Autonomous systems face a fundamental challenge of needing to understand where they are positioned as they move through the world. Towards this end, roboticists have extensively investigated solutions to the problem of simultaneous localization and mapping (SLAM), whereby systems must infer both a map of their surroundings and their relative locations as they navigate through space (Cadena et al., 2016). Considering that these same challenges face any freely moving cybernetic system, natural selection has similarly exerted extensive teleonomical (i.e., illusory purposefulness) optimization in this direction (Dennett, 2017; Safron, 2019b), so generating mechanisms for enabling wayfinding and situating organisms within environments where they engage in multiple kinds of adaptive foraging. Perhaps the most sophisticated of all biological SLAM mechanisms is the hippocampal-entorhinal system (H/E-S), whereby vertebrates become capable of both remembering where they have been, inferring where they are, and shaping where they are likely to go next.

Here, we argue that the development of the H/E-S represented a major transition in evolution, so enabling the emergence of teleology (i.e., actual goal-directedness) of various forms (Safron, 2021b), ranging from governance by expected action-outcome associations to explicitly represented and reflexively modellable causal sequences involving extended self-processes. We focus on the implications of SLAM capacities via the H/E-S, and of evidence that this functionality may have been repurposed for intelligent behavior and cognition in seemingly non-spatial domains. We propose that all cognition and goal-oriented behavior (broadly construed to include mental actions) is based on navigation through spatialized (re-)representations, ranging from modeling abstract task-structures to temporal sequences, and perhaps even sophisticated motor control via SLAM with respect to body maps. Indeed, we would go as far as to suggest that the ubiquity of implicit and explicit spatial metaphors in language strongly points to a perspective in which cognition is centered on the localization and mapping of phenomena within both concrete and abstract feature spaces (Lakoff and Johnson, 1999; Bergen, 2012; Tversky, 2019).

In these ways, we believe Generalized Simultaneous Localization and Mapping (G-SLAM) may provide enactive groundings for cognitive science within the principles of ecological rationality (Todd and Gigerenzer, 2012). That is, we adopt a perspective in which cognition is traced back to its ultimate origins, wherein rationality is understood in terms of adaptations for shaping animal behavior in ways that further evolutionary fitness. Such ecological and ethological connections further provide bridges to optimal foraging theory and (generalized) search processes as ways of understanding cognition as a kind of covert behavior (Hills et al., 2013). While somewhat similar models of intelligence have been proposed (Hawkins, 2021), we suggest these other views may be somewhat misleading in neglecting to account for the central role of the H/E-S for realizing G-SLAM. In addition to providing an accurate viewpoint that grounds cognition in its cybernetic function as shaped over the course of evolution and development, G-SLAM will further allow rich cross-fertilization of insights between cognitive science and artificial intelligence. Given the particular functionalities enabled by the H/E-S, we propose this reverse-engineering project ought to be the central focus of cognitive science and machine learning, potentially constituting the most viable path forward towards realizing AI with advanced capacities for reasoning and planning (Bengio, 2017).

A thorough discussion of these issues is beyond the scope of a single manuscript. However, below we attempt to provide an overview of why we believe the G-SLAM perspective may provide a unification framework for cognitive science. First (in Section “LatentSLAM, a bio-inspired SLAM algorithm”), we review our work on biologically-inspired SLAM architectures for robotics. Then, we consider features of the H/E-S, including its functionality for localization and mapping in both physical and abstract domains. Finally, we discuss correspondences between features of SLAM and core aspects of cognitive functioning. We hope to explain how common principles may apply not only to the fundamental task of finding one’s way to desired locations in physical space, but for thought as navigation through abstract spaces. While much of what follows will necessarily be under-detailed and speculative, in subsequent publications, we (and hopefully others) will explore these issues in greater detail as we attempt to explain fundamental principles in neuroscience and artificial intelligence, while simultaneously seeking synergistic understanding by establishing conceptual mappings between these domains (Hassabis et al., 2017).

In the following section, we provide a high-level overview of LatentSLAM, which is also treated in greater detail in (Çatal et al., 2021a, b). While we believe many of these technical details may be relevant for explaining fundamental aspects of high-level cognition, a more qualitative understanding of this content should be sufficient for considering the conceptual mappings we (begin to) explore in this manuscript (Table 1). Section “The Hippocampal/Entorhinal System (H/E-S)” then summarizes current views on the H/E-S and its functioning in relation to spatial modeling and cognition more generally. Finally, Section “G(eneralized-)SLAM as core cognitive process” draws parallels between understanding in machines (using LatentSLAM) and humans (considering the H/E-S) and propose G-SLAM as a unification framework for cognitive science and artificial intelligence.

We realize that this may be a challenging manuscript for many readers, with some portions focused on describing a robotics perspective, and other portions focused on cognitive/systems neuroscience. Indeed, this article emerged from an ongoing collaboration between roboticists and a cognitive/systems neuroscientist, which has been both rewarding and challenging in ways that demonstrate why this kind of interdisciplinary work is both desirable and difficult. One of our primary goals for this manuscript is to provide a rough-but-useful conceptual scaffolding (i.e., an initial partial map) for those who would attempt such cross-domain research. In this way, interested readers ought not be overly concerned if some of the content is found to be excessively technical relative to their particular background. However, we believe readers who follow through with exploring these suggested mappings (which we only begin to characterize) may be richly rewarded for those efforts.

In brief, G-SLAM can be summarized as follows:

If the H/E-S is the kind of gateway to high-level cognition that it is increasingly suggested to be (Evans and Burgess, 2020; George et al., 2021; McNamee et al., 2021), and if it can be well-modeled as having been selected for SLAM functionalities that were later repurposed, then we believe the difficulty of exploring the following material will more than repay the effort of attempting to make the journey. We also ask readers to note places where spatial language can be found, only some of which was intentional. Indeed, we take such linguistic spatializations as supporting evidence for the G-SLAM perspective, which perhaps may be overlooked by virtue of its very ubiquity (cf. fish not noticing water). This is not to say that all spatial cognition points to a SLAM perspective. Yet we believe such spatial mappings are notable in affording opportunities for localization and mapping with respect to such domains. We leave it up to the discernment of our readers to assess how far one can go with following such paths through conceptual spaces, which may not only provide new perspectives on familiar territories on minds, but may even make inroads into discovering how we may follow similar paths to the destination of creating artificial systems with capacities that were formerly considered to be uniquely human.

Simultaneous localization and mapping (SLAM) has been a long standing challenge in the robotics community (Cadena et al., 2016). For autonomous functioning, a robot must try to map its environment whilst trying to localize itself in the map it is simultaneously constructing (i.e., SLAM). This setup creates a kind of “chicken and egg” problem in that a well-developed map is required for precise localization, but accurate location estimation is also required for knowing how to develop the map by which locality is estimated. This challenge is rendered even more difficult in that not only must the system deal with the seemingly ill-posed problem just described, but the inherent ambiguity of the environment is made even more difficult by sources of uncertainty from sensors and actuators. A fundamental challenge (and opportunity) with localizing and mapping is the detection of loop-closures: i.e., knowing when the robot re-encounters a location it has already visited. The challenge is due to the circular inference problems just described, and the opportunity is due to the particularly valuable occasion for updating afforded by the system having a reliable reference point in space. Such loop-closures have a further functional significance in allowing experiences to be bound together into a unified representational system where updates can be propagated in a mutually-constrained wholistic fashion, so providing a basis for the rapid and flexible construction and refinement of knowledge structures in the form of cognitive schemas that have both graph-like and map-like properties. With further experience, these schemata can then be transferred to the neocortex in the form of more stable adaptive action and thought tendencies, so forming a powerful hybrid architecture for instantiating robust causal world models (Hafner et al., 2020; Safron, 2021b).

SLAM has traditionally been tackled by Bayesian integration of sensor information within a metric map, typically expressed in terms of absolute distances and angles. In previous work, this amounted to keeping track of distances between the robot and various landmarks in the environment. Distance measurements were typically combined through Bayesian filtering, a principled way of combining heterogenous information sources through Bayesian inference. Modern successful metric SLAM solutions, however, combine lidar scans with the robots internal odometry estimate through Kalman filtering (Kalman and Bucy, 1961) into 2D or 3D occupancy grid maps (Mur-Artal et al., 2015; Hess et al., 2016). These occupancy maps (Figures 1B,C) keep track of locations of objects in the environment by rasterizing space and then marking certain grid locations as inaccessible—due to being occupied with physical obstructions—so creating a map that resembles what an architect would create to diagram a room (Figure 1A).

Variations on this scheme are popular and differ wildly, either substituting the integration algorithm or the type of metric map. A metric map is akin to a Cartesian grid with regular spacings. However, such spatial maps do not speak to the object identities within the space of interest, nor the particular relations between those objects. Thus, one of the downsides of using metric maps is that by extension all robotic reasoning must also happen on a metric level, any semantic information (i.e., the meaning of a certain cluster of grid-cell activations) needs to be added in later. Further, such metric spaces represent an instance of deviating from natural designs, as hippocampal/entorhinal system (H/E-S) mappings are not independent of the objects contained within these spaces, but instead induce distortions (e.g., expansions and compressions) of spatial relations, which are also modulated as a function of the salience of these entities for the organism/agent (Bellmund et al., 2019; Boccara et al., 2019; Butler et al., 2019).

Popular approaches for such spatiotemporal modeling use particle filters or extended Kalman filters as Bayesian integration methods (Thrun et al., 2005). Kalman filters are notable in that they allow for estimation based on a precision-weighted combination of probabilistic data sources, so allowing for synergistic power in inference and updating, which is also theoretically optimal in making use of all available data (weighted by relative certainty). As will be discussed in greater detail below, such integration may be implemented in the H/E-S via convergent activation in regions supporting high degrees of recurrent processing, such as the CA3 subfield of the hippocampus. However, not only does the H/E-S promote integrative estimation, but also pattern separation/differentiation via other subregions such as CA1, so allowing for attractors to take the form of sparsely-connected graphs—cf. hybrid continuous/discrete architectures based on Forney factor graphs and agent-designs based on independently controllable factors (Friston et al., 2017b; Thomas et al., 2017, 2018). Below we will also describe how such graph-like representations not only help to solve problems in navigating through physical spaces, but may also form a basis for the kinds of high-level cognition sought after in the domain of neurosymbolic AI (Bengio, 2017).

We do not internally represent the world in a metric map. For instance, none of our senses can naturally give us an accurate distance measurement. Neither are we very effective in following a metric description of a path. Hence, it makes more sense for minds like ours (and potentially for artificial agents) to represent a map intuitively as a graph-like structure (Figure 1D), where subsequent graph-nodes could represent subsequent high-level parts of the environment e.g., a node could represent a part of the environment containing a door at a certain rough location. Map traversal then becomes equivalent to the potentially more intuitive problem of graph-traversal or navigating between meaningful landmarks. Trajectories can then be expressed in terms of consecutive semantically meaningful directions. For example, the metrical path “move forwards 2 meters, turn 90 degrees clockwise and continue for 2 meters” could become “after going through the door go right towards the table.” (Note: in vertebrate nervous systems, such forms of navigation could either be based on H/E-S graphs/maps, or occur via canalized striatocortical loops implicitly mapping states to actions, possibly with functional synergy, and also enhanced robustness (and thereby learnability) via degeneracy/redundancy.)

In LatentSLAM (Çatal et al., 2021a), we proposed a bio-inspired SLAM algorithm which tries to mimic this kind of intuitive mapping. With this architecture, we built topological, graph-based maps on top of a predictive model of the world, so allowing for separation of the low-level metric actions of the robot and high-level salient paths. Instead of using raw sensory data—or fixed features thereof (Milford et al., 2004)—directly as node representations, LatentSLAM learns compact state representations conditioned on the robot’s actions, which are then used as nodes. This latent representation gives rise to a probabilistic belief space that allows for Bayesian reasoning over environmental states. Graph nodes are formed from trajectories on manifolds formed by belief distributions. That is, rather than utilizing static maps, our agents navigate through space by moving between landmarks based on expectations of which state transitions are likely to be associated with those kinds of percepts. As an underlying foundation, LatentSLAM adopts the Free Energy Principle and Active Inference (FEP-AI) framework to unify perception (i.e., localization), learning (i.e., map building) and action (i.e., navigation) as a consequence of the agent optimizing one sole objective: minimizing its (expected) free energy (Friston, 2010; Friston et al., 2017a). As will be described in greater detail below, we believe this is an apt description of thinking as the unfolding of a stream of consciousness, with a variety of somatic states being generated in various combinations as the agent perceives and imagines itself moving through space and time.

The hippocampal/entorhinal system (H/E-S) represents a major transition in evolution (Gray and McNaughton, 2003; Striedter, 2004), with homologs between avian and mammalian species suggesting its functionality becoming established at least 300 million years ago (Suryanarayana et al., 2020), with some of its origins potentially traceable to over 500 million years in the past with the Cambrian explosion (Feinberg and Mallatt, 2013), and potentially even earlier. It may be no overstatement to suggest that the H/E-S represents the core of autonomy and cognition in the vertebrate nervous system, with similar organizational principles enabling the potentially surprising degrees of intelligence exhibited by insects (Ai et al., 2019; Honkanen et al., 2019).

While their precise functional roles continue to be debated, the discovery of hippocampal place cells and entorhinal grid cells was a major advance in our understanding of how space is represented in the brain (O’Keefe and Nadel, 1978; Hafting et al., 2005). Similarly important was the discovery of head direction cells in rats, which were found to activate according to moment-to-moment changes in head direction (Sharp PE, 2001). Place cells have been modeled as representing a “predictive map” based on “successor representations” of likely state transitions for the organism (Stachenfeld et al., 2017), and grid cells have been understood as linking these graphs (or Markov chains) to particular events happening within a flexible (multi-level) metric tiling of space, so allowing for estimates of locations via path integration over trajectories. While we need not resolve the precise correspondences between these cell types here, there are intriguing developmental observations of place cells acquiring more mature functioning prior to grid cells, both of which likely depend on head-direction cells for their emergence (Canto et al., 2019; Mulders et al., 2021). In other contexts, place-specific cells have been found to index temporal sequence information, potentially functioning as “time cells” (Pastalkova et al., 2008), so providing a further means by which the H/E-S may provide foundations for coherent sense-making and adaptive behavior through the spatiotemporal organization of organismic information (Eichenbaum, 2014; Umbach et al., 2020).

In addition to place, time and grid cells, a variety of additional specialized cell types have been observed in the H/E-S. While it was previously assumed that these features represent innate inductive biases (Zador, 2019), increasing evidence suggests these specialized cell types may arise from experience-dependent plasticity, including models with similar architectural principles to the ones described here. In recent work from DeepMind (Uria et al., 2020), a recurrent system was used to predict sequences of visual inputs from (the latent space) of variational autoencoders. A natural mapping from egocentric information to an allocentric spatial reference frame was observed, including the induction of specialized units with response properties similar to head direction, place, band, landmark, boundary vector, and egocentric boundary cells. Similar results have been obtained with the Tolman-Eichenbaum machine (Whittington et al., 2020), including demonstrations of reliable cell remapping, so enabling transfer learning across episodes with the potential for the creative (re-)combination of ideas and inferential synergy. Other intriguing work on the emergence of specialized H/E-S functions through experience comes from work on “clone-structured cognitive graphs”, where various aspects of spatial maps are parsimoniously formed as efficient (and explanatory) representations of likely state transitions through the duplication and pruning of nodes in a dynamically-evolving sequence memory (George et al., 2021). While this evidence suggests a potentially substantial amount of experience-dependence in the emergence of the “zoo” of specialized neurons for spatiotemporal navigation, the development of these features still involve clear innate inductive biases (Zador, 2019). Specifically, specialized pathways ensure that the H/E-S receives neck-stretch-receptor information from the mamillary bodies and yaw/pitch/roll information from the vestibular apparatus (Papez, 1937; Wijesinghe et al., 2015), so providing bases for sensor-orientation with respect to head-direction and thereby the foundations of egocentric perspective.

As described above, the H/E-S and its functional relationships with the neocortex may be understood as implementing a kind of Kalman variational autoencoder (Fraccaro et al., 2017). In this capacity, the H/E-S may provide inspiration for developing advanced SLAM architectures. In its dual role as both memory and control system, the H/E-S has been further optimized for facilitating comparisons between largescale patterns (e.g., organismic states), which in machine learning terms may be understood as implementing something akin to energy-based contrastive learning (Marblestone et al., 2016; Richards et al., 2019; Mazzaglia et al., 2022). In this capacity, the H/E-S may provide inspiration for developing architectures capable of engaging in self-supervised learning, counterfactual modeling, and further enabling high-level reasoning abilities including analogical structure mapping (Gentner, 2010; Safron, 2019a), causal inference (Pearl and Mackenzie, 2018), and imaginative planning (Kaplan and Friston, 2018; Safron, 2021b).

As described above, and elsewhere (Safron, 2020b, c, 2021a, b), LatentSLAM’s dual-tier architecture provides an abstract cybernetic interpretation of the H/E-S as the highest (or most integrative) level of heterarchical control for embodied-embedded organisms as they move through physical and imagined worlds in the pursuit of valued goals, so providing a computational/functional account of agency in biological (and perhaps artificial) systems. Further, this hierarchical architecture provides a basis for meta-learning in which slower and more encompassing “outer loop” processes aggregate information over faster “inner loop” processes, so affording the much-desired goal of realizing synergistic inference and generalization of knowledge across experiences (or lessons in curriculums for lifelong learning). Even more, the upper levels of this kind of hybrid architecture may provide a basis for explicit symbolic reasoning (via abstract experience graphs) in addition to enactive couplings with the world (via adaptive control of poses/views), both of which are likely required for achieving the goal of robust autonomous functioning for artificial systems.

While attempting to navigate towards such destinations may seem excessively ambitious, we would note that work on the extended H/E-S was part of what inspired the formation of some of the world’s leading AI companies such as DeepMind, and continues to be a central part of their research programs (Hassabis and Maguire, 2009; Hassabis et al., 2017; Koster et al., 2018; McNamee et al., 2021). Indeed, it is increasingly being recognized that the spatiotemporal modeling properties of the H/E-S may constitute an invaluable integrative framework for understanding high-level cognition (Whittington et al., 2022). However, we believe a G-SLAM framing might be particularly notable in connecting to the context under which these systems were first selected/shaped by evolution (and development), as well as one of the primary functionalities of the H/E-S that continues throughout the lifespan of organisms. That is, our abilities to navigate both physical and conceptual worlds represent an ongoing challenge for as long as we live. We further suggest the connection between the practical necessities involved in engineering physical systems may provide a particularly valuable source of empirical traction for attempting to specify the roles of particular features of the H/E-S, in that we can draw upon the rich data generated as robots attempt to navigate through the world.

Further, by also drawing upon biological details in designing AI-architectures, we may find ourselves with access to invaluable inductive biases which might be otherwise overlooked. Two examples that come to mind include recent proposals by Bengio and LeCun with respect to “GFlowNets” and “Joint Embedding Predictive Architectures” (Bengio et al., 2022; LeCun, 2022). We believe these efforts in creating autonomous and generally intelligent systems may benefit by incorporating principles of G-SLAM, such as the creation of systems capable of handling loop trajectories as potentially enabling greater open-ended and life-long learning, or in looking towards hybrid systems similar to LatentSLAM as potentially allowing for explicit representations and symbolic processing. While it has often been said that the goal of AI is to create the “cognitive equivalent of an airplane wing,” we would suggest that the magnitude of the challenge may be far greater (more akin to building a fully functioning plane or space ship), and the problems of navigating through under-constrained architectural (and learning curricula) design-spaces may be unsurmountable without biological inspiration/grounding.

While LatentSLAM continues to be refined, we believe these kinds of architectures provide a general framework for understanding core elements of minds and brains. Indeed, to localize something within a spatialized reference frame—which itself is impacted by the entities it maps/graphs— may be what it means to “understand” and “explain” something (Lakoff and Johnson, 1999), and possibly even to experience anything at all (Safron, 2020a, 2021a, b). That is to understand is to be able to adopt a stance (or pose) from which elements and their inter-relations may be mapped (or localized), as if projected onto a plane whereby they are made visible for inspection (or navigation). We believe these etymological considerations on the nature of knowledge may be more than “mere” metaphors but could point to the fundamentally embodied nature of minds.

We not only suggest that all thought may be understood as navigating between representations that are being localized and mapped (or graphed) within an organizing conceptual domain, but all communication may be understood as the transmission of such structures (as trajectories) between minds (Zurn and Bassett, 2020). While the simultaneity of generalized localization and mapping in cognition may not be obvious upon introspection, this is more clearly the case when considering unfamiliar concepts. For such novel domains, relationships between concepts and broader organizing schemas involves the same kind of challenges of circular inference as found in SLAM. That is, when we are first attempting to understand a conceptual domain, we do not know how to effectively connect the entities whose shared features and relations motivate the construction of organizing schemas. However, without such higher-order abstractions and the predictive (or compressive) capacities they provide, it is unclear which features of and relations between entities are relevant for shared structure learning.

Heuristic algorithms may be invaluable in the bootstrapping process, such as the kind of clustering involved in the hierarchical Dirichlet process (Griffiths T. et al., 2019), models of category formation via analogical alignment (Kuehne et al., 2000), or concept derivation as abstraction over episodes (Mack et al., 2018). We agree that such accounts may speak to fundamental mental processes, but we also suggest that rather than static feature maps, such nonparametric (Bayesian) structure learning may apply to paths through mapped/graphed domains. This is part of why we emphasized our use of the Fisher distance measure above, as an information metric that naturally applies to trajectories may potentially provide the most valid (and potentially predictive) means of assessing similarity/dissimilarity between entities in feature spaces. Indeed, one of the most notable aspects of thinking is its sequential operation and sensitivity to path dependencies. While abstract conceptualization does allow for a good deal of cognitive flexibility, cognition is still largely defined by deriving knowledge via particular “chains of reasoning,” or “paths” through mental space.

Regardless of the particular routes by which we reach the heights of category learning, the formation of such abstract representations constitute what may be the most powerful aspect of our intelligence in terms of generalizable knowledge that can robustly transfer across particular episodes (Marcus, 2020). Such abstract categories further allow for the kinds of structured representations whose importance was emphasized in decades of work in (non-radically-enactive) cognitive science and “good old fashioned AI.” The significance of such knowledge structures may prove even greater in light of the advent of graph networks within the context of geometric deep learning (Battaglia et al., 2018)—and symbolic regression as potentially representing a further degree of abstraction (Cranmer et al., 2020). Such graphical representations are of increasing interest because of both their interpretability as well as their extraordinary efficiency for modeling physical systems. With our models of node duplication and graph-relaxation, LatentSLAM provides a biologically plausible and computationally-tractable account of how such cognitive schemas may be formed and modified through experience. This is notable in that finding principled means of creating and modifying particular structures for graph neural networks (GNNs) remains an ongoing challenge. But if such challenges can be surmounted, then we may achieve the promise of neurosymbolic AI in combining the power of connectionism with reasoning over explicitly represented (and related) symbols (Garcez and Lamb, 2020; Greff et al., 2020). More specifically, we believe that the ability of the H/E-S to create navigable spaces populated by high-level attracting states may also provide a basis for creating “ad hoc” (Barsalou, 1983) GNN structures for different purposes.

While the relationships between place cells in the H/E-S (or nodes in LatentSLAM) can be understood as a kind of GNN, we believe it would be more accurate to characterize these models as graph nets, in that they represent relations—or semantic pointers (Blouw et al., 2016)—for hierarchically lower graphs. While these details have yet to be incorporated into LatentSLAM, it has been suggested that heteromodal association cortices may constitute a shared latent space across (autoencoding) cortical hierarchies with quasi-topographic characteristics akin to those found with GNNs (Safron, 2020b, 2021a, b). While the H/E-S has significant interactions with the entire cortical heterarchy, connectivity is most substantial for deeper (or hierarchically higher) portions of cortex, consistent with its potential role as a kind of graph network. The degree to which these machine learning analogies may apply to brain functioning is yet to be determined, but they nonetheless represent a promising direction for creating artificial systems that recapitulate the properties of natural intelligences (Greff et al., 2020).

Intriguingly, the work in which brains were proposed to entail GNN-type computation was developed independently of LatentSLAM. However, similarly to how LatentSLAMs only uses views and proprioceptive poses for specifying particular experiences to be mapped (Figure 2), this other work proposed that sufficient bases for agentic world modeling may involve conjoined visuospatial and somatospatial modalities, potentially (but not necessarily) understood as respective grid and mesh-pose GNNs. In the model of episodic memory and imagination described above (Figure 6), H/E-S trajectories are used to orchestrate state transitions between these experiences as the “stream of consciousness” (James, 1890). While many aspects of cognition are unconscious, “thinking” and “reasoning” are usually considered to involve sequentially generated conscious operations. Notably, the formal conceptualization of computation may have been largely inspired by Turing introspecting his own consciousness in the process of doing mathematics (Dehaene, 2014; Graves et al., 2014). Given that it is unclear that we can be conscious of anything that lacks grounding in somatic modalities and their abilities to change (and be controlled) through time, then all thinking/reasoning may potentially be understood as involving the kinds of action selection and modeling described by LatentSLAM.

Fully describing the potential correspondences between SLAM and high-level cognition is beyond the scope of a single publication (Table 1), but before concluding we will briefly comment on the importance of loop-closures and thresholds for graph-relaxation and node duplication. In brief, we may understand a (generalized) loop-closure event as a primary factor contributing to the feeling of understanding and insight (Gopnik, 1998; Fonken et al., 2020; Oh et al., 2020). After an initial period of relatively ambiguous exploration, the formation of a causal account (or trajectory through a concept space/graph) would allow for a rapid decrease in prediction-error (Joffily and Coricelli, 2013), or increase in compression (Schmidhuber, 2010). While some individuals may be relatively insensitive to these feelings of (potentially sudden) conceptual familiarity (Hou et al., 2013; Ben-Yakov et al., 2014), others may potentially be overly sensitive (e.g., “déjà vu” and other kinds of false positive inferences), with the specific functional tradeoffs involved depending on particular contexts (DeYoung, 2015; Blain et al., 2020; Safron and DeYoung, 2021).

Events in which this kind of cognitive closure is achieved provide special opportunities for updating H/E-S models (or categories) via graph-relaxation and node duplication. A variety of relevant parameters can be identified (Figure 3), whether in terms of thresholds for detecting loop-closures, the extent to which graphs may be relaxed, or the ease with which new nodes are created. However, in a G-SLAM context of trying to model cognition more generally, we may think of loop-closure recognition thresholds as sensitivity to cumulative prediction error increases/decreases, graph-relaxation as changing attractor dynamics within the H/E-S and neocortex on multiple scales, and node duplication as the establishment of new local ensembles of effective connectivity (cf. chained bump attractors)—and potentially (but not necessarily) involving neurogenesis, for which it is notable that the hippocampus is one of the few places where this phenomenon is reliably observed. These core SLAM processes may depend on multiple factors, including neuromodulators such as dopamine and serotonin (Safron, 2020c; Safron and Sheikhbahaee, 2021), as well as on Bayesian priors (or “yesterday’s posteriors”).

With respect to the previously described example of differential tuning thresholds for mapping the structure of aisles (Figure 5), we may potentially have a crucial source of individual differences in cognition. In theory, G-SLAM may be pointing to (or localizing) a cognitive spectrum (and potential basis for differential diatheses) spanning autism and schizophrenia (Byars et al., 2014; Crespi and Dinsdale, 2019). Theoretically, we may even expect to see these kinds of variations in SLAM maps in the drawings of autistic and schizophrenic individuals (Morgan et al., 2019; Philippsen and Nagai, 2020).

Speculatively, not only may the conceptual understanding of that which is being drawn be mapped and navigated by the H/E-S as SLAM system, but the eye movements (Wynn et al., 2020) and hand motions involved in skilled actions such as drawing could themselves be orchestrated according to hippocampal trajectories as a basis for chained equilibrium setpoints (Latash, 2010). Even more speculatively, it could even be the case that further degrees of sophisticated control—as inference (Kaplan and Friston, 2018; Friston et al., 2021)—are bootstrapped by simultaneously localizing and mapping the body itself as a kind of space/graph, so allowing for more rarefied and general SLAM capacities over the course of development. In this view, much of cognitive development would involve initial phases of using the H/E-S to learn intentional control over either overtly or covertly expressed motor patterns, which then become automatized (or amortized) by the thalamic-cerebellar system (Safron, 2021a; Shine, 2021) and dorsal striatal-cortical loops (Mannella et al., 2013), so freeing up the G-SLAM system for further high-level predictive modeling and control. Can the body itself be understood as a mapped spatial domain, or is this just a way of speaking without any useful technical correspondences? How far can we go with using these patterns of linguistic use as hypotheses regarding cognitive processes and underlying neural mechanisms? Could it even be the case that the phenomenology of embodiment involves navigation through and mapping of body maps via these cross-modal interactions, which when disrupted could potentially contribute to altered states of consciousness or potentially clinical conditions such as depersonalization (Safron, 2020c; Ciaunica and Safron, 2022)?

With respect to personhood, beyond its foundational role for autonomous functioning, widespread orchestration of value-canalized trajectories through biophysical phase space by the H/E-S also enables the development (and ongoing functioning) of the spatiotemporally-extended processes required for autonoetic and autobiographical self-consciousness. In addition to constituting major transitions in evolution, the advent of such self-reflective capacities may have been required for the construction of advanced social coordination and a (shared) symbolic order of being. While such rarefied processes may be well-beyond anything we are close to engendering in (abiotic) machines, it may be the case that we are forced to recapitulate these kinds of H/E-S functionalities if we are to successfully arrive at the destination of creating robustly autonomous and general artificial intelligences.

Indeed, G-SLAM parameters may constitute the most important source of variation we can identify both between and within individuals. To venture deep into unknown speculative territory, the H/E-S may be the source of key adaptations contributing to the evolution of cognitive modernity through (potentially proto-schizotypal) flexibly creative cognition and the birth of cumulative culture, which in time came to represent what may be the “secret of our success” as a species and the greatest of all major transitions in (generalized) evolution (Premack, 1983; Gentner, 2010; Hofstadter and Sander, 2013; Henrich, 2017; Safron, 2019b, 2020c; van den Heuvel et al., 2019; Dehaene et al., 2022). While such models extend far beyond domains of knowledge for which we have well-developed maps, we believe such possibilities are worthy of further exploration.

We have searched through broad and diverse terrains in considering the ideas above, covering a lot of ground. To try to come full circle, we have described technical details of a machine learning architecture for autonomous robot navigation, discussed particulars of biological systems for realizing these functionalities in brains, and started to explore how these principles may provide a framework for understanding all high-level cognition in terms of simultaneous localization and mapping in space (broadly construed to include conceptual spaces). We have only begun this journey, but we believe the destination is promising, and we invite others to join us in exploring this framework for understanding the nature of thought. Some might contend that “prediction” or “modeling” are more encompassing and fundamental than a generalized SLAM perspective, and we would not disagree. However, we believe that G-SLAM is unique in allowing for all these perspectives to be combined with the principles of ecological rationality that constituted the primary selective pressures for high-level cognition over the course of evolution and development. We suggest this neuroethological perspective will be invaluable in allowing us to “carve nature at its joints”, in terms of identifying the most important features of functioning for the hippocampal/entorhinal system and its connections to the rest of the brain [and body (and world)]. We further believe that G-SLAM is unique in the extent to which it connects to nature(s) of experience, where we do in fact exist in a spatial world through which we must navigate, and where it is difficult to find aspects of mind not impacted by this fundamental physical situatedness. In light of these sources of potential insight, we believe that G-SLAM represents the way forward for understanding complex minds, and potentially for building them, if we can find sustainable paths into the unexplored territory of the future.

The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author.

AS, OÇ, and TV conceptualized G-SLAM and the main ideas for this manuscript. OÇ and TV conceived and performed the LatentSLAM experiments. AS contributed his knowledge gained from his ongoing study of the hippocampal/entorhinal system literature. All authors contributed to the article and approved the submitted version.

OÇ is funded by a Ph.D. grant of the Flanders Research Foundation (FWO). This research received funding from the “AI Flanders” program of the Flemish government.