A Theory of Consciousness

Problems with connectionist approaches to consciousness

In the late 1960s, researchers found evidence that certain cells in the cortex of the brain exhibited a marked preference for very specific attributes of sensory stimuli and responded selectively to the presence of those attributes in the environment. On the basis of this finding, some researchers theorized that the attributes of multimodal complex stimuli are dispersed across the cortex and fractionated by these specialized feature extractors. These cells are organized into hierarchical networks in which simple feature detectors converge upon complex feature detectors, which in turn converge upon hypercomplex feature detectors. Thus, particular percepts might conceivably be represented by the firing of single cells at the peaks of such hierarchies. However, studies have shown that single neurons are unreliable reporters. Further, if a hypothetical hypercomplex cell at the peak of a hierarchy were to detect some percept for which the network is specialized, how could the nonspecialized remainder of the brain decide what exactly the firing cell is reporting? If such firing cannot be uniquely interpreted to indicate the presence of a specific object in the environmental scene, how does the brain obtain that information?

Clinical evidence for reproducible and discrete functional deficits in perception resulting from small brain lesions has fostered a belief in the localization of function in small ensembles of neurons, or cell assemblies, leading some researchers to search for a discrete locus of the neural correlate of consciousness. However, numerous studies indicate that such a locus seems unlikely. For example, more than 10% of the neurons throughout a cat's brain are significantly and differentially activated during a simple test of visual discrimination. In my theory of consciousness, I consider nonrandom firing in the neuronal population constituting a local cell assembly to be a “fragment of sensation.” How these dispersed fragments of sensation are integrated into meaningful perceptions remains unanswered, but the answer is central to any theory of consciousness.

Numerous research reports and theoretical articles indicate that subjective awareness and consciousness cannot be explained by processes in discrete neural pathways. Activation of widely dispersed brain areas by behavioral cues has been demonstrated in numerous brain-imaging and neurophysiological studies (Hoffman & McNaughton, 2002; John 2002). Thus, a conceptual shift is needed to explain how the brain accomplishes the transformation from distributed neuronal discharges to seamless global subjective awareness. Several areas of neuroscience provide evidence that consciousness is a property of an organized field of energy (John, 2002).

Dispersed nonrandom activity is information

Local field potentials (LFPs) are the synchronized fluctuations of voltage on cell membranes, caused by synaptic inputs modulating their probability of discharge, integrated across a huge number of neurons. Were such influences on individual cells in a feature-detecting neural ensemble randomly distributed in time, such random activity would provide no interpretable information, and the LFP would be zero. However, the activity of neurons within a cell ensemble becomes highly synchronized, or nonrandom, when that ensemble is processing information. I propose that nonrandomness (i.e., negative entropy) is information itself.

It is generally assumed that different sensory receiving regions within the brain selectively process stimuli in different sensory modalities (e.g., visual cues are processed in areas of the visual cortex). However, experiments with cats showed that electroencephalogram (EEG) and evoked potential2 responses to a novel and inconsequential visual stimulus appeared in many nonvisual regions upon initial presentation of that stimulus. Stimulus-locked EEGs and evoked potentials could be detected in almost every brain region as differentiated behavioral responses to that stimulus were established by conditioning (John, 1972). Appearance of stimulus-locked LFPs in a brain region is evidence of correspondingly synchronized neuronal discharges in the local cell population. Such observations support the premise that information about complex stimuli is widely distributed and is not restricted to the classical primary receiving areas of a specific sensory modality. On the basis of this evidence, I hypothesized that the brain encodes information statistically as synchronization of nonrandom activity in cell assemblies dispersed across many brain regions, rather than as activity selectively induced in dedicated cells in discrete, highly localized networks (John, 1972).

Influence of the endogenous system on perception

Experiments with cats that had been conditioned to respond in specific ways to specific stimuli illustrate how the endogenous system functions (John, 1972). As discrimination behaviors were acquired, responses synchronized to flashes at two different repetition rates that were the differential cues appeared in many spatially distributed brain regions, as was expected. However, a quite unexpected and informative phenomenon was observed when the animals made mistakes, performing incorrect behavioral responses to the differential cues. Occasionally, an animal would respond to a stimulus at one rate as if it were the different stimulus. During such behavioral misinterpretations of the meaning of the cue, sensory-specific brain regions would display responses at the repetition rate of the flash that was actually presented. At the same time, other brain regions displayed responses at the repetition rate of the absent alternative stimulus, which would have been the appropriate cue for the behavioral performance. The animal behaved as if its perception were largely influenced by endogenous readout from a spatially extensive representational system established by past experience (John, 1972). Further experiments established that synchronous patterned neuronal activity and LFPs associated with memory storage and retrieval were widely distributed in many different brain regions (Fries, Fernandez, & Jensen, 2003; Hoffman & McNaughton, 2002; John, 1972).

Electrical brain stimulation at different repetition rates, delivered to many brain regions to simulate the synchronous activity elicited by the differential cues, elicited appropriate differential performance of conditioned responses. This result confirmed the functional significance of the spatially distributed LFPs observed in these experiments (Kleinman & John, 1975). Taken together, the results suggested that the nonrandom synchronized neuronal activity in a spatially distributed set of regions represented the encoded information about the meaning of the stimulus (i.e., its behavioral significance). Further, electrical activation of the encoded pattern in one portion of the extensive representational system caused resonant activation in other parts of that system (John, 1972).

How is spatially dispersed nonrandomness bound?

An evoked potential is time locked to stimulus onset and persists for a period as long as 100 ms or more, called the response epoch. The weak oscillations in an evoked potential can be extracted from EEG records by averaging the response epochs to many stimulus repetitions, which yields an average evoked potential, or AEP. As the experimental cats described earlier acquired conditioned responses, the waveshape of the AEP elicited by the conditioned stimulus acquired a secondary peak late in the analysis epoch. When an error of omission occurred and the animal failed to perform the learned behavior, the late secondary peak was absent. Subtracting the AEP waveshape during such an error of omission from the AEP waveshape during correct performance, to isolate the readout process presumably represented by the secondary peak, yielded difference waveshapes that were essentially simultaneous among structures at many different anatomical levels of the brain. These distributed readout processes displayed corresponding peaks at identical latencies. This finding indicates that resonance occurs between these widely separated brain regions, because the peaks in these identical readout waveshapes were too closely synchronized to be explained by transmission of discharges along axons connecting the widely separated neuroanatomical structures.

Furthermore, it was possible to block behavioral performance selectively by timing disruptive cortical stimulation to coincide with the late secondary peak of the evoked potential, but not with the earlier primary peak. The intent was to block convergence of the endogenous and exogenous activity. During such concurrent stimulation, the animal failed to perform any conditioned response, although it remained alert and explored the experimental apparatus. Similar results in human surgical patients were later obtained by several investigators. Subjective awareness of a sensory event was blocked when disruptive thalamic or cortical electrical stimulation was delivered so as to coincide with arrival of the secondary component of the potential evoked by the peripheral stimulus. During such intracerebral stimulation, patients remained alert and aware of their surroundings but did not perceive the mild peripheral electrical shocks.

On the basis of these observations, I hypothesized that both exogenous activity (derived from sensory receptors) and endogenous activity (automatically released by associative processes from representational systems storing relevant past experiences) are elicited by an environmental stimulus. I proposed that these exogenous and endogenous activities converge via different pathways upon a widely distributed system of cortical coincidence detectors that compare incoming sensory input to the input from activated readout representing relevant past experience and present state. The level of firing from one cortical region to other cortical regions and to the thalamus is enhanced in those cells where significant coincidence detection is achieved, thus identifying cell assemblies containing items of information.

Coincidence detection causes local negative entropy

Coincidence detection between the present and the relevant past is proposed to act as a filter that identifies dispersed fragments of perceptions among the constant barrage of neuronal firing. These informational fragments constitute regions of improbable nonrandomness, relative to the noninformational random firing in cell assemblies. By some binding mechanism, such dispersed fragments are integrated into a unified perception that enters subjective awareness, endowed with a context that is a sort of “remembered present” (Edelman, 2001) so the brain can be guided to select adaptive behavior. The fragments of perception represent all of the momentary nonrandom synchrony that constitutes the momentary information content of the brain. These dispersed fragments of information constitute local negative entropy.

Recent neurophysiological studies have revealed a plausible mechanism that might perform such coincidence detection and facilitate global integration in the brain (Larkum, Zhu, & Sakmann, 1999). Pyramidal neurons are aligned perpendicularly to the surface in palisade-like arrays, with their cell bodies (soma) in deep cortical layers and with a branching structure of fine fibers or dendrites distant from the cell body but close to the surface in the uppermost layer of the cortex. Exogenous inputs to the cortex arrive on terminals located on the soma, while endogenous inputs arrive on the dendrites. Using stimuli delivered by micropipettes to selected terminals in a preparation of pyramidal cells, Larkum and colleagues found that when exogenous and endogenous inputs overlapped within a brief period, the output of discharges in cortico-thalamic (C-T) pathways released by the pyramidal cell was greatly enhanced, far exceeding the firing rate caused by either input alone. This C-T discharge was followed by a back-propagated (feedback) oscillation in the gamma frequency range (40–50 Hz). This suggests that a cortical pyramidal neuron upon which exogenous and endogenous inputs coincide might feed back to cause the thalamic cells from which the exogenous and endogenous inputs arose to repeat that input. The feedback establishes a local thalamo-cortical-thalamo-cortical (T-C-T-C) loop that reverberates every 20 to 25 ms. Thus, regions of negative entropy begin to reverberate.

LFPs mediate the binding process

In other studies, intracerebral recordings revealed that synchronized gamma oscillations, correlated with perceptual processing, occurred simultaneously with no delay (phase locked) in spatially separated brain regions. Sensory input or direct stimulation of the brain stem increased the amount of such synchronized activity. These simultaneous phase-locked oscillations with zero lag play a critical role in binding together dispersed regions containing the fragments of perception and may be the prerequisite for integrated perception (Rodriguez et al., 1999).

I propose that this phase locking across the distributed regions of nonrandom neuronal activity arises because rhythmic LFP waves superimpose a periodic extracellular modulation of voltage on the membrane potentials of the neurons in all cortical cell assemblies. This modulation is large enough to enhance excitability but cannot by itself cause neuronal discharge. However, every local cell assembly may contain a subgroup of neurons in which sufficient coincidence between exogenous and endogenous processes has occurred so that excitability has been enhanced. Some degree of membrane depolarization has occurred, but not enough to bring the cell to the threshold critical for firing. When the superimposed LFP further depolarizes the membranes of all cells in a brain region, those with already enhanced excitability will discharge. This discharge is simultaneous across all cortical regions.

When the modulating voltage arrives, a coherent C-T volley is released from all the dispersed fragments of perception whose excitability is sufficiently enhanced. This coherent C-T barrage arrives upon those cells in the thalamus from which came the contributions that excited the cortical coincidence detectors. This recursive feedback causes distributed T-C-T-C feedback loops to reverberate in the gamma range. I propose that, as these thalamic and cortical regions share a reverberation at the same frequency, electrotonic as well as synaptic interactions among them establish a resonating electromagnetic field of informational processes. This resonating field binds all the neurons engaged by these reverberations into a unified perception that becomes the content of consciousness. This information field constitutes global negative entropy, integrating all the information in the brain (John, 2002).

The ground state

Experimentally isolated slabs of cortical tissue display no rhythmic voltage oscillations, although the neurons in the slabs display random activity. If the activity of the neurons within a brain region were random and unsynchronized, LFP oscillations would approach zero amplitude. In order for a voltage to be detectable on the scalp, several square centimeters of neurons must be synchronously active in the underlying cortical region. When the eyes are closed, resting EEG arises from momentary synchrony within and interactions among brain regions, resulting from ongoing T-C-T-C reverberations.

Normative equations accurately predict the power at each frequency of the spontaneous EEG (QEEG) in functionally normal subjects of any age, resting with eyes closed, and also accurately quantify interactions among brain regions. Such QEEG analyses have extremely high test-retest reliability. Many studies using these norms have established their extremely high specificity, with false positive values only at the chance level in large groups of normally functioning individuals of any ethnic or racial background. The remarkable stability of these normative values and the fact that they are independent of ethnic or cultural background indicate that they reflect a biological constant. A complex homeostatic system regulates the synchronized transactions within and between different brain regions (Hughes & John, 1999).

This homeostatic system governs the structure but not the content of the neurophysiological interactions that occur in the brain as a resting individual sits with eyes closed and mental activity occurring at random. I propose that this regulated control system maintains a stable ground state and defines a stable equilibrium range of fluctuations of brain activity that may occur randomly around this set point under normal resting conditions.

Deviations from ground state are negative entropy

This resting ground state of brain activity is highly organized and defines a reference range of entropy. Because it is predictable, spontaneous resting activity within this range provides no information for the brain. Information consists of improbable departures, which cannot be predicted, from this dynamically regulated organization. Relative to this most probable, stabilized reference level, information can be considered as negative entropy (John, 2002), consisting of significant deviations from the local power spectra and altered regional interactions reflected in measures of coherence.

Transient localized shifts from the ground state occur during sensory stimulation, cognitive tasks, voluntary movements, and sleep. More generalized shifts occur during the action of substances on the brain. Sustained localized shifts of ground state are often seen in developmental disorders or psychiatric illness. Distinctive patterns of QEEG shifts from normative values provide a useful adjunct to psychiatric diagnosis. These persistent state changes imply distortion of those dimensions of perception mediated by the brain regions where they occur. Heterogeneous subtypes of state changes have been identified within groups of patients homogeneous with respect to their psychiatric diagnosis. Patients with different state changes respond differently to treatments (Hughes & John, 1999).

My colleagues and I have reported a reversible global displacement from the ground state that takes place during loss and return of consciousness during surgical anesthesia, independent of the anesthetic agent. Under every anesthetic, distinctive changes occur in the power spectrum, brain regions uncouple from each other, and brain organization simplifies. That is, the homeostatic regulation of the ground state is severely disrupted. Regional coordination is restored only in the gamma frequency as consciousness returns, followed by normalization of the regional power spectra. QEEG pattern recognition algorithms to recognize the distinctive features of the unconscious state have been incorporated into a QEEG monitor for the depth of anesthesia approved by the Food and Drug Administration (John et al., 2001).

Changes in the intrinsic set point of the homeostatic system can bring about more prolonged, difficult-to-reverse global departures from the ground state and loss of consciousness, such as after traumatic head injury. Progressive, irreversible shifts of brain state and concomitant alterations of consciousness have been characterized in dementing illnesses. In death, the brain state deteriorates to complete randomness and silence.

A regulated scanning process

The momentary variance of EEG voltage summed across the scalp is the global field power (GFP). GFP is simply a measure of the total variability of power across the scalp surface. Measurement of the GFP values at successive time points during an EEG recording reveals periodic peak values. Inspection of the topographic map of the voltage field across the scalp at the time point of any such peak reveals a particular topography or landscape, with a maximum positive voltage somewhere in the field that appears like a mountaintop and a gradient to some other point of maximum negativity that appears like a deep valley. Examination of this landscape at subsequent time points reveals that the mountaintop gradually becomes smaller and the valley less deep until the field looks almost flat. Abruptly, a mountain begins to emerge at some other point in the field, simultaneously with appearance of a new valley. The peak of the new mountain reaches a maximum positivity and the valley a greatest depth at the time of the next peak of the GFP. The process then repeats itself, with new mountains and valleys endlessly replacing each other. These shifting fields reveal sustained cooperative, synchronized activity patterns involving interactions among billions of neurons.

Remarkably, extensive study of scalp fields has revealed a limited set of landscapes, which have been termed microstates. These microstates flicker as if in a kaleidoscope and change whenever a peak occurs in the GFP. Automatic computer classification of lengthy EEG records from 500 normal 6- to 80-year-olds from Cuba, Switzerland, and the United States revealed that all the brain fields could be classified into only four3 different persisting landscapes (Koenig et al., 2002).

Microstate duration is very long relative to the discharges of any neuron. Such stability of field structure reflects a cooperative reinforcing process across multiple parallel, temporally offset channels, which sustain a spatiotemporal pattern independent of any neuron that participates in establishing that pattern. The average normative persistence of microstates found in this study was 80 6 2 ms, suggesting that the microstate reflects a regulated scanning process. These patterns reflect synchronized reverberatory interactions among cortical regions and between cortex and thalamus, indicating informational transactions (Koenig et al., 2002). Thus, it appears that a modulating voltage field scans the cortex at regular intervals, integrating the spatially distributed information.

I propose that this scan modulates the cortical sheet of comparators located in the arrays of pyramidal neurons. Consequently, in every cortical region, those neurons whose excitability has been enhanced by coincidence between exogenous and endogenous inputs discharge in a coherent C-T volley. Thus, the information dispersed as local negative entropy is assembled into global negative entropy, a moment of subjective awareness.

The long persistence of a microstate relative to the brief duration of a neuronal discharge suggests that T-C feedback consecutively recruits ensembles of neurons into a reverberatory maintenance of the microstate for this relatively long period. The mean duration of a microstate is approximately the same as the experimentally determined duration of a perceptual frame, the time interval within which successive events are perceived as simultaneous. Microstate cycling may be the basis for parsing of subjective awareness.

A field theory of consciousness

In summary, I propose the following theory of consciousness: