# Consciousness

Post-publication activity

Curator: Bernard J Baars

Figure 1:

# Conscious experiences

While conscious (cs) experience has been discussed throughout history, the late 19th century saw a rise in physicalistic reductionism, which, in its more extreme forms, declared "consciousness" and kindred terms to be unscientific. In the 1920s B.F. Skinner defined the goal of "radical behaviorism" as the complete elimination of mentalistic concepts from psychology --- about two-thirds of English content words. Skinner's influence dominated well into the 1970s, and during that time it was extremely difficult for scientists to openly study cs cognition, voluntary control, personal identity, and similar questions.

By the 1980s philosophers and scientists started to return to consciousness. Philosophers like Daniel Dennett and scientists like Francis Crick strongly supported a return to scientific research.

Since the 1980s a large literature has grown, with thousands of findings about visual awareness, coma and wakefulness, direct brain recording of binocular rivalry, and much more. The PubMed database shows almost 20,000 articles for the keywords "conscious brain."

Many scientific studies use experimental comparisons between similar cs and uncs events. This “contrastive analysis” approach has been very fruitful. Dozens of techniques now permit cs-uncs comparisons. Any method that permits such comparisons can enlarge our understanding. The resulting empirical harvest has been very large.

## Ambiguities

"Consciousness" (csns) has several meanings.

It is used in biomedical science to refer to the state of waking consciousness, as assessed by responsiveness to questions, commands, and mild pain, by the classical scalp EEG of waking, and by the ability to describe oneself and current events. We will use terms like “conscious state” or “waking state” for this specific meaning.

However, in scientific work "csns" is also used to refer to the "dimension of cs vs. uncs brain events" --- that is, as an experimental variable that allows us to study brain differences attributable to csns.

This usage is profoundly different from the first, since it involves a measurable dimension of variation. Yet “csns as an empirical variable” is still commonly confused with the waking state or with subjectivity. They are linked, but not the same.

Historically, many basic scientific concepts emerged from similar periods of ambiguity. In daily life heat refers to boiling pots and summer days. But with the rise of thermodynamics at the end of the 19th century, the scientific term heat came to mean a dimension of temperature variation with a true zero point at 0 degrees Kelvin, reflecting a wide quantitative range of random molecular activity in physical systems.

Thus in thermodynamic theory heat came to mean something very different from its everyday meaning.

Similarly, in scientific usage the topic of “consciousness” now often means an empirical dimension of variation in brain activity between matched cs and uncs events. When we talk about “the topic of consciousness” in science today, we do not mean subjectivity alone. Rather, we refer to the “cs-uncs dimension,” which includes subjective experiences, but always compared to close brain analogues that are not cs.

A classical example is binocular rivalry, where mutually incompatible stimuli are presented to the two eyes, which cannot be fused into a single gestalt. Only one of the two rivaling stimuli can be conscious (reportable) at a time, though the other often evokes some degree of visual processing. The Nikos Logothetis laboratory at the Max Planck Institute in Tuebingen, Germany, has used direct brain recording in macaques since the 1980s to trace both the conscious and unconscious processing stream. While many other methods are in use, the Max Planck group has performed the most coherent long-term research program of this kind.

Third and finally, the English language also uses "csns of something” to refer to the specific contents of mental life, or what we will call "cs cognitions" or “cs percepts.”

Given these ambiguities, it is essential to avoid vague language. Here we will use unambiguous expressions whenever possible.

(See thalamocortical system; and Part 4 in which major features of conscious states and contents are detailed).

## Conscious contents

In general, the reportable contents of csns include perceptual stimuli; inner speech, reportable dreams and visual imagery; the fleeting present and its fading traces in immediate memory; interoceptive feelings like pleasure, pain, anticipatory anxiety and excitement; the exteroceptive body senses, including external touch and pain; reportable emotions; autobiographical episodes as they are experienced and recalled; clear and immediate intentions, expectations and effortful voluntary control; explicit beliefs about oneself and the world; reportable feelings of knowing (FOKs; see below); novel skills as opposed to overpracticed ones; and concepts that are abstract but still reportable.

This basic list has not changed since Aristotle, and may be a cultural universal.

Feelings of knowing. Perceptual csns shows multiple levels of highly discriminable details --- tiny dots of light, contrast edges, inferred size constancy, object identity, etc. But “feelings of knowing” (FOKs) are classically said to be experienced as “vague” or “fringe-like.” FOKs include judgments (as opposed to percepts), feelings of familiarity, feelings of rightness and wrongness, and much more. William James made an empirically persuasive case for “the vague” in mental life in his Principles of Psychology (1890), and that evidence has only expanded since then. Bruce Mangan also includes intuitive feelings of beauty and goodness, which are highly reliable under experimental conditions. While the phenomenal experience of FOKs is “vague,” the actual cognitive processes involved can be very precise and complex. Introspection is not a reliable guide to the cognitive complexity of FOKs.

FOKs can be defined empirically as "reportable experiences that are verifiable, and which are reported with high certainty, but with very little descriptive detail."

In the case of a clearly visible coffee cup, we have both high confidence and the ability to experience numerous details. In contrast, FOKs often show high confidence and accuracy with no subjective experience of details. The range of FOKs is very large indeed.

## Observational definitions

Workers in the field are often asked to define "csns," but empirical science rarely starts with conceptual definitions. Rather, scientific concepts like "heat," "force" and "momentum" evolved over long periods of time, inductively, step by step, based on useful and reliable observations.

As pointed out, the thermodynamic definition of heat only emerged in the late 19th century. Beginning with the Renaissance, scientists from Galileo to Fahrenheit developed increasingly precise ways to measure “heat.” But without 19th century discoveries in the physics of molecular motion and the Kelvin scale, a clear theoretical definition was out of reach.

This basic point about inductive science is often misunderstood. Empirical science commonly starts with observational definitions that seem roughly correct. Refined theory tends to come much later.

Part 4 shows more than twenty reliable findings about the state of csns and its reportable contents. Each of these facts could be used as observational definitions. In practice, we generally start with a specific behavior--- the ability to accurately report stimuli or evoked mental events under careful experimental conditions. (see 1.4 below)

Today we know many “neural correlates” of csns. None of them are complete, but they give useful starting points.

(Emerging theoretical definitions. Note that theoretical definitions of the cs state and its contents have been proposed. See W.J. Freeman, M. Steriade, B.J. Baars, G. Tononi, G.M. Edelman, and others.)

## “Accurate report” as an observable index

The most widely used behavioral index of cs events is "accurate, voluntary report" under optimal conditions, such as minimal distraction and time delay. Reportability corresponds well with our everyday understanding. Indeed, the words "accurately reportable" can often be used instead of "cs." However, a purely behavioral index would cause us to miss something essential‚ the fact that accurate reports refer to a rich Umwelt of experiences that we all share.

In the sensory sciences accurate report has long been fruitful, beginning with Newton's prism experiments, and still in general use today. The study of color perception would have been impossible without accurate report. Clinical examinations of vision and hearing continue to rely on accurate report.

Failures of accurate report are just as important, as in the case of various types of color blindness. People who lack one, two or three types of retinal color receptors selectively confuse three-color with two-color pictures, and two-color pictures with grayscale copies. Just as perceptual difference reports are fundamental in psychophysics, equivalence reports or confusability between two physically different but subjectively identical stimuli has long been important in perception and cognition.

### Improving report measures

Observational definitions do not remain static. Improved measures often reveal unexpected facts, which may require a basic rethinking of the proposed concept. The history of science is filled with examples. Our understanding of conscious states and contents is evolving in precisely that way.

L. Jacoby and colleagues pioneered the use of “process dissociation,” as a measure of cs control. In process dissociation, subjects are instructed to try to stop automatic (over-learned) mental events, such as “word reading” rather than “color naming” in the classical Stroop task. Since subjects are skilled readers, the novel task of color-naming requires voluntary effort, to overcome automatic habits of reading for sound and meaning.

Such interference effects occur whenever an overlearned skill can be pitted against a new task. Errors and delays in performing the new task are thought to reflect a drop in “csly-mediated” control. Process dissociation bears on a major theoretical question, the interplay of conscious experiences with voluntary control.

The most dramatic empirical improvements have come with the brain imaging revolution. Recording techniques continue to improve year by year, with significant advances in our understanding of the cs brain.

## Contrastive analysis

There are two ways to study cs experiences experimentally. One is to compare them to each other, as in the sensory sciences. Content comparisons are used routinely in perception, recognition memory, mental imagery, short-term memory, and the like. A more recent approach is to compare cs events with closely matched uncs analogues. This approach dates back several decades, when researchers discovered convincing evidence for uncs but "intelligent" brain events, including sensory processing, memory maintenance, automaticity of complex skills, etc.

A simple example is to say a word mentally, and then let it fade; for about ten seconds afterwards, it can still be recalled. (The reader is encouraged to try this several times.) Our ability to retrieve the word after fading suggests that an uncs memory of the word must have been preserved. This informal experiment provides cs/uncs comparison conditions, even for a single word. Now we can try to answer the question, "What is the brain effect of our being cs of a word?"

In effect, we have a controlled experiment using cs access as an independent variable. Brain recordings show clear differences.

Numerous experiments like this have been published, over a range of phenomena. They give the most relevant body of evidence about cs experience "as such."

For an anatomical example, the human cerebellum, which has about the same number of neurons as cortex, does not seem to support cs contents directly. Patients with damage to the cerebellum are still cs of the same range of contents as before.

However, very local damage to sensory cortex can produce very specific deficits in cs experience, such as cortical color blindness and face blindness --- the inability to recognize a visual pattern as a face, in spite of being able to name eyes, noses, mouths, chins, and so on. Specific kinds of cortical blindness can be localized to specific regions of cortex, notably V3/V4 for color perception, and area IT (inferotemporal) for face and object identity. These phenomena have been replicated in the rhesus macaque, which has a strikingly similar visual brain to ours.

Depending on the specific aspect of csns being studied, an experimental comparison may show two or more values; it is often maintained that conscious events have high underlying dimensionality. The classical dimensions of color perception are one example, but visual perception in the natural world must have much higher dimensionality.

These are testable questions.

## A growing set of brain correlates

Part 4 (see below) shows 22 empirical correlates of cs states and their specific contents. These continue to be refined, clarified and expanded.

Historically, scalp EEG of the waking state was described as “irregular, low in amplitude, and fast.” By “irregular” one meant something close to random. When scalp EEG was averaged over multiple samples it reliably added up to zero volts. Yet scalp EEG is still a useful first index for defining waking consciousness in medicine and physiology.

Nevertheless, direct cortical recording improves the signal-to-noise ratio by a factor of 1,000. These “intracranial” recordings reveal a very different picture of cortical signaling, between specific arrays of neurons signaling via neurons spike firing, or by population oscillations ranging from <.1 to 200 Hz.

Visual cortex has more than 40 “visuotopical” arrays, ranging from V1, which resembles a pixelated screen with high spatial resolution, to area IT (inferotemporal) with much less spatial resolution and much higher gestalt organization. In V1 the retinal input from a human face is simply a 2D distribution of colored and grayscale pixels. About forty maps later, in IT, that input is seen in terms of facial features --- mouth, nose, eyes, ears, facial expressions and individual faces.

To a first approximation, the cortex is a vast collection of spatially organized neuronal arrays with six histologically different layers. Cortical arrays begin in sensory two-dimensional receptor arrays like the retina, and are ultimately converted to motor signals that trigger arrays of muscle cells. Signaling between arrays are often “point to point” so that a foveal neuron $$(x_1,y_1)$$ in the retina corresponds well to a similarly located thalamic neuron $$(x_2, y_2)$$ in LGN, followed by some 40 corresponding cells in cortical arrays V1 – V4 and ultimately areas IT and MTL. Each layered array sends axons to both higher and lower cells in other arrays. This bidirectional signaling gives rise to resonant excitatory activity, the typical signaling style of cortex.

The broad preservation of visuotopical and spatiotopical patterning across more than 40 visual maps is called labeled line coding. Signaling between these broadly similar arrays involves both neuronal spike firing and population oscillations, from <0.1 to 200 Hz. In general, cortical “close-up” recordings look radically different from traditional scalp EEG.

In retrospect, scalp EEG is misleading for detailed studies of cortex, since it suffers a thousand-fold loss of signal voltage compared to direct brain recordings. Direct recordings also solve potential artifacts of scalp EEG, such as scalp muscle activity, eye movement artifacts and poor source localizability. Such methods have yielded a harvest of new insights.

Independent variables are also much improved. Binocular rivalry is a classical method for comparing closely matched cs vs uncs retinal input. Thanks to three decades of rivalry studies in the macaque, we can now compare the fate of two visual input streams, one that is reportable, while the other is not.

These studies show that the cs stream shows significantly higher cortical activation, more oscillatory phase-linking, and wider spatial propagation than closely matched uncs stimuli. cs stimulus processing propagates not only within the visual cortex, but also to prefrontal and plausibly hippocampal regions. Direct brain recordings converge well with other techniques like fMRI, PET, MEG, etc. Compared to the cs processing stream, matched uncs events are not voluntarily reportable; they evoke less neuronal activity; show less oscillatory phase-linking; and tend to decay locally in visual cortex.

Several other experimental methods have been used, like visual backward masking, the attentional blink, inattentional blindness, and selective listening. There is reasonable agreement between stimulation and recording methods.

## Cortex enables conscious experiences: The cortico-thalamic system

A large and coherent body of evidence now shows that cortex and its resonant satellite, the thalamus, jointly constitute the brain basis of cs experiences. However, the states of waking, sleep and REM dreaming are switched on and off by basal brain nuclei that project neurochemicals to the cortex and thalamus.

This evidence comes from:

 a. direct brain recordings of both reportable and unreportable input b. brain imaging studies including fMRI, MEG, PET, etc. c. various deficits in cs perception and cognition, d. local brain stimulation and inhibition, either via the sensory pathways, or by electrical, magnetic, cooling, chemical, and other precisely localized interventions. e. pathological conditions, as in “silent” (unconscious) ischemia compared to severe ischemic pain; f. comparisons between waking-state stimulation vs. slow-wave (delta) sleep, coma and vegetative states, epileptiform loss of csns, syncope (fainting), and general anesthesia; g. Direct recordings of the C-T oscillatory system in animals and humans. Recently voltage-sensitive study whole-cortex regional activity in rodents.

These methods show that the cortico-thalamic system underlies the states and contents of csns, in contrast to cerebellum, basal ganglia and other subcortical regions.

However, to make that case clearly, we must consider a distinction between brain activities that show direct correspondence to cs experiences, compared to those that do not correspond directly with reported experiences.

## Direct correspondence with reported contents

A damaged brainstem reticular formation (BRF) can abolish the state of csns, and therefore its contents as well. But the BRF shows no correspondence to specific contents of csns --- such as color perception, fine visual resolution, or visual scene comprehension.

In contrast, for vision we can observe specific cortical “feature maps” whose activities correspond well to cs contents, as reported under optimal conditions. For vision, final gestalt formation may take place fin medial temporal lobe.

In sum, many lines of evidence converge to show that cortex underlies specific cs percepts, in resonance with corresponding thalamic nuclei.

There are a few qualifications:

1. Brain regions that do not show direct correspondence can still influence the C-T system, and therefore indirectly influence cs contents. A classical example is the cerebellum, which constant interacts with the C-T system. But cerebellum does not have neuronal feature fields whose activity patterns correspond to reported experiences. The same point applies to basal ganglia, motor tracts, amygdala, etc. The amygdala responds to fearful visual pictures, but the interoceptive feeling of fear depends on the body maps of the anterior insular cortex. Similarly, the vestibular (balance) system does not emerge in csns directly, though it shapes visual, sensorimotor, and interoceptive experiences.
2. Neurological impairments like visual neglect show that visual maps of the parietal cortex do not yield reportable cs events. However, these “reaching” maps do shape direct perception of visual feature maps in the ventral stream.
3. Paleocortex enables cs olfaction and taste. It includes hippocampus and rhinal regions, and evolved before the mammalian neocortex and thalamus. It has somewhat different visible anatomy. However, paleocortex and neocortex have strong similarities and are highly integrated in mammals.
4. “Feelings of knowing” (FOKs) may reflect cs gestalt formation in non-sensory cortex. A famous example is the “tip of the tongue” state, which shows BOLD activity in prefrontal cortex.

## Consciously mediated cognition

Stan Franklin has proposed the term “csly mediated” for cognitive processes that come in and out of csns. In perceptual experiments we commonly present isolated stimuli, but in natural conditions, cs moments emerge as part of a “stream of csns,” a series of intertwining topical threads that come to csns only intermittently, much like an internet chat room. Overlearned processes tend to be uncs, using highly practiced automatisms and memory traces. Cs cognition is needed for unpredictable, novel, significant, or effortful aspects of thinking.

A well-studied example is unconscious amygdala activation in response to fearful pictures. A conscious picture of a snake may trigger unconscious fear-related activity in amygdala, with very widespread activation of emotional, neurohormonal, and social-cognitive processing, while the conscious feeling of fears appears to require activation of the anterior insula. The conscious feeling of fear can then evoke a wide variety of other unconscious and conscious brain events, including fight and flight, but also rational planning.

Experimental evidence for the stream of csns comes from thought-monitoring studies, beginning with Jerome Singer and John Antrobus some fifty years ago. These methods have been important in understanding clinical conditions like post-traumatic stress disorder.

## Voluntary control

Conscious events can evoke voluntary actions on request, notably accurate event reporting, memory storage, sensorimotor control, mental problem solving, and emotional reactions.

Voluntary control governs a remarkably wide range of motor and neural activities. Recent intracranial recordings in waking patients shows that even single temporal lobe neurons can be controlled, on request, using cs feedback from a firing cortical neuron. How the brain selects one among 86 billion neurons to control, given consistent cs feedback, is not known.

While these indirect effects of cs moments are obviously important, our focus here is on the direct brain basis for cs experiences, when a correspondence can be shown between neuronal feature maps and accurate report.

(See Global Workspace, Global Workspace Theory, Integrated Information Theory, Adaptive Resonance Theory and Neural Darwinism).

# Unconscious states

It is surprisingly difficult to prove the complete absence of csns. Sleep can vary in arousability from moment to moment, much like vegetative states and general anesthesia. Some mentation is often reported even when subjects are awoken from slow-wave sleep. Coma scientists have made major progress in assessing different degrees of csness after brain damage, showing that waking-like functions can be preserved in "behavioral coma."

One class of behaviorally inert patients are conscious but paralyzed, a condition called "locked-in" syndrome. Some locked-in patients have been trained to use voluntary eye fixations on a computer keyboard, allowing them to spell out messages in a nearly normal way. Since any voluntary response can be used, eye fixations can be as good as finger pointing or verbal report.

Even slow-wave sleep may not be entirely uncs. Deep sleep is characterized by massive, spatially synchronized delta waves, <2 Hz, in the cortex and thalamus. The trough of each delta wave involves widespread pausing of neuronal firing, while the peak shows a waking-style range of oscillations and neuronal firing. Even deep sleep may therefore be intermittently cs, during the peak of the repeating delta wave. Deep sleep is the most unconscious natural state, but some waking-like mentation has been reported by subjects awoken from deep sleep.

Physiologically, deep sleep is an active state, needed for consolidation of experiential memory traces via hippocampal-neocortical theta-gamma signaling. We do not know the full set of functions of deep sleep, but more than 200 types of gene expression are known to correlate with the sleep-waking cycle. Deep sleep therefore does not mean the absence of active, biologically important brain activities.

There are different kinds of unconscious processes. Features of uncs sensory processes do not apply to other uncs brain events, like highly practiced automatisms, cerebellar and striate activities, long-term synaptic memories, parietal representation of nearby space, autonomic activities, etc. Not all uncs brain events are alike.

# General conclusions

Contrary to popular belief, scientists have gathered a substantial body of evidence about cs experiences and their brain basis. In the centuries since Newton’s color experiments an immense body of facts has been gathered about perceptual csns, using accurate report measures to compare precisely defined stimuli to each other. The study of waking csns, sleep and pathological impairments of csns has also seen significant progress.

In addition, “contrastive analysis” experiments that compare cs events to closely matched uncs analogues have been very productive in recent decades. With advances in brain recording, the evidence has expanded quickly, so that a database search for “csns” and “cs brain” shows many thousands of peer-reviewed articles.

This wealth of evidence does not necessarily mean that we “know what csns is,” in the sense that physicists know what “heat” means in thermodynamic theory. Inductive science seldom leaps directly to high-level theory. Rather, we work inductively, using observational definitions.

In this more modest sense, a great deal is known about cs states and contents. Current theoretical efforts aiming to make sense of this evidence are tightly constrained.

# Major features of conscious states and contents

1. Raw scalp EEG signature of waking.

The raw scalp EEG signature of waking appears to be “irregular, low-voltage and fast” (12-70 Hz).

However, direct cortical recording improves signal-to noise by a factor of 1,000 over scalp EEG. These data show much more detailed, local, and content-specific signaling among small regions of cortex and thalamus.

Direct brain recordings in waking epileptic patients have been particularly important. Patients wearing implanted electrode grids superimposed on cortex show prominent activation of auditory speech perception areas (Wernicke’s area) when they are asked to listen to a word attentively. When they repeat the same word, both Broca’s and Wernicke’s area are activated, precisely as one would expect.

By analogy, the audience in a football arena may give out undifferentiated crowd noises from a distance, but a set of local microphones will pick up thousands of one-to-one conversations, which are not random but highly regular, event-specific and organized. The closer cortical activity is observed, the more it resembles coherent, task-driven signaling between small regions.

Newer methods, such as genetically coded voltage imaging of the mouse cortex, have very high spatiotemporal resolution and wide spatial coverage. These methods show the quiet waking state to be “near criticality,” with scale-invariant spatiotemporal patterns of neuronal activity spanning four orders of magnitude.

The near-criticality state of waking cortex implies a high level of readiness for a wide range of specific types of information processing, at multiple scales. Other typical signatures of nonlinear dynamical regimes have also been found.

Walter J. Freeman and colleagues have developed analytical and physiological tools for characterizing these phenomena.

Direct brain recordings suggest a very large spatiotemporal signaling vocabulary, linking small cortical regions in a task- and stimulus-specific fashion, with oscillations from <.1 - 200 Hz. Single neurons can join or compete with these population oscillations. Complex waveforms in the intracranial EEG are believed to reflect a phase-linked wave hierarchy from slow to fast oscillations.

The traditional understanding of scalp EEG therefore appears to be an artifact of older, low-resolution, and unanalyzed recording techniques.

2. cs vs. uncs vision.

Much visual cortical processing is not cs. Retinal input is conveyed to the optic nerve via ganglion cells, which terminate in the visual thalamus (LGN) without loss of spatial resolution. LGN signals the first cortical projection area, Area V1, with point-to-point accuracy. Long-lasting binocular rivalry is enabled using continuous flash suppression, which allows for single-neuron and multi-unit recording of visual neurons throughout the hierarchy. Neurons responding to the cs visual stream can be followed independently of those responding to the uncs stream. For almost three decades, Logothetis and coworkers have systematically traced the two visual streams through the macaque cortex.

While the optic nerve is unidirectional, all higher-level signaling among thalamic and cortical neurons is bidirectional, giving rise to reentrant signaling or adaptive resonance, the typical signaling mode of the cortex and thalamus. From Area V1 the visual signal percolates through more than forty topographical arrays, emerging as high-level, reportable gestalts near the top of the hierarchy. Recent findings show that reportable visual gestalts emerge in the Medial Temporal Lobe (MTL), and are propagated from there to prefrontal regions. cs visual contents therefore correspond well to convergent visual “gestalts” in MTL. This is an historic result, built systematically on almost thirty years of cumulative studies. ￼ It seems likely that auditory, rhinal and somatosensory gestalts may also emerge in high level sensory cortex in those modalities. Baars et al. (2013) propose that reportable Feelings of Knowing (FOKs) emerge in non-sensory regions of cortex, e.g., the dorsolateral prefrontal cortex for feelings of subjective effort. This is consistent with brain imaging evidence.

Sensory cs contents emerge in high-level posterior cortex, while abstract Feelings of Knowing, as in feelings of mental effort or familiarity, may do the same in prefrontal cortex. This evidence is now quite strong in the case of vision, while other modalities need further testing. Consistent reports describe conscious visual cortical activity (compared to unconscious) as higher in amplitude, wider in oscillatory phase-linking, and propagated more widely, including causal (Granger) propagation (Gaillard et al., 2012). There is ongoing debate whether prefrontal activity is necessary for conscious perception. Panagiotaropoulos et al. (2014), using direct cortical recording in the macaque, find that visual gestalt formation (integration) occurs in the medial temporal lobe, from where the gestalt is accurately propagated to lateral prefrontal cortex.

3. Deep sleep

Deep sleep is most natural uncs brain state, with widespread delta waves (<2 Hz) observable in the raw EEG as regular, high-amplitude, and slow waves. Delta waves reflect coordinated pausing and firing among billions of cortical and thalamic neurons. Delta sleep serves to consolidate episodic memories acquired in recent cs periods.

The brain behaves differently in the delta peak compared to the trough. The UP half of delta sleep allows for a rich mixture of microlevel oscillations and firing for about a second, while the DOWN half shows widespread neuronal pausing. The UP state may be a momentary waking period. The major difference between deep sleep and waking may be the regular disruption of processing during the troughs of delta waves. Steriade (2006) has shown that slow oscillations in deep sleep reflect large, traveling waves flowing from front to back via the interhemispheric fissure. Induced uncs states like general anesthesia are sometimes called “artificial comas,” and are thought to be different from natural slow-wave sleep.

4. REM dreams.

REM dreams resemble the waking state, with eyes closed and large, regular, and stereotyped eye movements. Scalp EEG during REM shows “irregular, low-voltage and fast activity,” suggesting waking-like spatiotemporal signaling in the C-T core. Subjects awoken during REM report rich cs imagery and dramatic scenarios, with narrative discontinuities after perhaps 10 seconds. During lucid (self-cs) dreaming humans can count to ten using eye movements as voluntary “start” and “end” signals, similar to the 10-30 second time of working memory.

During REM dreams sensory inflow is blocked at the level of the thalamus, so that cs dream contents reflect endogenous cortical activity. Dream content shows “day residues” from recent waking periods, as well as motivationally driven “current concerns.” fMRI in dreams shows high metabolic activity in visual and emotional regions of cortex.

5. Brain anatomy.

cs contents depends on the thalamocortical complex, with major circadian states switched on and off by brainstem neuromodulation. Regions outside of the C-T complex constantly interact with, but do not directly support reportable cs contents. Biologically, the C-T complex emerges with mammals (approx. 200 million years ago) and has anatomical homologues in the bird pallium. Pre-mammalian amniotes show paleocortex (rhinal cortex and hippocampus).

￼A great deal of new evidence on the connectivity pattern of the C-T system reveals small-world connectivity, which serves to optimize signal flow. The C-T system is the most highly parallel-interactive system in the mammalian brain, in marked contrast to the cerebellum, which has similar numbers of neurons but reveals parallel streams of processing. The structural road-map of the C-T system appears to be ideally suited for resolving focal uncertainties, ambiguities, and decision-points, a major biological challenge that is associated with cs states and contents. (See Baars et al., 2013).

Like the structural map of the C-T system, the power in EEG frequency bands follow an inverse frequency law ((1/f) EXP B).

Neuronal activity corresponding to cs gestalts propagates widely in the brain, as indicated by highly distributed long-term memory traces, implicit learning, and biofeedback training of autonomic and motor functions. The very fact of accurate reportability implies that executive regions like the prefrontal cortex receive accurate source information from posterior sensory regions.

7. A very wide range of cs contents.

The cs state has an extraordinary range of contents—sensory perception in the various senses, endogenous imagery, emotional feelings, inner speech, abstract concepts, action-related ideas and Feelings of Knowing such as familiarity, confidence judgments, feelings of effort, and much more. The set of cs contents may be open-ended.

Spontaneous cs mentation is believed to reflect “current goal-driven concerns.”

8. Informativeness. (See automaticity)

cs contents fade when input signals become redundant, a general phenomenon that can be observed across the senses. Sensorimotor skills also fade from cs access with repetition. Such redundancy effects apply to high-level semantics as well, as in the case of semantic satiation.

A classical example in vision involves stabilized retinal images. During normal vision the eyes exhibit a constant, fast tremor, so that incoming light edges and gradients do not fall on the same retinal receptors. If a tiny light projector is mounted on a contact lens so as to stabilize light input to the fovea, the image will rapidly fade from consciousness. Any change in the incident pattern will return the stimulus to csns.

These very general effects suggest that sensory csns thrives on informative input, rather than physical energy patterns as such. The senses are not simple energy transducers, but fast-adapting neuronal networks at multiple levels of input analysis. When a constant stimulus becomes predictable it is no longer csly perceived.

Thus “informativeness” seems to be a necessary condition for cs perception.

9. The fleeting nature of cs events.

Our experience of the sensory present may last a few seconds, while our cognitive present may be less than half a minute. In contrast, much uncs knowledge resides in long-term memory, interacting constantly with sensory and endogenous input.

10. Internal consistency.

cs contents are marked by a strong consistency constraint. For example, while multiple meanings of most words are active for a brief time after presentation, only one becomes cs at any given time. This point also applies to perceptual stimulation (as in perceptual ambiguities or competing inputs), and it may also apply to beliefs, as in the case of cognitive dissonance.

In general, of two mutually inconsistent contents, only one can become cs at a time.

11. Limited capacity and seriality.

Many careful experiments show that momentary cs capacity is only 1-4 separate “items.” When unpredictable items are presented visually while the subject repeats a short word like ‘the… the… the” this radically limited capacity appears consistently.

Thus the isolated capacity of cs contents at any given moment is radically limited. We can circumvent these narrow limits by chunking, stimulus organization, and learned associations to established knowledge and skills in long-term memory.

Yet the flow of isolated conscious items is serial, in contrast with the massive parallelism of the (mostly unconscious) brain when it is observed directly.

12. Multiple levels of binding (integration).

Sensory cortex is regionally specialized such that different cortical “visuotopical maps” respond to different visual features such retinal location, size, contrast, color, object identity and motion. One classical question is how these cortical arrays coordinate their activities to generate the integrated gestalts of everyday perception.

Recent evidence implicates the Medial Temporal Lobe (MTL) in gestalt integration.

14. Multiple problem-solving functions.

Because conscious cognition is involved in perception, learning and novelty-processing it appears to be necessary for (novel) problem-solving. Herbert Simon and Alan Newell’s studies of explicit problem-solving protocols beginning in the 1950s, demonstrated the radical limited capacity of conscious cognition, and led to their later work on cognitive architectures, which have both a small-capacity and large-capacity component. In explicit problem-solving, as in mathematical proofs, every needed step is conscious. But most real-life problems have predictable and automatic components – such as the act of reading this sentence. Most real-life problem solving involve alternating conscious and unconscious moments.

Based on thought monitoring studies it has been argued that the spontaneous flow of thoughts involves multiple “threads” of conscious and unconscious elements, intertwined with each other. Jerome Singer and colleagues have shown that spontaneous thought is mostly driven by “current concerns.” In healthy subjects the stream of spontaneous cognition is neither random nor dysfunctional. It appears to be goal-directed, often unconsciously.

Conditions like depression show repetitive ruminations on unrealistically self-critical thoughts. Cognitive behavioral therapy studies strongly suggest the existence of half-conscious or unconscious dysfunctional beliefs that drive negative moods.

Thought monitoring studies in traumatized subjects often show unwanted and intrusive thoughts, feelings, images, and intense emotions. These findings helped to establish Post-Traumatic Stress Disorder (PTSD) as a formal psychiatric category.

15. Implicit self-attribution – the observing I.

cs experiences are generally attributed to an experiencing but implicit self, the ‘‘observing self,’’ as William James called it. Self-functions appear to be associated with precuneus and orbitofrontal cortex in humans. However, it engages a wider network of cortical and subcortical regions, such as the vestibular balance system.

There is little evidence that the “observing ego” can report its own functioning, since it is implicit. Self-knowledge is a cognitive skill that varies radically between individuals. Informal beliefs about the implicit ego are generally inaccurate.

Rather, the perceiving “I” seems to involve an implicit framework within which cs events are defined, but which is not cs in and of itself. For example, the egocentric visual maps of the parietal cortex work in coordination with allocentric (object-centered) maps, and with motor cortex, much like gravitational lines of force around large masses in physics. Both hippocampal/entorhinal and parietal regions are thought to have egocentric and allocentric maps.

16. Accurate reportability.

cs contents are reportable by a wide range of voluntary responses, often with very high accuracy. The conventional operational index of csns is based on accurate reportability.

Notice that humans are also distinguished by a lush imaginative life, as shown by way of legends, other-worldly beliefs, dreams and waking fantasies. Thus our brains have dual specializations --- we can be extremely accurate (which is essential in dealing with dangers and opportunities) and we are also very much driven by imagined events.

17. Subjectivity.

csns is marked by a private flow of events available only to the reporting observer. Reportability applies to the classical senses, but not to the vestibular (balance) sense, for example, nor to parietal control of sensorimotor reaching, syntactic analysis, and much more. The brain bases of these uncs functions is not well-understood.

18. Focal-FOK distinction.

While cs contents tend to be viewed in terms of focal percept-like contents, Feelings of Knowing (FOKs) are common and important. Judgments, intuitions, abstract concepts, and vivid expectations are experienced as FOK’s, including the famous tip-of-the-tongue state, which involves complex semantic and linguistic knowledge. This may reflect the distinction between percepts and concepts.

19. “csly-mediated” learning is very effective.

The evidence is strong that non-trivial learning requires some time for cs access. Many thousands of subliminal perception experiments have tried to show non-trivial uncs learning, but robust learning of novel material has not been found.

New claims for uncs learning appear often, but seem to be limited to priming of pre-existing knowledge. Stored knowledge can be easily activated. For example, uncs pictures of snakes may trigger amygdala activation, though there is still debate on this point.

However, “csly mediated learning” is very effective. In a classical experiment by Standing (1977), subjects were shown thousands of cs pictures for less than 10 seconds each. They were not asked to memorize them, but only to pay attention. Several days later they were able to tell “old” from “new” pictures at more than 90 percent accuracy. Thus robust learning occurs merely when information becomes cs. Such high learning efficiency has never been reported for subliminal perception.

These “easy-learning” results may seem counterintuitive, because we are used to academic study, which is effortful and demanding. The difference is in the task conditions. Academic learning requires deliberate memorizing and recall on demand, while highly efficient incidental learning is found for easier retrieval tasks, such as old vs. new recognition and cued recall.

The term “implicit learning” is often misunderstood to mean “uncs learning,” but this is not correct. Implicit learning experiments always show a set of cs stimuli to the subjects. What is genuinely uncs about “implicit learning” is the process of inference that occurs, once the cs stimulus set is perceived and learned.

Implicit learning is vital for the acquisition of natural language, but there, too, infants hear numerous cs utterances before they implicitly infer the regularities of their native language. Adults naturally engage in “baby talk” with infants, imitating infant tonal melodies while using adult words. Language acquisition therefore seems to be csly mediated.

In sum, the weight of evidence shows that significant learning requires cs mediation. Compared to the weak effects of uncs learning, csly mediated learning is both effortless and effective.

20. Experienced stability of contents.

cs percepts are impressively stable, given the very high variability of proximal physical stimuli that are nevertheless perceived as constant. Even abstract cognitions, such as beliefs, concepts, and personality patterns, are extremely stable over the adult lifetime.

21. Allocentricity. (Object-centeredness).

cs percepts generally have allocentric character (object-centered), though they are also shaped by unconscious egocentric maps of the parietal cortex. After one-sided parietal damage to the visual dorsal stream, the opposite side of the visual field disappears from csns. Such “visual neglect” shows that parietal maps do not support cs vision by themselves, though they are needed for ventral feature maps to become cs. They are said to be “contextual.”

22. csly-mediated decision making.

cs mediation is useful for making decisions, for goal pursuit and problem-solving. These tasks require cs moments, intertwined with uncs periods for routine and predictable processes.

Similarly, voluntary decision-making involves both cs and ucs moments. Essentially all cognitive tasks are mixtures, since each kind of process has its own pros and cons.

# Consciousness in philosophy and science

A clear boundary between philosophy and science is recent, starting about 1900. Historically, philosophers made important empirical as well as conceptual contributions. Descartes studied the optics of oxen eyes and the bilateral organization of the brain. William James summarized an immense body of experimental facts in his Principles of Psychology of 1890. The Asian tradition of experiential observations and mental praxes may date back to the Indus Valley culture of some 4,000 years ago.

However, with the rise of positivism around 1900, philosophers focused on analyzing words and arguments, and avoided empirical topics. Philosophers of mind tend to study the mind-brain question, with its three traditional perspectives of physicalism, mentalism and dualism.

Common sense alternates readily between public “physical” observations (e.g., the sight of an aspirin) and “mental” ones (e.g., noticing that an aspirin cured a subjective headache). The sight of a physical aspirin can be directly shared with others, but a fading headache cannot.

Modern philosophers tend to discuss csns from a subjective perspective. The major question therefore becomes "what is it like to be a certain cs being?"

For scientists, Karl Popper's distinction between empirically falsifiable and non-falsifiable claims continues to be fundamental. Empirical falsifiability is necessary for science.

# Selected references

• Baars, B. J. (1988). A Cognitive Theory of Consciousness. New York: Cambridge University Press. [Kindle edition, 2012].
• Baars, Bernard J., Stan Franklin and Thomas Zoega Ramsoy (2013) Global workspace dynamics: cortical “binding and propagation” enables conscious contents. Frontiers in Psychology, 4. doi: 10.3389/fpsyg.2013.00200
• Freeman, W. J. (2007). Indirect biological measures of consciousness from field studies of brains as dynamical systems. Neural Netw. 20, 1021–1031.
• Dehaene, S (2014) Consciousness and the brain: How the brain codes our thoughts. Penguin Group (USA).
• Edelman, G. M., and Tononi, G. (2000). A Universe of Consciousness: How Matter Becomes Imagination. New York: Basic Books Inc.
• Gaillard, R., Dehaene, S., Adam, C., Clémenceau, S., and Hasboun, D. (2009). Converging intracranial markers of conscious access. PLoS Biol. 7:e1000061. doi:10.1371/journal.pbio.1000061
• Panagiotaropoulos, T. I., Deco, G., Kapoor, V., and Logothetis, N. K. (2012). Neuronal discharges and gamma oscillations explicitly reflect visual consciousness in the lateral prefrontal cortex. Neuron 74, 924–935.
• Steriade, M. (2006). Grouping of brain rhythms in corticothalamic systems. Neuroscience 137, 1087–1106.