Consciousness from Contemporary Systems Neuroscience (Biological Perspective I)

How Brain Dynamics Shape Conscious Experience

First Lecture (1/5)

The central claim of contemporary systems neuroscience is no longer that consciousness is “somewhere” in the brain but that it is a specific kind of event within the brain: a transient, dynamic, globally coordinated state that emerges when information processed in specialised modules is amplified, sustained, and broadcast across a recurrent thalamocortical network. This is not a return to Cartesian theatre but the recognition that the apparent unity and privacy of experience are achieved by mechanisms that are, in principle, observable, perturbable, and increasingly quantifiable. The past twenty-five years have seen an explosion of tools (single-unit and intracranial recordings in humans, optogenetics, dense EEG, magnetoencephalography, simultaneous fMRI-EEG, closed-loop TMS-EEG, and machine-learning decoding) that have transformed consciousness from a philosophical orphan into one of the most active empirical domains in all of neuroscience.

The foundational observation is that conscious perception is not correlated with the initial sensory volley. Early evoked potentials (P1 at 80–120 ms, N1 at 150–200 ms) are present whether or not a stimulus is consciously seen. What distinguishes seen from unseen stimuli is a late, sustained, non-linear ignition that begins around 250–300 ms, propagates from sensory regions to a frontoparietal network, and is marked by a broad P3b-like wave. This ignition is all-or-none: below a critical threshold of stimulus strength, attention, or task relevance there is local sensory processing but no reportable experience; above threshold the same sensory cortex suddenly participates in a brain-scale reverberatory state. Masking, attentional blink, and binocular rivalry paradigms have established this beyond reasonable doubt.

The global neuronal workspace hypothesis (GNW), developed by Stanislas Dehaene, Jean-Pierre Changeux, and collaborators since the late 1990s, remains the most empirically successful articulation of this idea. Conscious access occurs when a representation wins the competition for amplification via long-distance recurrent connections linking sensory processors to a distributed workspace anchored in prefrontal, parietal, and cingulate cortex. Once ignited, the content becomes available to verbal report, working memory, planning, and episodic encoding. Intracranial recordings in epileptic patients performing the attentional blink task show that stimuli that escape the blink elicit sustained high-gamma power and phase-locking across distant sites; blinked stimuli do not. Lesions or TMS disruption of dorsolateral prefrontal cortex or inferior parietal lobule abolish the P3b and conscious report while leaving early sensory responses intact. Optogenetic silencing of prefrontal neurons in macaques performing a similar task prevents reportability without eliminating the sensory representation itself.

Predictive processing frameworks add a crucial Bayesian dimension. The cortex is a hierarchical generative model that constantly emits top-down predictions about the hidden causes of sensory data. Perception is inference: the winning model is the one that best minimises prediction error while respecting precision weighting. Conscious contents correspond to the perceptual hypotheses that survive error minimisation at the highest levels of the hierarchy. Anesthesia studies provide striking confirmation: propofol, sevoflurane, and xenon preferentially disrupt descending predictions from prefrontal and parietal cortex to temporal and occipital regions, collapsing hierarchical depth. Under ketamine the opposite occurs: bottom-up prediction errors are no longer adequately suppressed, leading to hallucinatory content that nevertheless remains conscious because the workspace is partially preserved. The REBUS (Relaxed Beliefs Under Psychedelics) model of Carhart-Harris and Friston extends this insight to psilocybin and LSD, where the relaxation of high-level priors allows anarchic low-level activity to flood consciousness.

At the cellular level, the most promising mechanism for ignition is apical amplification in layer-5 pyramidal neurons. Matthew Larkum and others have shown that coincident bottom-up sensory input and top-down contextual input at the apical dendrite trigger a calcium spike that propagates to the soma, dramatically boosting axonal output and enabling long-range broadcast. This dendritic coincidence detection is ideally suited to implement the GNW threshold. Ketamine and propofol both block NMDA receptors on apical tufts, preventing the calcium spike and abolishing consciousness. Astrocytic calcium waves and neuromodulatory tone (noradrenergic from locus coeruleus, serotonergic via 5-HT2A receptors, cholinergic from basal forebrain) further gate the gain of this amplification. Loss of neuromodulation, as in coma or deep sleep, collapses the system into local processing.

The posterior hot zone (parietal, temporal, and occipital association cortex) has emerged as the minimal common substrate across waking perception, dreaming, psychedelic states, and minimal consciousness in disorders of consciousness. Francesco Siclari’s high-density EEG sleep studies show that dream experience is predicted by high-frequency power in posterior cortex regardless of sleep stage; low-frequency power predicts dreamless sleep. Lucid dreaming, ketamine dissociation, and psilocybin intoxication all show enhanced posterior gamma synchrony and increased perturbational complexity. Christof Koch and collaborators now speak of a “posterior cortical hot zone” rather than a single seat, acknowledging that different qualia recruit different subregions (colour in V4/VO, space in parietal cortex, bodily awareness in temporoparietal junction, self in posterior cingulate/precuneus) but that the enabling condition is sustained recurrent activity within this broad territory modulated by thalamic and prefrontal input.

Complexity metrics have moved the field toward quasi-objective markers. Giulio Tononi’s perturbational complexity index (PCI), derived from the Lempel-Ziv compressibility of TMS-evoked potentials, distinguishes conscious from unconscious brains with near-perfect accuracy across wakefulness, dreaming, anesthesia, and vegetative states. Values above ≈0.31 correspond to consciousness; below to unconsciousness. Recent multicenter validation (2021–2024) confirms that PCI tracks consciousness independently of behaviour or etiology. Psychedelics push PCI higher than normal waking, consistent with the expanded repertoire of experiencable states. The index is now used clinically to stratify disorders of consciousness and predict recovery.

Single-neuron data in humans sharpen the picture further. Rodrigo Quian Quiroga’s concept cells in medial temporal lobe fire invariantly to a person or object across modalities but only when the patient reports awareness. Subliminal presentations evoke no response. Itzhak Fried’s recordings in presurgical patients show that conscious perception of faces or houses correlates with explicit firing in fusiform or parahippocampal neurons that is absent during unconscious processing of the same stimuli. These findings suggest a distinction between unconscious feature binding and conscious phenomenal binding that occurs only when local activity is amplified into the global workspace.

Dreaming provides a natural dissociation: the brain can generate vivid, multimodal, self-containing experience with almost no external input and severely reduced prefrontal activation. Yet posterior gamma power and PCI are high, and the phenomenology is often more intense than waking. This demonstrates that consciousness does not require behavioural responsiveness, sensory input, or intact executive function; it requires only a critical level of recurrent processing in the posterior hot zone. The fact that REM sleep dreams can be decoded from posterior cortex with above-chance accuracy using machine-learning classifiers identical to those used for waking perception further blurs the boundary between “real” and “hallucinated” experience.

Disorders of consciousness (vegetative state/unresponsive wakefulness syndrome, minimally conscious state) have become the crucial test bed. High-PCI patients who show no behavioural signs of consciousness often recover, while low-PCI patients do not. Command-following detected by fMRI or EEG in apparently vegetative patients (Owen et al., 2006; Monti et al., 2010) reveals preserved islands of consciousness hidden behind motor incapacity. Recent closed-loop paradigms that use real-time decoding to provide feedback and rehabilitation are beginning to restore communication in some locked-in patients.

At the microcircuit level, new theories emphasise the role of cortical layers and cell types. Layer 5 pyramidal neurons with extensive apical dendrites are now seen as the primary broadcasters; layer 2/3 neurons provide local recurrence; inhibitory interneurons (especially parvalbumin-positive) shape the temporal structure of ignition. The balance between excitation and inhibition is critical: too much inhibition (propofol) collapses complexity; too little (ketamine at subanesthetic doses) produces hyper-complex but incoherent states. Recent work on dendritic computation suggests that conscious perception may literally occur in the apical integration zones of layer 5 cells, where contextual, predictive, and sensory signals converge.

Despite these advances, the hard problem remains untouched. Perfect correlation between a specific spatiotemporal pattern of neural activity and a specific conscious content is now within reach for crude categories (face vs. house, seen vs. unseen). Yet the explanatory gap—why this pattern feels like anything from the inside—persists. Integrated Information Theory (IIT) attempts to cross the gap by defining consciousness as causation on itself (Φ), but empirical difficulties abound: the cerebellum has high Φ yet is not considered part of the neural correlate of consciousness; split-brain patients have two streams of consciousness despite a single cerebrum; deep sleep and anesthesia reduce Φ yet dreaming and ketamine dissociation increase experienced richness. The newest version (IIT 4.0 with geometric algebra) mitigates some criticisms but still predicts consciousness in large inactive grids that most neuroscientists consider unconscious.

A more modest conclusion is warranted: consciousness depends on a specific kind of large-scale, recurrent, predictive, neuromodulated dynamics in the thalamocortical system. No single area, frequency band, or molecule is necessary or sufficient. The phenomenon emerges from the way information is simultaneously differentiated (to represent diverse contents) and integrated (to bind them into a single scene). Whether this dynamical signature is identical with experience or merely enables it remains, for now, a question neuroscience can bracket but not dissolve. The field has moved from ignorance to precise ignorance: we now know exactly which processes must be explained, and we know that none of the usual candidates (gamma synchrony, recurrence, workspace broadcasting, complexity) is individually sufficient. The next decade will likely see the integration of cellular, circuit, and systems levels into multi-scale models that predict not only reportability but the specific structure of phenomenal content. Whether such models will ever close the explanatory gap or merely push it to a deeper level is the open question that keeps the hard problem alive even as empirical progress accelerates.

Tanmoy Bhattacharyya