behavioral variability profiles. The distribution of neural activity closely matched the predictions of the bifurcation model. A linear mixed-effects model revealed a significant interaction between consciousness report and SNR, supporting the bifurcation model. Bayesian model comparison formally confirmed the superiority of the bifurcation model over unimodal models for the time window of 250-700 ms post-stimulus. The bifurcation model accurately predicted behavioral reports based on neural activity in the active sessions. In the passive sessions, late sustained activations persisted, albeit with a different topography than the P300 observed in the active sessions. Cross-classification analysis indicated that the late activity in the passive sessions reflected core conscious access mechanisms present in the active sessions, supplemented by decision-making mechanisms. Source reconstruction highlighted the role of frontal executive areas in task-related processing, while other brain areas showed sustained activity regardless of the task. Bifurcation dynamics were observed in the passive sessions, correlating with behavioral data from the active sessions. The bifurcation model successfully predicted the content of mind-wandering probes (sound-related thoughts versus other) in the passive sessions. These findings generalized to a control experiment using pure tones instead of vowels.

The findings support the global neuronal workspace theory, specifically highlighting the distinction between non-linear and bifurcation dynamics. The observed non-linear increase in early activity (before 200 ms) suggests preconscious processing, while the subsequent bifurcation dynamics (250-700 ms) mark the transition to conscious access. The results challenge the view that the P300 is a direct signature of conscious access; instead suggesting the P300 reflects decision-making processes closely intertwined with conscious access. The study proposes an updated model, suggesting that covert conscious access is subtended by a “global playground,” a subset of the global workspace, which is expanded to a full global workspace when a task is required. The findings also contribute to the debate about the frontal lobes' role in conscious perception, with some frontal areas potentially involved even in task-free conscious access.

This study demonstrates that bifurcation dynamics in brain activity, characterized by a burst of inter-trial variability around the perceptual threshold, provide a robust neural signature of conscious access independent of report. This signature persists even in the absence of an explicit task, suggesting a fundamental mechanism underlying both overt and covert conscious perception. The findings have implications for understanding the neural basis of consciousness and offer potential for clinical applications in diagnosing conscious processing in non-communicating patients.

The study used a relatively simple auditory paradigm. While a control experiment with pure tones supported the generalizability of the findings, further research is needed to explore the generality across other sensory modalities and more complex stimuli. The EEG source reconstruction is susceptible to misattribution; future studies using fMRI could provide a more precise localization of brain activity. The sample size, while sufficient to detect effects, could be further increased for improved statistical power.