Navigating Things That Are Not Places
How an Ancient Spatial Map Was Borrowed for Abstract Thought
Navigating Things That Are Not Places
How an Ancient Spatial Map Was Borrowed for Abstract Thought
We have built, over three chapters, a particular kind of machine. It is ancient — a derivative of the medial pallium, conserved from fish to human. Its function is prediction — using a stored model of the world to anticipate where the body’s needs can be met. And its mechanism is now concrete: place cells marking locations, grid cells laying down a metric, head-direction and boundary cells supplying compass and edges, all of it grounded in the body’s own motion and able to run offline, forward into journeys not yet taken. It is, in every respect we have examined, a machine for navigating physical space.
And yet you are about to use it to read this sentence, weigh an argument, locate a colleague in the social landscape of your department, and decide whether the conclusion of a syllogism is “close to” or “far from” its premises. Somewhere between the goldfish finding food and the undergraduate reasoning about abstractions, this spatial machine was put to profoundly non-spatial work. This chapter is about how that happened, and — just as important — about what it did not mean. Because there is a tempting misreading waiting here, and getting it wrong would undo the argument of the entire unit.
The misreading to avoid
The tempting misreading goes like this. We have just discovered that the “spatial” map also handles non-spatial things — concepts, relationships, abstract structure. So perhaps it was never really a spatial map at all. Perhaps it was always a general-purpose engine for relations of any kind, and physical space was simply the first and most obvious relational problem it happened to be applied to. On this view, space is not special; it is just one item on a long list, and the system was relational all along.
This is not a straw man. It is a serious and widely held position in the field, and you should understand its appeal. But this unit has spent three chapters assembling exactly the evidence that tells against it, and it is worth seeing why the “relational all along” reading cannot be right — or at least, why it is the weaker reading of the same facts.
If the system were a general relational engine that merely happened to do space first, then there would be nothing special about space. But space is not just another item on the list. It is the item that is built before the animal has any experience at all — the head-direction compass online before the eyes open, driven by the vestibular sense of the body’s own motion, as we saw in the last chapter. It is the item grounded in a literal sensorimotor signal — path integration over real physical movement, the metric assembled from the body’s displacement through the world. And it is the item that is demonstrably present and spatial in function half a billion years ago, in fishes whose hippocampal homolog, when removed, takes away the map. A general relational engine has no reason for one particular kind of relation — physical space — to be the innate one, the vestibular one, the evolutionarily ancient one. The relational-all-along view can describe the adult human data, but it cannot explain the developmental and evolutionary asymmetry. Something made space first, and made it from the body, and made it long ago.
So this book takes the other reading, and it is worth naming precisely, because the name carries the argument.
Exaptation, and its gentler cousin
In 1982, Stephen Jay Gould and Elisabeth Vrba gave a name to one of evolution’s most important moves. A trait evolves under selection for one function, and then — later, in a descendant lineage — it is co-opted for a different function it was not originally selected for. Feathers almost certainly evolved for thermoregulation and only later were recruited for flight. The move has a name because it is everywhere and because it dissolves a persistent confusion: the function a structure serves now need not be the function it was built for. Gould and Vrba called it exaptation.
This is the central claim of this chapter. The hippocampal-entorhinal map evolved, and was selected, for navigating physical space. In lineages with enough cortex to exploit it — most spectacularly our own — that same machinery was exapted: turned to the navigation of spaces that are not physical at all. Conceptual spaces, social spaces, the abstract landscapes of reasoning. The map was not built for these. It was borrowed for them. And like feathers pressed into the service of flight, it carries, into its new work, the unmistakable marks of what it was originally for.
It will help to distinguish exaptation from a gentler relative, because the human story involves both and they are not the same. Exaptation, strictly, is a change of function: the spatial map turned to genuinely non-spatial work, navigating concepts or kinship, where the job is no longer “find your way through the world.” Elaboration is something milder: the same function, extended and enriched without a real change of purpose. When we eventually turn to human episodic memory — the rich, detailed recollection of specific events — we will be looking at elaboration, because the scrub jay remembering what it cached where and when, and you remembering last night’s dinner, are doing recognizably the same prospective job, only scaled up in reach. That is a story for the final chapter. The story of this chapter is exaptation proper: the map doing work whose function has changed, navigating domains that are not places at all.
The map applied to concepts
What does it look like when the spatial map is turned on something abstract? Begin with the most direct evidence, because it is striking enough to settle the basic fact.
Suppose you teach a person an artificial “space” of stimuli — say, cartoon birds whose necks and legs vary continuously in length, so that any particular bird is a point in a two-dimensional space defined by those two features. There is nothing spatial about this in the ordinary sense; “neck length” and “leg length” are not directions you can walk. And yet, when people navigate this conceptual space — moving from one bird to another by smoothly changing the features — the entorhinal cortex produces a grid-like, six-fold-symmetric signal, the very same hexagonal signature the grid cells use to map a physical room. The finding, from Alexandra Constantinescu and Timothy Behrens and colleagues, has since been extended to other abstract continua. The brain appears to map the conceptual space using the machinery it evolved to map the floor of an enclosure.
This is the empirical heart of the exaptation claim, and notice what kind of evidence it is. It is not that people merely talk about concepts using spatial words — though they do, constantly, as we will see. It is that the actual metric structure of the spatial code, the hexagonal grid, reappears in the abstract domain. That is far more than metaphor. A metaphor is a way of speaking; this is the same neural geometry doing literal computational work on non-spatial information. Whatever else is true, the spatial machinery is genuinely being recruited, not merely alluded to.
We saw a cognate result in rodents at the end of the last chapter: hippocampal and entorhinal cells mapping the frequency of a tone an animal controls, with the same structured firing they use for location. Across species, then, the map-making apparatus can be pointed at a non-spatial variable and will impose its structure on it. The machine does not know, or care, that “concept space” is not a place. It does to abstract structure exactly what it does to a room: it lays down a metric and navigates.
The map applied to social and relational worlds
Physical and conceptual spaces are not the only ones the map has been exapted to handle. Consider the most consequential abstract landscape a social animal inhabits: the space of other individuals and one’s relations to them.
Human social life is organized along dimensions we describe, irresistibly, in spatial language — and the language turns out to track something real in the brain. We speak of people as close to us or distant; of someone’s high or low status; of being aligned with one person and opposed to another. These are not idle figures of speech. When people track their relationships to others along dimensions such as power and affiliation, the hippocampus represents that social position in a map-like way, with an individual’s location in “social space” encoded much as a location in physical space would be. The structure built to answer “where am I, and where is the food” has been turned to answer “where do I stand, and who stands with me.” The exaptation reaches into the most human of domains.
And this points at why the move was so evolutionarily valuable. A map is a device for prediction — that has been our refrain. A spatial map predicts where resources will be. A social map predicts how others will behave: who will help, who will compete, who is rising and who is falling. A conceptual map predicts which ideas imply which others, which inferences will hold. In every case the exapted map does, for an abstract domain, the same prospective work it did for physical space: it lets the animal anticipate. The reason a spatial-navigation system was worth borrowing for abstract cognition is that navigation and inference are the same shape of problem — finding a path through a structured space toward a goal — and an animal that already had a superb path-finder was a short evolutionary step from an animal that could think.
The tell: spatial fingerprints on non-spatial thought
Here we arrive at a question that turns out to be one of the most revealing in the whole unit, and it is worth posing as sharply as possible. If abstract thought really runs on borrowed spatial machinery — rather than on some purpose-built, natively abstract relational store — then it should bear the fingerprints of its spatial origin. It should carry distortions, biases, and structural quirks that make sense only if a spatial metric is doing the work, and that would be bizarre or impossible if the domain were genuinely abstract from the start. Exaptation predicts residue. A borrowed tool brings its old shape to its new job.
This is exactly the prediction that distinguishes exaptation from “relational all along,” and the residue is there to be found.
Consider the simplest case: number. There is nothing intrinsically spatial about quantity. And yet humans cannot seem to think about number without space. We lay the integers on a line; we represent magnitude along a left-to-right axis so reliably that small numbers are responded to faster on the left and large numbers on the right — a spatial bias attached to a non-spatial quantity. Numerical magnitude is processed in parietal regions adjacent to those handling spatial attention. Number does not need to be spatial. It is made spatial, forced onto a linear format it has no intrinsic requirement for. That is a fingerprint: the spatial system imposing its geometry on a domain that did not ask for it.
Consider abstract similarity judgments. Physical distance estimates are known to be systematically distorted — we misjudge them in lawful, geometric ways. The striking finding is that judgments of conceptual similarity inherit the very same distortions, including violations of the formal axioms a true distance metric must obey. If concepts lived in a natively abstract relational store, there would be no reason for them to suffer the specifically spatial errors of physical distance estimation. That they do is a sign they are being computed on spatial hardware, complete with that hardware’s characteristic mistakes.
And consider the grid signature itself, from earlier in this chapter. A hexagonal lattice is the optimal way to tile a plane; its sixfold symmetry is a fact about two-dimensional physical space. There is no abstract reason for “concept space” to be hexagonal. The appearance of the hexagonal grid in conceptual navigation is perhaps the purest fingerprint of all: the geometry of physical space, showing up, uninvited, in the representation of ideas.
The residue, in short, is everywhere, and it is exactly what the exaptation thesis predicts and the relational-all-along view does not. Abstract thought does not merely resemble spatial navigation; it malfunctions like spatial navigation, inherits the distortions of spatial navigation, wears the geometry of spatial navigation. These are the marks of a borrowed tool.
Where the borrowing fails — and why that matters most
There is a deeper version of the question, and it is the one that points beyond this unit entirely. If the spatial map is so readily exapted, is there any domain it cannot handle? Is there knowledge that simply will not fit on a spatial layout?
The question is worth taking seriously, because most of us have the opposite intuition. Nearly anything you can think about, you can sketch on a whiteboard — a timeline, a hierarchy, a flowchart, a web of related ideas. The apparent universality of the spatial format can feel like proof that knowledge simply is spatial. But be careful: that intuition may tell us less about knowledge than about the machinery doing the thinking. A mind equipped with a dominant spatial system will reach for spatial layouts compulsively, whether or not the material is natively spatial — which is precisely what exaptation predicts. The whiteboard’s reach is evidence of the borrowing, not of the borrowed thing’s nature. So the real question is sharper: are there structures that the spatial format actively cannot represent, where the whiteboard does not merely simplify but lies?
There are, and they are illuminating.
The spatial map is built on a particular kind of structure: continuous, metric, and transitive. Distances are graded; positions are consistent; if A is near B and B is near C, then A and C are constrained in their relation. Any structure that violates these properties strains or breaks the spatial format. The cleanest example is a non-transitive cycle. Rock-paper-scissors has no coherent spatial layout: rock beats scissors beats paper beats rock, a loop with no consistent “near” and “far,” no axis along which the three can be honestly arranged. Draw them anywhere on a whiteboard and the picture will misrepresent the relation, because space enforces transitivity and the relation is a cycle. The same defeat awaits non-transitive dominance hierarchies, voting paradoxes in which majorities prefer A to B to C to A, and any relational structure that loops. These do not fit — not because we lack ingenuity, but because the spatial metric cannot represent a structure that contradicts its own founding assumptions.
A second class of failure is genuinely high-dimensional structure. You can place a hundred-dimensional space on a whiteboard, but only by projecting it down to two dimensions — and the projection necessarily distorts, collapsing distinctions and falsifying distances. The spatial intuition does not fail cleanly here; it fails gracefully, degrading into confident falsehood, which is why high-dimensional reasoning is so notoriously counterintuitive and why we are so easily misled about it.
And a third class — the one that matters most for what comes next — is recursive, hierarchically nested structure. Consider the deep center-embedding of human language: the rat the cat the dog chased bit died. This is not a path through a space; it is a structure nested inside a structure inside a structure, and the spatial map handles it poorly. Recursion of this kind is not navigation. It is not a journey through a landscape, however abstract. It is a different shape of structure altogether — and it may be no accident that it is one of the things the human brain seems to require additional machinery, beyond the ancient map, to manage.
This is why the failures matter more than the successes. The places where exaptation breaks down are not embarrassments for the thesis; they are its sharpest confirmation. A general relational engine, the “relational all along” machine, should handle cycles and recursion as easily as it handles metric space — relations are relations. But the system does not handle them easily; it fails on exactly the structures that a spatial metric, built for a body moving through a continuous transitive world, would be expected to fail on. The failures reveal the machine’s true nature. It is not a general relational engine wearing spatial clothes. It is a spatial engine, doing as much abstract work as its spatial assumptions will allow — and faltering precisely where those assumptions give out.
And the frontier where they give out is the doorway to the final unit of this book. For if the ancient map could be exapted only so far — if it could navigate concepts and social worlds and magnitudes but stumbled on recursion, on cycles, on the genuinely high-dimensional — then the distinctively human achievements that lie beyond that frontier required something more than the borrowed map. The story of what had to be added, when exaptation reached its ceiling, is where we turn next.
Where we have arrived
The spatial map, we have argued, was not relational all along. It was spatial — built from the body, grounded in self-motion, ancient beyond the tetrapods — and it was exapted, in lineages rich enough to exploit it, for the navigation of spaces that are not places: concepts, relationships, abstract structure. We know it is a borrowing rather than a native general capacity because the borrowing leaves fingerprints — the hexagonal grid in concept space, the line beneath number, the spatial distortions of abstract similarity — and because it fails, revealingly, on exactly the structures a spatial metric should fail on. The map did not become a general engine. It remained a spatial engine, doing abstract work in a spatial way, with all the power and all the limits that implies.
This is the same lesson the whole book has been teaching, now reaching its most surprising application. Evolution does not build new organs for new purposes when an old one can be bent to the task. It took the ancient map that guided a fish to food and, without redesigning it, turned it loose on the landscape of human thought. We think, in part, by navigating — because the thing we think with was built, long ago, for finding our way.
What remains is the case that every other textbook puts first. We have built the machine from the bottom: its purpose, its deep history, its cells, and now its exaptation into the human mind. Only now, with all of that in place, are we ready to meet the patient whose damaged hippocampus has shaped half a century of memory research — and to see how differently his story reads when he arrives not as the beginning of our understanding, but as its culmination, and when the varieties of human memory are laid out at last against the ancient, predictive, spatial map from which they were built.