CONS: 00055

a0005 Neurobiological Theories of S Kouider, Laboratoire des Sciences Cognitives et Psycholinguistique, CNRS/EHESS/ENS-DEC, Paris, France ã 2009 Elsevier Inc. All rights reserved.

Cognitive Influences on Neurobiological Glossary s0010 Accounts g0005 Neural correlates of consciousness –They p0010 are defined by Christoph Koch as ‘‘The Regarding the influence of cognitive theories, minimal set of neuronal mechanisms or the majority of neurobiological accounts can be events jointly sufficient for a specific seen, in fact, as extensions of preexisting cognitive conscious percept or experience.’’ They theories (e.g., for instance global workspace the- allow to avoid the difficult problem of directly ories). Indeed, one of the main tasks exercised by looking for neural bases. neurobiologists in the last two decades has been to search for cerebral or neuronal equivalents to the g0010 Panpsychism – Reflects the philosophical functional elements constituting cognitive models doctrine that everything (in Greek, ‘pan’) has PROOF a mind (‘psyche’) and is therefore conscious. (e.g., the dorsolateral for volun- Some theories presented in this article tary control, or long range axons for connecting endorse a certain form of panpsychism in brain regions associated with ‘unconscious’ and which anything that transmits information is ‘conscious’ processing). Of course, many neuro- in a way conscious. biologists disagree with this approach. Conscious- ness, because it is a biological problem, should be g0015 The hard problem – It is the problem of explaining how and why we have the reframed the other way around, by focusing pri- subjective experience of consciousness. It is marily on its structural basis rather than relying on often contrasted with the easy problem, cognitive theories, often considered too specula- which consists of describing consciousness tive. Therefore, many neurobiologists consider that as the cognitive ability to discriminate, an ideal neurobiological science of consciousness integrate information, focus , etc. should focus on neural structures and mechanisms in order to understand how the organic matter constituting the brain creates consciousness.

The Hard Problem for a Neurobiology of s0015 s0005 Introduction VIER SECONDConsciousness p0005 Neuroscientists working on the issue of conscious- Yet, studying the neural mechanisms ‘leading to’ p0015 ness consider that it is a biological problem. They consciousness, trying to explain the ‘emergence’ assume that we will understand how and why we of consciousness, or focusing on how the brain are conscious by studying the cerebral and neuro- ‘creates’ consciousness, as often described in neu- nal features of the brain. These theories have robiological literature, sounds as if it involved an largely benefitedELSE from the recent advances in neu- immaterial soul that would magically arise from ropsychology, , and brain imaging the brain. This is not a new issue for philosophers in particular. However, neurobiologists have also who have also been wondering about the equiva- been influenced, on the one side, by cognitive lent mind–body problem since antiquity. More theories aimed at characterizing the psychological than a century ago, the contemporary ‘brain- determinants of consciousness, and on the other consciousness’ problem was well captured by side, by philosophical issues related to the mind– Thomas Huxley’s famous remark: ‘‘How it is that body problem. anything so remarkable as a state of consciousness

87 CONS: 00055

88 Neurobiological Theories of Consciousness

comes about as a result of irritating nervous tissue, assigning too much importance to the explanatory is just as unaccountable as the appearance of the gap might actually turn out to be counterproduc- djinn when Aladdin rubbed his lamp in the story.’’ tive and impedes rather than facilitates scientific The same issue applies today: understanding con- progress. Accordingly, neurobiologists have mostly sciousness as an ‘emergent’ property ‘arising’ from focused on the easy problem, considering that functional elements of the neurocognitive archi- this strategy will progressively get us closer to tecture, without falling on a dualistic position understanding the full issue. They extended the where consciousness lies somewhere outside of ‘contrastive analysis,’ originally put forward by the brain, poses serious epistemological difficul- Bernard Baars, from the cognitive to the neurobi- ties and leads to the so-called hard problem of ological domain. While this method initially con- consciousness. sisted in contrasting conscious and unconscious p0020 Indeed, many philosophers have concluded that processes in order to characterize their cognitive there is not one single problem, but actually two features, the neurobiological approach aims at problems that are faced by anyone trying to under- characterizing the neural features. A typical exam- stand consciousness: they distinguish between the ple consists in comparing the cerebral activity so-called easy problem and hard problem. In a when subjects are presented with subliminal sti- nutshell, the easy problem consists in relying on muli they cannot report (unconscious processing) objective measures of conscious processing in with that of visible stimuliPROOF they can report (con- order to explain how one is able to discriminate scious processing). sensory events, integrate information, report men- In other terms, the current first step in trying p0030 tal states, focus attention, etc. By contrast, the to understand the link between consciousness hard problem consists in explaining the first- and the brain consists in finding out which neural order, subjective nature of qualias and phenomenal components are specifically involved during con- states, the ‘what is it like to be conscious’ as well as scious processing, but importantly not during un- how and why we experience consciousness at all. conscious processing. Francis Crick and Christoph Addressing the hard rather than the easy problem Koch have coined the term ‘neural correlates of of consciousness constitutes an important episte- consciousness’ (NCC; see Glossary) in order to mological constraint put forward by philosophers. describe this epistemological approach. According In particular, contemporary philosophers such as to them, the best strategy for a neurobiological Joseph Levine and later have science of consciousness is to search for the NCC. argued that trying to resolve the hard problem Underneath this approach is the crucial principle leads to an ‘explanatory gap’ that science is unable that ‘correlates’ do not imply any relation of causality to cross, at least today. Indeed, they stress the fact between the occurrence of conscious mental events that it appears impossible to demonstrate that a and their associated physiological structure. Conse- neural structure leads to a conscious experience, quently, this strategy has the advantage of leaving while denying the reverse possibility.VIER In addition, SECONDaside, at least for the moment, the hard problem of given how different they are, reducing phenomenal finding the neural ‘bases’ of consciousness. states to neural states appears almost impossible. In the following sections, I will provide an over- p0035 view of the current most influential neurobiologi- cal theories of consciousness. These theories will be s0020 Looking for Neural Correlates, Not Neural Bases largely described in an independent manner, such ELSE that each of them can be understood individually, p0025 Should neurobiologists then give up on addressing that is, without having to frame it in the context of this issue? Most neurobiologists acknowledge the alternative accounts. Only later, in the section existence of a hard problem. However, they also labeled ‘Neurobiological standpoints on the hard endorse the principle that further scientific inves- problem’ will I evaluate their explanatory power by tigations will ultimately allow us to resolve it. confronting them in relation to some important Others explicitly deny the existence of a hard conceptual issues (e.g., dissociating access vs. phe- problem in the Chalmerian . For some, nomenal consciousness, dissociating attention vs. CONS: 00055

Neurobiological Theories of Consciousness 89

consciousness, panpsychism). I will conclude by remain to be demonstrated. I will further discuss emphasizing how promising these theories are in the speculative aspects of this theory later (see the getting us closer to resolve the issue of the hard section ‘Neurobiological standpoints on the hard problem. problem’). For now, I shall provide an overview of the core assumptions underlying this theory. In order to appreciate the specificity of the p0050 s0025 From Globalist to Localist Accounts of Consciousness Reentrant Dynamic Core theory, it is important to understand that, regardless of its explanation for p0040 Neurobiological theories of consciousness differ in consciousness, it offers an alternative view on brain many respects. One way to portray them in a coher- structures, considering the wiring of the brain into ent manner is to follow the large spectrum ranging neuronal assemblies as the result of variation and from globalist to minimally localist accounts. By selection mechanisms that are analogous to those globalist or localist I refer to the size of the brain underlying evolutionary theories. This macrolevel states that are assumed to be sufficient for conscious- account of brain development, formally developed ness (extended to large parts of the brain vs. focalized by Edelman in 1978, and called the theory of to specific and small brain areas, respectively). I will Neuronal Group Selection (also called the theory mainly restrict this review on the cerebral level and of Neural Darwinism) constitutes a key element exclude alternative theories that are more globalist, for understanding thePROOF development of conscious- in the sense that consciousness encompasses more ness. According to the Neuronal Group Selection than neural activity in the brain (e.g., theories pro- theory, the brain is assumed to be a selectionist posing that consciousness reflects modification in system in which variant groups of neurons are the electromagnetic field surrounding the brain, as selected over others in three steps. The first one, proposed by J. McFadden and by R. John), or, con- termed developmental selection, happens during versely, theories that are more localist, by focusing embryogenesis and early development, and it is on single neurons or even lower structural levels largely influenced by epigenetic factors. It consists (e.g., the quantum-level theory of microtubules by of several processes such as cell division, cell death, S. Hamerof and R. Penrose). Such theories remain extension, which has the consequence of connect- excessively speculative and unspecified to be inclu- ing neurons together into a large number of variant dedinaseriousreview,atleastforthemoment. neuronal circuits (labeled primary repertoires), and elimination, which targets unconnected neurons in particular. The second step is called experiential s0030 The Reentrant Dynamic Core Theory selection and lasts from early infancy through all p0045 The Reentrant Dynamic Core theory, proposed of the lifespan. It consists of reinforcing, through the by Gerald Edelman and Giulio Tononi, is argu- influence of behavior and experience with the envi- ably the most globalist account of consciousness. ronment, the synaptic connections of some variants Indeed, in this framework, consciousnessVIER is not SECOND to over others, leading to secondary repertoires of be localized in dedicated brain areas or with par- neurons. The final step, that of reentry, consists ticular types of neurons. Rather, it is the result of of the formation of massively parallel reciprocal dynamic interactions among widely distributed connections among distant maps of neuronal reper- groups of neurons in the entire thalamocortical toires, allowing them to exchange signals and be network (the ‘dynamic core’). This theory offers spatiotemporally coordinated. an interestingELSE tentative of unifying the hard and In the Reentrant Dynamic Core theory, con- p0055 easy problems, providing a neurobiological expla- sciousness results exactly from this mechanism of nation for qualias, and explaining both phyloge- reentry among distant groups of neurons within netic and ontogenetic aspects of the development the dynamic core of thalamocortical connections. of consciousness in humans and other species. The spatiotemporal coordination provided by reen- Yet, although this theory is very appealing, espe- try allows the binding of several elements into cially given its explanatory power, it is also highly a single and coherent object or event, providing a speculative, and based on several assumptions that solution to the binding problem. It also allows to CONS: 00055

90 Neurobiological Theories of Consciousness

explain why conscious experiences appear to be allows for a perceptual organization of the environ- unified. ment and whose characteristics have been presented p0060 Particularly important in this theory is the above, and higher-order consciousness that is reentry of information between groups of neurons possessed by humans, and which is related to lin- dealing with perceptual categorization (i.e., in guistic and symbolic mental activities. The latter posterior areas) and the more frontally located requires further reentry with additional brain systems responsible for value-category memory. regions such as those involved in language produc- Indeed, the latter will constrain the selectionist tion and comprehension. process by modulating or altering the synaptic In sum, this theory assumes that the brain is a p0065 connections within groups of neurons, as a conse- dynamic complex system in which consciousness quence of their influence on behavior (pleasure, emerges from the interactivity itself. Instead of pain, etc.). Another key aspect to this whole frame- involving top-down communication between ded- work is that groups of neurons are constantly in icated areas, the Reentrant Dynamic Core theory competition and their survival depends on creat- involves regions in the cortex in active communi- ing or reinforcing synaptic links with other groups, cation with one another and with associated nuclei such as to form large assemblies of reentrant neu- in the . Consciousness arises from the dif- ronal maps in the dynamic core. The victorious ferentiated integration of activity in these areas, as assembly (or ‘coalition’ in Crick and Koch’s terms) information is transmittedPROOF recurrently, with local will lead to consciousness, at least for a few hundred groups of neurons performing their specialized milliseconds, until a new coalition of neurons bypass and discriminatory function, while at the same it. Indeed, because the variant groups of neurons are time being unified with other neuronal groups of assumed to be degenerate, which means that differ- the dynamic core. Before concluding on the char- ent variants of neuronal assemblies can actually carry acteristics of this theory, it is important to point out the same function and have the same output, out that although the Reentrant Dynamic Core each integrated state in the dynamic core is followed can be considered as a globalist theory, in the by yet another and differentiated neural state in the sense that it can involve a large set of regions core. As such, coalitions of neurons leading to con- in the brain, it is also a gradual theory of conscious- sciousness are temporally transient by nature and ness that can result from minimal neural networks. widespread along variant regions of the whole dyn- Indeed, reentry among even a small number of amic core. Because interactions within the dynamic neurons, as far as it reflects differentiated integra- core are constantly moving, it explains the diversity tion, will induce consciousness to a certain degree. of consciousness, yet the constant integration thro- We will come back to this theoretical aspect below. ugh reentry explains the unity of consciousness. Note then that the victory of a coalition of neurons in leading to consciousness is not just a matter of its The Global Neuronal Workspace Theory s0035 size; rather it depends on itsVIER complexity, that is, SECOND its ability to generate at the same time an integrated The Global Neuronal Workspace theory proposed p0070 scene (i.e., appearing as a unitary event) and a differ- by Stanislas Dehaene with Lionel Naccache and entiated scene (i.e., which can be highly discrimi- Jean-Pierre Changeux is currently the most explicit nated from other scenes). Measures of complexity in and most functional account of the cerebral archi- relation to consciousness have been further devel- tecture underlying conscious access. Its function- oped byELSE Tononi in his recent Information Inte- ality has rendered this theory very popular in gration theory of consciousness (see the article neuroscience circles. Yet, as we shall see below ‘cognitive theories of consciousness’ in this volume). in the section ‘Neurobiological standpoints on According to Edelman and Tononi, these highly the hard problem,’ this theory has also been greatly discriminatory properties of neural complex sys- criticized for probing functionality at the price tems are the qualias that have been torturing philo- of sacrificing some phenomenological aspects of sophers, nothing more, nothing less! Finally, they consciousness. For now, let us focus on the main distinguish between primary consciousness that characteristics of this theory. CONS: 00055

Neurobiological Theories of Consciousness 91

p0075 This theory is a perfect example of the neuro- in direct interconnections, without requiring the biological extension of a cognitive theory, that of a use of workpace neurons and without involv- global workspace, originally proposed by Bernard ing consciousness. Another important aspect that Baars in 1988 (see the article ‘Cognitive theories of follows from this theory is that information com- consciousness’ in this volume). Dehaene’s theory puted in neural processors that do not possess assumes that neurocognitive architecture is com- direct long-range connections with workspace posed of two qualitatively distinct types of ele- neurons, will always remain unconscious. This ments. The first type is represented by a large idea is rooted in the work of Crick and Koch, network of domain-specific processors, in both postulating that neural activity in V1, because it the cortical and subcortical regions that are each does not project toward prefrontal neurons, does attuned to the processing of a particular type of not participate directly in visual consciousness. information. For instance, the occipitotemporal Note that for a mental object to become con- p0090 cortex is constituted of many such domain-specific scious, it is not sufficient that its activity gives processors, or ‘cerebral modules,’ where color pro- input to the global workspace. Two other condi- cessing occurs in V4, movement processing in tions have to be met. One is that the content of the MT/V5, face processing in the fusiform face area mental object must be represented as an explicit (FFA), etc. Although these neural processors can firing pattern of neuronal activity, that is, a group differ widely in complexity and domain specificity, of neurons that unambiguouslyPROOF indexes its relevant they share several common properties: they are attributes. A final and important condition is that triggered automatically (i.e., mandatorily, by oppo- the top-down amplification mechanisms mobilizing sition to voluntarily), they are encapsulated (their the long-distance workspace connections render the internal computations are not available to other content of consciousness accessible, sharpened, and processors), and importantly, they largely operate maintained. A mental object, even if it respects the unconsciously. two first conditions (explicit firing and accessibility p0080 Conscious access involves only the second type to workspace neurons) will still remain buffered in a of element, namely, the cortical ‘workspace’ neu- ‘preconscious’ (and thus a nonconscious) store until rons that are particularly dense in the prefrontal, it is attended to and its neural signal amplified. cingulate, and parietal regions. These neurons are Therefore, top-down attention in this framework characterized by their ability to send and receive is a necessary condition for consciousness. Whether projections to many distant areas through long- being conscious requires top-down attention con- range excitatory axons, breaking the modularity stitutes, as we will see below, one of the most of the and allowing the domain- debated aspects among neurobiological theories specific processors to exchange information in of consciousness. a global and flexible manner. The global work- space is thus a distributed neural system with The Duplex Vision Theory s0040 long-distance connectivityVIER that can potentially SECOND interconnect multiple cerebral modules through The Duplex Vision theory proposed by David p0095 workspace neurons. It offers a common communi- Milner and Melvyn Goodale postulates that visual cation protocol, by allowing the broadcasting of involves two interconnected, but dis- information to multiple neural targets. tinctive pathways in the visual cortex, namely, the p0085 One important aspect of this theory is that dorsal and the ventral stream. After being con- encapsulationELSE and automaticity are less rigid than veyed along retinal and subcortical (i.e., genicu- traditional modularist (i.e., Fodorian) accounts of late) structures, visual information reaches V1 and the cognitive system (see the article ‘Cognitive then involves two streams. The ventral stream theories of consciousness’ in this volume). Indeed, projects toward the inferior part of the temporal once a set of processors have started to commu- cortex and, according to this theory, serves to nicate through workspace connections, this multi- construct a conscious perceptual representation domain processing stream also becomes more of objects, whereas, the dorsal stream projects and more ‘automatized’ through practice, resulting toward the posterior parietal cortex and mediates CONS: 00055

92 Neurobiological Theories of Consciousness

the control of actions directed at those objects. of their attributes. In other terms, they can use Apart from these structural considerations, the vision for consciously perceiving objects in their two streams also differ at the computational and environment, but not for controlling real-time functional levels. On the one side, the ventral stream actions directed at those objects. On the other conveys information about the enduring (i.e., long- side, patients with damage to a ventral region lasting) characteristics that will be used to identify known as the lateral occipital complex are unable the objects correctly, and subsequently to link them to recognize everyday objects or even simple geo- to a meaning and classify them in relation to other metric forms, a deficit labeled ‘visual form agnosia.’ elements of the visual scene. Computing these Yet, such patients are strikingly efficient at grasping enduring characteristics involves relatively long objects correctly (for instance by opening their and costly computations. On the other side, the hand as a function of the object size, or by rotating dorsal stream can be regarded as a fast and online their hand according to the object orientation). visuomotor system dealing with the moment-to- Although such patients cannot consciously per- moment information available to the system, which ceive the object and even its visual attributes (size, will be used to perform actions in real time. shape, or orientation), they can use the same object p0100 It should be noted that this dissociation between attributes to control their object-directed actions. ventral and dorsal pathways has sometimes been One important consequence of this theory, p0110 misunderstood as equivalent to conveying the according to Milner andPROOF Goodale, is that because ‘what’ and ‘where’ information in the visual cortex, the processes performed by the dorsal stream are respectively. However, structural and spatial attri- very fast and largely automatic, they are largely butes can conjointly be used by both systems. For unconscious, while, by contrast, those conveyed by instance, visual information such as the size, geo- the ventral stream are assumed to constitute the core metrical structure, and location of a target object of conscious perception. The visual phenomenology might be computed both by the dorsal steam, in generated in ventral regions will in turn be trans- order to grasp and reach the object, and by the ferred to working memory components in order ventral stream in order to segregate the object to use information off-line, when the objects are from a complex visual scene. While the ventral not stimulating the visual system anymore. Accord- steam involves object perception by comparison ing to this theory, although we are typically aware with visual perceptual attributes stored in mem- of the actions we perform, no phenomenology is ory, the dorsal steam, because it is fast and updated associated with the visual information used by in real time, involves no storage of the visuomotor the dorsal stream to control those actions. Hence, attributes extracted from the object, nor the motor the neural computations performed in the dorsal program resulting from actions upon that object. stream remain quite inaccessible to consciousness. p0105 In other terms, while the ventral stream can be It is of note that when this theory was proposed p0115 seen as conveying vision for perception, the dorsal more than a decade ago, evidence from binocular stream is concerned with visionVIER for action. Both SECONDrivalry (in which conscious perception alternates systems work together in the production of adap- between two images, say a face and a house pre- tative behavior, but they can also function inde- sented to each eye) revealed a very strong correla- pendently as revealed by clinical studies. Indeed, tion with conscious perception in the ventral stream the development of this theory mainly results from (e.g., in the FFA for faces). Therefore, this evidence neuropsychological investigations, demonstrating misled many researchers at that time to deny the a doubleELSE dissociation between vision for percep- possibility of unconscious perception in the ventral tion and vision for action. On the one side, patients stream. However, Geraint Rees and colleagues with a lesion in the posterior parietal cortex, the found that patients with unilateral neglect, a deficit area terminating the visual dorsal stream, suffer in which they fail to pay attention and then report from ‘optic ataxia,’ a deficit in the control of reach- stimuli on half of their visual field, still exhibit ing and grasping objects. Despite this deficit, these FFA activity for faces in the neglected field. Since patients are perfectly able to verbally describe the neglected stimuli are perfectly reported when cued unreachable objects, either as a whole or in terms carefully, neglect is considered as an inability to CONS: 00055

Neurobiological Theories of Consciousness 93

report efficiently because of a lack of attention. Given and higher visual areas which, according to this the possibility that consciousness and attention might theory, lead to visual experience (i.e., phenomenal be distinct (see below), it still remains unclear consciousness, see below). The third and final whether. Rees’s finding reflected ventral neural activ- stage consists of ‘widespread recurrent processing,’ ity without consciousness or just without attention. which involves global interactions (similar to the Yet, more recent evidence with visual masking, global workspace model) and extends toward the obtained in my laboratory has revealed unconscious executive (i.e., the frontoparietal network) and lan- neural activity in ventral regions, including the FFA, guage areas. This final step also involves the atten- during subliminal face perception, thus clearly tional, executive, and linguistic processes that are demonstrating that the ventral stream is not exclu- necessary for conscious access and reportability of sively related to conscious perceptual processes. the . p0120 This type of evidence is problematic for the An interesting aspect of the Local Recurrence p0130 Duplex Vision theory, since this theory predicts theory is that it offers an explanation for the that conscious perception should be proportional difference between conscious and unconscious to neural activity in the ventral stream. Although perception in mechanistic rather than in architec- the possibility of unconscious ventral processing tural terms. Here, the distinction between sublim- was not taken into account in the original theory, it inal and conscious perception involves the same can be accommodated by assuming a threshold regions. Subliminal perceptionPROOF reflects the fact mechanism, as proposed by Zeki for the Minimal that the visual signal is fed forward to higher visual Consciousness theory reviewed below. However, areas without being fed back to early areas (for including this threshold leads the theory to lose instance a second stimulus replaces the first one in it former appeal, since consciousness is ‘only par- V1 and then prevents the setting up of recurrence tially’ correlated with activity in the ventral stream. between higher regions and V1). Another interest- Conversely, various recent evidences have shown ing aspect of this theory, although provocative and that the dorsal stream can, under some circum- speculative, is that consciousness should not be stances, be associated with the consciousness of defined by behavioral indexes such as the subject’s actions. Thus, although this theory might be effec- introspective reports. Instead, according to Lamme, tive for distinguishing the neural mechanisms that one should rely on neural indexes of consciousness, are responsible for vision for perception and vision one of which is neural recurrence. Indeed, Lamme for action, this dissociation might turn out to be is concerned with the question of defining phenom- orthogonal to the dissociation between conscious enological consciousness when a report is impossi- and unconscious processing. ble. According to Lamme’s theory, involvement of the second step (local recurrence) without involv- ing the third step (global recurrence) represents s0045 The Local Recurrence Theory VIER SECONDexactly this situation. We will come back to this p0125 The Local Recurrence theory put forward by aspect when discussing the neurobiological stand- Viktor Lamme is mostly concerned with vision points on the hard problem. for perception rather than action. It distinguishes In sum, the theory of Local Recurrence explic- p0135 between three hierarchical types of neural pro- itly stipulates that recurrent activity is sufficient cesses related to consciousness. The first stage for consciousness. Yet, one main difficulty with involves a ‘feedforward sweep’ during which the this theory is that is fails to take into account the informationELSE is fed forward from striate visual recurrent connections that exist between regions regions (i.e., V1) toward extrastriate areas as well that are not associated with consciousness (for as parietal and temporal cortices, without being instance between V1 and the thalamus). It remains accompanied by any conscious experience of the possible that consciousness involves local recur- visual input. Only during the second stage, involv- rence ‘between some specific cerebral components.’ ing the ‘localized recurrent processing,’ is the However, local recurrence cannot then be consid- information fed back to the early visual cortex. ered as a sufficient condition for consciousness It is these recurrent interactions between early anymore, since it requires the involvement of CONS: 00055

94 Neurobiological Theories of Consciousness

additional mechanisms for explaining why it only where it is postulated that microconsciousnesses applies to a restricted set of brain regions. are not only distributed in space, but also in time. A critical characteristic of the microconscious- p0150 The Microconsciousness Theory ness theory is that it considers these processing s0050 sites to be equivalent to perceptual sites, which p0140 The microconsciousness theory put forward by means that conscious perception of one specific Semir Zeki and colleagues is arguably the most attribute (e.g., color) is proportional to the strength localist account of the cerebral architecture under- of activity in its respective processing site (i.e., V4). lying consciousness. It is assumed in this theory According to this theory, a microconsciousness that instead of a single consciousness, there are associated with a processing site does not necessi- multiple consciousnesses that are distributed in tate top-down influences from higher (i.e., frontal) time and space. This theory has also been mainly areas although these regions might play a role in developed in the context of vision research and visual consciousness. Note that although the cor- reflects the large functional specialization of the relation between neural activity in a processing visual cortex. For instance, evidence from various site and conscious perception is predicted to be sources (clinical, brain imaging, etc.) have shown highly positive, it cannot be perfect in this theory. that while the perception of colors is associated Indeed, in order to take into account some clinical with neural activity in area V4, motion perception and experimental evidencesPROOF revealing unconscious reflects neural activity in MT/V5. In particular, processing, notably those obtained by Zeki him- neuropsychological investigations have demon- self, neural activity in a processing site must reach strated that the respective cerebral lesions in these a certain height for a conscious correlate to be two functional sites lead to dissociated forms of generated. Therefore, this theory does not deny conscious perception. Lesions in V4 lead to ‘achro- the existence of unconscious processing. In fact, matopsia,’ the inability to see the world in colors, a consequence of this postulate is that it obviates while motion remains intact, and conversely lesions the need to separate brain regions that are linked in MT/V5 result in visual ‘akinetopsia,’ the inabil- to consciousness from those that are linked to ity to perceive visual motion, while color percep- unconscious processing, as we saw above if the tion remains unaffected. Furthermore, because of Local Recurrence theory. The transition between the existence of direct connections between sub- unconscious and conscious perception reflects the cortical structures and MT/V5(i.e., without having crossing of a threshold ‘within’ the processing site, to be mediated by V1), patients with a lesion in V1 though the neural and behavioral characteristics of are unable to perceive objects, while they can still this transition remain to be specified. experience motion when these objects are moving. Note that this theory does not deny that the p0155 p0145 Zeki takes these findings as evidence that attributes of each processing site are bound together consciousness is not a unitaryVIER and singular phe- SECONDat one point. However, it assumes that binding is nomenon, but rather that it involves multiple con- a post-conscious process occurring ‘after’ con- sciousnesses that are distributed across processing sciousness of the specific attributes to be bound sites (also called essential nodes in this theory), together has taken place. This second step is which are independent from each other. Another termed ‘macroconsciousness.’ Although microcon- form of evidence in favor of this theory is that the sciousness involves only one perceptual attribute conscious perception of different attributes is not (e.g., color), macroconsciousness reflects the phe- synchronousELSE and can respect a temporal hierarchy. nomenal experience associated with the bound For instance, psychophysical measures have shown object (i.e., including its form, color, motion, etc.). that color is perceived a few tens of milliseconds It occurs higher and later in the hierarchy, and before motion, reflecting, according to Zeki, the depends upon the presence of the previous one. fact that neural activity during perception reaches In addition, Zeki proposes that there is also a third V4 before reaching MT/V5. This observation is and last level coined as ‘unified consciousness,’ congruent with the microconsciousness theory, reflecting a more global form of consciousness, CONS: 00055

Neurobiological Theories of Consciousness 95

which involves linguistic and communication skills. Neurobiological Standpoints on the s0055 This third level remains largely unspecified, but it Hard Problem is roughly equivalent to the brain regions leading to consciousness in the global workspace model. Now that we have seen the main characteristics of p0165 p0160 One main difficulty with this theory is that these competitive theories individually, the next any processing region in the brain should, at sections will describe how they face (or deny) first glance, constitute an NCC in the multiple- conceptual issues related to the hard problem of consciousness theory. As such, it remains unclear explaining conscious experience. why conscious perception is not associated with activity in most brain regions, including the cere- Relation to Access and Phenomenal s0060 bellum and subcortical regions, especially those Consciousness conveying visual signals (e.g., the Lateral Genicu- late Nucleus). Zeki takes this problem into account, The neuroscientific and philosophical issues of p0170 at least by admitting it, and tries to solve it by consciousness have never been so closely related. explaining that neural activity in a processing site A consequence of this close interplay is that concep- probably also necessitates the involvement of addi- tual issues raised by philosophers are progressively tional systems, in particular the reticular activating influencing neurobiological theories. Arguably, system maintaining arousal. Yet, the theory loses its the most influential issuePROOF in recent years has been force, since neural activity, in certain perceptual the potential distinction between phenomenal sites, is not a sufficient condition for microcon- and access consciousness proposed by Ned Block. sciousness anymore (not mentioning the fact that In short (and in the context of visual perception), the reticular system must be involved during this dissociative approach assumes that the NCC unconscious processing, since it also influences for phenomenal consciousness reflects the qualita- the subcortical regions). In addition, another diffi- tive subjective experience (i.e., the qualia) asso- culty for the Multiple Consciousness theory is that ciated with the percept. According to Block, the visual regions can lead to the binding of several phenomenological richness that one can experi- attributes in the absence of consciousness, as shown ence when seeing a complex visual scene goes far both by Dehaene and colleagues using visual mask- beyond what the observer can report. Hence, con- ing, and by Wojeiulik and Kanwisher in a patient scious access is assumed to reflect another NCC with a bilateral parietal damage suggesting that that is more linked to reportability, stimulus dis- the binding mechanisms that are supposed to lead crimination, introspection, etc. This distinction has to macroconsciousness can in fact operate in the been appealing to many neuroscientists in recent absence of consciousness. Therefore, empirical evi- years, and several research programs are currently dence contradicts the claim that binding has to be a investigating the possibility to dissociate different post-conscious process. Finally,VIER and maybe more SECONDNCCs for these two forms of consciousness. crucially, the inclusion of a threshold mechanism An important contrastive feature regarding p0175 between unconscious and conscious processing, the five theories reviewed above is whether they and hence the possibility of unconscious proces- accept or reject this dissociative approach. It turns sing, is far from being obvious in this framework. out that all of them, except the Global Neuronal This constitutes, as we saw above for the Duplex Workspace theory, have recently acknowledged Vision theory, an empirical constraint (i.e., there or even incorporated this distinction. In the has to beELSE a threshold even if its nature remains Reentrant Dynamic Core theory, the distinction unspecified) that implies a radical change in the between phenomenal and access consciousness is main message conveyed by these two theories (i.e., analogous to the one between primary conscious- consciousness is ‘only partially’ correlated with ness and higher-order consciousness, respectively. activity in the ventral/perceptual sites). In particu- In a recent extension of the Duplex Vision theory, lar, it then becomes difficult to understand the neural activity in the ventral stream has been frequent claim by Zeki that processing sites in the directly associated with phenomenal conscious- visual brain are also perceptual sites. ness by Goodale. Similarly, Zeki has recently CONS: 00055

96 Neurobiological Theories of Consciousness

linked micro- and macroconsciousness with the Relation to the Distinction between s0065 phenomenal consciousness of specific attributes Attention and Consciousness (colors, contrasts, etc.) and bound objects, respec- Although attention and consciousness have long p0185 tively, while unified consciousness is analogous to been considered to be similar, if not identical phe- access consciousness. In the Local Recurrence the- nomena, it is largely acknowledged today that ory, phenomenal consciousness has been a very they can be dissociated. Yet, a double dissociation central concept at the origin of the construction remains to be demonstrated. Indeed, since the of this framework. It is assumed to be a by-product seminal study by Naccache, Blandin, and Dehaene of local recurrent loops, while access conscious- showed that top-down attention modulates sub- ness involves additional widespread loops between liminal priming, there is now a general consensus posterior and anterior regions of the brain. Impor- regarding the fact that attention is independent tantly, all these theories assume that the NCC of consciousness and that subjects can attend to for access consciousness involves more or less the objects even if they do not consciously perceive network of brain regions constituting the workspace them. However, the reverse dissociation where in Dehaene’s theory. Therefore, the Global Neuro- consciousness is itself independent of attention nal Workspace theory is often considered as a remains highly debated. More precisely, the possi- restrictive theory of access consciousness, and has bility that subjects can be conscious of objects been criticized for confounding the subjective expe- without any attentionPROOF is at the center of a very rience of consciousness with the ‘subsequent’ exec- intense controversy. utive processes that are used to access its content. Neurobiologists acknowledging Block’s disso- p0190 p0180 Dehaene and his colleagues not only reject ciative approach actually assume that top-down this dissociation in terms of two NCCs, but they attention is driven by workspace regions and cor- also put doubts on its psychological reality. In respond to a central mechanism, which, along with particular, they deny the possibility of phenomenal language and working memory, define access con- consciousness without access and consider this sciousness. For instance, in the case of visual per- dissociation to be flawed for at least two important ception, the signal in the posterior visual regions reasons. First, following Larry Weiskrantz, they ass- (i.e., in the occipitotemporal cortex) is assumed ume that reportability is the only reliable index to be attentionally amplified in a top-down fashion of consciousness and thus stimuli that cannot be by the same widespread parietofrontal areas as reported/accessed are by definition not conscious the workspace regions in Dehaene’ theory. Thus, in any way. Second, following Kevin O’Regan and similarly to the possibility of phenomenal conscious- Alva Noe¨, they consider that apparent cases of a rich ness without access, all the neurobiological theories phenomenal experience without access actually seen above, except that of the Global Neuronal reflect the so-called illusion of seeing. For instance, Workspace, acknowledge the possibility of con- in ‘change blindness’ situations, observers are usu- VIER SECONDsciousness without attention. Some a priori support ally overconfident about their capacity of seeing an for the dissociative approach can be found in studies entire visual scene, while actually, they fail to notice showing that visible (i.e., supraliminal) though unat- when important modifications occur at unattended tended stimuli do not involve workspace regions, locations. As such, the illusion of phenomenal rich- but rather an increase of activity in posterior ness reflects the fact that observers think that regions, as I have recently evidenced myself. Yet, they canELSE see more than they actually do. Because as we will see below, it remains extremely difficult to of this overconfidence, it remains unclear how one demonstrate that there is any form of consciousness can empirically probe observers to describe the for visible but unattended stimuli. content of unattended or inaccessible perceptual Here also, Dehaene and colleagues disagree p0195 events. As one can see here, this conceptual issue of with this dissociation and consider, on the con- phenomenal consciousness without access is closely trary, that attention embodies conscious proces- linked to the empirical issue of consciousness with- sing. According to them, only attended stimuli out attention. This latter issue is addressed in the can be reported and are thus consciously next section. CONS: 00055

Neurobiological Theories of Consciousness 97

processed: demonstrating that an observer is con- It is interesting to note that since the recent p0205 scious of an unattended stimulus, without relying inclusion of the preconscious stage in the Global on any sort of report by the subject, seems extre- Neuronal Workspace theory in 2006, it is becom- mely difficult or even impossible. In other words, it ing harder to distinguish it from the Local Recur- appears that in order to assess consciousness of the rence theory on a purely structural basis. Indeed, a stimulus, one necessarily needs to direct the obser- tripartite taxonomy is provided in both accounts. ver’s attention on the stimulus, leading to the con- The initial stage reflects local feedforward activity clusion that consciousness without attention is an at the neural level and corresponds to unconscious illogical possibility! It is of note that a few attempts, (i.e., subliminal) processing, while the third stage notably by Koch and colleagues, have been made to involves a global form of recurrence in the brain demonstrate consciousness without attention. In and reflects conscious access. Yet, the intermediate particular, they have relied on situations termed stage is also structurally analogous in these two ‘near-absence of attention’ and where a stimulus theories, as it involves a local form of recurrence is presented in the periphery while the subject is in both cases. Rather, it is on the psychological performing the task on a central target. Under these dimension that these two theories diverge, Dehaene conditions, Koch and colleagues have found that considering this intermediate level to reflect a non- subjects can still consciously perceive the peripheral conscious stage, while Lamme takes it as reflecting stimulus, at least indistinctively. Yet, unfortunately,it phenomenal consciousnessPROOF before access. This remains impossible to demonstrate that there has not shows us how much neurobiological theories are been any residual attentional amplification in this still dependant on psychological (i.e., subjective) situation (additionally, the mere fact that this situa- indexes of consciousness. tion is called ‘near-absence of ’ and not ‘full-absence In sum, apart from proponents of the Global p0210 of attention’ appears to be highly symptomatic and Neuronal Workspace theory, there is an increasing suggestive of the difficulty to demonstrate con- tendency in most neurobiological accounts to con- sciousness without attention). sider that there is an intermediate level associated p0200 Importantly, the majority of neurobiologists with phenomenal consciousness without access. including Dehaene and colleagues, acknowledge Consequently, new indexes that do not rely on that the processing of supraliminal but unattended subjective reports and that mainly reflect neural stimuli has a special status, in-between the first mechanisms are being proposed. For instance, step of subliminal processing, and the last step of Lamme proposes that consciousness should be conscious access. However, they totally disagree on indexed by the observation of local recurrent loops. a crucial question which is, ‘whether there is any In Edelman and Tononi’s theory, consciousness is form of consciousness associated with this inter- indexed in terms of neural complexity reflected by mediate level of processing or rather whether it is differentiated integration. Before addressing these just another form of nonconscious perception.’ approaches in the next section, it is necessary to Proponents of the GlobalVIER Neuronal Workspace SECONDremind the reader that the dissociation between theory have been very explicit on this issue and consciousness with and without attention, similarly consider that this intermediate level involves just to the distinction between access and phenomenal another, more elaborated form of nonconscious consciousness, remains highly difficult to tackle processing. They termed it the ‘preconscious’ stage. experimentally. This difficulty is mainly due to the According to them, a stimulus reaching this precon- fact that subjective reports, which necessarily involve scious stateELSE becomes potentially accessible (it could access/attention components, remain so far the best quickly gain access to conscious report if it was index of whether someone is conscious. Unless a attended), but it is not consciously accessed at the more reliable index of consciousness is found, these moment, mainly because attention is maintained dissociations might possibly turn out to be immune away (e.g., by processing a concurrent stimulus as to scientific investigation. Here also, maybe we should in situations of attentional blink, inattentional blind- wait and see; or maybe new and less constrained ness, or change blindness). epistemological directions should be explored. CONS: 00055

98 Neurobiological Theories of Consciousness

s0070 Relation to Neurocognitive Panpsychism difficulty is that this doctrine leads to the con- clusion that any system displaying these specific p0215 As we have seen in the previous section, there is an increasing tendency to consider that subjec- core mechanisms will experience consciousness, tive reports cannot be trusted when indexing con- whatever implementation supports it and, more sciousness. Among the theories we have seen importantly, whatever its size. This fact relates to above, proponents of two theories in particular, Chalmers’ argument that even a system as small as a that of Local Recurrence and that of the Reentrant thermostat, because it recurrently shares informa- Dynamic Core (and in particular the resulting tion between a temperature sensor and a controller Information Integration theory put forward by for switching the heater on or off, will experience Tononi), have proposed alternative indexes that consciousness to some degree. Indeed, the theory of are supposedly more reliable than subjective rep- Local Recurrence leads to the fact that any two orts. Both theories consider consciousness as an neurons in interaction would be sufficient for con- emergent property of any system sharing specific sciousness. The Reentrant Dynamic Core theory core mechanisms of either recurrence or differen- would require a few more neurons, but not that tiated integration, respectively. Therefore the most many to generate consciousness (precisely nine reliable index of consciousness should, according neurons would be sufficient according to Herzog, to these accounts, reflect the quantification of Esfeld, and Gerstner). Although such possibilities these core mechanisms. In particular, these neural can hardly be totally rejected,PROOF it remains as unclear indexes are assumed to offer a more reliable esti- how they could ever be demonstrated. mation of phenomenal experience, well above the A second main issue lies in the paradox of trying p0230 poor estimation provided by subjective reports. to prove that neural indexes are more respectable because they supposedly probe phenomenal and p0220 This approach can be appealing since it could potentially lead to a quantification of conscious- not access consciousness. Indeed, although it is ness in nonhuman species, in preverbal infants, clear that neural indexes offer interesting possibi- and in artificial intelligence systems. Yet, it leads lities when report is impossible (for instance in to a form of panpsychism (see Glossary) that one cases of locked-in syndromes or for prelinguistic can term as ‘neurocognitive panpsychism.’ Both babies), they still cannot be taken as reflecting the Local Recurrence theory and the Reentrant more than conscious access. This principle follows Dynamic Core theory are leading to a similar, from the fact that neural indexes of any sort have though more restrictive panpsychist views: any to be validated by confronting them at one point system in the world can be conscious as long as it with some kind of report, hence with access and not shares information following the core mechanisms phenomenal consciousness. For instance, demon- that are specific to these theories. They constitute strating that recurrent processing is sufficient for neurocognitive versions of panpsychism, because consciousness initially requires consciousness to be these theories are primarilyVIER aimed at offering neu- SECONDindexed by probing whether the system is indeed robiological accounts of consciousness, while they conscious, and the best way to do so is still to rely on can also be extended to any cognitive system in the subjective reports in humans. Then, a neural index world that shares the same mechanical properties can only be validated when a correlation has been for information-processing. As such, both theories established with subjective reports. Assuming that follow the principle that consciousness arises when one can demonstrate that recurrence, even among informationELSE is transmitted among neurons inter- two neurons, correlates with some degree of con- acting along the core mechanisms of these theories. sciousness, this can be achieved only through report and that degree of consciousness will only p0225 Yet, there are several conceptual difficulties with these theoretical approaches (the term conceptual reflect a partial form of access, not phenomenal difficulties and not impossibilities is preferred here consciousness. Once again, the interesting aspects because none of these issues is impossible to resolve of this approach cannot be neglected, since it can per se; however, they might turn out to be extremely allow to subsequently probe whether the core mech- difficult to demonstrate as well). A first conceptual anisms represented by this index can be found in CONS: 00055

Neurobiological Theories of Consciousness 99

systems or species that cannot report, suggesting that We have seen that, actually, consciousness can they‘mightprobably’beconscious.However,neural hardly be restricted to neural activity in the ventral indexes cannot be taken as reflecting a superior or stream, nor is it simply reflecting recurrent activity richer phenomenal form of consciousness since they or minimal activity in processing sites. It remains can only be validated through reports. In other terms, unclear whether consciousness reflects or does not these theories paradoxically emphasize that their reflect differentiated integration within a dynamic core mechanisms do not reflect access consciousness, core, as emphasized by Edelman and Tononi’s while actually only this form of consciousness can speculative, but powerful theory, or whether it be used to demonstrate their validity. simply reflects activity in workspace regions, as p0235 A third conceptual issue is that even if it turns outlined by Dehaene’s well-specified, but appar- out that one of these neural indexes turns out to be ently restrictive account of conscious experience. perfectly correlated with consciousness, and thus One main difficulty in disentangling this issue becomes a perfectly reliable measure of conscious- and contrasting these theories, is that it remains un- ness, then one might still ask whether we have known whether neurobiological frameworks should made any progress. This issue is less problematic be restricted to conscious access and subjective and might turn out to be resolved in the future. reports, or whether they should extend to other However, it is worth mentioning, since it remains (i.e., inaccessible, unattended, phenomenal) forms unclear whether we are searching in the right of consciousness. Therefore,PROOF future research will direction. Indeed, although neural indexes might unavoidably be bound to decide whether neuro- turn out to be very useful in many circumstances, biological accounts should take into account or reject they do not escape the limitation of NCCs. They the hard problem of understanding a subjective form would still primarily inform us on whether someone of conscious experience that cannot be simply (or something) is or is not conscious. Yet, they would defined as conscious access. Thus, facing the hard not directly help us understand why they are con- problem will probably not be any more a main issue scious. If it turns out that recurrence or differen- on its own. Instead, it will be the problem of deciding tiated integration are sufficient for consciousness, whether there is a hard problem at all, leading we would still have to accept that consciousness is to a ‘super-hard problem’ of consciousness and obvi- ‘emerging’ from the brain each time these core ously complicating the whole issue. Whether we mechanisms come into play. This issue is more should stay focused on variants of the NCC until moderate because once we have found a perfect the solution pops out and reduces the hard problem neural index of consciousness, it might get us to an easy one, or whether we actually need new laws closer to the explanatory gap. Yet, it remains (of physics, life, or anything else) to cross the explan- unclear how it will allow us to cross that gap. atory gap, only time will tell. Yet, although none of the neurobiological theories provided so far have been able to provide a ‘conclusive’ explanation VIER SECONDregarding this matter, the search for a neurobio- s0075 Conclusion logical explanation of consciousness still constitutes one of the most exciting challenges of contemp- p0240 By the end of this article, one might wonder where orary science. we stand with regard to the original issue of how the brain leads to consciousness. Over the last two See also: (00030); The Neurochemistry of Conscious- decades, dozensELSE of theories of the NCC have been ness (00056). proposed (leading David Chalmers to ironically speak about an ‘NCC zoo’) and a kind of trial- and-error strategy has been applied to evaluate Suggested Readings them, and in most cases, to reject them one by one. Here we have focused more specifically on Baars BJ (1988) A Cognitive Theory of Consciousness. New York: Cambridge University Press. five recent neurobiological frameworks that are Block N (2005) Two neural correlates of consciousness. the most popular accounts in this field of research. Trends in Cognitive Science 9: 46–52. CONS: 00055

100 Neurobiological Theories of Consciousness

Chalmers D (1996) The Conscious Mind. New York: Oxford Blackwell Companion to Consciousness, pp. 616–627. University Press. Oxford: Blackwell. Crick F and Koch C (1990) Towards a neurobiological theory Koch C (2004) The Quest for Consciousness: of consciousness. Seminars in Neuroscience 2: 263–275. A Neurobiological Approach. Denver, CO: Roberts. Dehaene S, Changeux JP, Naccache L, Sackur J, and Koch C and Tsuchiya N (2007) Attention and consciousness: Sergent C (2006) Conscious, preconscious, and Two distinct brain processes. Trends in Cognitive subliminal processing: A testable taxonomy. Trends in Sciences 11: 16–22. Cognitive Sciences 10: 204–211. Kouider S and Dehaene S (2007) Levels of processing during Dehaene S and Naccache L (2001) Towards a cognitive non-conscious perception: A critical review. Philosophical neuroscience of consciousness: Basic evidence and a Transactions of the Royal Society of London B 362: workspace framework. 79: 1–37. 857–875. Edelman GM and Tononi G (2000) A Universe of Lamme VA (2006) Towards a true neural stance on Consciousness: How Matter Becomes Imagination. consciousness. Trends in Cognitive Sciences 10: New York, NY: Basic Books. 494–501. Gallese V (2007) The ‘‘conscious’’ dorsal stream: Embodied Tononi G (2004) An information integration theory of simulation and its role in space and action conscious consciousness. BMC Neuroscience 5: 42. . Psyche 13(1): 1–20. Zeki S (2007) A theory of micro-consciousness. In: Goodale M (2007) Duplex vision: Separate cortical Velmans M and Schneider S (eds.) The Blackwell pathways for conscious perception and the control of Companion to Consciousness, pp. 580–588. Oxford: action. In: Velmans M and Schneider S (eds.) The Blackwell. PROOF Biographical Sketch

Sid Kouider is a cognitive neuroscientist working at the Ecole Normale Supe´rieure (Paris, France) on the neurobiological and psychological foundations of consciousness. His work focuses on contrasting conscious and unconscious processes, both at the psychological and neural level, using various behavioral and brain imaging methods. Recently, he extended this line of research to study the neural correlates of consciousnessVIER in prelinguistic SECOND babies.

ELSE