We propose a definition of representations derived from a unified theory of long-term memory, encompassing both its explicit and implicit divisions (Squire and Zola Morgan, 1991; Squire, 1992; Sridhar et al., 2023; see Box 2). Evidence for multiple memory systems comes from behavioral dissociations. Lesion studies have demonstrated that certain types of brain damage impair only implicit performance, while others impair only explicit performance (Gabrieli et al., 1995).

In addition to its integrative qualities, multiple memory theory offers a favorable perspective for reconciling different views in neuroscience and philosophy of mind, particularly concerning concepts such as consciousness, intentionality, and the “hard problem of consciousness”.

Memory representations, as introduced schematically by their content and state, refer to the “knowledge elements” of both the outer and the inner world, as well as to anticipations and the execution of overt actions (Kok, 2020). They are linked to distributed networks within the brain, connecting local modules via long-range reciprocal pathways. These networks appear primarily in the anterior and posterior cortices, connecting to subcortical areas. Together, they constitute a significant portion of our functional cerebral space, encompassing various information-processing functions that utilize memory representations. The content of memories can be expressed as conscious awareness or not, depending on the involved type of memory and the corresponding activating neural ensembles.

This chapter provides a detailed description of the various structures and pathways underlying both explicit and implicit memory. It also provides arguments for how these two distinct forms of memory can interact. We here present a summary of the factors justifying their distinction and interaction.

Conscious access or not. Several factors may explain why explicit and implicit memories give rise to conscious access or do not. Why is implicit memory triggered without conscious awareness? Perhaps the most obvious explanation is that various forms of implicit memory share the commonality that they do not depend on the medial temporal cortex. This criterion has also proven to be a powerful tool for distinguishing performance domains where amnesic patients succeed and fail. A more substantive argument is that behaviors dependent on goal-directed learning of action-outcome contingencies and reinforcement-based learning would inherently depend on the biologically efficient and fast striatal-cortical pathways. Memory representations controlled by these pathways would likely remain “dispositional” (Damasio, 1994, 1999), that is, not directly dependent on conscious experience or internal models of the environment or action repertoires. In addition, intended actions realized by these pathways usually have the character of routine skills triggered without the intervention of the retrieval mechanism of the prefrontal cortex (Goschke and Kuhl, 1996). Research has further shown that in dual-task performance, motor skills do not interfere with tasks requiring conscious control (Ewolds et al., 2017), supporting the view that they do not rely on the limited capacity of consciously controlled brain structures. The same argument holds for the structures and routes of reinforcement-based learning, which regulate affective and “instinctive” forms of behavior via the dopamine pathways. In short, they utilize biological routes shaped by evolution to serve the immediate goals of survival.

Integrating implicit and explicit memories. Implicit-explicit memory interactions are often integrated as they occur in our daily performances. Unlike controlled processes, implicit and automatic functions do not require limited processing resources. This entails that they can be appended to functions without interfering with their control in systems with limited capacity, such as working memory. For example, driving a car in normal traffic conditions (typically a sensorimotor skill) can be time-shared with a conversation with your passenger (typically depending on explicit memory). Another spin-off of the fusion of implicit and explicit functions is that implicit memories, such as skill learning and priming, become available for conscious exploration (Dayan and Cohen, 2011). This will be especially true in tasks that involve elements such as providing feedback on performance, detecting errors, and identifying mismatches between novel stimuli and representations in long-term memory.

The Theory of Global Workspace (TGW) has attracted significant interest among researchers in neuroscience and cognitive psychology. In short, this theory states that top-down attention is a prerequisite for global workspace neurons to become conscious (Baars, 1988, 2003; Dehaene et al., 1998; Dehaene and Naccache, 2001). The global workspace links distant areas of the brain, including the prefrontal cortex, which inhibits surrounding neurons outside the workspace. Our model shares some similarities with TGW. For example, implicit sensorimotor skills may gain conscious access when they are fused with explicit memory within a multilevel representational network. On the other hand, in our model, conscious access and conscious experience occur within the broader context of representations retrieved from explicit long-term memory via routes 2b and 4. Importantly, our model also incorporates a principle lacking in TGW: intentionality, the capacity of elements in memory representation to interact with the external world (see section 6.2 for a further discussion).

The multimodal multilevel network model proposed by Pennartz (2022) suggests that conscious representations emerge from low-level neuronal configurations at the top of a hierarchical network. His model, however, does not seem to differentiate between explicit and implicit forms of learning and memory and would (in our view) apply only to the multilevel network of the hippocampal-cortical loop, creating full, conscious representations of objects integrated in space. Our model, on the other hand, allows for the differentiation of explicit and implicit forms of learning, both of which utilize the architecture of multimodal hierarchical structures. For example, an activated memory representation of a complex skill, such as catching a ball from the air by a professional baseball player, would not necessarily be accompanied by conscious experience, even if it were at the top of a hierarchy recruiting low-level neural configurations in the basal ganglia. The only route for implicit representations to reach conscious access is through their “fusions” with elements of explicit memory in an overarching representational network as sketched in Figure 3.

In the following paragraphs, we will elaborate on which neural principles underlie the computational capacity of memory presentations to interpret or “model” aspects of the outside world (as briefly introduced in section 3). We first sketch how representations may emerge in multilevel hierarchical networks—from local networks with precise coding to more global networks with increasingly coarser coding forms—and how these representations give rise to conscious experience in the association cortices. Prediction coding, then, is assumed to provide the representations at the top of the hierarchy with the mechanism for interpreting or modeling properties of the outside world. Our second focus is on another keystone of explicit memory, namely, how the content of episodic memories acquires a subjective-affective quality through interaction with structures in the limbic system.

Once we appreciate all the non-mysterious ways in which the brain can create benign “user illusions”,¹² we can begin to imagine how the brain creates consciousness (Dennett, 2005).

In its simplest definition, consciousness is not problematic at all. It means being aware, or “knowing,” of objects or states, either internal to oneself or in one's external environment (James, 1981). Consciousness can manifest itself in three distinct ways: as a state of alertness or wakefulness, as a process of orienting to significant events, or as an internally controlled process, such as thinking or deliberately focusing on a specific attribute when stimulated or faced with a demanding task. These various manifestations are a product of a hierarchical thalamic-cortical system with different levels controlling different manifestations (Kok, 2020; Whyte et al., 2024).

Philosophy of mind relies strongly on the introspection of phenomenal experiences, which is something within one's mind that becomes “introspectively conscious” (Chalmers, 1995, 1996, 2005; Dehaene, 2019). These insights and definitions align with aspects of the “explanatory gap” (Levine, 1983, 2009), a concept discussed earlier in this review. In the philosophy of mind, the phenomenal contents or the so-called qualia of consciousness are inherently subjective attributes of experience. By most accounts, they are considered non-representational in the physical sense (see Korf, 2014, for a review).

Challenging these views, Dennett (2011) argued that any neurobiological theory based on an experience/function division cannot be empirically confirmed or falsified and is thus outside the scope of “science”. Can we then develop falsifiable theories of consciousness grounded in the brain's properties? The answer is: yes. One example is neurocognitive theories of selective attention, as elaborated by LaBerge (1995). In addition, various studies with a predominant neuroscience orientation have recently presented an integrative, multiscale view on neural theories of consciousness (Flanagan, 1992; Lamme, 2010; Ellia and Chis-Ciure, 2022; Seth and Baine, 2022; Storm et al., 2024; Whyte et al., 2024; Mudrik et al., 2025, for reviews).

A theory that has attracted considerable interest in philosophy of mind and neuroscience is based on the global workspace, introduced earlier (Baars, 1988; Dehaene and Naccache, 2001; Dehaene et al., 1998). This theory describes the processing requirements that give rise to conscious experience as a dynamic interaction between local processors and a global workspace. Lamme and colleagues offer a different theoretical perspective on consciousness, based on research into primate vision (Lamme, 2010; Lamme and Roelfsema, 2000). They stated that recurrent or reentrant processing (RP) is essential and perhaps sufficient for visual conscious experience. In their extensive review of consciousness, Storm and associates commented on RP by suggesting that RP alone is insufficient for consciousness (Storm et al., 2024). An alternative recent theory, already briefly discussed in our model of memory representations, is neuroresentationalism (NREP), formulated by Pennartz (2015, 2022). NREP proposes that consciousness, in the sense of conscious experiences, may emerge on top of a hierarchy of conceptually distinct levels, with single neurons at its base and progressing upward through assembly-level, unimodal, and multimodal network representations. It was further proposed that representation is “a key brain function in realizing a plethora of sensory, motor, and cognitive processes, some carried out consciously and others not” (Storm et al., 2024, echoing other recent definitions of consciousness from a neuroscience perspective).

Our approach was not to introduce a comprehensive new theory of consciousness. Instead, we followed the tradition in the memory literature to re-frame consciousness as a “mode” of implicit vs. explicit processing, associated with a variety of information processing functions such as perception, memory, motor functions, and emotions (Kok, 2020; Cohen and Dennett, 2011; Tulving, 1985; Squire and Zola Morgan, 1991; Eichenbaum and Cohen, 2001; Sridhar et al., 2023). Within this framework, we defined conscious experiences as the product of two mechanisms: conscious access, which occurs through the process of encoding and consolidation of new information, and conscious experience, which results from the retrieval of encoded information in explicit long-term memory. In the same framework, retrieval incorporates any condition that would ignite slumbering representations in long-term memory, such as significant external events, task performance, but also spontaneous thoughts and mind wandering (Christoff et al., 2016).

In conclusion, the neurobiological processes that underlie conscious processes and subjective states are efficient but imperceptible, in contrast to the transparency of our subjective experiences. The “price we pay” for this biological efficiency is thus that we cannot understand the brain's inner workings through introspection (Musacchio, 2005). Still, the physical embodiment of information in the brain, reflected in patterns of action potentials, is intimately tied to information messages. The meaning of these messages could lie at the very basis of qualia.

In the tradition of dualism, philosophers of the mind have long struggled with the problem of intentionality, a concept introduced by Franz Brentano in the late nineteenth century (Textor, 2017). Intentionality is the power of minds and mental states to be about things, in short, that a subjective mental state includes something as an object within itself (Dennett, 1996; Mendelovici, 2018; Mendelovici and Bourget, 2023). Furthermore, stating that an individual's mental states have intentionality amounts to saying that they are mental representations or have contents. In philosophy, there is a tight connection between concepts such as intentionality and representation, which are discussed in more abstract terms within the overarching Representational Theory of Mind (Tye, 2009; Lycan, 1996).

The position taken in this article is that intentionality is a fundamental element of consciousness, residing in neural assemblies that incorporate representations. Representations harness the capacity of neural assemblies to predict or interpret events in the external world, utilizing the built-in modeling capacity of memory representations to transform physical reality into subjective experience (see also Pennartz, 2015, 2022; Giotakos, 2023). They underlie perceptual discrimination, memory, awareness, preparation of motor activity, and imaging of familiar visual scenes. Similar views were expressed by Alan Baddeley (2003), who referred to a “visuo-spatial sketchpad”, and by Anton Damasio, who spoke of “dispositional representations”, which are not raw copies of the outside world, but rather the potential of ensembles of neurons in the brain to fire and reconstruct or model the original events (Damasio and Damasio, 1994).

Intentionality also refers to how our thoughts and language are represented in the mind, with semantics concerned with the meaning encoded in language itself (Brandom, 2014). In this context, binding theories suggest that semantic knowledge involves a broadly distributed network of neural representations, particularly involving the inferotemporal and posterior inferior parietal regions. These regions enable increasingly abstract, multimodal representations of perceptual experience, supporting a variety of conceptual functions.

In summary, the theory of memory representations outlined earlier has the potential to incorporate the notion of intentionality, as elaborated in the philosophy of mind.

In philosophy, emergence often means irreducibility. The idea is that although mental properties arise from and supervene on the brain's physical properties, they can never be reduced to them. The concept formed a part of Roger Sperry's account of mind-brain interaction in his theory of emergent interactionism (Chezik, 1990). Neuroscientists have recently proposed that these processes emerge from hierarchical structures in complex systems, in which lower levels combine to form higher levels. New features emerge in the system as more levels are added or local assemblies in the brain become integrated into a much larger network with long-ranging connections (Feinberg and Mallatt, 2019; Dehaene, 2019).

Integration, in this sense, is a somewhat abstract principle that does not touch the richness and subjective character of our conscious experiences. A conscious experience can often be a vivid sensory image accompanied by a feeling, suggesting that your frontal cortex retrieves a slumbering subjective representation from visual memory. In other words, the subjective nature of consciousness depends entirely on the momentary locus of control within our representational systems (Donald, 1991). In episodic memory, consciousness is mostly situation-bound and concrete, with the locus of control at the top of the episodic system. Emergence in this context comes close to the content of phenomenal consciousness, in short, to “qualia”.

Qualia and memory representations. Philosophical discussions on the nature of intentionality have focused more strongly on the connection between intentionality and phenomenal consciousness, emphasizing that “intentionality is none other than phenomenal consciousness, the subjective, felt, or qualitative aspect of mental life” (see Kriegel, 2013a,b). In philosophy of mind, subjective conscious experiences, also defined as “qualia, are a core element of the hard problem of consciousness”, explaining how physical processes in the brain give rise to qualitative feelings like the redness of red. Chalmers (2005) argues that subjective experiences are fundamental to reality and may not be fully explained by purely materialist or reductionist models. Conscious perceptual experiences are, by essence, private and subjective; when I look at the blue sky with clouds, it can trigger a specific feeling that is difficult to describe and share with others (Searle, 2000). Similarly, Dennet (2002) defined qualia as “ineffable, intrinsic, private, directly apprehensible properties of experience” memory.

These definitions are heavily influenced by a traditional, strict version of phenomenology, initially formulated by Edmund Husserl, who posited that a perceived object or event should be described precisely as it is perceived. Husserl used the word “hyletic”, meaning “about matter, the raw, material data of sensory impressions. His view clearly contradicts representationalism, which Husserl fervently rejected (Box 3). According to contemporary versions of representationalism in the philosophy of mind, phenomenal qualia—specifically, the subjective “what it is like” aspect of our memory experiences—are part of its epistemic content (Shim, 2011; Liu, 2024a,b; Perrotta, 2025).

In our model of memory representations, explicit memory provides a framework, grounded in both neuroscience and cognitive psychology, for defining subjective quality (see section 5). This proposal is confirmed by influential studies of episodic elements of explicit memory, corroborating the idea that qualia are reflections of the subjective “fingerprints” of the brain situated in transient neural networks activated in episodic memories (Edelman et al., 2011; Ward and Guevara, 2022). In addition, as previously argued, emotions contribute to creating a highly subjective “affectively colored” content of episodic memory through interactions between the amygdala and the hippocampus.

Taken together, the arguments listed here justify the conclusion that although qualia may subjectively appear to be irreducible qualities of experience, they can also be understood to arise from neurocognitive mechanisms of memory without any form of reduction, and while preserving the richness of our memory representations as described in our theoretical model.

Given these considerations, we further conclude that there is no need to assume a separate ontology for qualia and subjective experiences, as long as they can be conceptualized within the framework of cognitive neuroscience theories (Feinberg and Mallatt, 2019; Fazelpour and Thompson, 2015). Such a theory is not reductionistic but rather materialistic, proposing that memory representations are rooted in neural assemblies that give rise to conscious experiences.

Box 4 summarizes the principal conclusions derived from our theory of memory representations, which we shall here briefly recapitulate. We propose that consciousness can be reframed in terms of access and phenomenal consciousness, referring, respectively, to the encoded and retrieved aspects of representations in explicit memory. In this context, consciousness is defined as the subjective experience of retrieved episodic elements of explicit memory, also referred to as phenomenal consciousness.

Regarding intentionality, this article argues that it is a fundamental element of conscious experience, residing in neural assemblies that incorporate memory representations. Representations harness the capacity of neural assemblies to predict or interpret events in the external world, utilizing the built-in modeling capacity of memory representations to transform physical reality into subjective experience. These views clearly confirm strong links with current definitions of phenomenology, emphasizing that every act of consciousness we perform, every experience we have, is intentional, i.e., essentially “consciousness” (Fahrenfort and Lamme, 2012).

From a similar perspective, emergence does not imply irreducibility. Instead, it can be conceived as a process realized in a bottom-up fashion by successive levels of single neurons forming small local networks, which are combined at subsequent levels into unimodal and multimodal multilevel networks, thereby creating (“emerging”) conscious representations of objects integrated in space (Pennartz, 2022).

According to our model of memory representations, explicit memory also provides a framework for defining subjective quality, as supported by studies of episodic memory in the human brain. These studies corroborate the idea that qualia reflect the subjective “fingerprints” of the brain, located in transient neural networks activated during the retrieval of episodic memories (Box 5).

Subjective experiences are intrinsic to our brain's neural circuits, emerging from interactions between emotion and cognition. In sum, phenomenal experiences (“qualia”) are anchored in episodic memory networks that possess a personal signature and affective flavor. In conclusion, there seems to be no need for qualia as a separate ontological category to produce a state of subjective consciousness.

In his book Darwin's Dangerous Idea, Dennett developed the idea of a Darwinian process, “involving variation, selection and retention, as a generic algorithm that is substrate-neutral and could be applied to many fields of knowledge outside of biology” (Dennett, 1995). Similarly, Bering and Schackelford proposed that several Darwinian selection principles could be grounded in representational systems, in which conscious motives have inserted themselves at the level of the gene, fundamentally changing the nature of hominid evolution (Bering and Schackelford, 2004).

A powerful concept directly relevant to a theory of memory representation is predictive coding, also known as predictive processing. This theory posits that an internal model, formed through learning and stored in long-term memory, serves as a model or template in the brain to compare input with existing representations. Predictive coding could also take the form of more complex predictive abilities, such as planning, anticipating future events, and imagination. Although the theory suggests that these functions rely on highly complex cognitive processes, they could also have “emerged gradually during evolution from simpler predictive and error correction loops in the brains of our earlier evolutionary ancestors” (quote from Pezzulo et al., 2021). A related hypothesis concerns the evolution of language abilities in the human species. These could have evolved from more basic human abilities to recognize and remember sequences, a crucial evolutionary step toward human language and a key trait in the development of human culture and thought (Jon-And et al., 2023).

Indeed, the dense distributed networks of our brain that underlie the functional operational space of our memory system could have evolved in such a way as to optimize adaptations to a world where symbolic communication became equally important as the direct need to survive in a physical sense (Donald, 1991; Buckner and Krienen, 2013). Symbolic reference systems are intrinsic to the brain's neural architecture and gradually became grounded in neural representations of our declarative memory system (Deacon, 1997). A larger brain enabled the human brain to evolve into a unique organ, distinct from those of other mammals. Manifested not only in a larger volume relative to body size, but also in specific adaptations of its connectome (Ardesch et al., 2019). This likely involved a larger volume of association areas than of primary areas, as well as stronger connectivity of long-range reciprocal fibers in the dorsal and ventral routes of the brain. Together, these neural architectures would enable easier access to the areas involved in attentional control and consciousness, as well as the area involved in language processing.