Reorganization of the connectivity between elementary functions as a common mechanism of phenomenal consciousness and working memory: from functions to strategies

Abstract

In the present communication, phenomenal consciousness, access consciousness and the closely related concept of working memory are presented in the context of a neurocognitive model—the REF (reorganization of elementary functions) framework. The REF framework is based on connectionist networks within which the ‘units’ are advanced processing modules called elementary functions (EFs). In this framework, the focus is on dynamically changeable ‘strategies’—based on reorganizations of the connectivity between EFs—rather than on the more traditional ‘cognitive functions’. The background for the REF framework and especially how the neural correlate of consciousness is understood within these models is summarized. According to the REF framework, phenomenal consciousness cannot ‘overflow’ availability of information for action. Phenomenal consciousness may, however, overflow working memory because working memory in the present context is seen as a surface phenomenon reflecting underlying dynamic strategies—influenced by both experience and situational factors.

This article is part of the theme issue ‘Perceptual consciousness and cognitive access'.

1. Introduction

The neural correlate of consciousness (NCC) may be best understood and modelled as part of a more comprehensive and complete neurocognitive modelling of the brain (e.g. [1]). An understanding of consciousness-related neural processes within such a comprehensive and integrative neurocognitive model may also offer new approaches to the conceptualization of a number of theoretical issues; for instance, the distinction between access consciousness and phenomenal consciousness [2]. According to this distinction, phenomenal consciousness is to be understood as subjective experience, whereas access consciousness has been conceptualized in a few different ways—originally as information available for rational control of action and speech [2,3]. Block and others have debated the proposal that phenomenal consciousness ‘overflows’ access consciousness, meaning that we can subjectively experience more than the information that is in working memory (e.g. [4–7]). Ned Block, however, has in most recent publications interpreted his own concept of access as being identical to the contents of ‘working memory’ [5,8].

As elaborated elsewhere, there seems to be no experimental evidence available to argue that phenomenal consciousness overflows access consciousness under the condition that access is interpreted as information available for control of action of speech [3]. Arguably, such evidence is impossible to provide, as the only way we can come to know about phenomenal consciousness is by talking about it, acting on it or in some other way take action or speech as the point of departure.

However, if access consciousness is taken to be identical to the contents of working memory, the situation may be more optimistic for the overflow hypothesis—as we here just have to show that certain actions do not require working memory capacity, and that those actions represent consciousness. Bronfman et al. [9] presented coloured letters to experimental participants in a Sperling paradigm (see [10]). Participants were asked to estimate the colour diversity in either the cued or an uncued row. It was shown that participants are able to report about the colour diversity outside the cued row without expense with regard to the ability to correctly report the letters in the cued row. Arguably, the finding shows that colour diversity judgements do not involve any working memory load, and, thus, represent phenomenal consciousness without access consciousness.

The finding can be interpreted in different ways and more experiments are needed in order to conclude that phenomenal consciousness may indeed overflow working memory content. Regardless, however, given the very fact that different empirical hypotheses can be raised for different interpretations of access, it is worthwhile to look into the consequences hereof for a theoretical understanding of the general relation between consciousness and varieties of cognitive access. How are we to understand that it is conceivable that there may be conscious experience without the experienced content being represented in working memory—yet still being available for action or speech?

In the present communication, we focus on the neurocognitive framework known as the REF (reorganization of elementary functions) framework. One reason for doing so is that—as will be elaborated below—the REF framework offers a number of re-conceptualizations of cognitive ‘functions’ (e.g. working memory) relevant to the relationship between access consciousness and phenomenal consciousness. We initially present the background for the creation of the REF model (e.g. [11–14]) and a brief account of the model. We further describe the subsequently developed REFCON (reorganization of elementary functions and consciousness) [15,16] as well as REFGEN (the general reorganization of elementary function) models (e.g. [1]). Within this neurocognitive framework, we then describe ways to conceptualize the access and phenomenal consciousness as well as the (in our opinion) closely related concept of working memory.

2. Background for the reorganization of elementary functions framework

The primary motivation for the original development of the REF model was to obtain a neurocognitive understanding that can simultaneously account for both of the two apparently contradictive phenomena of functional localization and functional recovery after focal acquired brain injury. The fact that the regional specialization of the brain involves at least a certain degree of cognitive functional localization has been well demonstrated in numerous human studies applying neuroimaging (e.g. [17,18]) or the symptoms associated with focal brain injury (e.g. [19–26]). Frequently, human studies of the consequences of acquired brain injury may be flawed by the lack of information regarding pretraumatic performance, the often limited group sizes and the need to resort to suboptimal control groups. Obviously, human studies are essential (not least when dealing with phenomena such as language and/or consciousness), but there is a need to supplement such studies with animal models that can provide better-controlled circumstances in which to analyse the consequences of traumatic brain injury—and more optimal methods regarding the analysis of posttraumatic neurocognitive processes. Thus, animal model-based research has provided a further and, in some instances, even better documented demonstration of the existence of a cognitive functional localization within the brain—a regional specialization pointing to specific functions being mediated by specific regions of the brain (e.g. [27–32]).

While the functional localization is, thus, well supported by studies across methods as well as across species, the phenomenon of posttraumatic functional recovery after acquired brain injury may appear to contradict the existence of such a functional localization. In patients as well as in animal models, it has repeatedly been demonstrated that via posttraumatic training individuals will become able to either perfectly or at least to a certain extent perform a task which was originally rendered more or less impaired by the acquired brain injury (e.g. [19–25,27–33]). If such a functional recovery should be compatible with the most common view of functional localization, it would be required that during the posttraumatic recovery process, the original neural substrate is recreated. As extensively discussed elsewhere [11], there is amble evidence to demonstrate that although extensive posttraumatic plastic processes are initiated by all types of acquired brain injury, the neural substrate of pretraumatic task mediation will not be re-created within any part of the brain.

In the terminology of the REF model, all behavioural and/or mental phenomena are surface phenomena. This includes, for example, task solution, thoughts and conscious awareness. To step beyond the apparent contradiction between functional localization and posttraumatic functional recovery, it is necessary to conduct research that in detail can analyse the neural and cognitive processes allowing an individual to posttraumatically regain the surface phenomena originally lost to the injury in question. Such research has mainly been performed in animal models where the ability to ‘challenge’ the ongoing recovery process at both the neural and cognitive levels allows a ‘mapping’ of the neural and functional reorganizations associated with improved task performance (e.g. [34–36]). Extensive studies conducted along such principles in both animal models and humans (e.g. [27–32,37]) have revealed two important principles regarding the posttraumatic recovery process.

• (a) Comparing a given surface phenomenon before it is impaired by acquired brain injury with after a posttraumatic recovery process, it becomes clear that the surface phenomenon is mediated by dissimilar neural substrates and dissimilar computational processes. This is the case even when the surface phenomenon has posttraumatically reached a quality similar to what was seen pretraumatically (e.g. demonstrating task solution of similar proficiency).

• (b) In the subject demonstrating a recovered surface phenomenon, individual substructures within the uninjured part of the brain contribute to task mediation in the form of processes of a ‘modular’ nature. Contributions to task performance by such substructures do not seem to be either task-specific or specific to any of the cognitive domains. Rather, a given substructure contributes the same type of information analysis within a multitude of different contexts and tasks (e.g. [11–14]).

As extensively discussed elsewhere (e.g. [11–14]) not the least based on such results, the original REF model was developed.

3. The reorganization of elementary functions framework

The REF model relates the neural processes and states to the mental/behavioural processes and states via computational processes. In doing so, it agrees with the arguments of, for instance, Carandini [38] and Overgaard & Mogensen [39] where it is stressed that without an ‘intermediate level’ of a computational nature, it will remain unrealistic to relate neural and mental/behavioural processes, respectively. Such arguments may be seen as further developments of the classic levels of analysis presented by Marr & Poggio [40] and Marr [41].

On the basis of the results summarized above, however, it is tempting to operate with two such computational levels—rather than just one. One of these levels is the computations provided by local networks within substructures of the brain—while the ‘higher’ level of computations is a level where the computational processes are mediated by connectionist networks combining the specific modules of the previous level (figure 1).

Figure 1.

The subdivision of the computational and neural levels of the REF framework (see text for details).

With such an emphasis on both local and strictly modular information processing and the combination of these ‘modules’ into connectionist networks, the REF model includes crucial elements from both the modular models of brain organization (e.g. [42–45]) and the neurocognitive models emphasizing distributed connectionist networks (e.g. [46–49]). A detailed discussion of the relationship between the REF model and the modular and connectionist neurocognitive models can be found in, for instance, Mogensen & Overgaard [1]. In general, it should be emphasized that modular models are well suited to account for functional localization while being, in general, unable to account for functional recovery—while connectionist models are well suited to account for functional recovery—but can rarely if ever account for the degree of functional localization and (relatively uniform across individuals) regional specialization found in numerous studies.

The REF model (e.g. [11–14]) can simplified be described as a connectionist network within which the ‘units’ are advanced processing modules called elementary functions (EFs). This is in sharp contrast with connectionist networks within which the ‘unit’ is a functionally ‘indifferent’ ‘neuron’.

The neural substrate of a given EF is localized within a restricted subdivision of a neural structure. Every EF performs a fixed information processing on whatever input it is given and provides an output for further processing. The information processing performed by individual EFs cannot be characterized according to traditional ‘functional/cognitive’ terms. It can best be described in mathematical terms. A given EF will typically simultaneously be involved within information processing associated with several if not numerous traditional cognitive domains. Every structure/substructure of the brain (e.g. the hippocampus) mediates hundreds or thousands of unique EFs.

The internal synaptic connectivity of a given EF remains constant. That connectivity represents the (fixed) network mediating the information processing provided by that EF. In contrast with this unchanging nature of the intra-EF connections, the inter-EF connections (the synapses potentially connecting one EF to another) are able to change dynamically—some doing so frequently and vary rapidly. The reorganization of connectivity between EFs is a crucial mechanism within the REF framework.

According to the REF model, the ‘bridge’ between the strictly localized and low-level information processing EFs and the surface phenomena in the form of problem-solving and/or mental phenomena are the algorithmic strategies (ASs). ASs consist of numerous interacting EFs and are distributed—in the sense that the neural substrate of an AS includes both the neural substrate of the individual EFs and the neural connections mediating the interactions between these EFs. Typically, an AS will include EFs within numerous parts of the brain. A given AS is the neurocognitive mechanism mediating a given surface phenomenon.

Whenever an individual meets a situation calling for a surface phenomenon for which there is no available AS (either because the situation has not been encountered before or due to previous mechanisms being impaired by brain injury), a ‘best guess’ AS will be applied. As illustrated in figure 2, backpropagation processes (e.g. [50–52]) will modify the connectivity between EFs according to the attempted AS application and the obtained result. It is, in other words, the obtained feedback (external as well as internal) from the attempted AS application that determines potential reorganizations of ASs. Thus, ASs will gradually be modified and novel ASs emerge. As discussed elsewhere (e.g. [1]), it is the modification of the inter-EF synaptic connections that plays a crucial role in the experience-dependent and backpropagation-mediated reorganizations of the neurocognitive mechanisms. As briefly mentioned below, some of these synaptic modifications remain rather stable, while others are short-lasting and constantly changing. This reorganization of the connectivity between EFs is the mechanism allowing a functional recovery after acquired brain injury. The neural structures lost to injury were the substrate of typically numerous EFs, and consequently many ASs were lost. But the preserved EFs within the rest of the brain may be able to be combined into ASs able to mediate a surface phenomenon of equal proficiency to what was seen pretraumatically. If so, a ‘full recovery’ will seemingly occur. It should, however, be noted that such a full recovery does not involve a recreation of either the original neural substrate or the original cognitive mechanisms (e.g. [14]).

Figure 2.

Simplified illustration of the mechanisms of selection and evaluation (with consequent backpropagation-mediated modifications) of ASs (e.g. [13] for details).

These mechanisms allowing an individual to potentially recover fully after acquired brain injury are also the mechanisms originally establishing most if not all ASs allowing surface phenomena such as task solution and mental phenomena. Furthermore, the feedback and backpropagation-mediated modification of ASs lead to a more or less constant flexibility and modification of the mechanism of such behavioural and cognitive phenomena.

At a level between ASs and EFs are the algorithmic modules (AMs). AMs are similar to ASs in consisting of a number of EFs and the patterns of interconnection between these EFs. And like ASs, AMs are connectionist ‘programs’ providing cognitively relevant information. In contrast with ASs, however, an AM cannot in itself be the mechanism of any surface phenomenon. AMs are typically created as a form of ‘common area’ between ASs and can be seen as higher-level ‘building blocks’ common to a number of ASs. As an illustrative example, it can be mentioned that while the interpretation of a linguistic message or the production of a spoken sentence will be mediated by an AS, an AM may mediate grammar (e.g. [37]). Grammar will obviously lend important aspects to practically every linguistic-oriented ASs—but grammar in itself is never a surface phenomenon.

While the original REF model primarily focused on the mechanisms of problem-solving as well as the basic neurocognitive processes, the subsequent expansion in the form of the REFCON model [15,16] (figure 3) was mainly developed to address perceptual processes and not the least consciousness in the form of perceptual awareness.

Figure 3.

Simplified illustration of the processes of the REFCON model from sensory input (activation of specialized EFs called perceptual EFs) to integration into SAS (see [15] for details).

The REFCON model is based on the same units and dynamic principles as the original REF model. It does, however, introduce a number of new entities and concepts. The two most essential such components are the perceptual algorithmic modules (PAMs) and the situational algorithmic strategy (SAS). The structures and mechanisms of the REFCON model are described in detail by Overgaard & Mogensen [15,16] and summarized by Mogensen & Overgaard [1].

PAMs represent the external entities being perceived. PAMs are hierarchically ordered. The lower-level PAMs represent features rather than objects, while PAMs at progressively higher levels represent more complex entities and eventually individual objects. PAMs are selected in a ‘mutual competition’ process—and a perceptual process consists of the selection of PAMs of constantly higher levels. Normally continuing until what is being perceived has been fully identified in the form of selection of a PAM of the highest level.

SAS is a highly specialized, dynamic network reflecting the current status of the individual. Thus, SAS (which is to be seen as distributed across practically all of the brain—including EFs from virtually all brain structures) represents not only the current perceptual situation but also the general internal—including mental—status of the individual.

Being AMs, PAMs (even those of the highest level) cannot in themselves mediate a mental phenomenon—including perceptual awareness. By contrast, PAMs of the highest level (and under certain circumstances—further discussed elsewhere [1,15,16]—also PAMs of lower levels) are integrated into SAS. And as parts of SAS, PAMs (and for that matter other AMs) become available for conscious awareness and/or action.

Perceptual awareness has traditionally been reported in a dichotomous way—assuming either complete presence or absence of consciousness. The use of more refined methods such as the perceptual awareness scale (PAS) [53] has demonstrated that consciousness is better understood as a process that is present in degrees—ranging via a number of intermediate steps from totally absent to totally present (e.g. [54–57]). According to the REFCON model, it is the degree of PAM integration into SAS that determines to what extent a perceptual entity is available for perceptual awareness. Thus, the perceptual process will—via selection of PAMs of progressively higher levels and eventual integration into SAS—result in a SAS configuration with a more or less integrated PAM of the highest level representing what is being perceived.

The most recently developed model within the REF framework, the REFGEN model (figure 4), focuses on the global neurocognitive organization and tries to avoid the short-comings (e.g. in the form of underdefined entities, etc.) that practically always will be the consequence of more domain-specific models (e.g. the original REF model and the REFCON model).

Figure 4.

The major components of the REFGEN model (see [1] for details).

Again, based on the basic principles of the REF model, the REFGEN model includes all components from both the original REF model and the REFCON model. Additionally, however, the REFGEN model adds two specialized ASs: the goal algorithmic strategy (GAS) and the comparator.

Like SAS, GAS is a highly distributed and extremely dynamic AS. But while SAS represents the current situation of the individual, GAS represents in the broadest sense the ‘goals’ of the individual. Such goals may be the need immediately to rearrange aspects of the environment, the need to gather information within a particular part of the visual field—or they may be more general and long-term goals.

Comparator is a specialized AS that constantly compares SAS and GAS. Based on such comparisons, comparator can initiate action, either in the form of externally oriented behaviour (e.g. via the activation of ‘ordinary’ ASs) or via backpropagation-mediated reorganization of GAS and/or SAS. Additionally, comparator can (again based on the combined configuration of GAS and SAS) initiate backpropagation-mediated reorganizations of other ASs or AMs.

It has repeatedly been demonstrated that first-order consciousness and introspection, respectively, differ regarding both cognitive (e.g. [58–60]) and neural [61] mechanisms. In an attempt to provide a neurocognitive model that is compatible with the available data, we recently [62] published the Integrative Model. The Integrative Model is, to our knowledge, the first model to suggest a parallel rather than a serial relationship between primary consciousness and introspection. In an account closely related to what is described in the REFCON model, the Integrative Model presents a stepwise and hierarchical analysis of perceptual information up to the highest level of analysis. This analysis is in parallel available within what is called network A and network B. While information within network A becomes available to primary consciousness and action, the information within network B becomes available to introspection and action based on introspection. While the Integrative Model is not explicitly part of the REF framework, it is closely related and based on the same principles. As discussed briefly by Mogensen & Overgaard [1], the principles of the Integrative Model will become integrated into the REFGEN model. This will be in the form of a modified REFGEN model in which SAS has been split into a SAS-A and a SAS-B—SAS-A being the equivalent of network A of the Integrative Model and SAS-B the equivalent of the network B of the Integrative Model. Such a modified REFGEN model is presently being prepared for publication (Mogensen & Overgaard [63]).

4. Implications for consciousness and cognition

A major difference between the REF framework and most, maybe all, other neurocognitive approaches is that the REF framework focuses on dynamically changeable ‘strategies’ rather than on functions. Surface phenomena in the form of overt behaviour or mental phenomena are mediated by either SAS or other ASs. And, in general, when analysing the mechanisms of surface phenomena, a focus on traditional functions has been replaced by an analysis based on ASs—not the least SAS, GAS and/or comparator. Traditional cognitive functions can of course be identified as surface phenomena. But the underlying mechanisms are within the REF framework analysed in terms of strategy-based approaches. An example elaborated by Mogensen & Overgaard [1] illustrates an aspect of this approach. If there is within GAS a representation of a goal calling for optimal integration of information from a given quadrant within the visual field, comparator will (via a comparison between SAS and GAS) detect the discrepancy between the current situation in SAS and the desired higher degree of information integration. As a consequence of this discrepancy, comparator will (via backpropagation mechanisms) modify the connectivity within the appropriate part of SAS in such a way that novel inputs within the given quadrant is more efficiently integrated into SAS. Consequently, PAMs associated with the quadrant in question will obtain a prioritized analysis. The process in question, thus, calls for a comparator-mediated analysis of both SAS and GAS followed by a SAS reorganization by comparator. But viewed as a surface phenomenon, the process can obviously also be described as selective attention given to the visual field quadrant in question.

The experimental approach and theoretical analyses within the REF framework have repeatedly resulted in modified delineations regarding cognitive functions. Spatial orientation has traditionally been conceptualized as being either allocentric (e.g. [29]) or egocentric (e.g. [30]). Research applying the experimental and conceptual methods presently presented has, however, demonstrated that allocentric spatial orientation is best understood as subdivided into allocentric orientation of the mapping type and non-mapping type, respectively—the two types differing regarding both neural mechanisms and cognitive processes (e.g. [28,29,32]).

Another example of the ways in which the AS and AM-based REF framework provides an alternative approach to traditional functions can be found within perceptual processes. PAMs are dynamically being changed according to experience. And PAMs—especially at the higher levels—are typically not only multi-modal but may also include, for instance, emotional components associated with the item in question. Consequently, the REF framework offers an alternative approach to the debate regarding whether emotions always are secondary to cognitive processes or can occur independently (e.g. [64–67]). When an emotion-associated stimulus or situation is perceived, it is in the REF framework associated with SAS integration of the relevant PAM (or PAMs) of the highest level. Such a PAM simultaneously represents both the ‘cognitive’ and emotional dimensions of the percept in question. Consequently, the model represents ‘cognitive’ and ‘emotional’ components as integrated elements within the same PAM. Thus, the REF framework stresses integration rather than separate processes with the potential of being either primary or secondary.

While some ASs and AMs are relatively stable and may undergo reorganization of the constituent EFs only rarely, others undergo practically constant dynamic changes as a consequence of situational changes as well as the internal adjustments between SAS, GAS and comparator. The synaptic basis for reorganizations of the connectivity of individual EFs are likely to differ between relatively rare and long-term modifications and the rapid dynamic changes within, for instance, SAS, GAS and comparator. Potential mechanisms for rapid reorganizations may (as further discussed by Mogensen & Overgaard [1]) be prewired ‘latent’ synapses (e.g. [68–71]) and/or modifications of dendritic spines (e.g. [72–79]).

The comparator-mediated ongoing reorganizations of the connectivity of SAS and GAS are of crucial importance for the processing of information—both regarding externally originating stimuli and in the form of ‘purely internal’ information processing in the form of, for instance, thinking, mental rehearsal etc. (the latter issues will be more elaborately analysed in a detailed treatment of the relative roles of SAS-A and SAS-B (Mogensen & Overgaard [63]).

As already indicated in the brief reference to the ‘attention’ example, the analysis of external stimuli is typically a combined consequence of ‘bottom-up’ processes primarily associated with the selection of adequate PAMs and ‘top-down’ influences in the form of comparator-mediated restructuring of adequate aspects of SAS.

It is, however, obvious that further analyses based on both experimental and theoretical developments will be needed to fully account for the ways in which information—e.g. in the form of various types of AMs—is being integrated into SAS. Recently, Fazekas & Overgaard [80] have published an extensive analysis of both cognitive and neural aspects of subjective experiences and perceptual representations. It is being demonstrated that both conscious experience and information at the representational level can be degraded along multiple dimensions. Such degradation is also associated with specific physiological processes. The analysis [80] emphasizes the fact that when addressing information integration into SAS, it is necessary not only to focus on the ‘degree of integration into SAS' but also on a variety of other qualitative aspects. Within the REF framework, we have so far primarily addressed the fact that the ‘degree of integration’ into SAS is the mechanism of the degree of availability to conscious awareness and/or action. But, it is obvious that a given level of integration into SAS can be realized in a multitude of ways—with different cognitive consequences. Further experimental and theoretical developments will expand the REFGEN model in order to accommodate such differences. It is presently premature to suggest any details regarding such processes. But there can be no doubt that a crucial element is the combined effects of external stimulus quality (bottom-up influences) and ‘top-down’ influences growing out of the comparator-mediated comparisons between current configurations within GAS and SAS.

The REF framework is different from other existing theories of mind–brain relations in general and consciousness in particular. According to the model, information that is available for action and is integrated into the SAS is conscious. This means that the model does not predict any specific NCC other than the correlate relating to the strategy processing the relevant informational content that is integrated into SAS. Any neural activation that realizes information available in the above-mentioned fashion is an NCC. This is different from many theories of consciousness, e.g. the global workspace model or higher order theories of consciousness (e.g. [81,82]) where the prefrontal cortex is considered a necessary prerequisite of consciousness. However, it is also different from theories arguing that primary regions of the brain are principally necessary for consciousness. That may empirically be true—but, according to this model, not as a matter of principle (e.g. [83–85]). Other ‘integrative’ models in neuroscience will share many predictions with the REF framework without agreeing in all aspects. One example could be Friston's predictive coding framework (e.g. [86]) which sees all perceiving as ‘top-down predictions'. Whereas our model also involves a constant ‘rewriting’ of representations based on error, it does not attempt to explain all brain events within a Bayesian framework, interpreting everything as ‘prediction’ and ‘error’. Similarly, the prediction coding framework seems rather neutral with respect to the theory of consciousness that is entailed in the REF framework. Thus, although several empirical predictions and interpretations are shared between the REF framework and several other current theories of how the mind relates to the brain, core statements of the models are unique to the REF, REFCON and REFGEN models.

5. Working memory according to the reorganization of elementary functions framework

As elaborated elsewhere [1,15], within the REF framework, consciousness is seen as fundamentally linked to information that is available for action by way of this comparator-mediated relation between SAS and GAS. As it seems at least intuitively reasonable to argue that information that is available for action is also represented in working memory, it could be seen as a challenge to the understanding of consciousness within the REF framework if phenomenal consciousness does indeed overflow access (meaning working memory). Nevertheless, in the experiment by Bronfman et al. [9], the evidence of overflow consisted of participants' reports of the colour diversity—meaning that the experience of diversity was in fact available for action (in this case, a report). How could this be interpreted?

As should be evident based on the perspectives above, the usual understanding of working memory as a specific function, consisting of very specific subfunctions, relating to particular localizations in the brain is conceived differently within the REF framework. Here, again, the dynamic interactions between various ASs is central: information represented in working memory has to be integrated sufficiently into SAS—in the form of, for instance, a PAM or other AM—in order to be available for action. Such an integration is in the case of working memory practically always dependent on situational (e.g. task-dependent) factors. Thus, the SAS integration of ‘items’ within working memory is strongly influenced by not only bottom-up factors but also ‘top-down’ factors related to the comparator-mediated interaction between SAS and GAS. According to this view, typical models of working memory showing a specific functional organization between subsystems (e.g. [87]) only represent one possible strategy. Whereas this strategy may be the most common, or maybe even the most effective strategy under what we consider as typical conditions, different strategies could be implemented under special conditions. One may consider synaesthetic associations of different modalities as an example of such an ‘unusual’ strategy. Furthermore, according to the REF framework, the strategy typically modelled as working memory may only represent one possible way that information may be available for action.

The repeatedly demonstrated ‘limited capacity’ of working memory may raise the issue why only a limited number of items can be available for conscious awareness and/or action at a given time. It might seem more likely that SAS could have the capacity for more or less unlimited information integration. In much literature on attention or working memory, the capacity limitation is explained by simply pointing out that perception would be ‘too overwhelming’ without it [88]. However, although this of course does appeal to intuition, it does not really explain why the ‘threshold’ of ‘being overwhelmed’ is as it is. Rather, the argument appears circular, as it is because of the limitation that any amount of information that exceeds it appears ‘overwhelming’—it is not because of ‘overwhelmingness’ that the limit is as it is.

The most likely explanation for such a phenomenon is, however, that limitations are found not in SAS itself but in the configuration of GAS. If the working memory limitations are interpreted as a reflection of the basic biological abilities of a given individual—for instance: how many ‘entities of information’ can the individual respond to at any given moment?—it becomes clear that GAS may be the factor dictating apparent limitations onto levels of information integration into SAS. We do not want to exclude the possibility of genetically dictated basic configurations of GAS calling for at least a certain level of such limitations. But because GAS is also constantly reconfigured according to experiences, dynamic influence on GAS are likely also to influence the limitations on working memory capacity. The influence of such dynamic processes would agree well with findings that individual differences in working memory capacity may be strategy-dependent [89] and that working memory capacity may be influenced by training procedures (e.g. [90,91]). In this way, elements of what is normally modelled as a ‘working memory function’ are seen as natural consequences of how SAS and GAS work.

6. Access consciousness and phenomenal consciousness in the context of the reorganization of elementary functions framework

As mentioned above, access consciousness (working memory) will within the REF framework be a reflection of the described interactions:

• (i) between bottom-up factors related to the stimuli and top-down factors related to the comparator-mediated interaction between SAS and GAS—and not least,

• (ii) the ‘internal’ and strategy-based interactions between the current configurations within SAS and GAS.

Phenomenal consciousness, too, has to be seen as reflecting not only the basic stimulus analysis (via the progressive levels of PAM analyses), but also interactions between SAS and GAS. As described in the REFCON and REFGEN models, the level of AM integration is a crucial factor regarding the level of phenomenal consciousness. As phenomenal consciousness is interpreted as tied to availability, the situational and strategy-related interactions between SAS and GAS may not only influence the level of experienced consciousness but also qualitative factors such as those addressed by Fazekas & Overgaard [80].

In this way, according to the REF framework, phenomenal consciousness cannot ‘overflow’ availability of information for action [4,5]. However, the REF framework does not make any claims regarding the likelihood that phenomenal consciousness may overflow working memory—because (like a number of other of the more ‘traditional’ cognitive functions) working memory is in the present context seen as a surface phenomenon reflecting underlying dynamic strategies (influenced by both experience and situational factors). Or, put differently, information may be available for action in other ways than by way of working memory.

7. Conclusion

As presently illustrated via examples regarding access consciousness, phenomenal consciousness and closely related processes such as working memory, we believe that important insights into both the NCC and the relationship between some types of consciousness can be gained by addressing such issues within the context of a comprehensive and integrative neurocognitive model. Presently, we offer the REF framework as such a model.

We believe that the REF framework already contributes significant insights to these issues. But the importance of the models is also to inspire new empirical, computational and theoretical approaches within this and other fields—approaches that are likely in turn to contribute to the further development of the REF framework.

Data accessibility

This article has no additional data.

Authors' contributions

Both authors contributed significantly to both the described theoretical developments and the writing of the present paper.

Competing interests

We declare we have no competing interests.

Funding

J.M. was supported by a grant from the Danish Council for Independent Research and by a Programme of Excellence grant from the University of Copenhagen.