The analysis of the neuronal prerequisites required for the maintenance of consciousness has a long history and has only recently been considered part of consciousness research. The reason for this is that the criteria used for distinctions between conscious and non-conscious states or altered states of consciousness can be tested in both humans and animals. Examples of these criteria are reactivity to noxious stimuli, the ability to move intentionally, and the ability to accomplish a number of well-defined cognitive tasks, involving attention, short and long term memory, recognition, and decision making. Thus, the plethora of studies performed both on animals and humans on the neuronal mechanisms underlying arousal, attention, wakefulness, sleep, anaesthesia, and coma all contribute to our understanding of the neuronal prerequisites of states permitting conscious processing. Accordingly, it is well-established that brain functions characteristic of the conscious state require that neuronal networks operate in a critical dynamical range. This range is regulated by half a dozen globally-acting modulatory systems that originate in deep and evolutionary ancient brain structures. The adjustable neuronal parameters are essentially the balance between excitatory and inhibitory effects and the time- and length constants of dendritic integration. These adjustments lead to marked modifications of the system’s dynamics. These modulations are reflected by changes in the prevailing frequencies of oscillatory activity, the degree and spatial granularity of synchronisation (also addressed as correlation length), and the propagation of signals across the network.
Classical brain theories have not attributed much attention to the significance of these dynamic variables for processing and assume that loss of consciousness in sleep and anaesthesia is essentially due to reduced excitability and signal transmission. However, in more recent theories, brain dynamics are thought to play a crucial role in information processing. This novel framework provides much more specific explanation of the breakdown of consciousness in sleep, anaesthesia, and coma. These theories posit that oscillations and the concomitant variables, such as synchronisation, phase locking, phase relations, and cross frequency coupling, are relevant for signal selection by attention, binding operations, and the representation of nested semantic relations (for review see Singer 1999; Buzsáki et al. 2013). In addition, these complex dynamics have been proposed as a substrate for the generation of the high-dimensional coding space required for the storage and superposition of priors, the matching of stored information with sensory evidence, and the segregation of patterns for classification (for review see Singer 2013). The basis of these operations is the transformation of low-dimensional input patterns into high-dimensional dynamic states, in order to perform the necessary computations in this space and to then retransform the results into low-dimensional output signals. The advantages of performing computations in high-dimensional dynamic space are currently explored in the conceptual framework of “reservoir computing” or “liquid state or echo state machines” (Bertschinger & Natschläger 2004; Buonomano & Maass 2009; Jaeger 2001).
Recent analysis of the properties of recurrent networks, such as those realized in neuronal systems and in particular the cerebral cortex, indicate that such high-dimensional dynamic states can indeed be generated in delay-coupled networks (Lazar et al. 2009; Buonomano & Maass 2009; Soriano et al. 2013; for review see Singer 2013). In the present context it is important to recall that the dynamics required for such computations can emerge only when the networks are in the appropriate state. The optimal state has been identified as the edge of chaos, slightly below self-organised criticality, the so-called SOC state, because in this state the dimensionality or the complexity of the system are very high. Computationally this range is optimal because it offers a maximum of possible bifurcation points and storage capacity. (Plenz & Thiagarajan 2007). In this conceptual framework, computational results should consist of substates with reduced dimensionality. Experimental evidence indicates that the high-dimensional resting states are actually reduced by sensory input, imagery, recall of memories, or focused attention. These processes are all associated with enhanced correlation between neuronal responses due to the induction of synchronized high-frequency oscillations—where enhancing correlations reduces dimensionality (for review see Singer 2013). The notion that SOC states are optimal prerequisites for processing also fits with the robust evidence that states compatible with consciousness are characterized by “desynchronized” brain activity, i.e., states characterized by uncorrelated activity, such as are typical for wakefulness and arousal. If, and evidence suggests this to be the case (for review see Singer 1999, 2013), establishment of lower-dimensional synchronous substates, e.g., the formation of transiently-synchronized assemblies of neurons, is an integral part of the computations, then dynamic states characterized by global, large scale synchrony would be inappropriate as background for computations underlying higher cognitive functions.
As outlined in the target paper and above, higher cognitive functions require fine-grained binding operations among semantically-related contents that need to be encoded in ad hoc-formed neuronal assemblies. Such concatenation of multiple assemblies by partial correlations and perhaps also cross-frequency coupling would be impossible in networks that are already highly synchronized to begin with and hence exhibit low complexity and dimensionality. The well-established notion that deep sleep, anaesthesia, and most forms of coma are associated with brain states that exhibit slow oscillations synchronized over considerable distances agrees with this interpretation. In agreement with the prediction that low-dimensional brain states are incompatible with sophisticated processing are also the recent stimulation experiments cited by Noreika. It is to be expected that stimulation of a dynamic system that is in a low-dimensional state and at an overall reduced level of excitability will elicit only a spatially-restricted responses of low complexity—in particular if the stimulus is itself very low-dimensional, as is the case for a TMS pulse.
Considering more recent theories on brain functions, it appears as if the prerequisite or the NCC of a conscious state is a dynamic state that assures a high degree of complexity and high-dimensionality of resting-state dynamics. It is only in this state that the higher cognitive functions can be realized that one expects from a conscious brain.
It should be noted, however, that this operational definition of consciousness makes no inferences about the subjective contents of consciousness or the awareness of particular qualia of experience. According to this definition, consciousness is simply a brain state that allows animals and humans to accomplish higher cognitive functions that include not only perception but also decision making, planning of actions, generation of procedural and episodic memories, and last but not least intentionality and reasoning. Thus, one would expect consciousness, defined in this way, to be a graded phenomenon. If the state of the brain changes towards reduced complexity and dimensionality, there should be a graded deterioration of functions. Those requiring integration of widely-distributed assemblies should become impeded first, while simple reactions to salient sensory stimuli would persist for much longer. This seems to be in perfect agreement with the gradual deterioration of cognitive functions as the brain state shifts from high levels of alertness to drowsiness and sleep.