chapter 26 neurodynamics of consciousness 26 neurodynamics of consciousness diego cosmelli,...

P1: KAE0521857430c26 CUFX049/Zelazo 0 521 85743 0 printer: cupusbw November 6, 2006 15 :57

C H A P T E R 2 6

Neurodynamics of Consciousness

Diego Cosmelli, Jean-Philippe Lachaux, and Evan Thompson

Abstract

One of the outstanding problems in the cog-nitive sciences is to understand how ongo-ing conscious experience is related to theworkings of the brain and nervous system.Neurodynamics offers a powerful approachto this problem because it provides a coher-ent framework for investigating change, vari-ability, complex spatiotemporal patterns ofactivity, and multiscale processes (amongothers). In this chapter, we advocate a neu-rodynamical approach to consciousness thatintegrates mathematical tools of analysis andmodeling, sophisticated physiological datarecordings, and detailed phenomenologicaldescriptions. We begin by stating the basicintuition: Consciousness is an intrinsicallydynamic phenomenon and must thereforebe studied within a framework that is capa-ble of rendering its dynamics intelligible. Wethen discuss some of the formal, analyticalfeatures of dynamical systems theory, withparticular reference to neurodynamics. Wethen review several neuroscientific propos-als that make use of dynamical systems the-ory in characterizing the neurophysiologi-

cal basis of consciousness. We continue bydiscussing the relation between spatiotem-poral patterns of brain activity and con-sciousness, with particular attention to pro-cesses in the gamma frequency band. Wethen adopt a critical perspective and high-light a number of issues demanding furthertreatment. Finally, we close the chapter bydiscussing how phenomenological data canrelate to and ultimately constrain neurody-namical descriptions, with the long-term aimbeing to go beyond a purely correlationalstrategy of research.

The Intuition

The central idea of this chapter is the notionof dynamics – the dynamics of neural activ-ity, the dynamics of conscious experience,and the relation between them.

Dynamics is a multifarious concept. In anarrow sense, it refers to the change a cir-cumscribed system undergoes in some time-dependent descriptive variable; for example,a neuron’s membrane voltage (Abarbanel& Rabinovich, 2001). In a broad sense,

72 9


730 the cambridge handbook of consciousness

dynamics indicates a field of research con-cerned with non-linear dynamical systems.Such systems range from mathematicalmodels to experimental problems to actualconcrete world systems (Van Gelder, 1999).Finally, in a more intimate sense, dynamic-srefers to the temporal nature of our obser-vations themselves and thus to our consciousexperience and how it is deployed in time(Varela, 1999). The interplay of these dif-ferent senses of dynamics is at the heart ofthis chapter.

What exactly are the properties ofdynamical systems, and why are they ofinterest in relation to consciousness? Theentirety of this chapter is concerned withthis question, but to begin addressing it, wewish, in this introductory section, first to givean overview of the basic intuition underlyingdynamical approaches to consciousness.

Briefly stated, a complex, non-lineardynamical system can be described, at anytime, by a position in a high-dimensionalstate space (Nicolis & Prigogine, 1989). Then coordinates of such a position are the val-ues of the set of n variables that define thesystem. This position changes in time andthus defines a trajectory, which will tend toexplore a subregion of the total state space.One can then measure the distance betweenany two points of the trajectory and showthat under certain circumstances the tra-jectory can exhibit spontaneous recurrence:Small portions of the state space will beexplored over and over, but never along theexact same path. When perturbed by exter-nal events, such a system will change its tra-jectory in a way that is never quite the sameand that depends on its position in the statespace at the time of the perturbation.

Given this feature, plus the system’sextreme sensitivity to initial conditions, thesystem’s response to perturbations will beunpredictable in practice. It is thereforequite difficult to control such a system andconstrain its movement along a predefinedtrajectory. For example, in the case of chaoticsystems, such control involves applying acontinuous succession of carefully chosen,delicate inputs, brute force usually being

inefficient (think of what happens when youtry to catch a fish out of water; Garfinkel,Spano, Ditto, & Weiss, 1992 ; Schiff et al.,1994). In general terms, there will be acertain degree of dissociation between theobserved behavior of these systems and thepatterns of external constraints that can beimposed on them. In other words, these sys-tems exhibit a certain degree of autonomy:When external perturbations cease, the sys-tem goes on; when external perturbationsbecome stationary, the system does not.Somehow we intuitively recognize such sys-tems as animated or alive, in contrast to sim-pler systems that respond in a linear and pre-dictable way to external control (e.g., a stonethat flies twice the distance when throwntwice as strong).

Thus, such systems exhibit an intrinsicvariability that cannot be attributed to noise,but appears to be constitutive of their func-tioning. Moreover, in the case of certaintypes of complex dynamical systems, onecan reveal a characteristic spatiotemporalbalance of functional segregation and coop-erative integration.1 This balance dependson the actual architecture of the system(its internal connectivity, for example), andis revealed in the transient establishmentof distributed couplings among separatedsubsystems that in themselves present localencapsulated dynamics. Finally, some ofthese systems display what has been termed‘self-organization’; that is, the emergence ofcollective coherent behavior starting fromrandom initial conditions. This last feature,although not necessary for a system tobe considered dynamical, has proven espe-cially interesting when dealing with bio-logical phenomena (Haken, 1983 ; Kelso,1995).

The brain is a major case in point.The nervous system is a complex dynam-ical structure, in which individual neuronshave intrinsic activity patterns and cooper-ate to produce coherent collective behav-ior (Llinas, 1988). The explosion of neu-roimaging studies in the last 15 years, as wellas the substantial amount of data producedby electrophysiological techniques since the


neurodynamics of consciousness 731

beginning of the 20th century, has shownthat the brain is never silent, but alwaysin a state of ongoing functioning (Wicker,Ruby, Royet, & Fonlupt, 2003). The ner-vous system has a domain of viability, ofallowed functioning, but within this domainit explores a multiplicity of possible statesin a recurrent, yet always changing manner(Palus, 1996). Incoming events are not suffi-cient to determine the system’s behavior, forany incoming event will change the system’sactivity only as a result of how the system,given its current activity, responds to thatevent (Engel, Fries, & Singer, 2001).

If we now follow the thread of dynam-ics back to our own conscious experi-ence, we can immediately notice that ourconsciousness manifests subjectively as akind of continuously changing or flowingprocess of awareness, famously called the‘stream of consciousness’ by William James(1890/1981). Our experience is made up ofrecurring perceptions, thoughts, images, andbodily sensations; yet, however similar theseevents may be over time, there is alwayssomething new to each one, something ulti-mately unpredictable to every forthcomingmoment. We can try to plan our day asstrictly as we want, but the wanderings ofour minds and how we react to the encoun-ters we have in the actual world are thingswe cannot fully control. There seems to bean endogenous, spontaneous, ongoing flowto experience that is quite refractory toexternal constraints (Hanna & Thompson,2003). Indeed, this dissociation can easilybe made evident from the first-person per-spective. If you sit down and close the win-dows, turn off the lights, and close youreyes so that external stimulation is greatlyreduced for you, there is nevertheless stillsomething going on subjectively in you, withan apparent temporal dynamics all its own.2

Furthermore, at any moment, conscious-ness appears diverse, complex, and rich withmultiple, synchronous, and local contents(images, expectations, sounds, smells, kines-thetic feelings, etc.), yet it seems to holdtogether as a coherent and globally organizedexperience.

This intuitive convergence of complexdynamical patterns in experience and inbrain activity is highly suggestive. It sug-gests that the framework of dynamical sys-tems theory could offer a valuable way ofbridging the two domains of brain activ-ity and subjective experience. If we wish tostudy the neurobiological processes relatedto consciousness, then we must provide adescription of these processes that is (some-how) compatible with the dynamics of livedexperience. On the other hand, dynami-cal aspects of experience might serve asa leading clue for uncovering and track-ing the neurobiological processes crucial forconsciousness.

In the rest of this chapter, we explore thisguiding intuition through a discussion of thefollowing topics: formal dynamical systems,neurobiological theories based on dynamicalsystem principles, and the attempt to dis-tinguish dynamical structures within expe-rience that can constrain how we study theneurobiological basis of consciousness.

Neurodynamics

Dynamical Systems

Dynamical cognitive science has beendefined as “a confederation of researchefforts bound together by the idea that nat-ural cognition is a dynamical phenomenonand best understood in dynamical terms”(Van Gelder, 1999, p. 243). Within this con-federation, the job of the neurodynamicist isto model the neural basis of cognition usingthe tools of dynamical systems theory. Thus,the first thing we need to do is to define moreprecisely the notion of a dynamical system.

A dynamical system is a collection ofinterdependent variables that change intime. The state of the system at any timet is defined by the values of all the vari-ables at that time; it can be represented by aposition in an abstract ‘state space’, whosecoordinates are the values of all the vari-ables at t. The system’s behavior consists oftransitions between states and is described



geometrically by a trajectory in the statespace, which corresponds to the consecutivepositions the system occupies as time passes.

At a first level of complexity, in thecontext of neurodynamics, we can thinkof the variables as being the membranepotentials of each individual neuron of thenervous system.3 These membrane poten-tials are obviously interdependent. Thus, atthis level, the state of the nervous system atany time t would be defined by the value ofall the membrane potentials at time t.

Although a dynamical system, in the mostgeneral terms, is any system that changes intime, dynamical systems theory gives spe-cial attention to non-linear dynamical sys-tems. The behavior of such systems is gov-erned by non-linear equations; in otherwords, some of the mathematical func-tions used to derive the system’s presentstate from its previous states and possi-ble external inputs are non-linear functions(for neurobiological examples, see Abar-banel & Rabinovich, 2001; Faure & Korn,2001). Non-linearity can endow the sys-tem with certain interesting properties. Forexample, when convection cells in a horizon-tal water layer are submitted to a thermalgradient above a critical value, the motionsof billions of molecules spontaneously orga-nize into long-range correlated macroscopicstructures (Chandrasekhar 1961, cited in LeVan Quyen, 2003 , p. 69; see also Kelso,1995). Such properties of non-linear systemsled in the 1970s to an increased interest inthe mathematics of dynamical system the-ory (Nicolis & Prigogine, 1977).

Several elements condition the behav-ior of a dynamical system during a givenwindow of observation. First, the system’sbehavior is conditioned by the values of a setof so-called order parameters. By definition,these parameters determine the exact math-ematical equations that govern the system.This set of parameter values is a functionof the architecture of the system (e.g., thesynaptic weights between neurons), factorsexternal to the system (e.g., outside temper-ature), and so on. These parameters cannotnecessarily be controlled, and their dynam-ics is slower than that of the system itself.

They can be considered as constant duringthe given window of observation, but arepotentially variable across different observa-tion periods. Their dynamics thus contrastswith that of the external inputs, which canhave a dynamics as fast as that of the sys-tem. The governing equations of the systemare also a function of these inputs, but thetemporal evolution of the external inputscannot be predicted from those equations(otherwise they could be considered as statevariables of the system). Finally, all real sys-tems include a noise component, which alsocounts as a factor in the governing equationsand thus affects the trajectory of the system.

Neurodynamics and the DynamicalApproach in Neuroscience

Neurodynamics emerged from the proposal,which can be traced back to Ashby in the1950s (Ashby, 1952), that the nervous sys-tem can be described as a non-linear dynam-ical system. Although simple in appearance,this proposal deserves some attention: Whatdoes the nervous system look like from adynamical point of view, and why is it non-linear? The majority of the dynamical mod-els of the nervous system describe the tem-poral evolution of the membrane potentialsof neurons (Arbib, 2002). The behavior ofany neuron of the system is a function ofboth its own history of activation and thehistory of activation of every other neuron,thanks to the intrinsic connectivity of thenervous system. The precise influence ofa given neuron on a second one is deter-mined by the weight of the synapse that linksthem. Thus, the overall synaptic pattern inthe nervous system provides the main set oforder parameters in such models (see Arbib,2002).4 To this desciption, it must be addedthat the system is not isolated, but under theconstant influence of external sensory inputsthat shape the behavior of peripheral sensoryneurons.

There are many models available for themathematical functions that link the mem-brane potentials of individual neurons to thehistory of the larger system and to the exter-nal inputs (Arbib, 2002). At this point, it



is sufficient to state that these functions arenon-linear and that this is the reason whythe nervous system is described as a (spa-tially extended) non-linear dynamical sys-tem (for reviews, see Faure & Korn, 2001;Korn & Faure, 2003).5

Chaos in the Brain

As Le Van Quyen (2003 , p. 69) notes,there was little echo to Ashby’s original pro-posal to view the nervous system as a non-linear dynamical system, mostly becausethe appropriate mathematical methods andcomputational tools to pursue this proposalwere lacking at that time. The real boost toneurodynamics came later, in the 1980s and1990s, with the widespread emergence in thescientific community of interest in the prop-erties of chaotic systems.

Chaotic systems are simply non-linearsystems, with their parameters set so thatthey possess an extreme sensitivity to ini-tial conditions. Such sensitivity means thatif one changes the initial position of the sys-tem in its state space, however slightly, thesubsequent positions on the modified trajec-tory will diverge exponentially from whatthey would have been otherwise. Given thissensitive dependence on initial conditions,combined with the impossibility of deter-mining the present state of the system withperfect precision, the future behavior of achaotic system is unpredictable. The systemthus appears to have an inherent source ofvariability, for it will never react twice inthe same way to identical external pertur-bations, even in the absence of noise.6

The possible existence of ‘chaos in thebrain’ sparked much speculation and excite-ment. There were two related matters ofdebate: (i) whether the nervous system isactually chaotic (or whether there are sub-systems in the brain that are; Faure & Korn,2001; Korn & Faure, 2003) and (ii) whatuse the nervous system could make of suchchaotic behaviors (Faure & Korn, 2001; Korn& Faure, 2003 ; Skarda & Freeman, 1987;Tsuda, 2001).

The second question proved to be theeasier one. One property of chaotic systems

is that their dynamics can organize aroundthe presence of ‘strange’ attractors. A strangeattractor is a pattern of activity that capturesnearby states (Arbib, Erdi, & Szentagothai,1997): It occupies a subregion of the statespace, a manifold, and if the trajectory of thesystem comes into contact with this man-ifold, the trajectory will stay on it subse-quently, in the absence of external pertur-bations. The precise number and shapes ofthe attractors are determined by the param-eters of the system, such as the intrinsicconnectivity of the nervous system. Whichparticular attractor captures the system isdetermined by the system’s initial position.When the parameters define several strangeattractors, then there can be associationsbetween certain initial positions in the statespace and certain attractors (Tsuda, 2001).This association is the basis for chaos-basedperceptual systems: For example, in a com-mon non-linear model of olfactory process-ing (as reviewed in Korn & Faure, 2003),each odor is represented by a specific attrac-tor, such that when confronted with slightlydifferent olfactory stimuli, the trajectory ofthe system will converge onto the sameattractor, if the stimuli actually correspondto the same odorant.7 Thus, in this model,perception is based on several coexistingattractors in a multistable system. An addi-tional and important feature is that externalperturbations to the system can make thesystem jump from one attractor to another;therefore, chaotic systems should not bethought of as ‘static’ and unreactive to theenvironment. Moreover, chaotic systems canbe controlled; that is, they can be ‘forced’to stay within specific portions of the statespace via external perturbations. It mustimmediately be added, however, that theterm ‘forced’ is misleading, for the externalperturbations are nothing like brute force;rather, they must be thought of as like a seriesof subtle touches, carefully chosen to adaptto the system’s dynamic properties.

A sobering thought is that it is not clearwhether the activity of the nervous sys-tem, considered as a whole, is chaotic. Onerequirement for a system to be chaotic isthat its trajectory in the state space be



constrained within geometrical structuresthat have a lower dimension than the spaceitself (this requirement is needed mainly todistinguish between chaotic and stochasticprocesses; Wright & Liley, 1996). Unfortu-nately, the behavior of the nervous systemcannot be observed directly in its actualstate space, but only via limited sets of mea-surements that are crude projections of itsactual state (like the electroencephalogramor EEG, which retains only the average activ-ity of millions of neurons; Korn & Faure,2003).8

Nevertheless, the debate over the chaoticnature of brain activity proved productiveand brought out of the shadows some ideascrucial to neurodynamics. For instance, cen-tral to neurodynamical thought is the ideathat the variability of neural activity maybe an integral part of the nervous sys-tem’s dynamics. This notion is orthogonalto a number of traditional (and still largelydominant) approaches in neuroscience thatattribute this variability to ‘meaninglessnoise’. As a case in point, most brain imag-ing studies try to get rid of the variabil-ity of the neural activity by averaging brainrecordings over multiple repetitions of thesame process.9 These averaging proceduresmost likely give an oversimplified view ofbrain dynamics. In the near future, neurosci-entists will undoubtedly have to go to thetrouble of making sense of the neural vari-ability by finding its experiential and behav-ioral correlates. Fortunately, new approachesalong these lines are emerging, such as try-ing to understand the brain response to sen-sory stimulation in the context of the brain’sactive state preceding the stimulation andthus in relation to an active ‘baseline’ of neu-ral activity that is far from neutral (Lutz,Lachaux, Martinerie, & Varela, 2002 ; Engelet al., 2001)

This shift in the focus of brain imagingshould not be underestimated: The brain’sreaction is no longer viewed as a passive‘additive’ response to the perturbation, butas an active ‘integration’ of the perturbationinto the overall dynamics (Arieli, Sterkin,Grinvald, & Aertsen, 1996). In other words,the processing of an incoming sensory stimu-

lation is no longer viewed as the simple trig-gering of a systematic, prespecified, chain ofneural operations that would unfold inde-pendently of the brain’s current activity, asin a computer algorithm. For this reason,the neurodynamical approach is often pre-sented as a sharp alternative to the computermetaphor of the brain (Freeman, 1999a;Kelso, 1995 ; Van Gelder, 1998).

Self-Organization and the Emergenceof Spatiotemporal Patterns

As Crutchfield and Kaneko (1987) note,dynamical system theory has developedlargely through the study of low-dimen-sional systems, with no spatial extension. Tobe useful for neuroscience, however, dynam-ical system theory needs to consider thespecial properties conferred on the nervoussystem by its spatial extension.

Fortunately, there has been a recent coin-cidence between, on the theoretical side,the development of a theory of large-scale non-linear systems and, on the exper-imental side, the advent of multielectroderecordings and imaging techniques to mapprecisely the electrical activity of entire pop-ulations of neurons. This coincidence hasled to renewed interest, in the biologicalcommunity, in large-scale models of neuralactivity.

The study of large, spatially extendednon-linear systems is a field in itself, in whichthe interest in attractors shifts to the relatedone of spatiotemporal patterns. (We rec-ommend the reader spend a few minuteslooking for pictures of ‘cellular automata’with the Google image search engine tosee some beautiful examples of spatiotem-poral structures.) As a result, the neurody-namical community is now becoming lessfocused on chaos and more focused on theproperties of self-organization in non-linearsystems, and particularly the formation oftransient spatiotemporal structures in thebrain.10 As noted by Freeman, in the pref-ace to the second (electronic) printing of hisseminal book, Mass Action in the NervousSystem (Freeman, 1975): “The word ‘chaos’has lost its value as a prescriptive label and



should be dropped in the dustbin of history,but the phenomenon of organized disorderconstantly changing with fluctuations acrossthe edge of stability is not to be discarded”(Freeman, 2004).

Spatiotemporal structures are ubiqui-tous in the brain. Apart from the obviousphysical construction of the system, theycorrespond to the emergence of transientfunctional couplings between distributedneurons. For a given period of time, the activ-ity of a set of neurons shows an increasedlevel of statistical dependency, as quanti-fied, for example, by mutual information.11

In pioneering work, Freeman (reviewed inFreeman, 2000a) observed spatiotemporalactivity patterns in the olfactory bulb andinterpreted them within the framework ofdynamical system theory. In an influen-tial theoretical paper with Skarda, he pro-posed that sensory information was encodedin those patterns (Skarda & Freeman,1987).

The classic example of spatiotemporalstructures in the brain is the Hebbian rever-berant cell assembly, which Hebb (1949)hypothesized to be the basis for short-termmemory (see Amit, 1994).12 (This notion isalso closely related to Varela’s [1995] idea ofresonant cell assemblies, described below.)Reverberant cell assemblies are labile setsof neurons that transiently oscillate togetherat the same frequency, at the level of theirmembrane potential. They are the best-studied spatiotemporal structures in thebrain. Indeed, the cortex has sometimesbeen modeled as a lattice of coupled oscil-lators – in other words, as a juxtapositionof reverberant cell assemblies. One advan-tage of such models is that the behavior ofoscillator lattices has been abundantly inves-tigated, mainly using numerical simulations(see Gutkin, Pinto, & Ermentrout, 2003 ;Kuramoto, 1984 ; Nunez, 2000; Wright &Liley, 1996).

The formation of spatiotemporal struc-tures in such systems often takes the form ofphase-synchronization patterns between theoscillators (Le Van Quyen, 2003).13 In thebrain, phase synchronization of large pop-ulations of neural oscillators can produce

macroscale oscillations that can be pickedup by mesoscale recordings (such as localfield potentials) or macroscale recordings(such as an EEG). For this reason, syn-chronous oscillations have been the easiestform of spatiotemporal structure to measurein the brain and not surprisingly, the first oneto be observed (Bressler & Freeman, 1980;Gray, Konig, Engel, & Singer, 1989); seealso the discussion of functional connectivitybelow).

One reason why resonant cell assem-blies in particular, and spatiotemporal struc-tures in general, are so appealing to neuro-science is because they provide a flexible andreversible way to bind together distributedneurons that may be primarily involved invery different functional processes. This typeof binding has three fundamental features:(i) the ability to integrate distributed neuralactivities (integration); (ii) the ability to pro-mote, and virtually extract, one particular setof neural activities above the rest of the brainactivity (segregation); and (iii) the capacity toevolve easily through a succession of flexi-ble and adaptive patterns (metastability). Forexample, one resonant assembly could tran-siently bind together the different popula-tions of neurons involved in analyzing theshape, color, and motion of a visual object,and this temporary assembly would con-stitute a neural substrate for the transientperception of a visual object. This idea isthe starting point of a very active stream ofresearch that we discuss later in this chap-ter (for an overview, see Roskies, 1999).Some authors have even proposed that everycognitive act corresponds to the formationof such a transient spatiotemporal pattern(Varela, 1995).

In summary, dynamical system theoryproposes a precise framework to analyzethe spatiotemporal neural phenomena thatoccur at different levels of organization inthe brain, such as the firing of individualneurons and the collective dynamics of syn-chronous oscillations within large networks.The future challenge is to relate these prop-erties of self-organization to various aspectsof mental life. This endeavor is still in itsearly phases, but the future looks promising.



For example, recent analysis methods mak-ing use of the brain’s dynamical propertieshave been proposed as a means to antic-ipate epileptic seizures (Martinerie et al.,1998). There is now a strong general sensethat the properties of metastable, neural spa-tiotemporal patterns match crucial aspectsof conscious experience and that neurody-namics may provide the tools and conceptsto understand how the neural activity crucialfor consciousness temporally unfolds. Thistrend is patent in a set of influential neu-roscientific models of consciousness that wereview in the next section of this chapter.

Examples of NeurodynamicalApproaches to Consciousness

Introduction

Although neurodynamics is quite popularin the neurobiology of consciousness, it isstill not a widespread practice to formu-late theories about the relation between con-sciousness and the brain in purely dynamicalterms. Dynamical concepts are incorporatedto varying degrees by different researchersand used alongside concepts from informa-tion theory or functionalist models of cogni-tive processing. In this section, we reviewsome models that make use of dynamicalconcepts in attempting to explain the phe-nomenon of consciousness. The list of mod-els we cover is not meant to be exhaus-tive, but rather a small sample of a largespectrum of dynamical approaches to brainactivity and consciousness. Furthermore, wedo not intend to scrutinize these models indetail, but instead to highlight some com-mon aspects while providing an overview oftheir main proposals and hypotheses. Thereader is referred to the original sources formore details.

Neurodynamical Modelsof Consciousness

consciousness as order parameter

and dynamical operator

Among the different approaches to neuraldynamics, the pioneering work of Walter

J. Freeman stands out as one of the mostelaborate and truly dynamical theories ofbrain function (Freeman, 1975 , 1999a,b,2000a,b, 2004). His work is based mainlyon animal studies, in particular electrophys-iological recordings of the olfactory sys-tem of awake and behaving rabbits. Thisapproach can be summarized as follows:The point of departure is the neuronalpopulation. A neuronal population is anaggregate of neurons, in which, through pos-itive feedback, a state transition has occurredso that the ensemble presents steady-state,non-zero activity. When negative feedbackis established between populations, whereone is excitatory and the other inhibitory,oscillatory patterns of activity appear. Thischange implies a second state transition,where the resulting attractor is a limit cyclethat reveals the steady-state oscillation ofthe mixed (excitatory-inhibitory) popula-tion. When three or more mixed popula-tions combine among themselves by furthernegative and positive feedback, the result-ing background activity becomes chaotic.This chaotic activity now distributed amongthe populations is the carrier of a spatialpattern of amplitude modulation that canbe described by the local heights of therecorded waveform.

When an input reaches the mixed pop-ulation, an increase in the non-linear feed-back gain will produce a given amplitude-modulation pattern. The emergence of thispattern is considered to be the first step inperception: Meaning is embodied in theseamplitude-modulation patterns of neuralactivity, whose structure is dependent onsynaptic changes caused by previous expe-rience. Thus, the whole history of the ani-mal sets the context in which the emerg-ing spatiotemporal pattern is meaningful.Through the divergence and convergence ofneural activity onto the entorhinal cortex,the pulse patterns coming from the bulbare smoothed, thereby enhancing the macro-scopic amplitude-modulation pattern whileattenuating the sensory-driven microscopicactivity. Thus, what the cortex ‘sees’ is a con-struction made by the bulb, not a mappingof the stimulus.



Hence, in Freeman’s view, perceptionis an essentially active process, closer tohypothesis testing than to passive recoveryof incoming information. This active stanceis embodied in the process of ‘preafference’by which the limbic system (including theentorhinal cortex and the hippocampus inmammals), through corollary discharges toall sensory cortices, maintains an attentiveexpectancy of what is to come. The stimulusthen confirms or disconfirms the hypothesisthrough state transitions that generate theamplitude-modulation patterns describedpreviously.14 The multisensory convergenceonto the entorhinal cortex becomes the basisfor the formation of Gestalts underlying theunitary character of perception.

Finally, through multiple feedback loops,global amplitude-modulation patterns ofchaotic activity emerge throughout theentire hemisphere, directing its subsequentactivity. These loops comprise feedforwardflow from the sensory systems to the entorhi-nal cortex and the motor systems, and feed-back flow from the motor systems to theentorhinal cortex, and from the entorhinalcortex to the sensory systems. Such globalbrain states “emerge, persist for a small frac-tion of a second, then disappear and arereplaced by other states” (Freeman, 1999b,p. 153).

For Freeman, it is this level of emergentand global cooperative activity that is crucialfor consciousness, as these remarks indicate:“Consciousness . . . is a state variable thatconstrains the chaotic activities of the partsby quenching local fluctuations. It is an orderparameter and an operator that comes intoplay in the action-perception cycle as anaction is being concluded, and as the learn-ing phase of perception begins” (Freeman,1999a, p. 132). Furthermore: “[T]he globallycoherent activity . . . may be an objective cor-relate of awareness. . . . In this view, aware-ness is basically akin to the intervening statevariable in a homeostatic mechanism, whichis both a physical quantity, a dynamical oper-ator, and the carrier of influence from thepast into the future that supports the rela-tion between a desired set point and an exist-ing state” (Freeman, 1999b, p. 157).

dynamic large-scale integration

and radical embodiment

Another proposal that falls squarely withinthe neurodynamical framework is one for-mulated initially by Francisco J. Varela(1995) and then developed with his collabo-rators, especially Evan Thompson (Thomp-son & Varela, 2001). Varela proposes toaddress the question of how neural mech-anisms bring about “the flow of adaptedand unified cognitive moments” (Varela,Lachaux, Rodriguez, & Martinerie, 2001, p.229). The main working hypothesis is thata specific neuronal assembly underlies theoperation of every unitary cognitive act.Here a neuronal assembly is understood asa distributed set of neurons in the brainthat are linked through reciprocal and selec-tive interactions, where the relevant vari-able is no longer single-neuron activity, butrather the dynamic nature of the links thatare established between them. Varela andcollaborators propose that such dynamicallinks are mediated by the transient establish-ment of phase relations (phase synchrony)across multiple frequency bands, especiallyin the beta (15–30 Hz) and gamma (30–80 Hz) range (Varela et al., 2001). More-over, the transient nature of such dynamicallinks (and therefore of the neural assembliesthemselves) is central to the idea of large-scale integration, for it brings to the forethe notion that the system, rather than pre-senting a series of well-defined states (attrac-tors), shows metastable (self-limiting andrecurrent) patterns of activity: “In the brain,there is no ‘settling down’ but an ongo-ing change marked only by transient coor-dination among populations, as the attractoritself changes owing to activity-dependentchanges and modulations of synaptic con-nections” (Varela et al., 2001, p. 237). Large-scale integration through phase relationsbecomes fundamental for understandingbrain dynamics as coordinated spatiotempo-ral patterns and provides a plausible solu-tion to the problem of how to relate thelocal specificity of activity in specializedcortical regions to the constraints imposedby the connectivity established with otherdistributed areas. We see in the next two



sections below, as well as in the other dynam-ical approaches described in this section,that this balance of segregation and integra-tion has been considered the hallmark ofbrain complexity and a plausible prerequi-site for consciousness.

Thompson and Varela (2001) then qual-ify this view by placing it in a ‘radicalembodiment’ framework. They proposethat, although the neural processes relevantto consciousness are best mapped at the levelof large-scale, transient spatiotemporal pat-terns, the processes crucial for consciousnessare not brain-bound events, but comprisealso the body embedded in the environment.By taking into account the notion of emer-gent processes as understood in complex sys-tems theory (order parameters or collectivevariables and the boundary conditions theyimpose on local activities), they propose thatconscious awareness (as an order parameteror dynamical operator) is efficacious withrespect to local neural events (see also Free-man 1999a,b, and above) and that the pro-cesses crucial for consciousness so under-stood span at least three ‘cycles of operation’that cut across brain-body-world divisions(Thompson & Varela, 2001, p. 424). The firstis the regulatory organismic cycle, in whichthe maintenance of internal variables withina viable range is achieved “through sensorsand effectors to and from the body that linkneural activity to the basic homodynamicprocesses of internal organs.” This cycle issupposed to be the basis of the “inescapableaffective backdrop of every conscious state,”also called ‘core consciousness’ (Damasio,1998, 1999) or ‘primary-process conscious-ness’ (Panksepp, 1998). The second cycleis sensorimotor coupling between organismand environment, whereby what the organ-ism senses is a function of how it movesand how it moves is a function of whatit senses. Here, “transient neural assembliesmediate the coordination of sensory andmotor surfaces, and sensorimotor couplingwith the environment constrains and mod-ulates this neural dynamics.” The third is acycle of intersubjective interaction involvingthe recognition of the intentional meaningof actions and (in humans) linguistic com-

munication. This last type of cycle dependson various levels of sensorimotor coupling,mediated in particular by the so-called mir-ror neuron systems that show similar pat-terns of activation for both self-generated,goal-directed actions and when one observessomeone else performing the same action(Rizzolatti & Craighero, 2004).

As a final aspect of this proposal, Thomp-son and Varela hypothesize that “conscious-ness depends crucially on the manner inwhich brain dynamics are embedded in thesomatic and environmental context of theanimal’s life, and therefore there may be nosuch thing as a minimal internal neural corre-late whose intrinsic properties are sufficientto produce conscious experience” (Thomp-son & Varela, 2001, p. 425 ; see also Noe &Thompson, 2004a).

cortical coordination dynamics

Based on extensive work in human motorcoordination, J. A. Scott Kelso has developeda detailed dynamical framework for under-standing human cognition (Kelso, 1995).His main focus is the appearance of self-organized patterns caused by non-linearinteractions between system components, atboth the neural and motor levels, as well astheir role in human behavior. Kelso views thebrain as fundamentally “a pattern formingself-organized system governed by poten-tially discoverable, non-linear dynamic laws”(Kelso, 1995 , p. 257). He proposes that cog-nitive processes “arise as metastable spa-tiotemporal patterns of brain activity thatthemselves are produced by cooperativeinteractions among neural clusters” (257).He then goes one step further, proposingthat “an order parameter isomorphism con-nects mind and body, will and brain, mentaland neural events. Mind itself is a spatiotem-poral pattern that molds the metastabledynamic patterns of the brain” (288).

What are the specific neural mecha-nisms underlying the establishment of suchself-organized patterns? Kelso, in collabo-ration with Steven Bressler, proposes thatthe answer lies in the notion of ‘coordi-nation dynamics’ (Bressler & Kelso, 2001).Coordination dynamics is presented as an



integrative framework, in which the mainissue is “to identify the key variables of coor-dination (defined as a functional orderingamong interacting components) and theirdynamics (rules that govern the stabilityand change of coordination patterns and thenon-linear coupling among components thatgive rise to them)” (Bressler & Kelso, 2001,p. 26). Using this framework, Bressler andKelso address the question of how inter-acting, distributed cortical areas allow theemergence of ongoing cognitive functions.On the basis of previous studies of biman-ual coordination, they propose more specif-ically that the relevant collective variableis the relative phase (the continuous phasedifference) among the given neural struc-tures, which are themselves considered tobe accurately described by non-linear oscilla-tors. They argue that this coordination vari-able is adequate because (i) it reveals thespatiotemporal ordering between interact-ing structures, (ii) changes in the relativephase occur more slowly than changes inthe local component variables, and (iii) rel-ative phase shows abrupt changes duringphase transitions or bifurcations. When thetwo coordinated local neuronal populationshave different intrinsic frequencies, the rel-ative phase shows a metastable regime inthe form of ‘attractiveness’ toward preferredmodes of coordination, without settling intoany unique one. Accordingly, Bressler andKelso propose that “a crucial aspect of cog-nitive function, which can both integrateand segregate the activities of multiple dis-tributed areas, is large-scale relative coor-dination governed by way of metastabledynamics” (Bressler & Kelso, 2001, p. 30).

the ‘dynamic core’ hypothesis

Gerald M. Edelman and Giulio Tononi havedeveloped an account of the neural basisof consciousness that aims to explain twofundamental properties of conscious expe-rience, which they call ‘integration’ and‘differentiation’ (Edelman & Tononi, 2000;Tononi & Edelman, 1998). Integration refersto the unitary character of conscious expe-rience, whereby the multiplicity of aspects,such as color, taste, audition, kinesthetic

sense, etc., come together in a unique coher-ent experience. Differentiation is the capac-ity to experience any of a vast numberof different possible conscious states. Thiscapacity is intimately tied to what Edel-man and Tononi call the informativeness ofconscious experience, where each consciousstate would be highly informative, giventhe reduction in uncertainty that is accom-plished by the selection of one among apotentially infinite number of possible states.

Edelman and Tononi stress that con-sciousness is not a thing, but a process, andtherefore should be explained in terms ofneural processes and interactions and notin terms of specific brain areas or localactivities. More specifically, they postulatethat to understand consciousness it is nec-essary to pinpoint neural processes thatare themselves integrated, yet highly differ-entiated. Their answer to this problem iswhat they call ‘the Dynamic Core hypoth-esis’ (Edelman & Tononi, 2000; Tononi &Edelman, 1998). They describe this hypoth-esis as follows:,

1) a group of neurons can contribute directlyto conscious experience only if it is part of adistributed functional cluster that achieveshigh integration in hundreds of millisec-onds. 2 ) To sustain conscious experience,it is essential that this functional cluster behighly differentiated, as indicated by highvalues of complexity. We call such a clusterof neuronal groups that are strongly inter-acting among themselves and that have dis-tinct functional borders with the rest of thebrain at the time scale of fractions of a sec-ond a ‘dynamic core,’ to emphasize bothits integration and its constantly changingcomposition. A dynamic core is thereforea process, not a thing or a place, and itis defined in terms of neural interactions,rather than in terms of specific neural loca-tions, connectivity or activity (Edelman &Tononi, 2 000, p. 144).

In addition, they argue that “the dynamiccore is a functional cluster: its participat-ing neuronal groups are much more stronglyinteractive among themselves than with therest of the brain. The dynamic core mustalso have high complexity: its global activity



patterns must be selected within less thana second out of a very large repertoire”(Tononi & Edelman, 1998, p. 1849). Theyhypothesize that the dynamic core achievesintegration on the basis of reentrant inter-actions among distributed neuronal groups,most likely mediated by the thalamocor-tical system. Specifically, for primary con-sciousness to arise, interactions are requiredbetween sensory cortices in different modal-ities and value-category and memory sys-tems in frontal, temporal, and parietalareas.15

Edelman and Tononi claim that thedynamic core provides a “neural referencespace for conscious experience” (Edelman &Tononi, 2000, p. 164). They depict this spaceas an n-dimensional neural space, wherethe number of dimensions is given by thenumber of neuronal groups that are partof the dynamic core at that moment. Suchneuronal groups would be segregated intoneural domains specialized for various func-tions, such as form, color, or orientation dis-crimination, proprioceptive or somatosen-sory inputs, and so on, and they would bebrought together through re-entrant inter-actions. The local activities of these groupswould therefore need to be understood inrelation to the unified process constitutedby the functional cluster; that is, the entiredynamic core: “The pure sensation of redis a particular neural state identified bya point within the N-dimensional neuralspace defined by the integrated activity ofall the groups of neurons that constitute thedynamic core. . . . The conscious discrimina-tion corresponding to the quale of seeing redacquires its full meaning only when consid-ered in the appropriate, much larger, neuralreference space” (Edelman & Tononi, 2000,p. 167).

Related Models

Several other authors have advanced mod-els in which dynamical system concepts arepresent, yet appear less explicitly. Neverthe-less, these approaches also aim to describethe formation of spatiotemporal patterns ofbrain activity that are crucial for action, per-

ception, and consciousness. Therefore, webelieve that it is important to keep thesemodels in mind as part and parcel of thewider research program of neurodynamics.

the cortico-thalamic dialogue

Rodolfo Llinas and his collaborators haveproposed a model of how consciousnessis related to brain activity, in which thenotion of emergent collective activity playsa central role (Llinas & Pare, 1991; Llinas& Ribary, 2001; Llinas, Ribary, Contreras, &Pedroarena, 1998). In particular, Llinas pos-tulates that consciousness arises from theongoing dialogue between the cortex andthe thalamus (Llinas & Pare, 1991). He callsattention to the fact that most of the input tothe thalamus comes from the cortex, ratherthan from peripheral sensory systems. Onthis basis, he proposes that the brain be con-sidered as a ‘closed system’ that can gen-erate and sustain its own activity thanksto the intrinsic electrical properties of neu-rons (Llinas, 1988) and the connectivity theyestablish. The interplay of these two maincharacteristics underlies the establishmentof “global resonant states which we know ascognition” (Llinas & Ribary, 2001, p. 167).

A crucial feature of this proposal is theprecise temporal relations established byneurons in the cortico-thalamic loop. Thistemporal mapping is viewed as a ‘functionalgeometry’ and involves oscillatory activity atdifferent spatial scales, ranging from individ-ual neurons to the cortical mantle. In par-ticular, 40-Hz oscillations that traverse thecortex in a highly spatially structured man-ner are considered as candidates for the pro-duction of a “temporal conjunction of rhyth-mic activity over large ensemble of neurons”(Llinas & Ribary, 2001, p. 168). Such gammaoscillations are believed to be sustained bya thalamo-cortical resonant circuit involvingpyramidal neurons in layer IV of the neo-cortex, relay thalamic neurons, and reticu-lar nucleus neurons. In particular, tempo-ral binding is supposed to be generated bythe conjunction of a specific circuit involvingspecific sensory and motor nuclei projectingto layer IV and the feedback via the reticularnucleus, and a non-specific circuit involving



non-specific intralaminar nucei projecting tothe most superficial layer of the cortex andcollaterals to the reticular and non-specificthalamic nuclei. Thus, the ‘specific’ system issupposed to supply the content that relates tothe external world, and the non-specific sys-tem is supposed to give rise to the temporalconjunction or the context (on the basis of amore interoceptive context concerned withalertness). Together they generate a singlecognitive experience (Llinas & Ribary, 2001,p. 173).

timing and binding

Wolf Singer and collaborators have exten-sively investigated the issue of temporal cor-relations between cortical neurons and therole this phenomenon could play in solv-ing what has been called ‘the binding prob-lem’ (Engel & Singer, 2001; Gray et al., 1989;Singer & Gray, 1995). This is the problemof how the signals from the separate neu-ronal populations concerned with distinctobject features (color, shape, motion, etc.)are bound together into a unified percep-tual representation. The main idea behindtheir approach is that there is a “tempo-rary association of neurons into functionallycoherent assemblies that as a whole repre-sent a particular content whereby each indi-vidual neuron is tuned to one of the ele-mentary features of composite perceptualobjects” (Singer, 1998, p. 1831). The specifichypothesis is that neurons become mem-bers of such coherent assemblies through theprecise synchronization of their discharges;in other words, such synchronization estab-lishes a “code for relatedness” (Singer, 1998,p. 1837).

Recently, Singer and colleagues (Engel& Singer, 2001, Engel et al., 1999) haveextended this framework to address the issueof phenomenal awareness. Their argumentcan be summarized as follows: (1) Brainscapable of phenomenal awareness should beable to generate metarepresentations of theirown cognitive processes. (2) Metarepresen-tations are realized by an iterative process,in which higher-order cortical areas readlow-order (sensory) areas. (3) Combinatorialflexibility of metarepresentations is obtained

via dynamical cell assemblies. (4) Binding ofsuch assemblies is effected by transient syn-chrony that establishes a code for relatednessamong features and facilitates downstreamevaluation and impact. (5) Such assembliesneed desynchronized EEG16 (which corre-lates with phenomenal awareness in thewaking state and REM dream state andshows high-frequency beta and gamma oscil-latory activity), and are facilitated by atten-tion (Singer, 1998).

Engel, Fries, and Singer (Engel, Fries, &Singer, 2001; Engel & Singer, 2001) alsoexplicitly espouse a ‘dynamicist view’ ofbrain function. According to this view,brain processes are not passive, stimulus-driven, and hierarchical, but active, context-dependent, endogenously driven, and dis-tributed. In particular, “spatio-temporal pat-terns of ongoing activity . . . translate thefunctional architecture of the system and itspre-stimulation history into dynamic statesof anticipation” (Engel et al., 2001, p. 705).In this dynamicist account of top-downinfluences, relevant patterns are generatedas a result of continuous large-scale interac-tions, and these patterns can bias the saliencyof sensory signals by changes in their tempo-ral correlations. Endogenous, self-generatedactivity displays distinct spatiotemporal pat-terns, and these patterns bias the self-organizing process that leads to the tempo-ral coordination of input-triggered responsesand their binding into functionally coher-ent assemblies. This dynamicist approachthus stresses the importance of top-downinfluence in the form of large-scale dynam-ics that express contextual influences andstored knowledge in the system and thatcan modulate local processing and hence thedownstream effect of the impinging event(Engel et al., 2001).

the neural correlates of

consciousness

Francis Crick and Christof Koch haveemployed dynamical concepts in a seriesof proposals regarding the relation betweenneural activity and conscious perception(Crick & Koch, 1990, 1998, 2003). In theirview, the best way for the neuroscience



of consciousness to proceed is first touncover the neural correlates of conscious-ness (NCCs), in particular the neural con-tents of visual consciousness. They define aneural correlate of consciousness as a mini-mal set of neuronal events necessary and suf-ficient for a given state of phenomenal con-sciousness (see also Chalmers, 2000). Herewe summarize the version of their theorypresented in one of their last joint articleson consciousness (Crick & Koch, 2003).

They begin with the notion of an ‘uncon-scious homunculus’, which is a system con-sisting of frontal regions of the brain “lookingat the back, mostly sensory region” (Crick& Koch, 2003 , p. 120). Crick and Kochpropose that we are not conscious of ourthoughts, but only of sensory representa-tions of them in imagination. The brainpresents multiple unconscious processingmodules, mostly feedforward, that act as‘zombie’ modes. These modules presentstereotyped responses in a sort of ‘corticalreflex’, whereas conscious modes are nec-essary only to deal with time-consuming,less stereotyped situations that need plan-ning and decision making. The most impor-tant point, however, with regard to the NCCissue, is the existence of dynamic coalitionsof neurons in the form of neural assemblieswhose sustained activity embodies the con-tents of consciousness. Explicit representa-tions of particular aspects of the (visual)scene are present in special brain regions(‘critical nodes’), and these representationsare bound together in the dynamic neuralcoalitions. Additionally, Crick and Koch sug-gest that higher levels of cortical processingare first reached by the feedforward sensorysweep and that only through backpropaga-tion of activity from higher to lower levels dothe lower levels gain access to this informa-tion. They distinguish ‘driving’ from ‘modu-lating’ connections and suggest that the feed-forward sweep is mostly driving activity inthe frontal regions, whereas the feedbackreturn onto sensory cortices is mainly mod-ulating. In the specific case of conscious per-ception, they propose that it is not a contin-uous phenomenon, but rather that it workson the basis of a series of ‘snapshots’. Such

snapshots are possibly related to alpha andtheta rhythms and are the reflection of a cer-tain threshold that has been overcome (fora certain amount of time) by neural activ-ity, enabling it to become conscious. Con-scious coalitions would therefore be con-tinually “forming, growing or disappearing”(Crick & Koch, 2003 , p. 122). Crick andKoch propose that attention is fundamentalin biasing the competition among coalitionsthat share critical nodes. Attention producesthe effective binding of different attributesof the given conscious content by means ofshared “membership in a particular coali-tion” (Crick & Koch, 2003 , p. 123).

Although Crick and Koch recognize thatthe mechanism to establish such coali-tions probably involves synchronous fir-ing between distributed populations, theyexplicitly state that they no longer believethat 40-Hz oscillatory activity is a suffi-cient condition for consciousness. Finally,they propose that there is a set of neural pro-cesses that, although not part of the NCC,is affected by the NCC, both with respectto its actual firing and with respect to synap-tic modifications caused by previous expe-rience. This ‘penumbra’ (Crick and Koch,2003 , p. 124), could eventually become con-scious, if incorporated into the NCC.

consciousness as global workspace

Stanislas Dehaene, Jean-Pierre Changeux,and collaborators have explored an alterna-tive model of brain functioning that under-lies the accessibility to verbal report ofconscious experience (Dehaene, Kerszberg,& Changeux, 1998; Dehaene & Naccache,2001; Dehaene, Sergent, & Changeux,2003). The main proposal of their modelis the existence of “two main computa-tional spaces within the brain” (Dehaene,Kerszberg, & Changeux, 1998, p. 14529).The first computational space consists of aseries of functionally segregated and spe-cialized modules or processors that consti-tute a parallel distributed network (exam-ples of modular processors would be primaryvisual cortex (V1) or the mirror neuron sys-tem in area F5 of the premotor cortex). Thesecond computational space is not confined



to a series of brain areas, but rather is dis-tributed among multiple cortical regions.The main property of this second space ismassive reciprocal connectivity on the basisof horizontal projections (long-range cor-tico-cortical connections). Through descen-ding connections, this ‘global workspace’determines the contributions of the modularprocessors of the first computational spaceby selecting a specific set while suppress-ing another. Through this selective mobiliza-tion of the specialized processors into theglobal workspace. a ‘brain scale’ state can bereached, in which a group of workspace neu-rons are spontaneously coactivated while therest are suppressed. As a result, an exclusive‘representation’ invades the workspace and

may remain active in an autonomous man-ner and resist changes in peripheral activ-ity. If it is negatively evaluated, or if atten-tion fails, it may however be spontaneouslyand randomly replaced by another discretecombination of workspace neurons. Func-tionally, this neural property implementsan active ‘generator of diversity,’ which con-stantly projects and tests hypotheses (orpre-representations) on the outside world.The dynamics of workspace neuron activ-ity is thus characterized be a constantflow of individual coherent episodes ofvariable duration (Dehaene, Kerszberg, &Changeux, 1998, p. 14530).

This postulated workspace has access to theworld through ‘perceptual circuits’; ‘motorprogramming circuits’ enable action guida-nce’; ‘long-term memory circuits’ enableaccess to past experiences; ‘evaluationcircuits’ allow negative-positive judg-ments; and ‘attention circuits’ endow theworkspace with the capacity to alter its ownactivity separately from the influence ofexternal inputs. Through connections withmotor and language centers, the workspacemakes its resident representation avail-able for verbal report by the subject. Thus,Dehaene and Changeux see consciousness asa selective global pattern: “When a piece ofinformation such as the identity of a stimu-lus accesses a sufficient subset of workspaceneurons, their activity becomes self-sustained and can be broadcasted via long-

distance connections to a vast set of definedareas, thus creating a global and exclusiveavailability for a given stimulus, which isthen subjectively experienced as conscious”(Dehaene, Kerszberg, & Changeux, 2003).

summary

The majority of the approaches reviewedabove stress the importance of a certain typeof distributed, spatiotemporal pattern ofneural activity that ‘demarcates’ itself fromthe background activity of the brain. Suchpatterns are described as ongoing, transient,metastable coordination processes amongseparate neurons, and they are considered tobe crucial for the moment-to-moment emer-gence and formation of conscious experi-ence. Another related feature crucial to sev-eral of the above approaches is that thesespatiotemporal patterns reveal the interplayof two apparently fundamental principlesof brain organization and function; namely,functional segregation and cooperative inter-action or integration. This interplay andthe dynamical properties of the brain’s spa-tiotemporal activity patterns are the focus ofthe following sections.

The Search for MeaningfulSpatiotemporal Patterns in the Brain

Introduction

Despite their significant differences, all theabove models agree that the constitution ofdynamic spatiotemporal patterns of neuralactivity plays a central role in the emergenceof consciousness. This section discusses thepractical aspects of the search for such pat-terns. After a short review of the connec-tivity of the brain, we discuss the detectionof such patterns in real brain data. A shortmathematical presentation leads us to theconcept of synchrony, which is the preferredcandidate to date for such patterns.

Connectivity in the Brain

The organization of the brain’s connectivityis what ultimately determines the form ofthe neural spatiotemporal patterns. For this



reason, it is useful to start with a review ofsome basic facts about this architecture.

The brain is probably one of the mostcomplex biological systems we know (Edel-man & Tononi, 2000). Its complexity is cer-tainly due in great part to its histologicaland morphological structure, and one of themost striking aspects of the brain as a sys-tem is the connectivity pattern it exhibits.This pattern is that of a compact but dis-tributed tissue, with local clusters of highlyconnected neurons that establish long-rangeinteractions. In general, two neurons in thebrain are always in interaction either directlyor via a certain number of intermediate cells.It is useful to distinguish two levels of con-nectivity in the brain.

Local connections: Several types of neu-rons coexist in the neocortex. Within agiven portion of neocortex, a complexarrangement of pyramidal, spiny stellate, andsmooth stellate cells can be found. Thisarrangement of collateral axons, dentritictrees, and cell bodies gives rise to clustersof interconnected neurons that extend overa fraction of a millimeter. Neurons tend toorganize into radial clusters that share func-tional characteristics, known as functionalcolumns. These structures are particularlyevident in somatic sensory cortex and visualcortex and are believed to play a fundamen-tal role in basic discriminative capacities.

Long-range connections: In addition to thebodies and dentritic trees of the local neu-rons, axons from deep structures and othercortical regions terminate at different pointsin the six-layered structure of the neocor-tex. Likewise, pyramidal neurons in a givenregion of the neocortex have axons thatextend into the white matter and reach bothdeep structures and other cortical regions.At least four patterns of long-distance con-nectivity can be distinguished in the brain(Abeles, 1991): (i) between cortical neu-rons within one hemisphere, (ii) betweencortical neurons of different hemispheres,(iii) between cortical neurons and deepnuclei, and (iv) between brainstem modu-latory systems and extended areas of thecortex. In general, long-range connectivityobeys a reciprocity rule (Varela, 1995): IfA projects to B, then B projects to A. This

rule clearly favors the establishment of recur-sive loops. Nevertheless, some basal gan-glia nuclei present a slightly different con-nectivity structure: Although they receiveaxons from cortical neurons, they projectonly through the thalamus into frontal loberegions (Edelman & Tononi, 2000).

Interestingly, however, no one zone in thebrain can be distinguished as the ultimatehighest level, at least in terms of the connec-tivity patterns. Indeed, the massively inter-connected nature of the brain suggests thatdynamic relations between local and dis-tant activities will necessarily be establishedwhatever the observed origin of a givenactivation is. On the other hand, it is truethat clusters of more strongly interconnectedregions are evident. Stephan and collabora-tors recognize at least three main clustersin the primate cortex: (i) visual (occipito-temporal); (ii) somatomotor (mainly pre-and post-central, but extending into pari-etal regions); and (iii) orbito-temporopolar-insular (Stephan et al., 2000). It is inter-esting that the overall structural connectiv-ity (Hilgetag & Kaiser, 2004) and functionalconnectivity (Stephan et al., 2000) show a‘small world’ architecture. Networks hav-ing such an architecture display remarkableproperties, such as reduced average lengthpath (reaching any node from any othernode is accomplished in a minimal numberof steps), high synchronizability, enhancedsignal propagation speed, and stability (onecan randomly eliminate links without affect-ing substantially the network properties;Watts & Strogatz, 1998).

One of the most important conclusionsof the study of the brain as a system is that,despite its massive interconnectedness, thebrain shows a strong segregation into clus-ters at both structural and functional lev-els. The interplay of these two characteristicfeatures of the brain lies at the basis of oneof the most interesting issues in contempo-rary neuroscience – the large-scale integra-tion of brain activity and its role in the uni-fied nature of experience (James, 1890/1981;Varela, 1995 ; Von der Malsburg, 1981).

The combination of extensive neuropsy-chological studies since Broca and the explo-sive use of imaging techniques in the last 15



years has highlighted two main principles ofbrain functioning (Edelman & Tononi, 2000;Friston, 2002a,b, 2005). On the one hand, afunctional encapsulation is evident: Distinctregions in the brain contribute differentiallyto different aspects of adaptive behavior;for instance, bilateral damage to the humanhomologue of V5 /MT (in the middle tem-poral area) can lead to a restricted impair-ment in the capacity to discriminate move-ment (akinetopsia; Zihl, Von Cramon, &Mai, 1983). On the other hand, for a givencognitive task, it is rarely the case that onlyone isolated region shows significant activa-tion. For example, directing attention to aparticular location of the visual field corre-lates with the concomitant activation of sev-eral cortical regions, preferentially right pari-etal, anterior cingulate, and occipital cortices(Mesulam, 1999).

Indeed, as we saw above, the connectiv-ity pattern of the mammalian brain reveals acomplex structure of recursively connecteddistant areas (Hilgetag & Kaiser, 2004 ;Stephan et al., 2000). Although the cortico-cortical connectivity pattern is paradigmatic,the structure of recursive connections isreflected most prominently in the thalam-ocortical matrix (Edelman & Tononi, 2000;Llinas & Ribary, 2001). For instance, the lat-eral geniculate nucleus (LGN) of the tha-lamus receives only around 5–10% of itsinputs (not more than 20%) from the retina,whereas the remaining connections comefrom local inhibitory networks, descend-ing inputs from layer VI of the visual cor-tex, and ascending inputs from the brain-stem (Sherman & Guillery, 2002). Yet theLGN is the major relay in the visual path-way from the retina to the cortex. Such acomplex structure of recursive, re-entrant,and interconnected networks that pervadethe mammalian brain (Edelman & Tononi,2000) strongly suggests the existence of con-stitutive cooperative interactions, and there-fore integrative activity, among differentregions.

Nevertheless, the presence of anatomicalconnectivity is not enough to explain effec-tive interactions among separate regions(Friston, 2002a,b). Indeed, in addition tobeing connected, it is necessary that such

regions establish interdependent activationto account adequately for the integrationof functionally separate activity (Bressler,1995). Thus, one task facing the neurody-namicist is to detect such interdependentactivation from the brain recordings avail-able today.

Detecting Interdependent Activationsfrom Real Brain Recordings

the data

The activity of the brain can be recordedat several different spatial and temporalscales. The neurodynamicist will be primar-ily interested in those techniques that arefast enough to follow the formation of spa-tiotemporal patterns in the time scale ofhundreds of milliseconds. Because the con-struction of such patterns often involvesactivities at the millisecond time scale(e.g., in the case of fast neural oscilla-tions), the desired temporal resolution is onthis order of milliseconds. In practice, thisexcludes the neuroimaging techniques basedon slow metabolic measures, such as func-tional magnetic resonance imaging (fMRI)or positron emission topography (PET).17

Millisecond temporal resolution is accessi-ble through direct intracellular and extra-cellular measurements of individual neu-rons and recordings of local field potentials(LFPs) or of the electromagnetic fields oflarge neural populations that produce theelectroencephalographic (EEG) and mag-netoencephalographic (MEG) signals. LFPsare the summation of the membrane poten-tials of populations of neurons. The size ofthe populations depends on the site and pre-cision of the recordings: Local microelec-trodes can record small populations, extend-ing over less than a square millimeter oftissue, whereas scalp EEG electrodes orMEG sensors (and optical imaging) recordthe average activity of several square cen-timeters of cortex. At an intermediate level,intracranial recordings from human patientscan record from a couple of square millime-ters of cortex (Lachaux, Rudrauf, & Kahane,2003). Except in those rare situations justi-fied by therapeutical reasons, human record-ings are almost exclusively non-invasive and



performed therefore at the centimeter-widespatial resolution of MEG or EEG.

With this panel of recording techniques,spatiotemporal patterns can in principle beobserved at three levels: (i) as interactionsbetween simultaneous recordings of multi-ple individual neurons, (ii) as interactionsbetween simultaneous recordings of multi-ple individual LFPs, and (iii) in single LFPrecordings. The third level is intermediatebetween the first two: Because an individualLFP records from a single neural population,the average activity of the LFP is sensitiveto the spatiotemporal organization of activ-ity within this population. For example, ifall the neurons are coactive periodically, theaverage activity in the LFP will be a massiveoscillation, much stronger than if the neu-rons are not synchronous.

some simple mathematical

considerations

As we have seen, the organization of thebrain suggests that interactions among dis-tributed neuronal groups are bound to occur,given their massive interconnectedness. Wealso mentioned that recordings of neuronalactivity can be obtained by a diversity ofapproaches and at several levels of spatialresolution. With these points in mind, let usreturn to the central question of this section:How does one detect neural spatiotempo-ral patterns from real brain data? The defi-nition of the ‘dynamic core’ by Tononi andEdelman (1998) provides a useful startingpoint: “The dynamic core is a functionalcluster: its participating neuronal groupsare much more strongly interactive amongthemselves than with the rest of the brain.”The challenge for the neurodynamicist istherefore to find neurons or groups of neu-rons with particularly strong (but transient)interactions.

In keeping with the dynamical approach,we can usefully consider this question ingeometrical terms. Consider the n simulta-neous measures of brain activity that onecan record in a typical electrophysiologi-cal setting (e.g., 64 measures from 64 EEGchannels). At any time t, the n simultane-ous measurements define a position in an n-

dimensional state space, and the evolutionof this position in time defines a trajectory.If the measurements are independent fromeach other, then the trajectory will progres-sively completely fill a hypercubic portion ofthe state space, leaving no hole. In contrast,if there are interactions among the mea-sured neuronal populations, then the trajec-tory will fit into a restricted portion of thefull space and be constrained onto a manifoldwith a (fractal) dimension less than that ofthe state space.18 What this means in infor-mational terms is that, for at least one pair ofthe measured neural populations, measuringthe activity of the first population providessome information about the activity of thesecond one. The probability distribution ofthe activity of Population 2 (the probabil-ity p(y) that this activity is y), given that theactivity of Population 1 is x, is different fromwhat it would be if the activity of Popula-tion 1 were x′. Consider this metaphor: Ifwe know where John will spend the after-noon, we can predict with some accuracythat his wife Ann will spend the afternoon inthe same city, but we cannot predict whereJane, unrelated to John, will be. Certain mea-sures, such as mutual information (David,Cosmelli, & Friston, 2004), quantify exactlythis sharpening of the probability distribu-tion.

The transient nature of neural inter-actions, however, makes general measuresbased on such geometrical formulations dif-ficult to apply. The main problem is that,to know whether or not the trajectory fillsup the whole space, the experimenter needsto observe it during time windows that aretypically orders of magnitude longer thanthe typical lifespan of cell assemblies. Thisdifficulty can be avoided if the researcherassumes a priori what will be the shapeof the manifold onto which the trajectoryis constrained. Because spatiotemporal pat-terns can potentially take an infinite num-ber of shapes in the state space, a possiblesolution is to assume a specific shape and tobuild a special detector for this shape (surf-ing on the advances of signal processing). Itis easy to see that it takes fewer measure-ments to test whether the trajectory stays on



a circle or whether it follows some general,unknown, geometrical structure.

In the absence of noise, three succes-sive measurements are sufficient to knowwhether the trajectory stays on a straightline, which would correspond to a linearrelationship between the recorded activities.Synchrony between quasi-periodic oscilla-tors – that is, the transient phase-locking oftheir oscillations – is a good example of suchan interaction. But linear relationships canbe extended to other trajectories constrainedon simple manifolds of dimension 1,19 suchas a circle. This is the case for two oscilla-tors rotating at the same frequency with aconstant phase lag.

Synchrony: Perhaps Not the BestCandidate, but at Least the Simplest

The practical reason that synchrony has sofar been the best-studied (if not the only)type of transient interaction between neu-ral populations is the ease with which itcan be detected. Furthermore, since thedevelopment of the EEG, it has been evi-dent that oscillations are ubiquitous in thebrain. This fact, combined with the rela-tion between coordinated oscillatory activ-ity and several important cognitive functions(discussed below), has also contributed tothe development of approaches that seek todetect the occurrence of synchrony from realneurobiological signals.

In its original neurophysiological formu-lation, ‘synchrony’ refers to a positive cor-relation between the spike timing of a setof neurons. In other words, if we considertwo neurons within a synchronous popula-tion, the probability of the first neuron tofire a spike is significantly higher at specificdelays relative to the spikes of the otherneuron.20 In the simplest case, this delay iszero, which means that the neurons havea high probability of firing simultaneously.In general, this probability, as well as theeventual delay, is quantified by the cross-correlogram between the spike trains of thetwo neurons (Perkel, Gerstein, & Moore,1967). One speaks of oscillatory synchrony,or synchronous oscillations, if the neurons

tend to fire at periodic latencies. Numerousanimal studies conducted over the past 20

years have now established that synchrony isubiquitous in virtually all sensory and motormodalities. It has often been found to berelated to perception, memory, and motorprogramming (see Roskies, 1999, for a groupof excellent reviews that summarize theseresults).

Synchrony has also been studied in hu-mans as an instance of spatiotemporal pat-terns of interdependent neural activity,although at a different level from animalstudies. Here an important distinction needsto be made between the local recordingsof individual neurons, almost always acces-sible only in animals, and the more globalrecordings of entire neural populations,accessible in humans through scalp EEG orMEG. EEG and MEG average across largeneuronal assemblies, and hence oscillatorysynchrony between neurons shows up aschanges of power in particular frequencybands. The reason this happens is that groupsof synchronously firing neural oscillators canbe modeled as oscillators themselves, withthe amplitude of the oscillations dependingon the number of individual oscillators in thegroup and on the precision of the synchronybetween them. This point entails a furtherdistinction: On the one hand, oscillatoryactivity as recorded by an individual EEGelectrode or MEG sensor implies already acertain amount of local synchronous activ-ity. On the other hand, one can chooseto consider synchronization between oscil-lations produced by distant neuronal pop-ulations (separated by several centimeters)to describe distributed spatiotemporal pat-terns that occur at a more global level.In any case, when dealing with EEG andMEG recording one is always in the presenceof noisy data. Hence, any interdependencemeasure must be understood in a statisticalsense throughout a given temporal window(Lachaux et al., 2002 ; Lachaux, Rodriguez,Martinerie, & Varela, 1999; Le Van Quyenet al., 2001).

Synchrony at the more regional or locallevel has been demonstrated repeatedly inhumans in relation to integrative mecha-



nisms in language, memory, attention, andmotor tasks and in virtually all the sensorymodalities. For example, the perceptionof coherent objects in humans is specifi-cally associated with synchronous oscilla-tions in the gamma range (above 30 Hz),the so-called induced gamma response (fora review, see Tallon-Baudry & Bertrand,1999). This response, although not com-pletely time-locked to the stimulus presen-tation, typically starts in posterior brain areas(over the occipital cortex) around 200 msafter the stimulus and then returns grad-ually to the prestimulus level when thestimulus does not require further analysis(Lachaux et al., 2000, 2005 ; Tallon-Baudry& Bertrand, 1999).

As we mentioned above, oscillations pro-duced by two neural populations can alsobe synchronous within larger cell assem-blies. This synchrony can be detected by atransient phase-locking between the oscilla-tions of the two local fields (Lachaux et al.,1999, 2002 ; Rodriguez et al., 1999). Suchlong-range synchrony between distant neu-ral populations has been suggested as a plau-sible candidate to mediate the integrationof activity in functionally specialized anddistinct brain regions (Bressler, 1995 ; Varelaet al., 2001). For example, Tallon-Baudry andcolleagues have shown in humans that dur-ing the maintenance of a complex shapein visual short-term memory, two function-ally distinct regions within the ventral visualpathway, the fusiform gyrus and the lat-eral occipital sulcus, produce synchronousoscillations around 20 Hz (Tallon-Baudry,Bertrand, & Fischer, 2001).

Synchronization is a complex conceptthat can cover several possibly distinct typesof temporal relations, such as coherence, fre-quency synchronization, phase synchroniza-tion, generalized synchronization, as wellas others (Brown & Kocarev, 2000; Friston,1997; Pikovsky et al., 1997). Here we havefocused on synchrony as either occurringbetween stochastic point-processes, such asspike trains, or in terms of phase rela-tions between oscillatory processes (phase-locking synchrony). As mentioned above, wechose this focus mainly because of techni-

cal limitations in the estimation of gener-alized measures of synchronous activation.With this point in mind it is possible to saythat synchrony, as presented here, appearsas a simple measure of precise temporalrelations between neural processes that canenable one to follow the formation of spa-tiotemporal brain patterns relevant for con-sciousness. Not surprisingly, this mechanismis referred to in several of the dynamicalmodels reviewed above. In the next sec-tion, we review more specifically a set ofresults concerning the relation between con-sciousness and the current ‘crowd’s favorite’among neurodynamicists – synchrony in thegamma range.

The Crowd’s Favorite:The Gamma Band

Evidence for a Relation Between GammaSynchrony and Consciousness

We have mentioned that synchrony amongoscillating neural populations is a plausiblecandidate to mediate functional connectiv-ity and therefore to allow the formationof spatiotemporal structures, such as thosereviewed in the previous sections. In this sec-tion, we return to this hypothesis in moredetail, with a particular focus on gammaband oscillations, which have been repeat-edly associated with consciousness in the last15 years.

The putative role of gamma band oscil-lations in the formation of conscious experi-ence was proposed by Crick and Koch (Crick& Koch, 1990), shortly after Singer and col-leagues (Gray et al., 1989) had completed aseries of observations in the cat visual cor-tex showing that neurons tend to synchro-nize their spiking activity when stimulatedwith parts of the same visual object, such asa moving bar (whereas they do not synchro-nize when stimulated with features that can-not be part of the same object). Those obser-vations matched theoretical predictions byVon der Marlsburg (1981) that synchronycould be used to achieve figure/ground seg-mentation during perception of the visual



scene. Thus, synchrony was assumed toprovide a solution to the visual binding prob-lem (the problem, discussed above, of inte-grating distinct visual features into a unifiedand coherent perception; see Roskies, 1999).

In their 1990 paper, Crick and Kochpushed this idea further by stating thatvisual consciousness of the object occursonly when its features are bound together asa result of this type of synchronous activity.This hypothesis was in good agreement withthe feature-integration-theory, proposed byAnne Treisman, suggesting that attention isnecessary to bind together the features ofobjects (Treisman & Gelade, 1980).21 Hence-forth, a close relation between gamma syn-chrony and attention and consciousness hasensued (Fell et al., 2003 ; Varela, 1995). Thisassociation has been very appealing for neu-rodynamicists addressing the consciousnessissue, because it provides ground materialfor the neural spatiotemporal patterns theyassociate with consciousness on the appro-priate time scale. Indeed, several dynamicmodels among those reviewed above, specif-ically those by Singer, Llinas, Varela, andtheir respective collaborators, consider syn-chronous activity in high-frequency ranges,most preferentially the gamma range, as cru-cial for conscious experience.

Fortunately, the association between thegamma band and attention, vigilance, andconsciousness is not just based on its the-oretical appeal but also on sound exper-imental evidence. For instance, it is wellknown that the precise synchronization ofneuronal discharges is more prevalent dur-ing states characterized by arousal and more-over that gamma oscillations are particularlyprominent during epochs of higher vigilance(Herculano-Houzel, Munk, Neuenschwan-der, & Singer, 1999; Rodriguez, Kallenbach,Singer, & Munk, 2004). In cats, for exam-ple, gamma synchrony is stronger after thestimulation of the mesencephalic reticularformation (Munk, Roelfsema, Konig, Engel,& Singer, 1996). Furthermore, EEG/MEGgamma-band activity is present both duringREM sleep and awake states, with a muchstronger amplitude than during deep sleep(reviewed in Engel et al., 1999).22

Several studies have also demonstratedthat the presentation of sensory stimuli elic-its stronger gamma synchrony when atten-tion is focused on the stimulus than whenattention is diverted away. This findingwas observed in monkeys for somatosensorystimulations and found again recently forneurons in area V4 of monkeys presentedwith small visual gratings (Fries, Reynolds,Rorie, & Desimone, 2001).

There is also evidence for a more directrelation between gamma activity and con-sciousness. Lachaux and colleagues haverecently shown that the perception of facesis associated with strong gamma oscillationsin face-specific regions along the ventralvisual stream (Lachaux et al., 2005). Epilep-tic patients with intracranial electrodes thatrecord directly from the fusiform face area(a region along the ventral visual path-way particularly associated with the percep-tion of faces) were presented with high-contrast ‘Mooney figures’ representing faces.Because the figures were presented briefly,for 200 ms, they were consciously perceivedas faces only half of the time. The authorsreported that the gamma band response tothe images was significantly stronger whenthe figures were actually consciously per-ceived as faces than when they were not.This high-resolution study followed a pre-vious one (Rodriguez et al., 1999), using thesame protocol in normal subjects with non-invasive scalp EEG recordings; this studyshowed that gamma oscillations tend to syn-chronize across widely separated brain areas(typically frontal versus occipital) only whenthe figures are perceived as faces.

Fries and colleagues (Fries, Roelfsema,Engel, Konig, & Singer, 1997) have shown aneven more direct relation between gammasynchrony and consciousness. They showedthat, during binocular rivalry in cats, thelevel of synchrony between visual neuronsfollows in time the shift of perceptual dom-inance. Cats were presented with two visualpatterns moving simultaneously in differentdirections: One pattern was presented to theleft eye and the other to the right eye. Undersuch circumstances, the visual percept can-not encompass the two contradictory



patterns and instead alternates betweenthem (hence the term ‘binocular rivalry’).The results of this study showed that neu-rons stimulated by the perceived stimuluswere strongly synchronized, with stronggamma oscillations, whereas cells stimulatedby the suppressed visual pattern showedonly weak synchrony. This experiment ishighly relevant to the study of visual con-sciousness, because conscious perceptionis decoupled from the drive of the sensoryinputs (the physical stimulus remains con-stant while perception does not) and gammasynchrony is used as an indicator of whichpattern is being consciously perceived bythe cat.

Gamma synchrony has been further asso-ciated with consciousness in the context ofthe attentional blink effect. The attentionalblink occurs when a subject must detect twotargets in a series of rapidly presented pic-tures (at a rate of about 10 per second). Typi-cally, the second target is detected (and con-sciously perceived) less frequently when itcomes within 500 ms of the first target, asif the subject had ‘blinked’. Fell and col-leagues have argued that the blink could bedue to the suppression of gamma synchro-nization shortly after the response to thefirst target (Fell, Klaver, Elger, & Fernandez,2002). Once again, gamma synchrony wouldbe necessary for conscious perception.

This proposal is consistent with a recentobservation from Lachaux and colleagues,in the face perception paradigm detailedabove (Lachaux et al., 2005), that parts ofthe primary visual cortex shut down, withrespect to gamma activity, after the presen-tation of a Mooney figure: There is a drop ofenergy in the gamma band, below the base-line level, which lasts a couple of hundredsof milliseconds, and is simultaneous withthe induced gamma increase in the fusiformface area. This drop in gamma activity couldbe the trace of a transient deactivation of theprimary visual cortex that could cause thetransient attentional blink after a meaning-ful visual stimulus. The visual cortex wouldbe transiently ‘unavailable’ while processingparticularly meaningful stimuli, as in a reflexprotective mode.

Further hints about the role of gammasynchrony come, albeit indirectly, fromthe experimental contributions of BenjaminLibet (Gomes, 1998; Libet, 2002). In aseries of classic experiments in patientsmixing direct intracranial electric stimula-tions and peripheral somatosensory stimu-lations, Libet revealed a number of interest-ing properties of somatosensory awareness:(1) An electrical cortical or thalamic stimu-lus requires a duration of more than 250 msto be felt, whereas a skin stimulus of 20 msis sufficient. (2) If a direct cortical (elec-trical) stimulus occurs within 250 ms aftera skin stimulus, it can suppress or enhancethe felt perception of the latter stimulus.(3) For a skin stimulus to be felt as syn-chronous with a non-overlapping corticalstimulus, the skin stimulus must be delayedabout 250 ms relative to the latter stimu-lus. Interestingly, all three properties matchquite closely the known temporal dynamicsof the cortical gamma response induced bysensory stimuli. This match is particularlyintriguing considering the fact that Libetused rhythmic electrical stimulations in thegamma range (typically 60-Hz trains ofelectric pulses).

If the induced gamma response isinvolved in the conscious perception of asensory stimulus, then one would indeedexpect that a rhythmic train of electri-cal stimulations in the gamma range couldmimic the effect of the induced gammaresponse, if it possesses the same temporalproperties; that is, if it starts roughly 250 msafter the mimicked stimulus onset and lastsfor at least 250 ms. Then it should be felt assynchronous with a corresponding skin stim-ulus and possibly interfere with perceptionof that latter stimulus. In brief, Libet’s obser-vations can readily be interpreted via theinvolvement of the sensory-induced gammaresponse in sensory awareness, at least in thecase of somatosensory awareness.

In summary, the previous studies cer-tainly build a strong case for the role of res-onant assemblies, oscillating in the gammarange, as neural correlates of sensory aware-ness. Nevertheless, this assessment is not theend of the story, for a number of arguments



make it difficult to equate gamma synchronyand consciousness.

Problems Concerning the Link BetweenGamma Synchrony and Consciousness

The first problem to mention is that gammasynchrony can be observed in unconsciousanesthetized animals, although it is strongerwhen animals are awake (see Sewards &Sewards, 2001, for arguments against therole of gamma synchrony in conscious-ness). Sewards and Sewards further arguethat gamma oscillatory activities have beendetected in structures that most likely donot participate in the generation of sen-sory awareness, such as the hippocampalformation: “Obviously hippocampal activi-ties could not contribute to sensory aware-ness since lesions to that structure do notresult in purely sensory deficits of any kind”(Sewards & Sewards, 2001, p. 492). Thisargument, as well as others, leads them toconclude that “while synchronization andoscillatory patterning may be necessary con-ditions for activities to participate in gener-ating awareness, they are certainly not suffi-cient” (Sewards & Sewards, 2001, p. 492).23

This point echoes the conclusions from astudy by Revonsuo and colleauges (Revon-suo, Wilenius-Emet, Kuusela, & Lehto,1997). In this study, they recorded thegamma band response of normal subjectsduring the fusion of random-dot stere-ograms. They observed that, although 40-Hzsynchronized oscillations seemed to partici-pate in the construction of the unified per-cept, they were not maintained during thecontinuous viewing (and conscious percep-tion) of the same stimulus once it hadbeen constructed. Lachaux (unpublishedfindings) repeatedly confirmed this obser-vation with human intracranial recordings:The gamma response induced by durablevisual stimuli in the visual system oftenstops before the end of the stimulus pre-sentation, despite the fact that the sub-jects still fixate the images and consciouslyperceive them.24

These considerations indicate that otherspatiotemporal structures may participate

in the emergence and the stabilization ofthe conscious percept. The presence of suchstructures is especially the case for short-term memory, which has been proposedas a central component of consciousness(Baars & Franklin, 2003). In visual short-term memory, when an individual has tomaintain a conscious representation of acomplex visual shape, using mental imagery,for a couple of seconds, synchrony occurs notin the gamma range, but in the lower betarange (between 15 and 20 Hz), between dis-tributed sites of the ventral visual pathway(Tallon-Baudry et al., 2001). Therefore, res-onant cell assemblies in the beta range mayalso subserve continuous visual perception(if only in its imagery aspect).

The above studies emphasize the pointthat gamma synchrony may be necessary forthe emergence of a conscious perception,but perhaps only in this emergence. Onceformed, the percept could then continue viaother cell mechanisms, in the form of othertypes of spatiotemporal structures.

Nevertheless, even at the initial level ofthis emergence, the role of gamma syn-chrony needs to be clarified. As we haveseen, gamma synchrony occurs in anes-thetized animals and is therefore not suffi-cient for consciousness (Sewards & Sewards,2001). One interesting possibility, in thecase of the visual system, is that gammasynchrony could be involved in the forma-tion of visual objects. Visual objects arethe preferred targets of visual attention,and yet they present themselves to us onlyvia conscious perception. Furthermore, asargued by the Feature Integration Theory(Treisman & Gelade, 1980), visual objectsseem to require visual attention to form.The question thus arises of which comesfirst: objects or attention. One solution tothis problem is that in the absence of atten-tion there are only ‘pre-objects’; that is,bundles of features that are object candi-dates and that are sufficient to attract atten-tion, which would then finish the construc-tion and remain grabbed by them (Wolfe &Bennett, 1997).

Engel and Singer (Engel et al., 1999) pro-pose that gamma synchrony may mediate



this mechanism. According to this proposal,proto-objects, based on their physical fea-tures and Gestalt properties, assemble in theform of nascent cell assemblies via gammasynchrony. This synchrony corresponds tothe kind of ‘automatic’ synchrony observedin anesthetized animals. This nascent syn-chrony is reinforced in awake animals, suchthat there is a formation of the visual object.This process corresponds to the grabbing ofattention by the object and is simultaneouswith the object’s actual formation for per-ception. In this model, attention and gammasynchrony become two sides of the samecoin, as long as one is ready to extend theconcept of attention (usually associated withconscious perception) to a general selec-tion mechanism that includes an uncon-scious preselection mechanism. This pres-election mechanism is the one observed inanesthetized animals. Attention, in its clas-sic ‘conscious’ sense, is thus envisioned as thetip of the selection iceberg.

Can we therefore relate the full formationof resonant gamma assemblies to the emer-gence of consciousness? The answer wouldseem to be yes, in a certain sense; namely,that the content that is correlated with theformation of the resonant gamma assem-bly is accessible to verbal report, workingmemory, and so on. On this view, gammasynchrony is necessary for any kind of sen-sory awareness. This view gains support fromEngel and Singer’s observation that syn-chrony is related to all of the four presumedcomponent processes of awareness: arousal,segmentation, selection, and working mem-ory (Engel et al., 1999). In the, however, weexamine certain problems with this idea thatlead us to qualify it.

Consciousness and DynamicalStructures: Some Qualifications

Introduction

Throughout this chapter we have exploredthe view that consciousness seems to requirethe formation of distinct, dynamic spa-tiotemporal structures in the brain. This

view is, after all, one of the main pointsof agreement among the different neurody-namical proposals we reviewed earlier. Inthis section, we take a more critical stanceregarding this central issue and put forthsome qualifications we believe are importantto keep in mind.

In several of the neurodynamical theo-ries we have discussed, the notion of a dis-tributed neuronal assembly, understood assome kind of synchronous pattern of acti-vation, is central to explaining the neuronalbasis of consciousness. As we saw in the pre-ceding section, the gamma band has been apreferred region of the frequency domain,in which such assemblies have been studied.Whether restricted to this frequency band orspanning multiple frequencies, an emergentand stabilized spatiotemporal pattern is seenas a prerequisite for conscious experience tohappen.

This viewpoint, however, raises at leasttwo related questions. On the one hand, ifsuch patterns are necessary for conscious-ness, and if we can distinguish them ashaving a certain spatiotemporal unity, whathappens between patterns? Are we con-scious during such transitions? Or is con-sciousness a sequence of snapshots, in whichthe apparently seamless fusion of succes-sive moments into the ongoing flow ofexperience is achieved by some additionalmechanism?

On the other hand, can we define a stableconscious moment within the flow, and arewe therefore entitled to suppose that duringsuch a moment, the assembly will ‘hold’ or‘contain’ a certain unity, even though duringthat moment one can distinguish a change(or changes) in one’s experience? Recall thatdynamic assemblies are supposed to last forseveral hundreds of milliseconds, but oursensory experience can change within thatduration. Suppose, for example, you are sit-ting in a train, staring out of the window,and as you look out into the countryside,trees, electricity poles, and other objects swi-fly cross your visual field, without your beingable to grasp them fully and stably. Yet youknow they are trees, electricity poles, andother objects. Does your rapid experience



of each of these objects correspond to a dis-tinct assembly? Or is it rather a matter ofone global assembly, in which various localassemblies ‘ride’? In several neurodynamicalproposals, as we have seen, an experience ofan object is supposed to depend on the for-mation of distinct, coherent brain patterns.But a conscious moment can include full-fledged objects as well as less definite visualpatterns that, although conscious to a cer-tain extent, cannot be completely describedas stable entities.

These two interrelated features – ongoingflow and fleeting experiences – need to beaddressed by any neurodynamical approachto consciousness. In the remainder of thissection, we discuss both features and pro-pose a simple distinction that may help clar-ify the issues at hand.

Ongoing Flow and Fleeting Experiences

The issue of the ongoing, fluid nature ofconscious experience is certainly not new.25

William James, in his famous chapter on“The Stream of Thought” (James, 1890/1891,Chapter IX), provides a detailed descriptionof the structure of this flow. He distinguishesat least two fundamental aspects – ‘substan-tive’ stable moments, in which one is actu-ally conscious of something, and ‘transitive’fleeting moments, in which one passes fromone content to another. He describes con-sciousness as like a bird’s life, for it seems tobe made up of an alternation of flights andperchings. James remarks that substantivemoments can be recognized as such, whereastransitive moments are quite difficult to pin-point accurately. They present themselves astendencies and changes between states, andnot as distinct contents immediately defin-able in themselves, save by some retrospec-tive exercise.

How do these phenomenological obser-vations relate to the neurodynamical pictureof the brain and its relation to consciousness?As we have seen, most neurodynamicalproposals stress that each conscious statedepends on a specific neural assembly oremerging dynamic pattern, but the issue ofhow transitions between states take place

and what they mean in terms of the expe-riencing subject is addressed less frequently.With regard to this issue, the proposalsof Varela and Kelso are the most explicitand developed.26 These authors stress themetastable nature of such patterns, so thatsuccessive moments of distributed neuralcoherence combine in a continuous andongoing fashion, in contrast to a sequence ofclear-cut states.27 These approaches presentattractive alternatives that seem to fit nicelywith James’s intuitions. They also allow fora different interpretation of what counts asa meaningful dynamic pattern. Rather thanseeing these patterns as individual assem-blies that arise, maintain themselves for abrief period, and then subside, they can beviewed as one itinerant trajectory, and thusas one pattern (Friston, 1997, 2000; Varela,1999) in which the rate of change is theonly internal definition of the stability of agiven moment. In any case, neurodynami-cal approaches must deal explicitly with thisissue of the apparent unity of the flow ofconsciousness,28 as opposed to the unity ofmoment-to-moment experience.

The second question to which we wishto draw attention is related to the stabil-ity of actual perceived objects during a con-scious moment. As we mentioned above, thenotion of an assembly implicitly incorpo-rates a notion of stability during the lifespanof the pattern in question. Our sensory envi-ronment, however, can be subject to rapidchange in time windows lasting less than sev-eral hundred milliseconds, and yet we are,to a certain extent, aware of the change astaking place. This fact would seem to posea difficulty for any theory that postulatesa neural assembly, organized on a slowertime scale, as necessary for conscious expe-rience. On the other hand, not every objectof the visual scene is perceived as stably asone might naively think. This fact is espe-cially clear in inattentional blindness experi-ments (Simons, 2000). In such experiments,subjects are asked to focus on a particulartask and set of stimuli in a visual scene. Ifan additional stimulus appears unexpectedlyin that scene, the subjects are often unableto report it afterward. What is particularly



striking with such ‘inattentional blindness’is that it can happen even for very distinc-tive and salient objects. In one famous exam-ple (described in Simons, 2000), subjectswatch people passing basketballs. Three peo-ple wearing white T-shirts pass a ball toeach other, while three other people wearingblack T-shirts pass another ball to each other.The subjects have to count the number ofpasses between the white players, whichoccur at a fast enough rate to require thefull attention of the viewer. After 45 s of thedisplay, a man in a gorilla suit walks acrossthe scene, stops for a moment in betweenthe players, waves his hands in the air, andthen exits through the other side 5 s later.It is well documented that a high portion ofthe viewers fail to report seeing this gorilla.

In models like the one advocated bySinger and collaborators (see above) thereis a strong correspondence between a fig-ure/ground distinction (and therefore anobject) and the formation of a synchronousassembly. This correspondence would seemto imply that only fully formed assem-blies can ‘support’ some type of percep-tual recognition of the object in question.As discussed above, however, both phe-nomenological observation of one’s ownexperience and experiments such as theunnoticed gorilla suggest that a great dealof experience may be unstable and fleet-ing. Where would such fleeting experiencesof quasi-objects fall in the framework ofdynamic neural assemblies? Lamme (2003 ,2004) has proposed that such fleeting expe-riences belong to ‘phenomenal conscious-ness’ (i.e., are subjectively experienced, butnot necessarily accessible to verbal report),whereas more stable experiences belongalso to ‘access consciousness’ (i.e., are avail-able to verbal report and rational actionguidance; see Block, 1997, 2001, for this dis-tinction between phenomenal consciousnessand access consciousness).29 Neurodynami-cal models need to be able to account for thisevanescent aspect of conscious experience ina more explicit way.

More precisely, we propose that the sta-ble/fleeting duality be considered a struc-tural feature of consciousness experience(see also the next section) and dealt with

accordingly. In a certain sense, this dual-ity mirrors the access/phenomenal distinc-tion, but without assuming that there canbe fleeting phenomenally conscious expe-riences that are inaccessible in principleto verbal report. In endorsing the need tomake this stable/fleeting distinction, we alsostress the need to consider the possibilityof the more ephemeral aspects of experi-ence as being accessible to verbal report,if approached with the appropriate first-person and second-person phenomenologi-cal methods (Petitmengin, in press; Varela &Shear, 1999).

Given this structural distinction betweenstable and fleeting aspects of experience, itwould be interesting to see how a neuro-dynamical theory that relates the formationof well-defined spatiotemporal patterns inbrain activity to conscious experience woulddeal with the intrinsic mobility of any givenperceptual act. For example, the feedfor-ward stream (or sweep, FFS) is defined as theearliest activation of cells in successive areasof the cortical hierarchy. In the visual modal-ity, it starts with the retina, the LGN, V1,and then the extrastriate visual areas and theparietal and temporal cortex. Thorpe andcolleagues (Thorpe, Fize, & Marlot, 1996)have shown that the FFS is sufficient tocarry out complex visual processing, such asdetecting whether a natural scene presentedfor 20 ms contains an animal. It is temptingto relate the more stable aspect of experi-ence to the formation of spatiotemporal pat-terns, in the sense of dynamic neural assem-blies mediated by recurrent neural interac-tions, whereas the fleeting, unstable aware-ness could be embodied through the rapidFFS that modulates and continuously affectsthe formation of such assemblies while notbeing fully excluded from a certain levelof perceptual experience. This proposal ishighly speculative, but is intended simplyas a way to highlight the necessity of deal-ing with the stable/fleeting structure thatappears to be inherent in each and every con-scious moment.

To conclude this section on qualifica-tions to the dynamic approach, we wouldlike briefly to draw the reader’s attentionto another aspect of consciousness that is



significant in light of the preceding discus-sion and the overall topic of this chapter.This aspect is the subjectivity or subjec-tive character of consciousness. For exam-ple, Damasio (1999) has stressed that, inaddition to understanding the neurobiolog-ical basis for the stream of object-directedconscious experiences, it is also necessary tounderstand the neurobiological basis for “thesense of self in the act of knowing” (Parvizi &Damasio, 2001; see also Panksepp, 1998, fora convergent argument, and Wicker et al.,2003). The sense of self with which Dama-sio is concerned is a primitive kind of con-scious self-awareness that does not dependon reflection, introspection, or possession ofthe concept of a self. In phenomenologi-cal terms, it corresponds to the fundamental‘ipseity’ (I-ness or selfhood, by contrast withotherness or alterity) belonging to subjectiveexperience (see Chapters 4 and 19).

In a related line of argument, Searle(2000) has suggested that a major draw-back of current attempts to uncover theneural correlates of consciousness in humanbeings is that they begin with already con-scious subjects. He advocates a ‘field ofconsciousness’30 viewpoint, in which theperceptual experience of an object arises asa modification of a pre-existing conscious‘ground-state’ that is unified, subjective, andqualitative. In this context, the transitionbetween conscious states need not be punc-tuated by a radical gap in consciousness, butcan rather be a modulation of a more basicstate of background consciousness, whichaccounts for the fact that even such transi-tive moments are felt as belonging to one-self. Here dynamic patterns in the formof transient and distributed coactive assem-blies would mainly reflect the nervous sys-tem’s own homeodynamic activity; that is,its maintenance of a range of internal regular-ities in the face of its ongoing compensationfor the systematic perturbations to which itis exposed from both the sensory environ-ment and the internal bodily milieu (Dama-sio, 1999; Maturana & Varela, 1980).

Nevertheless, it remains difficult to seehow metastable assemblies of coactive neu-rons could by themselves account for thiscrucial aspect of the subjectivity of con-

sciousness. This crucial feature is often putto the side as something to deal with oncethe issue of the neural correlates of percep-tual consciousness has been resolved (e.g.,Crick & Koch, 2003). Our view, however, isthat unless the subjectivity of consciousnessis adequately confronted and its biologicalbasis understood, proposals about the neuralcorrelates of perceptual consciousness willprovide limited insight into consciousnessoverall. Thus, the issue of subjectivity is anon-trival matter that any neurodynamicalapproach must confront sooner or later if it isto become a cogent theory of consciousness.We briefly pick up this thread in the whendiscussing how to relate phenomenologicaldescriptions to neurodynamical accounts.

The Future: Beyond Correlation?

Introduction

So far we have dealt primarily with the issueof meaningful spatiotemporal patterns in thebrain and their relevance to the study ofconscious experience. It may have becomeincreasingly evident to the reader, however,that the issue of how to relate such pat-terns to experience as a first-person phe-nomenon has been left untouched. Indeed,one of the major challenges facing the cogni-tive sciences is precisely how to relate thesetwo domains – the domain of third-personbiobehavioral processes and the domain offirst-person subjective experience. What isthe right way to conceptualize this rela-tion, and what is the best way to approachit methodologically? These questions havenot yet received anything near a satisfactoryanswer from the neuroscientific community.We do not intend to propose an answerto them here. Rather, we wish to highlightsome conceptual and practical issues in thequest to understand the relation betweenthese two domains while keeping in mindthe dynamical insights we have gained fromthe previous exposition.

Correlation and Emergence

The first question that comes to mind is theextent to which the entire neurodynamical



approach rests on a merely correlationalstrategy. In coarse terms, one isolates a giventarget experience, say the perception of afigure; one determines the neural patternsthat correlate with the moment the subjectsees the figure; and one then concludes thatthe conscious experience depends on suchneural patterns.31 In the last decade or sothis correlational approach, in the form ofthe search for the neural correlates of con-sciousness, has undergone important devel-opments and become more sophisticatedwith regard to its conceptual formulation,methodological commitments, and empiri-cal results (Block, 1996; Crick & Koch, 1990,1998; Rees, Kreiman, & Koch, 2002). Herethe central idea is that, rather than formu-lating explanatory principles about the rela-tion between neural activity and experience,what has to be done first is to determinethose neural processes that can count as a“specific system in the brain whose activitycorrelates directly with states of conscious-ness” (according to the Association for theScientific Study of Consciousness, cited byChalmers, 2000, pp. 17–18). Once such pro-cesses have been found, then one can turnto the issue of how they are causally relatedto experience itself.32

Neurodynamics as a research programis devoted, at least methodologically, tothis correlational strategy and in this senseremains closely linked to the NCC program.Of course, this commitment is due to thefact that, in the scientific tradition, establish-ing a relation between two target events orphenomena is mainly approached by estab-lishing a correlation in their occurrence.Causal relations can then be assessed on thebasis of altering one of the target eventsand observing whether and how the otherchanges. This ‘interventionist’ strategy canbe employed in the case of brain functioningand consciousness by using microstimula-tion during surgery or transcranial magenticstimulation (TMS). Nevertheless, by itselfthis strategy does not guarantee the eluci-dation of the underlying causal mechanisms.

Several of the proposals reviewed above,however, formulate explicit links betweenthe neural and the experiential in terms of

the notion of emergence or emergent phenom-ena and thus can be considered as attemptsto go beyond a purely correlational descrip-tion. Although ‘emergence’ is a complexconcept subject to multiple interpretations(see Keslo, 1995 ; Thompson, 2007; Thomp-son & Varela, 2001), in simple terms it canbe defined as follows: A process is emergentwhen (i) it belongs to an ensemble or net-work of elements, (ii) it does not belongto any single element, and (iii) it happensspontaneously given both the way the ele-ments interact locally and the way thoseinteractions are globally constrained and reg-ulated. Thus, an emergent process cannot beunderstood at the level of local componentstaken individually, but depends rather on therelations established between them. Further-more, an emergent process not only dependson the local components but also constrainstheir degrees of freedom, a two-way pro-cess that has been termed ‘circular caual-ity’ (Haken, 1983). Especially in Freeman’sand Varela’s approaches, conscious experi-ence is considered to be an emergent process.The difference between their views is thatwhereas Freeman (1999a,b) proposes thatconsciousness is a global brain state, Varelaproposes that consciousness may encompassmultiple cycles of organismic regulation thatare not fully restricted to the brain (Thomp-son & Varela, 2001). Neverthless, althoughprinciples of emergence have been clearlyformulated at the level of physical processesand molecular interactions (Nicolis & Pri-gogine, 1989), in the case of conscious expe-rience, such principles still need to be under-stood and formulated in a more rigorousway. We believe that the study of complexsystems offers a promising approach in thisdirection (Le Van Quyen, 2003 ; Thompson,2007).

Gaining Access to Experience

As mentioned earlier, any neurodynamicalapproach to consciousness must eventuallydeal with the issue of how to describeexperience itself. In the previous sectionsof this chapter we have discussed mainlyspatiotemporal brain patterns in relation to



consciousness, but we now turn to considerthe other side of the issue; namely, how togain access to experience itself and render itaccessible to scientific description.

On the more operational side of this ques-tion, one can ask how it is possible to setup an experimental paradigm that addressesthe issue of gaining access to experience ina way that allows us to study the underly-ing neuronal processes. Not all that is goingon in the brain is necessarily related to whatthe subject is consciously experiencing. It isknown that, during our conscious engage-ment with the world, a non-negligible partof our adapted behavior depends on non-conscious processes that are carried on with-out us being aware of their functioning. Forexample, do you have any feeling what-soever of the oxygen level in your bloodright now? Yet this bodily state of affairscan be crucial to your capacity to be hereright now reading this text. Although thisexample is extreme, carefully crafted exper-iments reveal that even perceptual informa-tion can be used to guide behavior in a non-conscious way. For example, when a subjectis presented with a small circle surroundedby larger circles, the small circle appearssmaller than if it is presented in isolation.Yet if the subject is asked to reach for it,his fingers adopt a grip size that is consis-tent with the true size of the circle andnot with its illusory dimension (Milner &Goodale, 1995). Another classic example isknown as blindsight (Danckert & Goodale,2000; Weiskrantz, 1990). In this neurologi-cal condition, conscious visual experience isimpaired due to damage in primary visualcortex, yet subjects can produce quite accu-rate motor actions, such as introducing anenvelope through a horizontal slot or point-ing to a target they claim not to see. Thus, theproblem arises of how to determine thoseneural processes that show some kind ofdirect relation to the actual conscious expe-rience of the subject, in contrast to those thatsustain ongoing and non-conscious adaptivebehavior in the world.

Several experimental approaches havebecome paradigmatic in this endeavor. Ingeneral, the rationale behind these experi-

ments is to dissociate what is presented tothe subject from what the subject sees inorder to distinguish the neural patterns thatare specific for conscious perception. Amongthese approaches, three stand out as themost well studied and influential. The first isvisual masking (Dehaene & Naccache, 2001;Hollender, 1986), in which short-lived visualstimuli flanked by meaningless masks are notperceived consciously, yet can alter futurebehavior (an index of non-conscious pro-cessing). The second is inattentional blind-ness and change blindness (O’Regan & Noe,2001; Simons, 2000), in which diverting thesubject’s attention can render major changesin the scene unnoticed. The third is binocu-lar rivalry (Blake, 1989; Blake & Logothetis,2002), in which the presentation to each eyeof a different image induces an alternation inconscious perception between the two alter-natives, despite the fact that both are alwayspresent.

This last experimental paradigm is par-ticularly relevant to the issue of gainingaccess to experience because it providesan ongoing, slow phenomenon that can bedescribed by the subject. In virtue of itsalternating character, the experience lendsitself to repetitive scrutiny, in order to bet-ter characterize ‘what it is like’ subjectivelyto undergo it. Finally, because both stim-uli do not change, yet perception changesdramatically, binocular rivalry evidences theendogenous and ongoing character of expe-rience and therefore calls for attending tothose neural processes that share this funda-mentally dynamical structure.33

These considerations suggest that, inaddition to using experimental paradigmsfor dissociating unconscious and consciousprocesses, we need to be able to capturethe dynamics of experience itself. Hence itis necessary for the experimenter to takemeasurements of each phenomenon – thedynamics of the brain and the dynamics ofexperience. Measurements should providepublic data; that is, information that can beshared with another observer. One recurrentproblem with consciousness is that the directobservation of experience is accessible onlyto the subject, and such observation is not



a public measurement in itself. The experi-ence therefore needs to be transcribed intopublic data in a subsequent step to provideso-called first-person data. What the statusof first-person data is and to what extent thesubject can play an active role in describ-ing his or her experience are matters ofactive debate in the science of consciousness(Jack & Roepstorff, 2003 ; Varela & Shear,1999). We cannot review these debates here.Rather, in the remainder of this section, wewish to explore two complementary lines ofinvestigation that are relevant to the issueof making experience more scientificallyaccessible.

A Topological Approachto First-Person Data

In a general sense, one expects measures tobe somehow organized in a universe, themeasurement universe, that is the set of allthe possible ‘values’ that measure can take.The term ‘universe’ must be understood inits statistical sense and simply refers to theset of possible values, states, or items thatcan be valid measurements. For instance, asingle word is one particular item among allthe possible words. The universe may be dis-crete (as for words or sentences) or continu-ous (as for magnetic fields). In any case, ameasurement will be the selection of oneparticular value allowed in a given universe,based on the present state of the observedphenomenon. For a given subjective experi-ence, this may correspond to the selection ofone description, among all the possible writ-ten descriptions that can be produced (say)in a couple of minutes.34

We mentioned that the measurementshould be ‘organized’. This means thatit should be provided with some sort oftopology: It should be possible to estimatea distance between two measures. Indeed, itshould be possible to say whether measure Ais closer to measure B than it is to measure C(see Fell 2004 for a convergent discussion).Without any kind of topology, it would bedifficult to compare the dynamics of the twophenomena. For instance, the notion of sta-

bility requires distance. Stability means thatthe phenomenon remains somewhat con-stant during a certain time interval; it fur-ther means that the distance between con-secutive measures is shorter now than whatit was in earlier observation windows. Thenotion of distance is also central to the con-cept of recurrence: If we find a certain neu-ral pattern that correlates with a consciousexperience, we expect this neural pattern torepeat when the same experience repeats.Because neither neural patterns nor experi-ences repeat in a perfectly reproducible way,we also need a way to know whether a cer-tain neural pattern or experience looks likeone that occurred in the past. This requires aquantification of resemblance between twomeasures; that is, a distance.

Note that this first definition is largeenough to include many possible measures.In fact, a dance could be considered as a mea-sure or a series of measures if each succes-sive body configuration constitutes by itself ameasure. A drawing could also be a measure.But to be actually useful, we insist that thesubject and the experimenter should agreeon a measure of distance, which enablesanybody to evaluate the degree of similar-ity between two measurements. The BasicRequirement (so called in the following) isthat the distance should be consistent withthe experience of the subject (as only thesubject can tell): If measure A is closer tomeasure B than to measure C, then the ele-ments of experience that led the subject toselect measure A should appear to him ascloser to the elements that led him to choosemeasure B than to the elements associatedwith measure C.35 This requirement directlyimplies, for instance, that recurrences in thesubject’s experience should translate intorecurrences in the measure.

Once provided with measures of (someelements of) the subjective experience andwith measures of neural phenomena, itshould be possible to establish a relationshipbetween the two phenomena by compar-ing the dynamics of those measures: Relatedphenomena should provide sets of mea-sures with compatible dynamics. That is,



once again, stability in experience should beassociated with stability (or stationarity) inthe neural dynamics, whereas moments ofchange should be correlated with changing(or non-stationary) neural processes.36

A ‘Structural Invariants’ Approachto First-Person Data

As a complement to the fine-grained topo-logical description presented above, it seemspossible to adopt what can be termed a‘structural invariant’ strategy. Here the mainaim is to obtain, through descriptions of thetarget experience, an account of that whichis invariant (or stable) as a feature of theexperience, regardless of whether it is oneor another subject that undergoes it. Theroots of this approach go back to the methodadopted in phenomenological philosophy(see Chapter 4). Here, through several rep-etitions of the same experience in differentcontexts, one can arrive first at a certain sub-jective invariant, and then, through contrastwith other subjects, intersubjective invari-ants that are present in the original expe-rience, no matter how many versions of itone tries and no matter how many differentsubjects engage in it. A traditional exampleis the structure of the visual field, in whichwhat one sees focally always appears as arelatively detailed center surrounded by anincreasingly less detailed region, which, atthe limit, fades into an ungraspable indeter-minacy. In the particular context of the neu-rodynamics of consciousness, the relevanceof this type of approach can be illustratedby recent work on the experience of binoc-ular rivalry (Cosmelli et al., 2004).

As we briefly described above, binocu-lar rivalry occurs whenever one is presentedwith dissimilar images, one to each eye. Thesubjective experience is that of an ongoingalternation between both possible images,with only one of them consciously perceivedat a time. If the images are large, then dur-ing the transition from one to the other, onecan distinguish a mosaic, patchwork patterncomposed of both images, but as a rule, ifthe adequate contrast and luminance con-

ditions are met, at any given point of thevisual image of only one of the images (orpart of it) will be seen (will dominate) in anexclusive way. In general, binocular rivalryis considered a clear-cut alternation betweentwo states, and average measures of the brainstate during one or the other dominanceperiod are contrasted. Most commonly, thesubject’s indication via a button press ofthe moment when the alternation takesplace is used to fix a rigid temporal referencearound which the average brain responsesare defined.

We recently used this experimental pro-tocol to investigate the underlying neuralpatterns, but with the specific objectiveof describing their spatiotemporal evolu-tion throughout extended periods and with-out presupposing a rigid two-state structure(Cosmelli et al., 2004). To do so, we workedwith a group of subjects who were exten-sively exposed to the experience and pro-duced free, ongoing descriptions of whatthey were seeing and how they were expe-riencing it. As conflicting stimuli we useda human face and a moving pattern withan intrinsic frequency (a frequency tag; seeBrown & Norcia, 1997; Tononi & Edelman,1998). This intrinsic frequency was incor-porated in order to tag a neural evokedresponse that could be followed by magne-toencephalography (MEG).

The descriptions produced by the sub-jects showed some interesting features: Inaddition to experiencing the well-knownalternation between both images, the sub-jects repeatedly described this alternation asextremely variable in the way it occurred.Although sometimes the alternation fromone image to the other started in the cen-ter of the field and progressed toward theouter limits, in other occasions it began onone side, from the top or the bottom, or evenfrom the external borders, and then progres-sively invaded the pre-existing image. Mostsubjects claimed that it was difficult to givea stable description of how these transitionstook place, because at each time they devel-oped in a different way. Nevertheless, allsubjects invariantly stated that dominance



periods would alternate and recur, no mat-ter what the subjects did or how much theytried to prevent it from happening.

At a first coarse level, these descriptionsalready provide us with some crucial aspectsof the experience of rivalry: This experienceis one of an ongoing flow of recurrent domi-nant periods, in which alternations are ex-tremely variable in the way they develop.This feature is indeed a hallmark of binocularrivalry that will be experienced by any nor-mal observer and is thus a structural invari-ant in the sense described above. Althoughthis descriptive feature is not particularlynovel, it nevertheless points toward a con-crete restriction in the methods we needto choose to analyze the underlying neuralprocesses (and consequently what we under-stand as the neural underpinnings of con-sciousness). If we wish to reveal neural pat-terns that are meaningful in the contextof this specific experience, then we cannotimpose a rigid temporal grid and supposethat there is such a thing as an average transi-tion from one image to the other. This point,however, is rarely acknowledged. We there-fore developed a statistical framework thatconsidered significant any neural activitythat is recurrent in time, without any restric-tions on the temporal pattern of activation.The result was an original description of anetwork of distributed cortical regions thatshowed synchronous activation modulatedin concert with conscious dominance peri-ods. Moreover, the dynamics of modulationof these brain patterns showed a striking sim-ilarity to the bell-type pattern that WilliamJames had predicted (more than a centuryago) would underlie the occurrence ofany given conscious moment (James, 1890/1981).

An important contribution of the struc-tural invariant approach is thus that it canserve as an effective constraint on howwe study the dynamic brain patterns. Basicphenomenological observation shows thatexperience (or the stream of conscious-ness) is at least (i) dynamic and ongoing;(ii) continuous;37 (iii) able to be parsed,so one can distinguish in a given subjec-tive experience components or aspects that

are more visual, or more auditory, for ins-tance, and eventually segment it along suchdimensions; and (iv) recurrent, in the sensethat we recognize objects, feelings, thoughts,memories, etc., as seen or felt before, eventhough they are never experienced in thesame way. These properties, although cer-tainly not exhaustive of our conscious lives,do suggest that methods that allow for pro-cesses of compatible dynamics should bepreferred if we want to advance in our under-standing of the neural underpinnings of con-sciousness.

In addition to this methodological con-straint, however, the structural invariantsapproach can potentially make a furthercontribution. As we mentioned above, oneof the most prominent structural invari-ants of consciousness is precisely its subjec-tive character, in the sense of its fundamen-tal prereflective and preconceptual ‘ipsiety’(see Chapters and ; see also Zahavi, 2005 ,for an extended discussion). This backdropof consciousness pervades the occurrenceof specific states of perceptual conscious-ness. It would appear to call for an expla-nation not so much in terms of the dynamicbehavior of the system (e.g., only in termsof the dynamical properties of the nervoussystem’s patterns of activity), but rather interms of how a certain self-referring perspec-tive can emerge from a certain dynamical orga-nization (Rudrauf et al., 2003 ; Thompson,2007). Whether this type of account isbeyond the domain of neurodynamics aswe have defined it here is an empiricalissue. The crucial point is that if, throughsome enriched neurodynamical plus organis-mic plus biological approach (e.g., Damasio,1999; Varela, 1979), one could account forthe conditions of possibility of a minimallysubjective system, then transcending a purelycorrelational strategy would become a realpossibility.

Can We Avoid the Pitfallsof Introspectionism?

One recurrent question, when discussingthe use of first-person data, is howto avoid the pitfalls of introspectionism.



Introspectionism was an attempt to useintrospection as a scientific method to elab-orate psychological theories. It was the mainscientific approach to mental phenomena atthe beginning of psychology, but was laterdismissed by the scientific community infavor of behaviorism (reviewed by Vermer-sch, in Depraz, Varela, & Vermersch, 2003).The main problem with introspection, asused at that time, was that it provided con-flicting theories. The root of the problemwas in fact methodological: It was neverpossible to ascertain whether the introspec-tive reports met the Basic Requirementmentioned above, and there were seriousdoubts about the correspondence betweenthe descriptions of the experiences and theexperiences themselves. On the other hand,there was little explicit description of theintrospective method by which to proceedto explore and describe experience, andhence the actual testing and refinement ofthe research method, as opposed to the con-tent of its descriptions, remained underde-veloped (Varela, 1996). Consequently, animportant part of the cognitive science com-munity is generally reluctant to use first-person data. Is it therefore possible to build aneurodynamics of consciousness, given thatit must rely on first-person data?

Our position is that the whole issue isa technical one: If the measure providingfirst-person data meets the Basic Require-ment described above, then the measure isuseful. Alternatively, in the structural invari-ant approach, if a given invariant is stableacross all subjects for a given experimentalparadigm, it should be considered valid. Infact, the real question is not whether cogni-tive scientists should ‘trust the subject’, butin which conditions they can trust the sub-ject and what they should ask. First-persondata, defined as measures of the subjectiveexperience, are continuously being used inpsychophysics: When a subject presses abutton to indicate that he saw a blue square,and not a red circle, he provides a mea-sure of his immediate perceptual experiencein its simplest form. In this extremely sim-ple form, first-person reports are consideredas perfectly valid and trustworthy. At the

other extreme, first-person data about theprecise dynamic of subtle variations of emo-tions would probably be considered less reli-able (this means that they would not meetthe Basic Requirement – the same subtlevariations would not lead to the same first-person data, if repeated).38 So, in fact, thereal question concerning first-person datais, Where shall we draw the line betweenwhat is acceptable, perfectly good data, andwhat is not?39 A related question of equalimportance is whether this line is the samefor all individuals and whether it is fixedwithin a single individual or whether train-ing can move the line (see Chapter 19).40

We believe that this question should becomecentral in cognitive neuroscience in the nearfuture, especially in view of the adventof new fields, such as the neuroscience ofemotions or the neuroscience of conscious-ness itself. Such emergent fields rely heav-ily on trustworthy measures of subjectiveexperience.

Conclusion

In summary, the neurodynamics of con-sciousness is an attempt to relate twodynamical phenomena that take place in asubject – the formation of metastable pat-terns in the subject’s neural activity and thetransient emergence of dissociable elementsor aspects of his or her conscious experience.To establish such a relation, cognitive neu-roscientists need to observe systematic sim-ilarities between the dynamical propertiesof these two phenomena. In this sense, theneurodynamical approach works at the levelof correlations, albeit refined ones. On theexperiential side, this approach requires thesubject to provide first-person descriptionsthat can serve as ‘public’ measures of expe-rience, with at least two objectives. The firstobjective is to capture reliably the degreeof similarity (or disparity) between differ-ent subjective phenomena and produce tim-ings that can be compared to the timing ofneural measurements. The second objectiveis to produce descriptions of the structuralinvariants of the experience in question, in



order to constrain the methods that are cho-sen to determine which neural activity is tobe considered significant. It is not yet clearhow much of the complexity of conscious-ness can be revealed in this way, and thisquestion constitutes an important field ofinvestigation for the future.

Notes

1. We define in more detail the notions of func-tional segregation and cooperative interac-tion in the section, “The Search for Mean-ingful Spatiotemporal Patterns in the Brain”.Here we only say that they can be consideredanalogous to local specialization and collec-tive interaction, respectively.

2 . This situation is, of course, only suggestive.A dream state might be a more rigorous caseof a true sensory filter (Llinas & Pare, 1991).

3 . One might rightly consider other variables,such as the local concentrations of certainneurotransmitters. Quantitative measures ofthe glial system should probably also beincluded. In fact, there is no one single wayof choosing which variables to include inthe system. This choice is largely drivenby our current knowledge of the nervoussystem, which unfortunately remains quitelimited. As a starting point, one needs tokeep the following three elements in mindwhen choosing the variables of the system:(1) The time scale: If one candidate vari-able maintains a constant value during thetime of observation of the system, then itdoes not need to be counted as a variable,but rather can be considered as a parame-ter (see below). (2) The spatial scale: Thenervous system can be modeled at severalscales – molecules, neurons, neural popula-tions, etc. The variables should be meaningfulat the spatial level of investigation. (3) Theinterdependence within the system: If thevalue of one candidate variable is fully deter-mined by the values of the other variables,then it does not need to be included in thesystem.

4 . Alternative definitions of the nervous sys-tem as a dynamical system can include thesynaptic weights themselves among the vari-ables. At certain time scales, the weights area function of the evolution of the mem-

brane potentials; for instance, via long-termpotentiation (LTP) mechanisms. Models thatinclude a changing connectivity can quicklybecome unmanageable, however, both math-ematically and computationally (but see Ito& Kaneko, 2002).

5 . For further details on this subject, westrongly recommend one of the original andmost influential sources in neurodynamics byWalter Freeman (Freeman, 1975). See alsoBressler & Kelso, 2001.

6. For this reason, it would not necessarily bemeaningful to repeat the same perturbationover and over to study the average reactionof a chaotic system, for there is no guaran-tee that this average reaction would have anymeaning. Yet, such averaging procedures arethe basis of almost all the imaging studies ofthe nervous system.

7. Note that the exact reaction of the system –that is, the precise trajectory that the systemwill follow to converge on the attractor – canbe very different from one olfactory stim-ulation to another, even though the targetattractor is the same for all. Therefore, theexistence of attractors is compatible with theintrinsic variability of chaotic systems.

8. A recent review notes that “incontrovert-ible proof that EEG reflects any simplechaotic process is generally lacking. There aregrounds for reservation concerning reports ofthe dimensionality of EEG from direct mea-surement. Fundamental difficulties lie in theapplicability of estimation algorithms to EEGdata because of limitation in the size of datasets, noise contamination, and lack of signalstationarity” (Wright & Liley, 1996).

9. In human electrophysiology, for instance, thedominant paradigm is recording the EEG ofhuman subjects while presenting them withseries of similar sensory stimulations. The sig-nal studied is the evoked potential: the meanEEG response averaged over all the stim-ulations. The intertrial variability is consid-ered as noise and disappears in the averagingprocedure.

10. As Le Van Quyen (2003 , p. 69) notes, “Inphysics, what is usually referred to as self-organization is the spontaneous formation ofwell organized structures, patterns, or behav-iors, from random initial conditions. Typ-ically, these systems possess a large num-ber of elements or variables interacting in acomplex way, and thus have very large state



spaces. However, when started with someinitial conditions, they tend to converge tosmall areas of this space which can be inter-preted as a form of emergent eigenbehavior.”

11. Mutual information quantifies the ability topredict the behavior of one element in thesystem from the behavior of one or sev-eral other elements (David, Cosmelli, & Fris-ton, 2004). This measure is one of severaltools used to quantify statistical dependencewithin a system. Note that what counts asa spatiotemporal structure will depend onwhich measure of statistical dependence isused.

12 . “It seems that short-term memory may bea reverberation in the closed loops of thecell assembly and between cell assemblies”(Hebb, as cited in Amit, 1994 , p. 621).

13 . A precise definition of synchronization bet-ween chaotic systems can be found in(Pikovsky, Zaks, Rosenblum, Osipov, &Kurths, 1997, p. 680): “The phase synchro-nization of a chaotic system can be defined asthe occurrence of a certain relation betweenthe phases of interacting systems or betweenthe phase of a system and that of an exter-nal force, while the amplitudes can remainchaotic and are, in general, uncorrelated (seealso Brown & Kocarev, 2000; Rosenblum,Pikovsky, & Kurths, 1996).

14 . In Freeman’s words, “preafference providesan order parameter that shapes the attrac-tor landscapes, making it easier to captureexpected or desired stimuli by enlarging ordeepening the basins of their attractors. [ . . . ]corollary discharges do this by a macroscopicbias that tilts sensory attractor landscapes”(Freeman, 1999a, p. xxx).Au: Please

provide pageno.

15 . Edelman and Tononi distinguish primaryconsciousness from higher-order conscious-ness. The former involves the capacity toconstruct a mental scene to guide behav-ior without the semantic, linguistic, and self-reflective capacities unique to the latter.

16. Here we are referring to desynchronizedEEG in the classical sleep/wake cycle sense,in which desynchronized gamma and betafrequencies dominate the EEG of the wak-ing state, by contrast with the synchronizedslow-wave delta frequency EEG of sleep.This notion of desynchronized and synchro-nized EEG as a whole should not be confusedwith the synchronization and desynchroniza-tion of particular EEG signals, which is more

accurately termed phase-locking and phase-scattering respectively (see below).

17. Mathematical methods are available to de-tect correlations between localized meta-bolic activations as measured by fMRI andPET. The advantage of these methods is thatthey provide maps of functional connectiv-ity with a high spatial resolution in normalhuman subjects. These methods, however,measure interactions that occur on the timescale of a couple of seconds at best. Thistemporal resolution may be sufficient whenstudying the neural correlates of slow expe-riential patterns, such as the evolution of cer-tain emotions (Buchel & Friston, 1997).

18. The fractal dimension d of a trajectory canbe envisioned as follows: Imagine that eachpoint along the trajectory is in fact a smallball of lead. Then, the total mass of lead con-tained in a sphere centered on the trajec-tory will increase as a function of the sphereradius r proportionally to rd. If the trajectoryoccupies all the space, then d is equal to thedimension of the space. If the trajectory is astraight line or a circle, d equals 1.

19. See the previous note.20. Note that synchrony can occur between two

neurons without an actual direct relationbetween them, if they are driven by a com-mon driver. This fact reveals one of the limi-tations of the synchrony measure so far. Thethree-neuron system that includes the driver,however, can be seen as a larger spatiotem-poral pattern revealed by the synchrony mea-sure.

21. This theory was based, among other things,on the observation of false conjunctions inthe absence of attention: When presentedbriefly with a red square and a blue circleoutside of the scope of attention, a subjectwould sometimes report having seen a redcircle and a blue square. Such perception istypically an incorrect binding of the color andshape attributes.

22 . This has led to the suggestion that gamma-band synchrony is the trace of similarprocesses in the emergence of dreamingconsciousness in REM sleep and wakingconsciousness (Engel et al., 1999).

23 . We do not wish at this point to step intothe debate about which brain areas actu-ally participate in the generation of sen-sory awareness (see Rees, Kreiman, & Koch,2002).



24 . Letter strings presented to the subject for 1

s, for instance, generate an induced gammaresponse that lasts roughly only for thefirst 500 ms (Lachaux, unpublished observa-tions).

25 . For an extensive presentation of questionsconcerning the experience of time, we referthe reader to the notable work by CharlesSherover (1991).

26. Varela in particular proposed a neurodynam-ical account of Husserl’s phenomenologicalaccount of time consciousness (see Varela,1999, and for further extensive discussion,Thompson, 2007).

27. Tononi and Edelman (1998) do mention thattheir dynamic core is constantly changing,but they do not develop this point further.

28. The question of whether this unity is illusoryor real remains an unresolved problem (Van-Rullen & Koch, 2003).

29. Lamme’s distinction, however, is not com-pletely equivalent to Block’s initial pro-posal (Block, 1995). In its original formula-Au: Block 1996

or 1997? tion, phenomenal consciousness is subjectiveexperience, in the sense that there is some-thing it is like for the subject to be in the state.Access consciousness, on the other hand, is aninformation-theoretical concept that is sup-posed to account for the availability of con-scious information for further rational guid-ance of behavior, including reportability. Theconceptual and empirical validity of this dis-tinction are a matter of lively debate in thescience of consciousness (see Block, 1997,2000; see also the discussion in Thompson,Lutz, & Cosmelli, 2005).

30. The notion that consciousness has a unifiedfield structure goes back to A. Gurwitsch(1964).

31. We see below that this general characteriza-tion needs some important qualifications, inparticular at the level of determining whatcounts as a valid conscious experience andhow to contrast such a conscious experiencewith possibly unconscious processing in sim-ilar situations.

32 . The theoretical validity and empirical plau-sibility of this approach remain a matter ofextensive discussion. Rather than endorse orreject this approach, we wish to highlightit as an influential approach that can serveas a reference for further discussion. Theinterested reader is referred to several inter-esting publications (and references therein)

on this controversial and interesting question(Crick & Koch, 2003 ; Metzinger, 2000; Noe& Thompson, 2004a,b; Pessoa, Thompson, &Noe, 1998).

33 . This feature pertains to multistable andambiguous perception in general (seeLeopold & Logothetis, 1999).

34 . This way of defining measures of subjec-tive experience should be sufficiently gen-eral to include all the measures used inpsychophysics: Choosing to press one but-ton among two, or to press one at a particu-lar time, for example, fits into that definition.Psychophysics is indeed partly about first-person data. For example, the experimentershows a shape to a subject and asks him topress button A if what he sees looks morelike a circle, and button B if it looks morelike a square. The subject’s answer is basedon one particular element of his subjectiveexperience (he selects one particular actionin the universe of allowed responses, based onthe observation of his conscious visual expe-rience). The button press can therefore beseen as a (very crude) description of a con-scious content.

35 . In other words, what is needed here is firsta possible one-to-one monotonic correspon-dence between the phenomena under inves-tigation and the measurements. ‘Monotonic’is to be understood in its usual mathemat-ical sense: For three phenomena pa, pb,and pc and their corresponding measure-ments m(pa), m(pb), and m(pc), it wouldbe desirable that if D(pa,pb) > D(pa,pc) (Dbeing a subjective distance between experi-ential phenomena), then d(m(pa), m(pb)) >

d(m(pa),m(pc)) (d being the distance definedby the experimenter and the subject in theuniverse of measures;for a convergent per-spective see also Fell, 2004).

36. This relation implies an additional require-ment for the measures of the subjectiveexperience: They should be timed. Indeed,the dynamics of experiential phenomena canonly be accessed through series of consecu-tive timed measures (as simple as a seriesof button presses, for instance, or the timecourse of the pressure applied on a joy-stick). Therefore, to establish a strong rela-tion between the dynamics of an experienceand the formation of certain patterns of neu-ral activity, one should be able to say thatthe experience started at time t = 2s andfully developed between t = 5 s and t = 10s



(this is easy to understand in the case ofan emotional reaction to a sound, for exam-ple). It does not follow, however, that thetime as experienced must correspond preciselyto the timing of neural processes. The for-mer is a matter of the content of experienceand the latter of the neural vehicles that (inways we do not fully understand) embody orencode those contents. Within certain smalltemporal windows, a given neural vehiclecould encode one event as happening beforeanother event, even though that neural vehi-cle occurs after the neural vehicle encodingthe second event (see Dennett & Kinsbourne,1992).

37. ‘Continuous’ here is not meant as the oppo-site of discrete, but rather is used to mean thatconsciousness does not jump around withno connection whatsoever from one sort ofexperience to another.

38. Consider, however, the possibility of workingwith individuals who can produce and stabi-lize mental states more reliably (see Chap-ter 19). The issue of working with ‘experts’or trained subjects is important and con-troversial (Jack & Roepstorff, 2003 ; Lutz& Thompson, 2003 ; Varela & Shear, 1999;Chapter 19).

39. Cognitive psychologists sometimes ask sub-jects very difficult questions, so how can theytrust their answers? Why shall we trust thebutton presses of a subject during a binocu-lar rivalry experiment? The subject is askedto press the button as soon as one patterndominates completely, but how can one besure that the subject can actually do this taskreliably or that he has this sort of fine capac-ity to attend to his own visual experience andits dynamics in time?

40. There is a similar problem with the measureof neural events. For instance, with EEG, thenoise level is sometimes simply so strong thatmeasures of gamma activity cannot be made:The Basic Requirement is not met.

References

Abarbanel, H. D., & Rabinovich, M. I. (2001).Neurodynamics: Nonlinear dynamics and neu-robiology. Current Opinion in Neurobiology,11(4), 423–430.

Abeles, M. (1991). Corticonics. New York: Cam-bridge University Press.

Amit, D. J. (1994). The Hebbian paradigmreintegrated: Local reverberations as internalrepresentations. Behavioral and Brain Sciences,18, 617–626.

Arbib, M. A. (2002). Handbook of brain theoryand neural networks (2nd ed.). Cambridge, MA:MIT Press.

Arbib, M. A., Erdi, P., & Szentagothai, J. (1997).Neural organization: Structure, function, anddynamics. Cambridge, MA: MIT Press.

Arieli, A., Sterkin, A., Grinvald, A., & Aert-sen, A..(1996). Dynamics of ongoing activity:Explanation of the large variability in evokedcortical responses. Science, 2 73(5283), 1868–1871.

Ashby, R. (1952). Design for a brain. London.:Chapman-Hall.

Baars, B. J., & Franklin, S. (2003). How con-scious experience and working memory inter-act. Trends in Cognitive Science, 7(4), 166–172 .

Blake, R. A. (1989). A neural theory of binocularrivalry. Psychology Review, 96, 145–167.

Blake, R., & Logothetis, N. K. (2002). Visual com-petition. Nature Reviews Neuroscience, 3(1),13–21.

Block, N. (1996). How can we find the neural cor-relate of consciousness? Trends in Neuroscience,19(11), 456–459.

Block, N. (1997). On a confusion about afunction of consciousness. In N. Block, O.Flanagan, & G. Guzeldere (Eds.), The natureof consciousness: Philosophical debates (pp. 375–416). Cambridge, MA: MIT Press.

Block, N. (2001). Paradox and cross purposes inrecent work on consciousness. Cognition, 79,197–219.

Bressler, S. L. (1995). Large-scale cortical net-works and cognition. Brain Research Review,2 0(3), 288–304 .

Bressler, S. L., & Freeman, W. J. (1980). Frequencyanalysis of olfactory system EEG in cat, rab-bit, and rat. Electroencephalography and ClinicalNeurophysiology, 50(1–2), 19–24 .

Bressler, S. L., & Kelso, J. A. (2001). Corticalcoordination dynamics and cognition. Trends inCognitive Science, 5(1), 26–36.

Brown, R., & Kocarev, L. (2000). A unifying def-inition of synchronization for dynamical sys-tems. Chaos, 10(2), 344–349.

Brown, R. J., & Norcia, A. M. (1997). A methodfor investigating binocular rivalry in real-timewith steady-state VEP. Vision Research, 37 ,2401–2408.



Buchel, C., & Friston, K. J. (1997). Modulation ofconnectivity in visual pathways by attention:Cortical interactions evaluated with structuralequation modelling and fMRI. Cerebral Cortex,7(8), 768–778.

Chalmers, D. J. (2000). What is a neural corre-late of consciousness? In T. Metzinger (Ed.),Neural correlates of consciousness: Empirical andconceptual questions. Cambridge, MA: MITPress.

Chandrasekhar, S. (1961). Hydrodynamic andhydromagnetic stability. Oxford: Oxford Uni-versity Press.

Cosmelli, D., David, O., Lachaux, J. P., Mar-tinerie, J., Garnero, L., Renault, B., et al.(2004). Waves of consciousness: Ongoing cor-tical patterns during binocular rivalry. Neu-roimage, 2 3(1), 128–140.

Crick, F. C., & Koch, C. (1990). Towards a neu-robiological theory of consciousness. Seminarsin Neuroscience, 2 , 263–275 .

Crick, F. C., & Koch, C. (1998). Consciousnessand neuroscience. Cerebral Cortex, 8(2), 97–107.

Crick, F. C., & Koch, C. (2003). A framework forconsciousness. Nature Neuroscience, 6(2), 119–126.

Crutchfield, J. P., & Kaneko, K. (1987). Phe-nomenology of spatiotemporal chaos. In Direc-tions in chaos (pp. 272–353). World Scientific.

Damasio, A. R. (1998). Investigating the biologyof consciousness. Philosophical Transactions ofthe Royal Society London B: Biological Science,353(1377), 1879–1882 .

Damasio, A. (1999). The feeling of what happens:Body and emotion in the making of consciousness.New York: Harcourt.

Danckert, J., & Goodale, M. A. (2000). A con-scious route to unconscious vision. CurrentBiology, 10(2), 64–67.

David, O., Cosmelli, D., & Friston, K. J. (2004).Evaluation of different measures of functionalconnectivity using a neural mass model. Neu-roimage, 2 1(2), 659–673 .

Dehaene, S., Kerszberg, M., & Changeux,J. P. (1998). A neuronal model of a globalworkspace in effortful cognitive tasks. Proceed-ings of the National Academy of Sciences USA,95(24), 14529–14534 .

Dehaene, S., & Naccache, L. (2001). Towards acognitive neuroscience of consciousness: Basicevidence and a workspace framework. Cogni-tion, 79(1–2), 1–37.

Dehaene, S., Sergent, C., & Changeux, J. P.(2003). A neuronal network model linking sub-jective reports and objective physiological dataduring conscious perception. Proceedings of theNational Academy of Sciences USA, 100(14),8520–8525 .

Dennett, D. C., & Kinsbourne, M. (1992). Timeand the observer: The where and when of con-sciousness in the brain. Behavioral and BrainSciences, 15 , 183–247.

Depraz, N., Varela, F., & Vermersch, P. (2003).On becoming aware. A pragmatics of experienc-ing. Amsterdam: Benjamins.

Edelman, G. M., & Tononi, G. (2000). A universeof consciousness: How matter becomes imagina-tion. New York: Basic Books.

Engel, A. K., Fries, P., Konig, P., Brecht, M., &Singer, W. (1999). Temporal binding, binoc-ular rivalry, and consciousness. Consciousnessand Cognition, 8(2), 128–15 1.

Engel, A. K., Fries, P., & Singer, W. (2001).Dynamic predictions: Oscillations and syn-chrony in top-down processing. Nature ReviewsNeuroscience, 2 (10), 704–716.

Engel, A. K., & Singer, W. (2001). Temporal bind-ing and the neural correlates of sensory aware-ness. Trends in Cognitive Science, 5(1), 16–25 .

Faure, P., & Korn, H. (2001). Is there chaos in thebrain? I. Concepts of nonlinear dynamics andmethods of investigation. Les Comptes rendusde l’Academie des sciences III, 32 4(9), 773–793 .

Fell, J. (2004). Identifying neural correlates ofconsciousness: The state space approach. Con-sciousness and Cognition, 13 , 709–729.

Fell, J., Fernandez, G., Klaver, P., Elger, C. E.,& Fries, P. (2003). Is synchronized neuronalgamma activity relevant for selective attention?Brain Research: Brain Research Reviews, 42 (3),265–272 .

Fell, J., Klaver, P., Elger, C. E., & Fernandez, G.(2002). Suppression of EEG gamma activitymay cause the attentional blink. Consciousnessand Cognition, 11(1), 114–122 .

Freeman, W. J. (1975). Mass action in the nervoussystem. New York: Academic Press.

Freeman, W. J. (1999a). How brains make up theirminds. London: Weidenfeld and Nicolson.

Freeman, W. J. (1999b). Consciousness, inten-tionality and causality. Journal of ConsciousnessStudies, 6(11–12), 143–172 .

Freeman, W. J. (2000a). Neurodynamics: Anexploration of mesoscopic brain dynamics.London: Springer-Verlag.



Freeman, W. J. (2000b). Mesoscopic neuro-dynamics: From neuron to brain. Journal ofPhysiology Paris, 94(5–6), 303–322 .

Freeman, W. J. (2004). Mass action in the ner-vous system (2nd ed.). Retrieved from http://sulcus.berkeley.edu.

Fries, P., Reynolds, J. H., Rorie, A. E., & Des-imone, R. (2001). Modulation of oscillatoryneuronal synchronization by selective visualattention. Science, 2 91, 1560–1563 .

Fries, P., Roelfsema, P. R., Engel, A. K., Konig, P.,& Singer, W. (1997). Synchronization of oscil-latory responses in visual cortex correlates withperception in interocular rivalry. Proceedings ofthe National Academy of Sciences USA, 94(23),12699–12704 .

Friston, K. J. (1997). Transients, metastability,and neuronal dynamics. Neuroimage, 5(2), 164–171.

Friston, K. J. (2000). The labile brain. I. Neuronaltransients and nonlinear coupling. PhilosophicalTransactions of the Royal Society London B: Bio-logical Science, 355(1394), 215–236.

Friston, K. J. (2002a). Beyond phrenology: Whatcan neuroimaging tell us about distributedcircuitry? Annual Review of Neuroscience, 2 5 ,221–250.

Friston, K. (2002b). Functional integration andinference in the brain. Progress in Neurobiology,68(2), 113–143 .

Friston, K. J. (2005). Models of brain function inneuroimaging. Annual Review of Psychology, 56,57–87.

Garfinkel, A., Spano, M. L., Ditto, W. L., & Weiss,J. N. (1992). Controlling cardiac chaos. Science,2 57 , 1230–1235 .

Gomes, G. (1998). The timing of consciousexperience: A critical review and reinterpre-tation of Libet’s research. Consciousness andCognition, 7(4), 559–595 .

Gray, C. M., Konig, P., Engel, A. K., & Singer,W. (1989). Oscillatory responses in cat visualcortex exhibit inter-columnar synchroniza-tion which reflects global stimulus properties.Nature, 338(6213), 334–337.

Gurwitsch, A. (1964). The field of consciousness.Pittsburgh: Duquesne University Press.

Gutkin, B., Pinto, D., & Ermentrout, B. (2003).Mathematical neuroscience: From neurons tocircuits to systems. Journal of Physiology Paris,97(2–3), 209–219.

Haken, H. (1983). Synergetics: An introduc-tion. Nonequilibrium phase transitions and self-

organization in physics, chemistry, and biology.New York: Springer-Verlag.

Hanna, R., & Thompson, E. (2003). The spon-taneity of consciousness. Canadian Journal ofPhilosophy, 2 9, 133–162 .

Hebb (1949) [reference to come]Herculano-Houzel, S., Munk, M. H., Neuen-

schwander, S., & Singer, W. (1999). Preciselysynchronized oscillatory firing patterns requireelectroencephalographic activation. Journal ofNeuroscience, 19(10), 3992–4010.

Hilgetag, C. C., & Kaiser, M. (2004). Clusteredorganization of cortical connectivity. Neuroin-formatics, 2 (3), 353–360.

Hollender, D. (1986). Semantic activation with-out conscious identification in dichotic lis-tening, parafoveal vision, and visual masking:A survey and appraisal. Behavioral and BrainSciences, 9, 1–66.

Ito, J., & Kaneko, K. (2002). Spontaneous struc-ture formation in a network of chaotic unitswith variable connection strengths. PhysicalReview Letters, 88(2).

Jack, A., & Roepstorff, A. (2003). Trusting thesubject (Vol. 1). Journal of Consciousness Stud-ies, 10(9–10).

James, W. (1981). The principles of psychology.Cambridge, MA: Harvard University Press.(Original work published 1890)

Kelso, J. A. S. (1995). Dynamic patterns: The self-organization of brain and behavior. Cambridge,MA: MIT Press.

Korn, H., & Faure, P. (2003). Is there chaos in thebrain? II. Experimental evidence and relatedmodels. Comptes rendus biologies, 32 6(9), 787–840.

Kuramoto, Y. (1984). Cooperative dynamics ofoscillator community: A study based on lat-tice of rings. Progress of Theoretical Physics,79(Suppl.), 223–240.

Lachaux, J.-P., George, N., Tallon-Baudry, C.,Martinerie, J., Hugueville, L., Minotti, L., et al.(2005). The many faces of the gamma bandresponse to complex visual stimuli. Neuroim-age, 2 5(2), 491–501.

Lachaux, J. P., Lutz, A., Rudrauf, D., Cosmelli,D., Le Van Quyen, M., Martinerie, J., et al.(2002). Estimating the time-course of coher-ence between single-trial brain signals: Anintroduction to wavelet coherence. Neurophys-iology Clinics, 32 (3), 157–174 .

Lachaux, J. P., Rodriguez, E., Martinerie, J.,Adam, C., Hasboun, D., & Varela, F. J. (2000).



A quantitative study of gamma-band activityin human intracranial recordings triggered byvisual stimuli. European Journal of Neuroscience,12 (7), 2608–2622 .

Lachaux, J. P., Rodriguez, E., Martinerie, J., &Varela, F. J. (1999). Measuring phase synchronyin brain signals. Human Brain Mapping, 8(4),194–208.

Lachaux, J. P., Rudrauf, D., & Kahane, P. (2003).Intracranial EEG and human brain mapping.Journal of Physiology Paris, 97(4–6), 613–628.

Lamme, 2003 , 2004

Le Van Quyen, M. (2003). Disentangling thedynamic core: A research program for a neuro-dynamics at the large-scale. Biological Research,36(1), 67–88.

Le Van Quyen, M., Foucher, J., Lachaux, J.-P.,Rodriguez, E., Lutz, A., Martinerie, J., et al.(2001). Comparison of Hilbert transform andwavelet methods for the analysis of neuronalsynchrony. Journal of Neuroscience Methods,111(2), 83–98.

Leopold, D. A., & Logothetis, N. K. (1999). Mul-tistable phenomena: Changing views in per-ception. Trends in Cognitive Science, 3(7), 254–264 .

Libet, B. (2002). The timing of mental events:Libet’s experimental findings and their impli-cations. Consciousness and Cognition, 11(2),291–299; discussion 304–233 .

Llinas, R. (1988). The intrinsic electrophysiologi-cal properties of mammalian neurons: Insightsinto central nervous system function. Science,2 42 (4886), 1654–1664 .

Llinas, R., & Pare, D. (1991). Of dreaming andwakefulness. Neuroscience, 44(3), 521–535 .

Llinas, R., & Ribary, U. (2001). Consciousnessand the brain. The thalamocortical dialoguein health and disease. Annals of the New YorkAcademy of Sciences, 92 9, 166–175 .

Llinas, R., Ribary, U., Contreras, D., &Pedroarena, C. (1998). The neuronal basis forconsciousness. Philosophical Transactions ofthe Royal Society London B: Biological Science,353(1377), 1841–1849.

Lutz, A., Lachaux, J. P., Martinerie, J., & Varela,F. J.. (2002). Guiding the study of brain dynam-ics by using first-person data: Synchrony pat-terns correlate with ongoing conscious statesduring a simple visual task. Proceedings of theNaionatl Academy of Sciences USA, 99(3),1586–1591.

Lutz, A., & Thompson, E. (2003). Neuro-phenomenology: Integrating subjective experi-ence and brain dynamics in the neuroscience ofconsciousness. Journal of Consciousness Studies,10, 31–52 .

Martinerie, J., Adam, C., Le Van Quyen, M.,Baulac, M., Clemenceau, S., Renault, B., et al.(1998). Epileptic seizures can be anticipated bynon-linear analysis. National Medicine, 4(10),1173–1176.

Maturana, H. R., & Varela, F. J. (1980). Autopoiesisand cognition: The realization of the living. Bostonstudies in the philosophy of science, Vol. 42 .Dordrecht: D. Reidel.

Mesulam, M. M. (1999). Spatial attention andneglect: Parietal, frontal and cingulate contri-butions to the mental representation and atten-tional targeting of salient extrapersonal events.Philosophical Transactions of the Royal SocietyLondon B: Biological Science, 354(1387), 1325–1346.

Metzinger, T. (Ed.). (2000). Neural correlates ofconsciousness: Empirical and conceptual ques-tions. Cambridge, MA: MIT Press.

Milner, D., & Goodale, M. (1995). The visualbrain in action. Oxford: Oxford UniversityPress.

Munk, M. H., Roelfsema, P. R., Konig, P., Engel,A. K., & Singer, W. (1996). Role of reticularactivation in the modulation of intracorticalsynchronization. Science, 2 72 (5259), 271–274 .

Nicolis, G., & Prigogine, G. (1977). Self-organization in nonequilibrium systems. New-York: Wiley.

Nicolis, G., & Prigogine, I. (1989). Exploringcomplexity: An introduction. New York: W. H.Freeman.

Noe, A., & Thompson, E. (2004a). Are there neu-ral correlates of consciousness? Journal of Con-sciousness Studies, 11, 3–28.

Noe, A., & Thompson, E. (2004b). Sorting outthe neural basis of consciousness. Authors’reply to commentators. Journal of Conscious-ness Studies, 11, 87–98.

Nunez, P. L. (2000). Toward a quantitativedescription of large scale neocortical dynamicfunction and EEG. Behavioral and Brain Sci-ences, 2 3(3).

O’Regan, K., & Noe, A. (2001). sensorimotoraccount of vision and visual consciousness.Behavioral and Brain Sciences, 2 4(5), 939–973 .



Palus, M. (1996). Nonlinearity in normal humanEEG: Cycles, temporal asymmetry, nonstation-arity and randomness, not chaos. BiologucalCybernetics, 75 , 389–396.

Panksepp, J. (1998). The periconscious substratesof consciousness: Affective states and the evo-lutionary origins of self. Journal of Conscious-ness Studies, 5 , 566–582 .

Parvizi, J., Damasio, A. (2001). Consciousness andthe brainstem. Cognition, 79, 135–159.

Perkel, D. H., Gerstein, G. L., & Moore, G. P.(1967). Neuronal spike trains and stochas-tic point processes. I. The single spike train.Biophysical Journal, 7(4), 391–418.

Pessoa, L., Thompson, E., & Noe, A. (1998). Find-ing out about filling-in: A guide to perceptualcompletion for visual science and the philos-ophy of perception. Behavioral and Brain Sci-ence, 2 1(6), 723–748.

Petitmengin, C. (in press). Describing one’s sub-jective experience in the second-person. Phe-nomenology and the Cognitive Sciences.

Pikovsky, A., Zaks, M., Rosenblum, M., Osipov,G., & Kurths, J. (1997). Phase synchroniza-tion of chaotic oscillations in terms of periodicorbits. Chaos, 7(4), 680–687.

Rees, G., Kreiman, G., & Koch, C. (2002). Neuralcorrelates of consciousness in humans. NatureReviews Neuroscience, 3(4), 261–270.

Revonsuo, A., Wilenius-Emet, M., Kuusela, J., &Lehto, M. (1997). The neural generation of aunified illusion in human vision. Neuroreport,8(18), 3867–3870.

Rizzolatti, G., & Craighero, L. (2004). Themirror-neuron system. Annual Review of Neu-roscience, 2 7 , 169–192 .

Rodriguez, E., George, N., Lachaux, J.- P., Mar-tinerie, J., Renault, B., & Varela, F. J. (1999).Perception’s shadow: Long-distance synchro-nization of human brain activity. Nature,397(6718), 430–433 .

Rodriguez, R., Kallenbach, U., Singer, W., &Munk, M. H. (2004). Short- and long-termeffects of cholinergic modulation on gammaoscillations and response synchronization inthe visual cortex. Journal of Neuroscience,2 4(46), 10369–10378.

Rosenblum, M., Pikovsky, A., & Kurths, J. (1996).Phase synchronization of chaotic oscillators.Physical Review Letters, 76(11), 1804–1807.

Roskies, A. L. (1999). The binding problem[Special issue]. Neuron, 2 4(7–9).

Rudrauf, D., Lutz, A., Cosmelli, D., Lachaux,J.-P., & Le Van Quyen, M. (2003). Fromautopoiesis to neurophenomenology: FranciscoVarela’s exploration of the biophysics of being.Biological Research, 36, 27–66.

Schiff, S. J., Jerger, K., Duong, D. H., Chang, T.,& Spano, M. L. (1994). Controlling chaos inthe brain. Nature, 8, 615–620.

Searle, J. (2000). Consciousness. Annual Reviewof Neuroscience, 2 3 .

Sewards, T. V., & Sewards, M. A. (2001). Onthe correlation between synchronized oscilla-tory activities and consciousness. Consciousnessand Cognition, 10(4), 485–495 .

Sherman, S. M., & Guillery, R. W. (2002). Therole of the thalamus in the flow of informa-tion to the cortex. Philosophical Transactions ofthe Royal Society London B: Biological Science,357(1428), 1695–1708.

Sherover, C. (1991). The human experience of time.Evanston, IL: Northwestern University Press.

Simons, D. J. (2000). Attentional capture andinattentional blindness. Trends in Cognitive Sci-ence, 4(4), 147–155 .

Singer, W. (1998). Consciousness and the struc-ture of neuronal representations. Philosophi-cal Transactions of the Royal Society London B:Biological Science, 353(1377), 1829–1840.

Singer, W., & Gray, C. M. (1995). Visual fea-ture integration and the temporal correlationhypothesis. Annual Review of Neuroscience, 18,555–586.

Skarda, C. A., & Freeman, W. J. (1987). Howbrains make chaos in order to make sense ofthe world. Behavioral and Brain Sciences, 10(2),161–195 .

Stephan, K. E., Hilgetag, C. C., Burns, G. A.,O’Neill, M. A., Young, M. P., & Kotter, R.(2000). Computational analysis of functionalconnectivity between areas of primate cerebralcortex. Philosophical Transactions of the RoyalSociety London B: Biological Science, 355(1393),111–126.

Tallon-Baudry, C., & Bertrand, O. (1999). Oscil-latory gamma activity in humans and its rolein object representation. Trends in CognitiveScience, 3(4), 15 1–162 .

Tallon-Baudry, C., Bertrand, O., & Fischer, C.(2001). Oscillatory synchrony between humanextrastriate areas during visual short-termmemory maintenance. Journal of Neuroscience,2 1(20), RC177.



Thompson, E. (2007). Mind in life: Biology, phe-nomenology, and the sciences of mind. Cam-bridge, MA: Harvard University Press.

Thompson, E., Lutz, A., & Cosmelli, D. (2005).Neurophenomenology: An introduction forneurophilosophers. In A. Brook & K. Akins(Eds.), Cognition and the brain: The philosophyand neuroscience movement (pp. 40–97). NewYork: Cambridge University Press.

Thompson, E., & Varela, F. J. (2001). Radicalembodiment: Neural dynamics and conscious-ness. Trends in Cognitive Science, 5(10), 418–425 .

Thorpe, S., Fize, D., & Marlot, C. (1996). Speed ofprocessing in the human visual system. Nature,381(6582), 520–522 .

Tononi, G., & Edelman, G. M. (1998). Conscious-ness and complexity. Science, 2 82 (5395), 1846–1851.

Treisman, A. M., & Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psy-chology, 12 (1), 97–136.

Tsuda, I. (2001). Toward an interpretation ofdynamic neural activity in terms of chaoticdynamical systems. Behavioral and BrainSciences, 2 4(5), 793–810; discussion 810–748.

Van Gelder, T. J. (1998). The dynamical hypoth-esis in cognitive science. Behavioral and BrainSciences, 2 1, 1–14 .

Van Van Gelder, T. J. (1999). Dynamic app-roaches to cognition. In R. Wilson & F. Keil(Eds.), The MIT encyclopedia of cognitive sciences(pp. 244–246). Cambridge MA: MIT Press.

VanRullen, R., & Koch, C. (2003). Is percep-tion discrete or continuous? Trends in CognitiveScience, 7(5), 207–213 .

Varela 1979

Varela, F. (1995). Resonant cell assemblies: Anew approach to cognitive functioning andneuronal synchrony. Biological Research, 2 8(81–95).

Varela, F. J. (1996). Neurophenomenology: Amethodological remedy for the hard problem.

Journal of Consciousness Studies, 3(4), 330–349.

Varela, F. (1999). The specious present: Theneurophenomenology of time consciousness.In J. Petitot, F. J. Varela, B. Pachoud, &J. M. Roy (Eds.), Naturalizing phenomenology(pp. 266–314). Stanford: Stanford UniversityPress.

Varela, F., Lachaux, J.-P., Rodriguez, E., & Mar-tinerie, J. (2001). The brainweb: Phase synchro-nization and large-scale integration. NatureReviews Neuroscience, 2 (4), 229–239.

Varela, F. J., & Shear, J. (1999). First-personmethodologies: Why, when and how. Journalof Consciousness Studies,

Von der Malsburg, C. (1981). The correlation the-ory of brain function. Gottingen: Germany: MaxPlanck Institute for Biophysical Chemistry.

Watts, D. J., & Strogatz, S. H. (1998). Collectivedynamics of ‘small-world’ networks. Nature,393(6684), 440–442 .

Weiskrantz, L. (1990). The Ferrier lecture, 1989.Outlooks for blindsight: Explicit methodolo-gies for implicit processes. Philosophical Trans-actions of the Royal Society London B: BiologicalScience, 2 39(1296), 247–278.

Wicker, B., Ruby, P., Royet, J. P., & Fonlupt, P.(2003). A relation between rest and the selfin the brain? Brain Research: Brain ResearchReview, 43(2), 224–230.

Wolfe, J. M., & Bennett, S. C. (1997). Preattentiveobject files: Shapeless bundles of basic features.Vision Research, 37(1), 25–43 .

Wright, J. J., & Liley, D. T. J. (1996). Dynamicsof the brain at global and microscopic scales:Neural networks and the EEG. Behavioral andBrain Sciences, 19, 285–320.

Zahavi, D. (2005). Subjectivity and selfhood: Inves-tigating the first-person perspective. Cambridge,MA: MIT Press.

Zihl, J., Von Cramon, D., & Mai, N. (1983). Selec-tive disturbance of movement vision after bilat-eral brain damage. Brain, 106(2), 313–340.

chapter 26 neurodynamics of consciousness 26 neurodynamics of consciousness diego cosmelli,...

Documents