This article was originally published in the Encyclopedia of published by Elsevier, and the attached copy is provided by Elsevier for the author's benefit and for the benefit of the author's institution, for non- commercial research and educational use including without limitation use in instruction at your institution, sending it to specific colleagues who you know, and providing a copy to your institution’s administrator.

All other uses, reproduction and distribution, including without limitation commercial reprints, selling or licensing copies or access, or posting on open internet sites, your personal or institution’s website or repository, are prohibited. For exceptions, permission may be sought for such use through Elsevier's permissions site at:

http://www.elsevier.com/locate/permissionusematerial

de Gardelle V and Kouider S (2009), Cognitive Theories of Consciousness. In: William P. Banks, (Editor), Encyclopedia of Consciousness. volume 1, pp. 135- 146. Oxford: Elsevier.

Author's personal copy

Cognitive Theories of Consciousness

V de Gardelle and S Kouider, De´ partement d’Etudes Cognitives, CNRS/EHESS/ENS-DEC, Paris, France

ã 2009 Elsevier Inc. All rights reserved.

motivated by the will to tackle this issue in a scien- Glossary tific way. As such, developing a scientific theory of Connectionism – Connectionism is a consciousness has been the Holy Grail of psychol- framework in cognitive science, according to ogy since its earliest days. Today, consciousness which all of the processes achieved by the continues to be a central topic of interest, extend- mind can be modeled by parallel and ing its interest across almost all disciplines of cog-

distributed processing among simple nitive science. operational units. It is mostly based on the Studying consciousness, however, has not always development of artificial neural networks, and been an acceptable question in psychology. Indeed, it has been traditionally opposed to the the issue of consciousness was totally rejected position that mental processes are based on by the dominant behaviorist school during the

symbolic computations. first half of the twentieth century. The failure of Functionalism – Functionalism is a doctrine earlier psychologists, who were stuck in unsolvable in cognitive science, according to which a debates related to introspection, led the behavior- mental state is defined by its functional role, ists to reject consciousness as a plausible scientific rather than by its intrinsic structure and its issue. Instead, behaviorists sought to bring psychol-

implementation. In other words, a functional ogy into the scientific domain by restricting it to model of the mind (or of an operation that is objective data and reproducible methods. Even the achieved by the mind) involves mental states so-called ‘cognitive revolution’ that transformed that are causally related to sensory inputs psychology during the second half of the twentieth and other mental states, and behavior. century did not favor a renewal of interest in con- Homunculus – Literally, ‘little man,’ in Latin, sciousness. Indeed, the purpose of the cognitive that is in the context of cognitive theories of approach was to reintroduce the notion of internal consciousness, a conscious observer, which representations or mental states, not the notion of is at the top of the cognitive system. consciousness per se. Furthermore, an important A Homunculus is a hypothetical construct new assumption was that internal representations that operates (i.e., with no further were largely unavailable to consciousness. None- explanation) the very operation that is theless, in this new information-processing per-

supposed to be explained. Thus, a theory spective, cognitive scientists developed several that relies on a homunculus at some point is key elements that largely influenced the forthcom- incomplete in the same extend. ing cognitive theories of consciousness. These pre- cursors included new models of attention and working memory, and also new concepts, such as

the distinction between modules and central pro- Introduction cesses, or that between automatic and controlled

processes. Consciousness is probably the most privileged In the present article, we will first overview topic in psychology. The study of consciousness the precursors that allowed the development of is considered to be at the origin of the separat- cognitive theories of consciousness. Then we will ion between psychology and philosophy during present a selection of influential contemporary the nineteenth century, as psychologists were accounts of consciousness. These theories will be

135

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

136 Cognitive Theories of Consciousness

grouped according to three main themes: theories a huge quantity of information becomes available that consider consciousness to result from specific to the rest of the system. However, the central architectural elements within the cognitive system; system is very limited in terms of computational theories claiming that some features of conscious- resources. Hence, a selection mechanism is needed ness are in fact illusory; and theories that focus to prevent overload. As such, attention operates on the relation between consciousness and learn- by selecting the most relevant information and by ing. We will conclude this article by emphasiz- filtering out that which is irrelevant. Then, the most ing the common challenges that current cognitive relevant information, which is under the focus models of consciousness have to face: the pressure of attention, becomes the target of the central sys- from the philosophically defined hard problem, on tem and can thus benefit from deeper and more one side, and the pressure from neurobiological enriched processing. Once again, although con- evidences, on the other side. sciousness was not the main concern, one conse-

quence of attentional selection was that it allowed the target information to become conscious. In this Precursors for a Cognitive perspective, attention and consciousness are two

Perspective on Consciousness tightly related notions. The notion of short-term memory put forward

In this section, we highlight the most influential by George Miller and later extended to the notion precursors for a theory of consciousness. Most of of working memory is also an important precursor. these elements emerged during the cognitive rev- For example, in their model of working memory, olution. Although they were originally sketched Baddeley and Hitch relied on a central executive out in light of an information-processing perspec- system, which has top-down control over the dis- tive, those elements are now largely linked to the tinct specific subsystems, namely the phonological dissociation between conscious and unconscious loop and the visuospatial sketchpad. Here, the processing. content of working memory may be roughly equated with the content of consciousness, an aspect that will also be important for future cogni- Attention and the Central Executive tive theories of consciousness. Various influential models developed in the 1960s Norman and Shallice, in turn, proposed a model referred to a central processor, a central executive of action selection implicating a supervisory atten- system, or a supervisory system. Processing within tional system. This central system receives sensory the central system can be considered as analogous to evidence and determines the appropriate beha- conscious processing, even if the word consciousness vior by selecting instruction schemes for action was still largely banished in the scientific commu- mechanisms. In addition, the supervisory atten- nity. This system is at the top of the hierarchy in the tional system can be modulated by the goals of cognitive architecture: it is involved in higher-order the organism, and it is primarily involved when a computations (decision, monitoring, planning, etc.) new or critical situation appears. Here too, the and leads to selection and control over lower-level central part of the model shares some properties subsystems. As in many contemporary accounts of that are associated with consciousness, namely consciousness, the central system was considered flexibility, reactivity regarding unexpected situa- the most integrative element of the cognitive system, tions, decision, and control over behavior. In sum, these influential early models depicted granting flexibility and control over behavior. Another key element was the simple but power- the global architecture of the cognitive system by ful metaphor of attention as a filtering mechanism emphasizing the following components: sensory that was put forward by Broadbent. In a nutshell, inputs in the periphery that are processed in par- peripheral processors in this theory provide sen- allel in multiple channels, attention that performs sory information to the central system dealing selection upon these sources of information, a with control and decisions. Because multiple sen- working memory component that keeps tracks sory channels are continuously acting in parallel, of the selected information, and finally a central

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

Cognitive Theories of Consciousness 137

system that acts as a supervisor. But one major were assumed to be conscious. These dichotomies limitation of this view is that it falls into the between central and peripheral, controlled and homunculus trap, when it comes to the question automatic, flexible and hard-and-fast processes of consciousness. Indeed, if this central supervisor have provided the ground for the distinction is governing the whole cognitive system, one may between conscious and unconscious processing, ask who is in turn governing the central supervisor! which has been central in the development of That is,if we were to rephrase this question by current cognitive theories of consciousness. It is focusing on consciousness, it would be problematic of note that the difficulty of studying conscious- to rely on a hypothetical little man in our head (i.e., ness did not arise only when researchers decided a homunculus) that has consciousness, which is the to face it. Although a few serious attempts have same property we are supposed to explain. This been made to propose functional description of the approach unavoidably leads to an infinite regres- central system, such as in the Adaptive Control of sion. Because consciousness was not the main issue Though (ACT) theory by John Anderson, this for these early models, this crucial issue was left notion was itself often unspecified and often con- out or even denied during the development of sidered as a mysterious but needed component. early cognitive models with a central supervisor. For instance, Fodor strongly defended the idea As we will see below, current theories of con- that although the program of cognitive science sciousness will overcome this limitation by pro- was to understand how modules work, we would posing various cognitive architectures, sometimes surely be in an impasse when trying to address the including a central system, that take into account nature of central processes. the homunculus issue.

Architectural Accounts of Specialized Modules versus Flexible Consciousness Integration

Closely related to this distinction between central We present in this section three of the most influ- and peripheral processes is the very influential the- ential cognitive theories of consciousness. For each ory of modularity developed by Jerry Fodor. In this of them, consciousness is grounded in an informa- framework, modules are fast and efficient devices tion processing system. Baars’ global workspace that process inputs in an automatic and mandatory theory uses the metaphor of global broadcasting fashion. They are tuned to a particular kind of to describe conscious processing, Jackendoff and computation on a particular kind of information. Prinz’ intermediate level theory emphasizes the

In other words they are functionally specialized need for consciousness to be focused on interme- and they constitute the small computational bricks diate representations, and Tononi’s information of cognitive architecture. While modules operate in integration theory proposes to relate conscious- the periphery in this architecture, they have been ness with complexity in the cognitive system. classically opposed to central processes that can be The three accounts all share the same will to link slow but flexible, and can integrate inputs from consciousness with a particular representational differentmodalities. Closely related to this architec- aspect of the cognitive system. These theories dif- tural dissociation is Posner and Snyder’s dissociation fer, though, in many respects and thus provide a between, on the one side, automatic processes that diversified sample of what cognitive accounts of consciousness can be. are mandatory and fast and, on the other side, con- trolled processes that are assumed to be strategic and voluntary. The of Here again, although this was not explicitly Consciousness acknowledged in these various works, the fast and automatic processes operated by modules were Grounded on the distinction between conscious assumed to reflect unconscious processing, while and unconscious processes, Bernard Baars’ global the control processes involving the central system workspace theory is one of the most influential

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

138 Cognitive Theories of Consciousness

cognitive theories of consciousness. This theory conscious observer in the system, and thus there is relies on the metaphor of a theater. In this theater, no issue of infinite regression with a homunculus unconscious specialized processors (equivalent to inside another homunculus. Instead, consciousness modules) are assumed to be the actors and the is considered to reflect the global broadcasting audience. While the audience represents the set of information to an audience of unconscious of passive processors, actors represents active pro- processors. As the audience is unconscious, unsu- cessorsplaying on the stage of the theater (i.e., the pervised, and receptive rather than attending to workspace). These actors are engaged in a compe- the information, it does not constitute an internal tition for being seen by the audience: by broadcast- homunculus. ing their information they actually compete for more broadcasting. Active processors with the The Intermediate Level Theory of highest coherent activity can form local coali- Consciousness tions that strengthen them in this competition process. The strongest coalition in this competi- The intermediate level theory originally proposed tion dominates the workspace, in a winner-take-all by Ray Jackendoff and further defended and spe- fashion, and corresponds to the content of con- cified by Jesse Prinz proposes that within the hier- sciousness. The workspace is equated by Baars to archy of representations that are used to describe working-memory, in which only the most active the cognitive system, conscious experience occurs content becomes conscious. Additionally, the dom- only for specific levels of representation. inant coalition benefits from global broadcasting, The theory is rooted in Jackendoff ’s analysis of which allows it to recruit new processors from the different cognitive systems such as vision, lan- audience in order to form a global coalition. Here, guage, and music and the subsequent observation consciousness allows for the integration of compu- that consciousness does not arise anywhere within tational resources in a large-scale coordination and these systems. According to Jackendoff, conscious- for the exchange of information among processors ness is not associated with low-level, nor with that would otherwise remain separated. In this high-level representations, but rather with those theory, each processor can operate in the conscious implying intermediate levels of processing. For mode if it benefits from global broadcasting instance, in the domain of object recognition, it is through the workspace, or it can operate in the assumed that the visual system comprises a low level unconscious mode when disconnected from the with local computations of visual features, an inter- workspace. mediate level reflecting binding and object recog- An important feature of the global workspace nition, and a higher level computing viewpoint theory is the presence of contexts as stable coali- invariance and representing abstract categories. tions shaping access to the workspace. Contexts are According to Jackendoff and Prinz, conscious expe- constituted of unconscious processors reflecting, rience is not comprised of a disunified picture in a hierarchical manner, our expectations, our of visual features, nor is it represented by view- beliefs, our goals, and ultimately our self. In par- invariant categories. Rather it is composed of bound ticular, attention is implemented as a goal context and specific instances of objects that are assumed to in this theory. It is described as a mechanism that be computed at the intermediate level of represen- controls access to the workspace, acting as a filter tation. In an analogous manner, speech perception and biasing the competition process toward a par- can be decomposed into three levels: an acoustic ticularset of actors. At any given moment, the representation of speech sounds at the lower level, dominant coalition is under the spotlight of atten- a high level involving abstract lexical and syntac- tion, and its informational content becomes the tic categories, and in between a word recognition content of conscious experience. level relying on phonological representations. This A crucial aspect of Baars’ theory is that it avoids theory explains why the conscious experience asso- the problem of the homunculus by reducing it to ciated with speech perception mostly involves pho- an audience of multiple unconscious processors. nological representations, rather than other levels Here, there is no need for a hypothetical single of representations. In Jackendoff and Prinz’ theory,

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

Cognitive Theories of Consciousness 139

the privileged role of the intermediate level of pro- from the interactions of diverse subsets within this cessing is based on the need for real-time computa- system. To account for integration, Edelman and tional efficiency. Indeed, this level of representation Tononi relied on the notion of neuronal reentry is assumed to be the most relevant one regarding within a thalamocortical dynamic core. ecological and functional needs. The information integration theory, formulated

Another important aspect of this theory con- more recently by Tononi, is more concerned with cerns the central role of attention during conscious how any physical system, brain or machine, with experience. Here, attention is defined as a selec- both integrated and differentiated information can tion process that acts as a gate to working memory lead to conscious experience. In this theory, con- mechanisms. It performs the function of selecting sciousness is a property of a system that can inte- the relevant information that is amplified after- grate differentiated information: the more one ward and then becomes conscious. Indeed, Prinz system exhibits integrated and differentiated states, acknowledges that activation of an intermediate- the more it is conscious. Accordingly, Tononi pro- level representation on its own cannot be a suffi- posed to measure information integration by cient condition for consciousness, given that those means of a function labeled F, whose value allows representations can be activated during subliminal one to assess the degree of consciousness within the perception. However, this theory makes the crucial system. This function F takes high values for sys- postulate that the amplification of intermediate- tems with high complexity, such as small-world level representations by attention is a necessary architectures where connectivity patterns between and sufficient condition for consciousness. In sum, units are heterogeneous. Conversely, it has low for each domain of processing, the content of con- values for simple and feedforward systems. Impor- sciousness at a particular moment is supported tantly, Tononi gives an operational method for the F by a representational structure of intermediate computation of in a given system, based on level for that domain, which is selected to enter decomposition of the system into its subsets. As short-term memory, and enriched by attentional such, he also puts forward the notion of a complex processing. in a system: a complex is mathematically defined as a subset of the system that is not part of a subset of

higher F value. Importantly, according to the infor- The Information Integration Theory of mation integration theory, the content of con- Consciousness sciousness at a given moment corresponds to the ‘The information integration theory of conscious- information processed in the complex, which exhi- ness’ has been proposed by Gulio Tononi to bits the highest F value, called the main complex explain how consciousness arises from dynamic of the system. As the system processes informa- complex systems. It originates from Tononi’s tion dynamically, interactions between the differ- work with Gerald Edelman and their observation ent parts of the system are continuously changing. that conscious states share two fundamental prop- Thus, the main complex changes accordingly, and erties: they are both differentiated and integrated. so does the content of consciousness.

Conscious states are highly differentiated in the One important aspect of this approach is that it sense that the occurrence of a particular conscious considers consciousness to be a quantitative and state results from its selection among a huge rep- graduate variable. Furthermore, as consciousness ertoire of possible conscious states. As such, a is only determined by the F measure, it is only a matter of system complexity in any system. Con- conscious state carries an important amount of information, as it reflects a large reduction in sequently, animals or mechanical systems exhibit- uncertainty. At the same time, conscious states ing similar properties can be considered as having are integrated as a unified experience. For a certain degree of consciousness. Still, although instance, one does not have the experience of the the value of F can be computed in theoretical color of a particular shape independently from the situations, with fully specified systems, one obvi- experience of the shape itself. A given state in a ous difficulty is the measure of F in natural sys- system is considered to be integrated if it results tems. The decomposition of the mind into relevant

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

140 Cognitive Theories of Consciousness

subunits is still a matter of research, and the assess- multiple drafts are edited in parallel and continu- ment of information processed by these subunits ously revised within the system. In addition, these has been to date an untargeted issue. drafts have different fates: some will be read by the rest of the system and will affect subsequent behaviors, while others will simply fade out. In

Illusory Features Accounts of Dennett’s model, cerebral celebrity makes a par- Consciousness ticular draft conscious (or ‘fame in the brain’), that

is, the extent to which it affects other processes in Several approaches have claimed that some fea- the system, and eventually subsequent behaviors tures associated with conscious experience are in and responses. In particular, by introspecting our- fact illusory. Here, we present the most popular selves and thus directing our attention to one views on this matter. A first perspective is repre- particular thread, we let the content of this thread sented by ’s affect our behaviors and thus become conscious. of consciousness, where the appearance of a uni- Introspection can also have the consequence of fied stream of consciousness reflects an illusion modifying the content of the draft itself. For produced during introspection. A second view is instance, if a thread is probed too late, the asso- held by Daniel Wegner whose theory of apparent ciated draft will not be available anymore, or it will mentalcausation claims that free will and the fact be totally reconstructed on purpose. Conversely, if that we consciously determine our actions is illu- the thread is probed to early, its process is inter- sory. A third account, the sensory-motor theory of rupted, and the draft that becomes conscious will consciousness by Kevin O’Regan and Alva Noe, not reflect further edition. also takes phenomenal experience as a retrospec- The multiple drafts model of consciousness is an tive illusion. However, this theory also associates early and influential cognitive theory of conscious- consciousness with a learning process, and thus ness, developed with the will to eradicate problem- so it will be addressed in the next section on atic homunculus assumptions. Indeed, the theory ‘Learning process accounts of consciousness.’ emphasizes that there is not a single observer that would receive all the information and provide a

single and unified narrative stream of conscious- The Multiple Drafts Model of ness. Rather there are multiple on-going processes, Consciousness from which some drafts have sufficient impact to The quest for a conscious subsystem in the brain influence behaviors and lead to consciousness one has been overtly criticized by Dennett who explic- after the other. According to Dennett, the illusion itly related it to the homunculus assumption. of a single narrative stream stems from the fact Instead, he proposed a multiple drafts model of that the story is continuously revised in order to consciousness in which information does not need be more plausible. Though less specified than more to be represented in front of a conscious observer recent accounts that are similar in principle (e.g., withinour heads. In this model, the stream of Baars’ global workspace theory), this theory remains consciousness is neither unified nor is it produced an interesting instance of a strongly reductive view, by a single narrative system. Instead, what makes which offers to replace the central homunculus the stream of consciousness apparently unified is a system by parallel and distributed processing in a retrospective reconstruction involving multiple network of threads or daemons. In this approach there is nothing more to consciousness than the drafts describing the story. In the multiple drafts model, the cognitive sys- causal impact that one particular content has on tem continuously processes information in parallel subsequent processing and behavioral reports. How- in different threads, either in perceptual, concep- ever, one possible criticism linked to this feature is tual, or motor domains. In fact, threads look like that the explanation provided by Dennett is a theory

Fodorian modules or specialized processors of the of report rather than of conscious experience. This global workspace, and their computations in prog- latter argument reflects more generally the critical ress are logged in a temporary draft. As such, problem of assessing consciousness without relying

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

Cognitive Theories of Consciousness 141

a priori on some kind of report (we return to this point in the about the experience of conscious will. conclusion of this article). Nonetheless, one might wonder why then would

we experience this illusion? Wegner proposes that it may help the subject to maintain his goals The Theory of Apparent Mental Causation through consciousness or to build a better repre-

In everyday life as well as in scientific accounts, sentation of the world, in which his own contribu- consciousness is usually associated with the deter- tions are marked as such. This theory, however, mination and control of appropriate behaviors. suffers from an important difficulty, as it is In his apparent mental causation theory, Daniel expressed in terms that remind us of the homun- Wegner takes a different view in which conscious- culus problem, as pointed out by Dennett. Indeed, ness and will are actually determined by uncon- there is still in Wegner’s account one self: someone scious causes, and have no real causal role in who is conscious, someone who attributes causality return. This view is also called epiphenomenalism, to conscious thoughts, someone who is experien- as it considers that conscious experience is an epi- cing the illusion of conscious will, and who has in phenomenon that accompanies unconscious pro- fact the properties of a homunculus. cesses, but has no functional role. In Wegner’s theory, our conscious thoughts do not necessarily cause our behaviors. Rather, both Learning Process Accounts of conscious thoughts and behaviors are caused by Consciousness unconscious mechanisms. These underlying uncon- scious causes of thoughts and the unconscious Here, we present three theories that emphasize the causes of behaviors are different, though they can influence of learning on consciousness. In their influence each other. Because of the mutual influ- sensory-motor theory of consciousness, Kevin ence between these two types of unconscious causes, O’Regan and Alva Noe¨put forward the notion of their effects (i.e., conscious thoughts and conscious learnt sensory-motor contingencies. In both Axel behaviors) are correlated as well. Because conscious Cleeremans’ radical plasticity thesis and Hakwan thoughts happen just before conscious behaviors, Lau’s higher-order Bayesian decision theory, con- they are taken to be the causes of initiated actions. sciousness results from the ability of the cognitive Here, the attribution of a causal role to conscious system to learn about its own internal states. thoughts is an illusion based on what is apparent to consciousness, not on what really happens. In addi- The Sensory Motor Theory of tion, the theory specifies the condition under which Consciousness this illusion occurs: thoughts have to appear just before an action (priority), they have to be consis- Most models of vision are based on internal tent with the action (consistency), and they have to detailed representations that are active when a be the only possible explanation of the action (exclu- particular stimulus is present in the visual world. sivity). When these conditions are satisfied, con- The sensory-motor theory of consciousness pro- scious thoughts contain a predictive model of the posed by Kevin O’Regan and Alva Noe¨ takes forthcoming action, and when the action is rea- an alternative view in which there is, according lized in agreement with the predictions, we grant to them, no need for detailed representations authorship for it and we experience ourselves as in the brain, and in which conscious experience is causal agents. produced by the mastery of sensory-motor contin- The theory of apparent mental causation does gencies. not aim at explaining how consciousness arises in In normal situations, the observer knows that he a cognitive system. Rather, it explains how our only has to direct his eyes or his attention toward conscious experience of will is an illusion that it in order to obtain detailed information. In stems from our ignorance of actual unconscious other words, the world is an external memory, causal chains. This approach has the advantage of and the information it carries is usually sufficient trying to eliminate a false but still well-established for action. As a consequence, rather than relying

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

142 Cognitive Theories of Consciousness

on internal representations that would be at the be represented. What is rejected is the notion of origin of conscious experience, this theory consid- continuous detailed representations of the outside ers consciousness to be an active and exploratory world, and the fact that having these representa- process in between the observer and the external tions could be sufficient to create consciousness, environment. In support of this theory, several without making use of it, in the sense of exploring studies, including some experiments by O’Regan it through sensory-motor contingencies. Regard- and Noe¨, have shown that normal observers ing this question, one interesting argument put can suffer from ‘change blindness,’ a situation pre- against this theory was the issue of dreams or dicted by the idea that our memory lies in the mental imagery. Since those phenomena provide outside world. In this situation, observers have compelling intuitive support for the existence of the illusion that they are conscious of the whole such detailed internal representations, how does visual scene while, actually, they fail to notice the sensory-motor theory deal with that? The important modifications in the scene. Importantly, answer provided by O’Regan and Noe¨ is that these changes are noticed when participants direct there are still some differences between normal their eyes or their attention to the critical location. visual experience and dreaming or visual imagery

These findings show that observers have an illu- situations, which correspond to the fact that in the sory and overconfident estimation of their visual latter cases the subject cannot make use of all the capacities. The sensory-motor theory of con- sensory-motor contingencies. Additionally, they sciousness also proposed to explain some features deny that dream-experiences are stable in the of conscious experience on the basis of the char- details, as they miss the stability of the world as a acteristics of the sensory-motor contingencies, i.e., memory. the principles that link exploration acts to sensorial consequences. In vision, for instance, a saccade to The Radical Plasticity Thesis of the left will shift the visual input on the retina Consciousness accordingly, but even if the position of the object in front of you has changed on your retina, you While many cognitive models use symbolic and would still feel that this object has not moved: this discrete representations, connectionist models rely principle is embodied in your sensory-motor con- on sub-symbolic and distributed units in artificial tingencies. Besides, the different sensorial modalities neural networks. In these models, representations are different means for exploring the environment, are patterns of activation over processing units. and among these modalities, the differences in the Following this perspective, Axel Cleeremans sensory-motor contingencies (e.g., optical laws differ proposed a conceptual framework termed the from acoustical laws) are the basis for the differences Radical Plasticity Thesis that put a strong empha- in the structure of conscious experience. Impor- sis on the link between conscious awareness tantly, these contingencies apply at different levels and learning. This theory is based on three main of abstraction: some relate to the physical apparatus principles. of the stimuli in a given modality, while others relate The first main idea states that learning is a to features or categorical attributes. When we look mandatory consequence of information processing, at a particular object from a changing viewpoint, leading the cognitive system to develop represen- the visual image changes but the category of the tations of higher quality. Here, the quality of a object remains constant. representation is assessed by the stability and The most original idea expressed by the sen- strength of activation in the dynamic network, sory-motor theory is that external stimuli do not and distinctiveness, which is equivalent to differ- have to be represented in detail in the brain. entiation in the theory put forward by Tononi

Importantly, however, O’Regan and Noe¨ are (see ‘The information integration theory of NOT against any form of representation, or any consciousness’ above). The second important idea storage of information in the brain. Following their is that consciousness reflects the quality of own terms, they grant at least that the knowledge representations within the cognitive system. In of the laws of sensory-motor contingencies have to this theory, the more representations achieve high

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

Cognitive Theories of Consciousness 143

quality (i.e., high strength, stability, and distinc- The Higher-Order Bayesian Decision tiveness), the more they participate in conscious Theory of Consciousness experience. Hence, in this theory, conscious expe- Signal detection theory and Bayesian frameworks rience is a graded and continuous variable. Given have recently undergone a great renewal of interest those two principles, learning is associated with among cognitive scientists. These conceptual tools higher quality representations, which are in turn bring useful insights in the description of behavioral more likely to be conscious. Finally, the third prin- performance, such as discrimination, detection, and ciple highlights the implication of metarepresenta- decision. In a nutshell, signal detection theory pro- tions for self-consciousness. Cleeremans proposes poses that discrimination between target and noise that high-quality representations are efficient detec- relies, on the one hand, on the objective distance tors of a particular content, and that they can be the (discriminability) between their two signal distribu- target of metarepresentations. These metare- tions on a psychophysical continuum and, on the presentations capture the associations between other hand, on the particular setting of a decision first-order representations, which are developed threshold (criterion or bias) on that continuum. through learning and past experience. The theory Bayesian decision theory, in turn, proposes a way also proposes that a metarepresentation helps the to optimize the setting of the decision threshold, first-order representations on which it is focused to through prior learning over time of the probability achieve higher quality. In other words, one system distributions of the noise and target signals. can support consciousness insofar as it is able to In many empirical studies on consciousness, par- learn about its environment and create internal ticipants’ awareness of a given stimulus is equated representations, and also be able to learn about its with their performance on discrimination tasks (i.e., own representations and increase their quality. discriminability). Conversely, chance level perfor- Here, the more the system knows about its own mance on a discrimination task is often assumed to rules, the more it is assumed to be conscious. imply that the participant is completely unaware Cleeremans further distinguishes between dif- of the feature targeted by the task. Hakwan Lau’s ferent aspects of conscious experience and higher-order Bayesian decision theory of conscious- describes how these aspects correlate with the ness uses empirical dissociations between perfor- increase, through learning, in the quality of repre- mance and awareness to support the idea that sentations. The formation through learning of consciousness may not always be associated with internal representations is depicted in three stages. an increase in discriminability. Rather, the hypothe- The first one relates to implicit cognition: a poor- sis defended here is that it is related to the setting quality representation can influence behavior, but and the maintaining of the criterion threshold used it is not strong enough to let the subject know about for the perceptual decision. This theory associates these influences or to have much control over some features of the higher-order thought theory them. Through exposure and learning, the repre- with the Bayesian decision framework. More pre- sentation achieves higher quality and becomes cisely, it proposes that while the lower-order system explicit. In this second step, the availability to con- implements discriminability, the higher-order sys- trol and the potential impact on the cognitive sys- tem, in turn, implements the decision threshold. In tem also increase dramatically and reach a this view, the lower-order system performs a certain maximum. When the representation is sufficiently number of discriminations upon external signals, learned, it becomes automatic. According to and the higher-order system learns about the dis- Cleeremans, this third stage is associated with tribution of states of the lower-order system, so as high-quality representations readily available to to interpret the signal, and to be able to set the conscious awareness, though the subject has less threshold in an optimal manner. While Lau’s theory control over their influences as they operate in a is to date clearly not developed as far as other mandatory way. Thus, in this final idea, Cleere- proposals, this work provides a new idea to the mans takes a view that can be contrasted with the current theoretical landscape. It addresses a theoret- classical assumption that automatic processes are ical issue related to signal detection theory, which unconscious.

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

144 Cognitive Theories of Consciousness

is a methodological tool of increasing importance consciousness proposed by the philosopher Ned in the field of consciousness. Block. Access consciousness relates to the global

use of conscious information, and the possibility through consciousness to trigger complex and Conclusion integrated processes such as reasoning, control of

actions, decisions, and verbal reports. In contrast, In thisarticle we have presented an overview of phenomenal consciousness refers to the mere sub- the most representative cognitive accounts of con- jective experience, the ‘what it is like’ question sciousness. Most of these theories radically differ expressed also by Thomas Nagel. Both Chalmers in their conception of what consciousness is. While and Block defend phenomenal consciousness and some consider that it reflects the activation the hard problem, claiming that conscious experi- of attended intermediate level representations ence of a stimulus is not reducible to its information

( Jackendoff, Prinz), or the involvement of com- processing and its causal influences in the system. plexes in a system (Tononi), others would equate It turns out that most cognitive models are consciousness with global broadcasting (e.g., Baars, expressed in information processing terms, and as

Dennett), and still others would associate it with such they are bound to take a reductive approach learning upon ones’ own representations (e.g., when trying to explain phenomenal consciousness.

Cleeremans, Lau) or upon sensory-motor contin- Intrinsically, they favor functionalist perspectives gencies (O’Regan and Noe¨). As such, it is obvious whereby information processing is all there is to that consciousness is not yet a well-defined notion. conscious experience. On a more general perspec- In addition, cognitive accounts are now facing two tive, science deals with measurements and measure- epistemological constraints that impose important ments are by definition targeted to a piece of pressure on their development. The first one has information that is measured. Hence, in most cogni- been put forward by philosophers and corresponds tive accounts, the hard problem is either reduced to the need to focus on the ‘hard problem’ rather to the easy one or even completely denied. For than the ‘easy problem’ of consciousness. The sec- instance, some will stand that it might be in fact ond one is related to the increasing amount of necessary to revise our definition of what conscious- empirical evidence resulting from the study of ness is, in order to eradicate any reference to some the brain. We conclude this article by focusing on mysterious ‘phenomenal’ properties of the mind these two constraints. (e.g., Dennett). Indeed, one should not overlook The philosopher David Chalmers termed a the possibility that phenomenal and access con- dichotomy between the easy problem and the hard sciousness are two notions that have to be disso- problem of consciousness in order to delineate the ciated only conceptually. They are not easy to two major features related to consciousness. On the dissociate experimentally since any measure of one hand, consciousness offers a processing advan- phenomenal consciousness can hardly be disso- tage, as it allows for the information in working ciated from the involvement of access consciousness. memory to be processed in a long-run by multiple Indeed, measuring phenomenal consciousness in an devices. On the other hand, consciousness carries experimental perspective must be based on some the qualitative property of subjective experience. form of report, hence on access consciousness.

Chalmers states that the first issue is in fact an The other limit of purely functional accounts of interesting though only a computational problem; consciousness comes from the brain. Indeed, as they remain distant from the biological implemen- hence it is easy to study scientifically. The second one, however, is much more mysterious. How can tation, purely functional or philosophical perspec- the subjective quality of experience arise from tives on consciousness are now likely to miss this squishy organic matter is a question that seems crucial dimension, and source of evidence. Conse- to go far beyond our possible understanding. The quently, they will lack the same amount of explan- distinction between the hard problem and the atory power. Indeed, consciousness has become easy problem can be mapped onto the dichotomy for many scientists a biological problem whose between access consciousness and phenomenal answers will be found by studying the brain. In

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

Cognitive Theories of Consciousness 145

fact, some authors go a step further in arguing that Chalmers D (1995) Facing up to the problem of the operational definition for the scientific study of consciousness. Journal of Consciousness Studies 2(3): 200–219. consciousness should be expressed in neural terms. Chalmers DJ (1996) The Conscious Mind: In Search of a In that perspective, the psychological tools that we Fundamental Theory. Oxford University Press. use may all be discarded in favor of a more physi- Cleeremans A (2005) Computational correlates of consciousness. Progression in Brain Research 150: ologically grounded approach. Even if we do not 81–98. want to go that far, it is a matter of fact that today Cleeremans A (2008) Consciousness: The radical plasticity basic observations of brain processes might help, thesis. Progression in Brain Research 168: 19–33. Dehaene S and Naccache L (2001) Towards a cognitive by providing new concepts that would help direct- neuroscience of consciousness: Basic evidence and a ing research, and new critical test that would help workspace framework. Cognition 79(1–2): 1–37. discarding unfitting theories. Memory is a good Dennett DC (1991) . Little, Brown & example. Functional accounts of memory had to Company. Edelman GM and Tononi G (2000) A Universe of go back to the drawing board when the neurology Consciousness: How Matter Becomes Imagination. of memory began to be understood in a more New York: Basic Books. precise way. On the other hand, functional and Fodor J (1983) The Modularity of Mind: An Essay on Faculty Psychology. MIT Press. neurological accounts have sometimes worked Jackendoff R (1987) Consciousness and the Computational together productively, as for instance Baars’ global Mind. Cambridge: MIT Press. Lamme VA (2006) Towards a true neural stance on workspace theory, which has been extended at the consciousness. Trends in Cognitive Sciences 10: neurobiological level by Stanislas Dehaene and 494–501. colleagues. Of course, without theoretical knowl- Lau HC (2008) A higher-order Bayesian decision theory of edge of the functions that are to be explained, a perceptual consciousness. Progression in Brain Research purely biological theory of cognition would be 168: 35–48. Mack A and Rock I (1998) Inattentional Blindness. impossible to construct. In other terms, the union Cambridge, MA: MIT Press. between psychological and neurobiological per- Miller GA (1956) The magical number seven, plus or minus spectives makes both approaches stronger, and two: Some limits on our capacity for processing information. Psychological Review 63: 81–97. future models of consciousness will be bound to Nagel T (1974) What is like to be a bat? Philosophical Review include brain evidences and hence to be trans- 83: 434–450.

Norman DA and Shallice T (1986) Attention to action: Willed formed into neurocognitive rather than purely and automatic control of behavior. In: Davidson R, cognitive accounts of consciousness. Schwartz G, and Shapiro D (eds.) Consciousness and Self Regu lation: Advances in Research and Theory, 2nd edn., See also: History of Philosophical Theories of vol. 4, pp. 1–18. New York: Plenum. Consciousness; Neurobiological Theories of Online Papers on Consciousness. Compiled by David Consciousness. Chalmers.http://consc.net/online/ Posner MI (1994) Attention: The mechanism of consciousness. Proceedings of the National Academy of Sciences of the USA 91(16): 7398–7402. Posner MI and Snyder CRR (1975) Attention and cognitive Suggested Readings control. In: Solso RL (ed.) Information.

Prinz J (2005) A neurofunctional theory of consciousness. In:

Anderson JR (1983) The Architecture of Cognition. Brook A and Akins K (eds.) Cognition and the Brain: Cambridge, MA: Harvard University Press. Philosophy and Neuroscience Movement, pp. 381–396. Baars BJ (1988) A Cognitive Theory of Consciousness. Cambridge: Cambridge University Press. New York: Cambridge University Press. Rumelhart DE and McClelland JL (1986) Parallel Distributed Baddeley AD and Hitch GJ (1974) Working memory. In: Processing: Explorations in the Microstructure of Cognition, Bower GA (ed.) The Psychology of Learning and Vol. 1: Foundations. Cambridge, MA: MIT Press. Motivation , pp. 47–89. Academic Press. O’Regan JK and Noe¨ A (2001) A sensorimotor account of Block N (1995) On a confusion about a function of vision and visual consciousness. Behavioural and Brain consciousness. Behavioural and Brain Sciences 18: Sciences 24(5): 883–917. 227–247. Schneider W and Shiffrin RW (1977) Controlled and Block N, Flanagan O, and Gu¨ zeldere G (eds.) (1997) The automatic human information processing:

Nature of Consciousness: Philosophical Debates. Vol. I. Detection, search, and attention. Psychological

Cambridge, MA: MIT Press. Review 84: 1–66. Broadbent DE (1958) Perception and Communication. Tononi G (2004) An information integration theory of Pergamon Press. consciousness. BMC Neuroscience 5: 42.

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146

Author's personal copy

146 Cognitive Theories of Consciousness

Tononi G and Edelman GM (1998) Neuroscience – Wegner DM (2002) The Illusion of Conscious Will. MIT Press. Consciousness and complexity. Science 282(5395): Wegner DM (2003) The mind’s best trick: How we 1846–1851. experience conscious will. Trends in Cognitive Sciences 7: Velmans M and Schneider S (eds.) (2007) The Blackwell 65–69. Companion to Consciousness, pp. 225–235. Oxford, UK: Blackwell Publishing.

Biographical Sketch

Vincent de Gardelle is a doctoral student supervised by Sid Kouider at the Ecole Normale Supe´rieure (Paris, France). His research addresses the question of biased conscious perception, by examining the perceptual illusions that might result from a subject’s expectations under poor sensory evidence.

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

Encyclopedia of Consciousness (2009), vol. 1, pp. 135-146