An octopus carries roughly five hundred million neurons, which puts it in the neighborhood of a dog. Two-thirds of them are not in its head. They are distributed through the eight arms, in nerve cords that run the length of each limb, and the central brain that remains is arranged as a doughnut with the esophagus threaded through the hole, which means an octopus that swallows something too large risks damage to its own brain. Three hearts, copper-based blood that runs blue, no bones, no fixed body shape, a skin that can change color faster than most animals can turn around, and a lineage that split from ours somewhere in the neighborhood of five hundred and fifty million years ago, before there were vertebrates to split from. Whatever else the cephalopod nervous system is, it is the product of an experiment that ran independently of ours for the entire history of complex animal life.

The temptation with this animal is to reach for the word alien and stop there. It is the wrong move for two reasons. The first is that alien is an adjective standing in for a mechanism, and the mechanism is where everything interesting lives. The second is that the last few years of work have made the picture considerably more specific, and the specifics cut against the alien framing in one direction while pushing much harder in another. Where the cephalopod nervous system looked most exotic, in the supposedly autonomous arms, a 2025 anatomical study found a familiar solution nobody expected. And where nobody was looking, in the way these animals handle their own genetic information, they turn out to be doing something with no real parallel anywhere in the animal kingdom.

The cephalopod nervous system by the numbers

Coleoid cephalopods, which is octopuses, cuttlefish, and squid but not the nautilus, are the group that built the big nervous systems. There are roughly three hundred described octopus species and several hundred more squid and cuttlefish, and the neural investment across them is unusual for invertebrates by a wide margin: a common octopus runs about five hundred million neurons against a fruit fly‘s hundred thousand or so, a honeybee’s roughly one million, and a rat’s two hundred million.

The distribution is the striking part. Something like three hundred and fifty million of those neurons sit in the arms rather than the central brain, which inverts the vertebrate arrangement where the periphery is mostly wiring and the processing happens centrally. The central brain itself is divided into dozens of anatomically distinct lobes, wrapped around the esophagus in a ring, encased in a cartilaginous capsule that is the closest thing a soft animal has to a skull.

That architecture solves a specific problem. An octopus arm is a muscular hydrostat, a structure with no skeleton, which can bend at any point along its length in any direction, elongate, shorten, and twist. It has effectively infinite degrees of freedom. A vertebrate limb has joints, and joints are a form of compression: they reduce the control problem to a manageable number of angles. Remove the joints and the number of parameters a controller would need to specify explodes past anything a central brain could plausibly manage in real time for eight limbs at once.

So the solution was to push the control local, and for a long time the interpretation of the cephalopod nervous system was that the arms handle themselves and the brain issues something like high-level intent. A severed arm will still perform coordinated reaching movements and will still pass food toward where a mouth would be. That observation, reported repeatedly and reliably, generated the popular claim that an octopus has nine brains and that each arm thinks for itself, and it is that claim the recent anatomy has forced a revision to.

The energetic accounting is worth a moment because nervous tissue is the most expensive tissue an animal can build. Neurons are metabolically ruinous, running ion pumps continuously to maintain the gradients they then deliberately collapse in order to signal, and any lineage carrying half a billion of them is paying a bill that has to be covered by something. For most vertebrates that bill is paid over a long life with a slow reproductive schedule. An octopus pays it over roughly eighteen months while growing fast enough to increase its body mass by orders of magnitude, on a diet of crabs and bivalves it has to actively hunt. The cephalopod nervous system is not a luxury feature bolted onto a mollusc. It is the hunting apparatus, and the animal is running a high-metabolism, high-mortality, short-horizon strategy that the neural investment exists to service.

What the arms are actually running

In January 2025 a team at the University of Chicago published a study of arm anatomy in the California two-spot octopus, and the report on neuronal segmentation in cephalopod arms is the most consequential structural finding in this animal in years.

The axial nerve cord runs down the center of each arm, and rather than running straight it snakes back and forth, with every bend forming an enlargement sitting over a sucker. In cross section the arrangement follows the standard invertebrate pattern, with neuronal cell bodies in a layer wrapping around a central neuropil where the connections happen, and notably no separation of sensory from motor neurons, which are intermingled in a way vertebrate spinal cords are not.

Looking along the long axis, though, the cord is segmented. Neurons form discrete modules, each with its own adjoining region of neuropil. Between segments are septa, which are neuron-poor and contain the exits for nerves heading out to muscle, along with vasculature and collagen. Each sucker gets its own nerve supply, arranged as a spatial map in the cord, which the authors called suckerotopy, and which is the same organizing principle as the somatotopic maps in vertebrate cortex and the retinotopic maps in visual systems: adjacent bits of body map to adjacent bits of neural tissue.

The unexpected detail is that nerves exiting from neighboring septa take different trajectories, meaning multiple adjoining segments have to cooperate to innervate any given stretch of arm muscle. The segments are not independent units strung in a line. They overlap, and the overlap is presumably how a smooth traveling wave of bend propagation gets produced rather than a series of discrete jerks.

Comparative work in squid nailed the link down. Squid have eight arms bearing suckers along their whole length, plus two much longer feeding tentacles carrying suckers only on the club-shaped pad at the tip, and the segmentation pattern tracks the suckers rather than the limb: prominent in the sucker-laden arms, correspondingly different in the tentacles. Segmentation goes with flexible sucker-covered appendages, which is a functional argument rather than a phylogenetic one.

The sucker itself deserves noting as a sensory organ rather than a gripper. Each one carries chemical and mechanical receptors in enormous density, and octopuses possess chemotactile receptors that respond to compounds that do not dissolve well in water, which means an arm exploring a crevice is tasting surfaces by touch. An animal hunting in a reef at night is therefore running a sensory modality with no vertebrate equivalent, distributed across hundreds of independently steerable contact points, feeding into segmented local circuitry that can act on it without consulting the brain. The fish that hunt the same reefs by coordinating with other species are solving the same foraging problem with completely different equipment, which is the comparison the showcase exists to draw.

This is also the first documented example of nervous system segmentation in a mollusc, which is the part with the deepest implications. Segmentation is the organizational trick that annelids and arthropods and vertebrates all use for controlling elongated bodies, and it evolved independently here, in a limb rather than a trunk, for a soft-bodied animal that needed distributed control of a structure with no joints.

Nine brains is the wrong picture

Take the segmentation finding seriously and the popular framing collapses in a useful way.

An octopus does not have nine brains. It has one brain and eight segmented motor control systems with local sensory processing and topographic organization, which is a genuinely different thing. The arms are not deliberating. They are running the low-level implementation of movements the central brain requests, in the same way your spinal cord runs the details of a step without your cortex specifying which motor units fire in which order.

The reason this matters is that the standard telling gets the impressiveness backwards. The claim that each arm has a mind is a claim about distributed cognition, and it is not what the anatomy shows. What the anatomy shows is a solution to a control problem that vertebrates never had to solve, because our limbs come with joints that do the dimensionality reduction for free. The octopus took a body with no joints and built a segmented controller to make it tractable. That is a better story than eight little minds, and it has the additional advantage of being supported by the tissue.

There is a related finding worth holding alongside it. Evidence suggests the octopus central brain does not maintain a detailed map of arm position the way vertebrate somatosensory cortex maps the body, which is exactly what you would expect if the arms handle their own configuration and report upward in summary. The animal appears to know what its arms are doing at the level of outcome rather than at the level of posture, and there is behavioral work consistent with an octopus being able to guide an arm toward a visible goal without tracking the arm’s exact shape en route. If that holds, the cephalopod nervous system is running a control scheme in which the brain does not know, and does not need to know, where its own limbs are.

That combination, local segmented control plus low-resolution central representation, is a distinct engineering philosophy. It resembles nothing so much as the difference between a robot arm whose controller specifies every joint angle and one that offloads to compliant hardware and specifies only the endpoint.

That comparison is not decorative. Soft robotics has spent two decades trying to build controllers for continuum manipulators, and the octopus is the reference organism for the entire subfield, because it is the only existence proof that the problem is solvable at speed. The segmentation result reads, from an engineering standpoint, as a hint about how to architect such a controller: local modules with overlapping innervation fields, a topographic map from actuator to controller, and no attempt to maintain a high-resolution model of limb configuration centrally. Whether that transfers to hardware is an open question, and the engineering programs trying to read and write to nervous systems directly face a version of the same problem from the opposite direction, since a prosthetic limb has to be controlled by a brain that never evolved to specify its parameters.

RNA editing, and the trade nobody would have predicted

Here is where the cephalopod nervous system stops resembling anything else.

Adenosine-to-inosine RNA editing is a process where an enzyme called ADAR chemically modifies a base in an RNA transcript after it has been copied from DNA. Because inosine gets read as guanosine during translation, editing can change which amino acid ends up in the protein. Every animal does some of this. In humans, recoding of this kind affects a small fraction of genes, on the order of a few percent.

Coleoid cephalopods do it at a scale that has no parallel. They recode the majority of their neural proteins, at tens of thousands of sites, with the editing concentrated in nervous tissue and in genes involved in neural function. Work published in 2023 examined roughly sixty thousand known editing sites in California two-spot octopuses acclimated to warm or cold water and found the editing pattern shifted substantially with temperature, affecting over thirteen thousand codons.

The functional demonstration ran through kinesin, a molecular motor that hauls cargo along microtubules and which does the essential work of moving material down the length of an axon. The cold-associated edited variant of octopus kinesin behaves measurably differently from the unedited version: slower, with shorter run lengths, more inclined to stall. That is a protein being tuned for the temperature the animal currently finds itself in, on a timescale of days to weeks, without any change to the genome.

Now the trade-off, which is the part that makes this a real evolutionary story rather than a curiosity. ADAR needs double-stranded RNA structure to find its targets, and that structure depends on the sequence surrounding the editing site being able to fold back and pair with itself. Preserving thousands of these structures means preserving the underlying DNA sequence, which means those regions cannot drift the way neutral sequence normally does. Cephalopods appear to have purchased enormous proteomic flexibility at the cost of genomic evolvability, and the genomic regions around heavily edited sites show exactly the conservation that trade predicts.

Read that as an engineering decision and it is remarkable. Most lineages adapt by changing the genome across generations. Cephalopods built a system that adapts the proteome within an individual lifetime, in response to conditions, and paid for it by partially freezing the genome that supports it. For an animal that is mostly short-lived, mostly solitary, and cannot inherit behavioral solutions from its parents, buying within-lifetime flexibility is a coherent bet.

The temperature finding also has an uncomfortable forward-looking edge. A system tuned to reconfigure the neural proteome in response to water temperature is a system whose operating assumptions are set by the thermal regime it evolved in, and ocean temperatures are moving faster than any recent evolutionary baseline. Nobody has established what happens to an animal running temperature-dependent recoding when conditions go outside the range the editing repertoire was selected against, and it is the kind of question that will matter for the same reason the collapse of cold-water fish stocks mattered: a physiological system finely matched to conditions is a liability when conditions move.

Inside the central cephalopod nervous system

The octopus central brain is organized into dozens of lobes, and the one that matters most for learning is the vertical lobe system, a structure sitting at the top of the brain that functions as the animal’s principal learning and memory center. Lesion it and the animal retains basic sensorimotor function while losing the ability to form and retain learned associations, which is the same experimental signature that identified the hippocampus in mammals and the mushroom bodies in insects. Three lineages, three unrelated structures, one experimental result, which is roughly the strongest form of evidence comparative neuroscience is able to produce.

Architecturally the vertical lobe is a matrix: a large number of small amacrine interneurons receiving input and converging onto a much smaller number of output neurons, an arrangement that supports the kind of high-dimensional expansion useful for separating similar patterns. That general layout, many-to-few fan-out followed by convergence, appears in the cerebellum, in the insect mushroom body, and in the avian pallium. Nobody inherited it from anybody. It keeps getting rebuilt because it works.

Genomic work added an unexpected wrinkle. Octopus and squid genomes carry an unusual expansion of transposable elements, jumping genes, and some of these are actively expressed in the learning and memory centers rather than being silenced there. Transposon activity in neural tissue also occurs in mammalian hippocampus, and one hypothesis holds that it contributes to the somatic diversity of neurons in memory-forming regions. That is a suggestive parallel and it remains a hypothesis, which is where an honest account should leave it rather than reaching for the conclusion the parallel invites.

The protocadherin story is firmer. Octopus genomes carry a large expansion of protocadherin genes, cell-surface molecules involved in specifying neural connectivity, which vertebrates also expanded and which most invertebrates did not. Two lineages building large nervous systems both hit on expanding the same family of wiring-specification molecules, independently, is exactly the kind of convergence the comparative approach exists to find.

What the octopus brain does not have is as informative as what it does. There is no cortex, no layered sheet of tissue, no obvious equivalent of the thalamic relay organization that structures vertebrate sensory processing, and no myelin, which means conduction velocities are achieved through axon diameter rather than insulation. The giant axon of the squid, thick enough to be visible without magnification, is the extreme version of that solution and is the reason squid became the preparation on which the action potential itself was first characterized. Much of what is known about how any neuron works, in any animal including us, was worked out on a cephalopod because its wiring was thick enough to push an electrode into.

Eyes that cannot see color, on an animal that matches it

The cephalopod eye is the textbook case of convergent evolution: a camera eye with a lens, an iris, and a retina, arrived at entirely independently of the vertebrate camera eye, and arguably better engineered, since the photoreceptors face the light rather than pointing backwards through the wiring, which means no blind spot.

And it is almost certainly colorblind. Octopuses and cuttlefish have a single photoreceptor type in the retina, which normally forecloses color vision, because distinguishing wavelength requires comparing outputs across receptors with different sensitivities. This sits badly against the fact that these animals match the color of their surroundings with startling accuracy.

Several explanations are in play and none is settled. The chromatic aberration hypothesis proposes that the animals exploit the fact that a lens focuses different wavelengths at different distances, so an animal with an off-axis pupil, a wide aperture, and the ability to change focal depth could extract spectral information from how sharply different parts of a scene come into focus. The geometry works and the pupil shapes are suggestive. Whether the animals actually do this remains unproven, and it is worth treating the idea as an interesting live proposal rather than an established solution.

The other route runs through the skin. Opsins, the light-sensitive proteins normally found in eyes, are expressed in cephalopod skin, and isolated skin preparations respond to light by expanding chromatophores. Whether this constitutes distributed spatial vision, as opposed to a light-level detector feeding local reflexes, is not established, and the popular formulation that octopuses see with their skin runs considerably ahead of what has been shown.

Camouflage as a motor problem

The color-change system deserves treatment as engineering because that is what it is.

Chromatophores are pigment sacs, each surrounded by radial muscles under direct neural control from the brain. Contract the muscles and the sac stretches into a disc, exposing pigment; relax them and it shrinks to a point. This is why cephalopod color change happens in milliseconds while a chameleon’s takes seconds to minutes: the cephalopod system is neuromuscular, not hormonal or chemical. Each chromatophore is effectively a pixel with a motor attached, and a large cuttlefish carries millions of them, every one of them wired back to the brain rather than operating on local chemistry, which is why the entire display can be redrawn in the time it takes a predator to turn its head.

Underneath sit iridophores, which produce structural color through stacked reflective platelets, and leucophores, which scatter light broadly and produce white. The full display is a stack: broadband scatterers at the bottom, wavelength-selective reflectors in the middle, pigment shutters on top. Then there are papillae, muscular projections that change skin texture from smooth to spiked, which get set and held with almost no ongoing energy cost through a catch-like mechanism.

The control burden is enormous. The brain is driving millions of independent actuators in patterns that have to match a visual scene the animal is assessing in real time, and the whole system is open-loop with respect to the result, since the animal cannot see its own skin from the outside. That an animal with a single photoreceptor class produces displays that fool color-sighted predators is one of the genuine unresolved problems in the field, and it is more interesting stated as an unresolved problem than papered over.

The display system does more than hide. Cuttlefish produce a moving band of dark pattern down the body, the passing cloud display, apparently used to startle or transfix prey. Giant Australian cuttlefish males in mating aggregations run split displays, showing courtship patterning on the side facing a female and female-mimicking patterning on the side facing a rival, which is a deception requiring the animal to track who is standing where. That the same actuator array serves camouflage, hunting, courtship, and deception makes it less a defensive adaptation than a general-purpose output channel, and the bowerbirds that build and decorate elaborate structures to be looked at are running the same problem through completely different hardware: making a specific visual impression on a specific viewer.

Cuttlefish, self-control, and memory that does not decay

The cognitive work has increasingly moved to cuttlefish, partly because they tolerate laboratory conditions better than octopuses and partly because they will sit still for a task.

Common cuttlefish were run through an adaptation of the marshmallow test, choosing between an immediately available but lower-quality prey item and a preferred one available only after a delay. They waited, tolerating delays in the range of fifty to a hundred and thirty seconds, which is comparable to what has been demonstrated in some large-brained birds and primates. More interesting, the individuals that waited longest also performed better in a reversal learning task, which is the first reported link between self-control and learning performance outside the primates.

Separate work found cuttlefish retaining what-where-when information about previous feeding events, adjusting foraging based on which food had been available where and how recently, which is the operational signature of episodic-like memory. And unlike essentially every vertebrate tested, that capacity did not decline in aged animals, even as other functions deteriorated, which makes cuttlefish a potentially useful comparative case for anyone studying why memory degrades with age in the systems where it does.

The mirror-mark test, which some corvids, elephants, and cetaceans have been reported to pass, has not produced a clean cephalopod result. Cephalopods clearly recognize their own arms as theirs, and there is work showing they use chemical cues to avoid grabbing themselves, but mirror-directed self-exploration of the kind the test looks for has not been demonstrated. That is a real negative result and it should be reported as one rather than explained away.

The social dimension has produced its own surprises, in an animal long assumed to have none. Aggregation sites off eastern Australia, given the inevitable names Octopolis and Octlantis, host unusually dense gatherings of gloomy octopuses around shell beds, where individuals interact repeatedly, display at each other with body posture and color, evict each other from dens, and have been documented propelling silt and shells at one another with jets of water in a manner that is at least sometimes directed at a specific recipient. This is not sociality in the sense that meerkat sentinel systems or cooperative hunting packs are social. It does establish that the solitary characterization was partly an artifact of where people had looked.

Sentience, and a legal category that moved

The evidence question became a policy question quickly, and the sequence is worth knowing because it is one of the few cases where a literature review changed a law.

In 2021 a team commissioned by the UK government reviewed more than three hundred studies against eight criteria covering neural architecture and behavioral markers: nociceptors, integrative brain regions, connections between them, responses to anesthetics and analgesics, motivational trade-offs, flexible self-protective behavior, associative learning, and valuing analgesia. Octopuses satisfied seven of the eight, the strongest score of any group assessed. The report recommended recognizing them as sentient, and the UK Animal Welfare Sentience Act 2022 was amended to include cephalopod molluscs and decapod crustaceans, the first legal recognition of these groups anywhere.

The updated assessment published in Biological Reviews in 2026 refines rather than repeats the conclusion, and the refinements are the useful part. Octopuses and cuttlefish now carry high or very high confidence on six of the eight criteria. Squid sit at five of eight. Nautilus sits at one of eight, which the authors treat as unknown rather than negative. Cephalopod is not a single category, and treating it as one was always a convenience. The nautilus, which never built the elaborate cephalopod nervous system its coleoid relatives did, is the control condition sitting inside the same class.

The pharmacological evidence carried a lot of the weight. Lidocaine abolishes injury-directed grooming behavior in octopus and reduces it in pharaoh cuttlefish, which matters because grooming a wound is a behavior that persists past the noxious stimulus and therefore indicates an ongoing state rather than a reflex. A 2023 study in the hummingbird bobtail squid provided the first evidence of systemic analgesia in a cephalopod, with three different drug classes affecting baseline nociceptive thresholds, peripheral nerve excitability, and behavior. Earlier work had shown octopuses learning to avoid a chamber where they experienced a noxious event and preferring one where they received relief, which is the standard test for the affective rather than merely sensory component of pain.

The downstream consequences are live. Octopus farming has been banned in Washington State and California, a federal bill has been introduced, and the research community has spent a decade building husbandry and anesthesia guidelines for animals that until recently fell outside most regulatory frameworks entirely.

The claims that do not hold up

An audit, because this animal attracts more nonsense per capita than almost anything else in comparative neuroscience.

Octopuses came from space is an actual published claim, in a 2018 paper arguing for panspermia partly on the grounds that cephalopod genomic novelty appeared too abruptly to be terrestrial. It does not survive contact with the phylogeny. Cephalopods sit exactly where they should among molluscs, with nautiluses and other molluscs as relatives, and the genomic novelties have identifiable origins in gene family expansion and transposon activity. The paper is a useful case study in how a journal’s peer review can fail and how quickly a good headline outruns a bad argument.

Nine brains is a slogan rather than a description, for the reasons the segmentation work makes clear. One brain, eight segmented controllers, and a great deal of confusion generated by a severed arm continuing to move, which a severed vertebrate limb would also do given intact circuitry and a stimulus.

Octopuses are as smart as dogs because they have similar neuron counts is neuron-count reasoning, and neuron count predicts less than people want. Two-thirds of the octopus total is doing motor control in the arms, the organization is entirely different, and comparing across that gap with a single scalar is the same error as comparing two companies by headcount.

Octopuses see with their skin overstates a real finding. Skin opsins exist and skin responds to light. Spatial vision through skin has not been demonstrated, and the gap between a photodetector and an image is the entire history of the eye.

The individual octopus is a genius who escapes tanks is selection bias with a good publicist. The escape stories are real, the animals are genuinely exploratory and genuinely strong, and the ones that do something remarkable get written about while the ones that sit in a corner do not. The same filter operates on every charismatic working animal whose individual exploits became the record, and it is worth applying deliberately rather than assuming the published anecdotes are representative.

Octopuses are the most intelligent invertebrate is a ranking claim that assumes a single axis. Jumping spiders plan detours to prey they can no longer see on roughly six hundred thousand neurons, honeybees perform symbolic communication, and the reef fish running cooperative interspecies hunts are doing something no cephalopod has been shown to do. Comparing any of these is comparing animals solving unrelated problems with unrelated hardware.

The one that deserves more attention than it gets is the lifespan problem. Most octopuses live one to two years, are semelparous, dying after a single reproductive event, and do not overlap meaningfully with their offspring. Whatever an octopus knows, it worked out inside a couple of years, alone.

The mechanism behind that schedule is known and it is bleak. Removal of the optic glands, which sit behind the eyes and function roughly as an endocrine control center, prevents the post-reproductive decline and extends life substantially, which establishes that senescence here is a programmed endocrine cascade rather than accumulated wear. The animal is built to die on schedule. That is a design decision with consequences for everything else about the lineage, and it forecloses the accumulation strategy that long-lived social animals whose oldest individuals carry the knowledge depend on entirely.

What cephalopods are actually evidence for

That last fact is the deepest one available, and it reframes the whole subject.

Every other lineage that built impressive cognition built it alongside social transmission. Corvids and parrots learn from conspecifics and pass local traditions down. Cetacean populations maintain vocal dialects and foraging techniques across generations, and the pods whose specific behaviors are transmitted from mothers to offspring demonstrate how much of what an animal knows can come from another animal. Great apes maintain tool traditions that differ between neighboring populations, the macaque troop whose food-washing spread through a population is the classic demonstration, and elephant matriarchs carry spatial and social knowledge that dies with them if the population loses its old females. Even the songbirds whose regional dialects can be mapped street by street are inheriting something from a tutor, and the cranes whose migratory route had to be taught by aircraft after the knowledge was lost demonstrate what happens to a species when the transmission chain breaks.

Cephalopods have essentially none of this. Mostly solitary, mostly short-lived, no parental care to speak of in most species, no generational overlap, no observed cultural transmission. Whatever the cephalopod nervous system delivers, it delivers within a single lifetime, from scratch, without a teacher.

That makes them the closest thing available to a controlled comparison on a question the rest of the field cannot isolate: how much of complex cognition requires accumulated social knowledge, and how much is what a sufficiently well-built nervous system can do on its own? The animals whose knowledge visibly moves between individuals are running the other arm of the experiment, and the contrast is the finding. It is also the question sitting underneath every attempt to build cognition in a system with no evolutionary history at all, where the amount of accumulated human knowledge poured into training is the entire variable under discussion.

It also explains the RNA editing bet. An animal that cannot inherit solutions from its parents and will not live long enough to accumulate many of its own has a strong incentive to build maximum flexibility into the hardware, and paying for within-lifetime proteomic adjustment with reduced genomic evolvability is a rational trade under exactly those constraints.

What the cephalopod nervous system therefore represents is not a stranger version of us. It is an independent trial of the same engineering problem, run with different materials, under different constraints, with a different answer to the question of what gets remembered across generations. Both trials produced animals that hunt by inference, learn by association, and manipulate objects with precision. Only one of them produced animals that teach.

So the honest summary is not that octopuses are aliens. It is that complex nervous systems have been built twice on this planet, from different starting material, on opposite sides of a five-hundred-million-year gap, and the second attempt produced something that converged on camera eyes, matrix memory structures, expanded connectivity-specification gene families, and segmented motor control, while diverging completely on where the neurons live, how the genome is used, and whether anything gets passed on.

Every element of that picture is a mechanism with a number attached rather than an adjective, which is the standard the 24-lecture Neurozoology course applies across the tree of life. Which is what it is built to make legible, running the first edition’s survey of nervous systems forward with the mechanism in front and the adjectives thrown out, and it is why an animal that threads its own esophagus through its brain is worth more than a paragraph of wonder. Five hundred million neurons, two-thirds of them in the arms, a segmented cord with a topographic map of suckers, chemotactile receptors that taste by touch, millions of neurally driven pigment cells redrawing a body pattern in milliseconds, and a proteome that gets rewritten when the water turns cold. Not one item on that list required the word alien to become interesting, and every one of them is a number somebody had to go and measure. The awe is in the numbers. It always was.

Put a desert ant on stilts and it will walk straight past its own nest.

The experiment is exactly as blunt as it sounds. Cataglyphis foraging ants wander a long looping path across featureless Saharan salt pan, find food, and then run a near-perfect straight line home, which is a genuinely remarkable feat of navigation in an animal with under a million neurons. Researchers glued pig bristles to the legs of ants that had already completed the outbound trip, lengthening their stride for the return. The stilt-walkers overshot the nest by a predictable margin. Ants with shortened legs stopped short. The animal was not consulting anything resembling a chart. It was counting steps, multiplying by stride length, and integrating that against a compass heading, and when the experimenters changed the stride length the arithmetic came out wrong in exactly the way the arithmetic should.

That result is the single most useful thing to hold onto in this subject, because it demonstrates the gap between spectacular navigation and an actual cognitive map. The ant is doing something extraordinary and it is not doing what the word map implies. It cannot take a novel shortcut from an arbitrary point. It has no representation of the relationship between two places it has never traveled between. It has a running estimate of the vector home, and if you corrupt the odometer the estimate fails gracefully and completely.

Seventy-odd years of work separates Edward Tolman’s proposal that animals build internal maps from the current state of the field, and the most interesting developments of the last decade have gone in two directions at once. The neural machinery turned out to be more geometrically specific than anyone expected, sitting on a mathematical structure that can be measured. And the thing it maps turned out not to be space.

What Tolman actually claimed, and what he got right by accident

In 1948 Edward Tolman published “Cognitive Maps in Rats and Men” into a field that did not want it. Behaviorism held that learning was a matter of stimulus-response chains strengthened by reinforcement, and a rat running a maze was assembling a sequence of turns, not a picture.

Tolman’s evidence was awkward for that account. In latent learning experiments, rats allowed to wander an unrewarded maze for days performed dramatically better than naive rats once a reward was introduced, which means they had been learning something with no reinforcement to strengthen anything. In place-versus-response experiments, rats trained to reach a goal from one starting point, then started somewhere else, went to the place rather than repeating the turn sequence. In the sunburst maze, rats trained on an indirect path to a goal, then offered a fan of new radial alleys, disproportionately chose the alley pointing at where the goal actually was.

None of that proves an internal map in any strong sense, and Tolman’s critics said so at length. What it establishes is that the animal learned spatial relationships it was never rewarded for learning and could use them in configurations it had never experienced, which is the operational signature people still test for.

The part that gets skipped is Tolman’s last section, where he took the idea somewhere strange. He argued that narrow, brittle maps produced by fear and frustration explained regression, fixation, displaced aggression, and social prejudice, and that the same mapping machinery organizing a rat’s maze also organized how people represent social and abstract relationships. In 1948 this read as a psychologist overreaching past his data, and for fifty years it was treated as the eccentric coda to an important paper. It is now the most vindicated part of it, for reasons Tolman had no way to anticipate.

The cell types, and what each one is actually doing

The neurobiology arrived in stages and it is worth knowing what each component contributes, because the popular version collapses them all into brain GPS and loses the mechanism.

Place cells, found by John O’Keefe and Jonathan Dostrovsky in 1971, fire when a rat occupies a particular location in a particular environment. Each cell has one or a few place fields, the population tiles the environment, and every location produces a distinctive pattern of active cells. Move the animal to a different environment and the cells remap, forming a new and largely unrelated assignment, which means place cells encode this place in this context rather than coordinates in any absolute frame.

Grid cells, reported by May-Britt and Edvard Moser’s group in 2005, sit upstream in medial entorhinal cortex and do something stranger. A single grid cell fires at multiple locations arranged in a hexagonal lattice tiling the entire environment, like a triangular tessellation laid over the floor. Grid cells cluster into modules with discrete spacings that increase along the dorsal-to-ventral axis, cells within a module share orientation and spacing while differing in phase, and the module scales follow an approximate power-law relationship. That arrangement is a plausible metric, a coordinate system with multiple resolutions, and it is why the 2014 Nobel Prize went to O’Keefe and the Mosers.

Then the supporting cast. Head direction cells fire when the animal faces a particular direction regardless of location, functioning as a compass and depending heavily on vestibular input. Border cells and boundary vector cells fire at a specific distance and direction from environmental edges, and they matter more than their billing because boundaries appear to serve as an error-correction reference for a grid system that otherwise accumulates drift. Speed cells encode running speed. Object vector cells fire at a set distance and bearing from discrete landmarks. Time cells fire at particular moments during a delay, tiling elapsed time the way place cells tile space, which was the first strong hint that the machinery was indifferent to whether the dimension it was tiling had anything to do with physical distance.

The 2024 review of grid cell mechanisms and function lays out where the mechanistic arguments now stand, and the short version is that continuous attractor network models, in which recurrent connectivity constrains the population to a low-dimensional state that gets pushed around by velocity input, have accumulated substantially more experimental support than the competing oscillatory interference accounts.

The torus, and why topology beat tuning curves

The result that settled a long-running argument came in 2022, and it is a good example of a finding that is hard to explain and worth the effort.

Continuous attractor models predict something specific and non-obvious. If grid cell activity in a module is generated by a recurrent network whose stable states form a two-dimensional sheet with periodic boundary conditions, then the population activity should live on a torus, a doughnut surface, regardless of what the animal is doing or where it is. Not the firing pattern on the floor, which everyone had already seen. The shape of the population state space itself.

Recording large numbers of grid cells simultaneously and applying topological data analysis, researchers found exactly that. The population activity of a grid module occupies a toroidal manifold, the torus persists across environments, it persists in darkness, and it persists during sleep when the animal is not navigating anything at all. That last point is the one that matters most, because it means the structure is intrinsic to the network rather than imposed by sensory input.

This is a different kind of neuroscience result than a tuning curve. It is a claim about the geometry of a neural population’s activity, tested with the mathematics of shape, and confirmed. It moves the grid system from a suggestive pattern to a mechanism with a demonstrated architecture, and it is the strongest evidence available that the brain implements something genuinely coordinate-like rather than merely producing coordinate-like output.

The persistence during sleep also connects the navigation system to something the animal is obviously not doing at the time, which is a recurring theme in this machinery. A structure that holds its shape with the eyes closed is a structure available for offline use, and offline use is where most of the interesting computation happens.

Path integration, vector memory, and everything a cognitive map is not

Back to the ant, because the field’s hardest problem is distinguishing a map from things that produce map-like behavior more cheaply.

Path integration, also called dead reckoning, requires a compass and an odometer and nothing else. Track your heading, track distance traveled, continuously update a single vector pointing home. It is computationally trivial, it works in featureless terrain, and it degrades predictably: errors accumulate with distance and never self-correct, which is why animals relying on it heavily also carry backup systems. Desert ants run a visual panorama-matching routine near the nest to clean up the final approach, because the integrated vector alone is not accurate enough to find a hole in the ground.

Vector memory is the next tier and it is still not a map. An animal that has traveled from A to B can store the vector and reuse it. Honeybees do this well, and the long argument over whether bees have map-like memory turns precisely on whether they can compute a novel route between two locations without having traveled it, which is a much harder claim than storing a library of learned vectors.

Beaconing is simpler still: head toward a detectable cue at the goal. Piloting means moving between recognized landmarks in sequence. Route following means reproducing a learned sequence of movements, and it can be extended almost indefinitely without ever becoming a map, which is how an animal can traverse a route of enormous length and complexity while remaining unable to deviate from it. Every one of these produces impressive navigation, and none requires the animal to represent the spatial relationship between places it has not connected by direct experience.

That last capability is the operational definition of a cognitive map, and it is why the evidentiary bar is so high. Novel shortcutting, and detour behavior when a familiar route is blocked, are the behaviors that cannot be produced by the cheaper systems. Everything else is compatible with an animal that has an excellent memory and no map at all.

The reason this taxonomy is worth memorizing is that it inverts the intuitive reading of almost every navigation story. A monarch butterfly crossing a continent to a grove it has never seen, four generations removed from the last butterfly that made the trip, is doing something a cognitive map could not accomplish, because there is nothing in its experience to build a map out of. It is running an inherited compass heading against a clock. That is less flexible than a map and far more impressive as a piece of engineering, and describing it as a map would make it sound easier than it is.

The same correction applies in reverse. A rat that has spent an hour in a small box, doing nothing anyone would call remarkable, may well be running the more sophisticated system, because it can be dropped anywhere in that box and head straight for a corner it has not approached from that angle before. Flexibility in unfamiliar configurations is the diagnostic, not distance covered or difficulty of terrain.

Compasses, magnetic maps, and the difference between them

Compass systems are separable from maps and animals stack multiple ones with a hierarchy of preference.

Sun compasses require time compensation, since the sun moves, and monarch butterflies solve this with circadian clocks located in the antennae feeding a sun-azimuth calculation in the central complex. Remove or paint the antennae and the compass fails while the clock in the brain keeps running. Many insects also read polarized skylight patterns, which persist under partial cloud and are detected by a specialized dorsal rim region of the eye.

Star compasses appear in birds, which learn the rotational center of the night sky as nestlings rather than inheriting specific constellations. Dung beetles orient by the Milky Way as a band, using it to roll a dung ball in a straight line away from competitors, which is the least dignified application of galactic astronomy on record and a genuine one.

Magnetic compasses are the contested case and deserve honest treatment. Two mechanisms remain live. The radical pair hypothesis proposes that cryptochrome proteins in the retina form spin-correlated radical pairs on photon absorption, with reaction yields sensitive to magnetic field orientation, giving a light-dependent inclination compass. Work on cryptochrome 4 from European robins showed magnetic sensitivity in vitro, which is real support and is not the same as demonstrating the mechanism operates in a living bird. The magnetite hypothesis proposes iron-mineral-based receptors transducing field intensity and direction mechanically, with the trigeminal nerve as a candidate pathway. Both may be true and serve different functions. Neither is closed, the field has a documented history of high-profile results that did not replicate, and anybody presenting magnetoreception as solved is ahead of the evidence.

The replication history deserves its own sentence because it is instructive about how a field can go wrong. Magnetoreception research has produced several widely cited candidate receptors that later analysis attributed to contamination, to iron-rich macrophages rather than sensory cells, or to effects that vanished under blind protocols. That record is not an argument that the sense does not exist, since the behavioral evidence for magnetic orientation is overwhelming across birds, turtles, fish, and insects. It is an argument that identifying the receptor is genuinely hard, that the incentives reward premature announcement, and that a reader should treat any new claim of a definitive magnetoreceptor with the same posture they would bring to a press release about room-temperature superconductivity.

Migratory birds appear to run all of these plus landmark memory, recalibrating one against another. The songbirds whose learning has been characterized in unusual detail navigate continental distances on their first attempt with no guide, while the cranes whose eastern migratory population had to be taught a route by aircraft demonstrate the opposite arrangement: a species where the route is culturally transmitted rather than inherited, and where losing the knowledgeable individuals means losing the route.

Here is a distinction that popular coverage almost never makes and that is central to the whole subject. A compass tells you which way is north. A map tells you where you are. These are different problems requiring different information, and an animal can have one without the other.

A magnetic map requires that the geomagnetic field vary predictably across the region in question and that the animal read at least two field parameters, typically inclination angle and total intensity, whose contours cross at an angle. Reading both gives a positional fix, in principle, in the same way that two intersecting lines of position give a fix in celestial navigation.

Sea turtles are the strongest case. Hatchling loggerheads exposed to magnetic signatures characteristic of specific locations along their migratory circuit orient in the direction appropriate for that location, in a laboratory tank, with no other cues, having never been there. That is positional information from the field itself. The current model holds that turtles imprint on the magnetic signature of their natal beach and use it to return decades later to nest, which explains an otherwise baffling homing feat and generates testable predictions about nesting distributions shifting as field lines drift.

More recent work has shown that loggerheads can learn to associate a magnetic signature with food, expressing a distinctive anticipatory behavior when placed back in the learned field conditions, which establishes that the magnetic sense is available to associative learning rather than being locked into a fixed navigational reflex. Salmon appear to use a similar geomagnetic imprinting mechanism for natal river homing, and spiny lobsters displaced to unfamiliar sites in ways that eliminated route-based cues still oriented homeward, which remains one of the cleaner demonstrations of true navigation in an invertebrate.

The relevance to marine mammals is unresolved and interesting. The orca populations whose ranging patterns have been tracked for decades follow routes stable enough to look like knowledge, bottlenose populations show location-specific foraging traditions, and the deep-diving species that cover ocean basins do something nobody has adequately characterized. The cod stocks whose migratory routes collapsed with the population raise the same question in a fishery context: if route knowledge is partly learned and partly social, removing the experienced animals removes the map.

Proving a cognitive map in the wild finally happened

Seven decades of cognitive map research produced enormous neurobiological detail and almost no field evidence from free-ranging wild animals, for a simple reason: ruling out the cheaper strategies requires knowing where an animal went, continuously, at high resolution, across a large area, for a long time, in many individuals.

That became possible with reverse-GPS. Instead of putting a receiver on the animal, a network of ground stations receives a signal from a lightweight tag and computes position by time-difference-of-arrival, which allows much smaller tags and much higher sampling rates than satellite GPS. Applied to Egyptian fruit bats in Israel’s Hula Valley, the resulting study of cognitive map-based navigation in wild bats tracked 172 individuals across 3,449 bat-nights over four years, producing more than eighteen million localizations, with every fruit tree in an eighty-eight-thousand-hectare study area mapped as a potential goal.

The findings: bats seldom searched randomly. They flew long, straight, goal-directed trajectories to specific trees, often ignoring closer trees of the same species. Of more than nine thousand recorded trajectories, several hundred were shortcuts between two known locations along routes the individual had never flown. Bats translocated to unfamiliar release points at the edge of their range returned along novel straight-line paths rather than searching or retracing. The analysis then worked through the alternatives systematically, using simulated tracks and trajectory analysis to rule out random search, beaconing, piloting, path integration, and following other bats.

A companion study tagged pups before their first outdoor flights and watched the map get built, with young bats making progressively longer exploratory excursions from the roost, extending their known area outward over months rather than arriving with it.

The scale of effort is the point worth extracting. It took four years, 172 animals, eighteen million position fixes, and a complete botanical inventory of an area larger than most cities to demonstrate something the field had assumed since 1948. That is what the evidentiary bar for a cognitive map actually costs.

The instrumentation point generalizes past bats. Reverse-GPS, miniaturized biologgers, wireless neural recording in freely moving and flying animals, and high-density silicon probes recording hundreds of neurons simultaneously have all arrived in roughly the same window, and each one converted a question that could only be argued about into a question that could be measured. The toroidal topology result was impossible before it became routine to record enough grid cells at once to reconstruct a population manifold. The bat map result was impossible before tags got small enough to fly on an animal that weighs about as much as a deck of cards. A meaningful fraction of what looks like conceptual progress in this field over the past decade is instrument progress arriving with a delay.

Three dimensions, and where the textbook was wrong

Almost all the foundational work was done on rats running on flat surfaces, which is a defensible simplification and turns out to have baked in an assumption.

Bats fly. Recording from bats in three-dimensional flight showed place cells with roughly spherical fields distributed through volume, and head direction cells organized in a three-dimensional scheme with toroidal topology covering azimuth and pitch. So far, so consistent.

Grid cells did not cooperate. The natural expectation was a three-dimensional analogue of the hexagonal lattice, something like a face-centered cubic packing tiling the volume. What flying bats actually showed was local order without global lattice structure: firing fields at characteristic distances from their neighbors, but no long-range periodic arrangement of the kind that defines two-dimensional grid cells. The metric is there. The crystal is not.

That is a real correction and it has consequences for how the code is understood. A locally ordered arrangement can still support distance estimation and path integration while giving up the elegant modular periodicity that made the two-dimensional grid so appealing as a coordinate system. It also suggests the hexagonal lattice may be partly a consequence of the constraint of moving on a plane rather than a universal solution to representing space.

There is a general lesson in it about the species you choose. A field built on animals that live on surfaces produced a theory suited to surfaces, and the correction arrived only when someone recorded from an animal that does not. The same critique applies to arboreal primates navigating a three-dimensional canopy, to fish, and to every marine animal in a volume rather than on a plane.

There is a second asymmetry in three-dimensional space that the flat-surface tradition never had to confront, which is that vertical is not equivalent to horizontal. Gravity provides an absolute reference, moving up costs more than moving sideways, and the range of vertical movement available to most animals is far smaller than their horizontal range. A representation optimized for that anisotropy should not be isotropic, and the locally ordered arrangement observed in flying bats may be exactly what an efficient solution to an anisotropic problem looks like. Whether animals that move freely in all three dimensions with less gravitational asymmetry, which is to say animals in water, use something different again is an open and largely untested question. The reef fish whose spatial and social behavior keeps outrunning expectations would be an obvious place to look.

Maps of things that are not places

This is the reframe, and it is the strongest reason to care about this subject even if you have no interest in navigation.

If grid cells implement a general coordinate system, there is no principled reason the axes have to be north and east. Human imaging work has found grid-like six-fold symmetric signals in entorhinal cortex while subjects navigate abstract spaces: in one influential study, participants learned to morph a bird stimulus along two continuous dimensions, neck length and leg length, and their entorhinal activity showed the same hexagonal signature as spatial navigation, in a task involving no space whatsoever. Comparable structure has been reported for social spaces defined by power and affiliation, for odor spaces, and for task and conceptual spaces generally.

The theoretical work followed. The successor representation frames the hippocampus as encoding predicted future states rather than current location, which reproduces place field properties while generalizing naturally to non-spatial sequences. The Tolman-Eichenbaum Machine and related models treat the hippocampal-entorhinal system as factorizing structural knowledge from sensory content, learning the abstract shape of a problem once and then binding new specifics onto it, which is why a familiar structure in a new environment is learned so much faster than a new structure.

There is a genuine competing account worth stating rather than burying. Some researchers argue the relationship runs the other way: that a domain-general clustering and concept-learning algorithm produces place-like and grid-like representations as a side effect when inputs happen to be uniformly distributed, as they are in an empty room, and produces more conceptual-looking representations when inputs are sparse and high-dimensional. On that reading the system was never a spatial map that got repurposed. It was always a general learning mechanism that we happened to discover in a rat in a box.

Either way, the practical implication holds. The machinery that lets a bat fly to a specific fig tree is the machinery that lets a person hold a family tree, an org chart, a chess position, or the relationship between concepts in a field they are learning. Tolman’s odd 1948 coda about maps of social relationships was not overreach. It was the part of the paper that took seventy years to catch up to.

One caution about how this reframe gets reported. Finding a six-fold symmetric signal in an imaging study of a conceptual task is considerably weaker evidence than recording grid cells directly, the analysis depends on methodological choices that have been contested, and the number of well-replicated non-spatial grid findings is smaller than the enthusiasm around them suggests. The direction of the evidence is consistent and the strength of any individual result is moderate, which is a normal state for a young idea and worth stating plainly rather than letting the accumulated citations imply more than any single study delivers.

Replay, sweeps, and a map that runs simulations

A map you can only read at your current position is a limited instrument. What makes the hippocampal system powerful is that it runs offline.

During sharp-wave ripple events, place cell sequences reactivate in compressed form, often during rest and sleep, sometimes forward and sometimes reversed relative to the original trajectory. Reverse replay after reaching a reward is well suited to propagating value backward along a path. Forward replay before movement looks like route selection.

More striking is what happens at decision points. A rat pausing at a maze junction, physically oscillating its head between the two options in a behavior called vicarious trial and error, shows place cell activity sweeping ahead down first one arm and then the other, representing locations the animal is not at and has not chosen. That is a spatial representation being used to evaluate hypothetical futures, which is a substantially different thing from a chart of where you are.

Replay also generates sequences for trajectories never taken, assembled from fragments of experienced ones, and the grid cell correlation structure is preserved during sleep, which is what allows the same machinery to run without sensory input. The system is not a map in the sense of a static document. It is closer to a simulator, and the navigational application is one thing it happens to be used for.

That framing also explains an otherwise odd clinical fact. Damage to the hippocampus produces amnesia, which is a memory disorder, and it also produces impaired navigation and an impaired ability to imagine novel scenes or plausible futures. Those look like three unrelated deficits under the map model and like one deficit under the simulator model: a patient who cannot assemble a coherent representation of a situation they are not currently in will fail at remembering the past, navigating to somewhere out of sight, and imagining tomorrow, for the same underlying reason.

The artificial agents trained on navigation tasks that spontaneously develop grid-like representations in their hidden layers are informative here without settling anything: a network optimized for path integration converging on something resembling a grid code suggests the solution is at least partly forced by the problem rather than by biology. That is a claim about computation, and the engineering work that reads and writes to these systems directly will eventually test it in a way simulation cannot.

What the map metaphor keeps getting wrong

An audit, because this subject generates unusually confident nonsense.

Humans have a GPS in their heads is the headline version of the Nobel work and it oversells in two directions. The system is not a global positioning system, since place cell assignments remap between environments rather than referencing an absolute frame, and it is not a receiver of external signal, since it constructs position from self-motion and landmarks. It is closer to an odometer plus a landmark-matching routine plus a relational memory, and calling it GPS imports the wrong intuitions about accuracy and absoluteness.

Grid cells are the cognitive map is a category error the field itself sometimes commits. Grid cells provide a metric. Place cells provide context-specific location. Boundary cells provide error correction. The map, if the word means anything, is the emergent product of the interaction, and no single cell type is it.

The London taxi driver result is usually flattened. Licensed drivers who completed the Knowledge showed larger posterior hippocampal grey matter volume than controls, with the effect scaling with years of experience, and trainees who qualified showed changes that those who failed did not. That is genuine evidence of experience-driven structural plasticity. It is not evidence that navigation training makes you generally smarter, and the same studies found the drivers performing worse on some other memory tasks, which suggests a reallocation rather than an upgrade.

A good sense of direction is a single trait does not survive testing. Individual variation in navigation performance is large, partly cultural, and factors into somewhat separable abilities involving path integration accuracy, landmark memory, and the ability to adopt an allocentric perspective at all. Some people appear to navigate almost entirely by route memory and do fine.

Satellite navigation is destroying our cognitive map is the newest entry and the evidence is thinner than the confidence around it. Studies have found that turn-by-turn guidance reduces hippocampal engagement during the task itself and that heavy lifetime use correlates with worse performance on some spatial tests, which is real and also exactly the pattern you would expect from any offloaded skill. Whether it produces durable structural change, whether the correlation runs the direction people assume, and whether it matters for anything beyond navigation are all unresolved, and the cleanest available reading is that not practicing a skill makes you worse at that skill.

Animals that navigate impressively must have cognitive maps is the error the ant kills. Path integration, vector memory, compass orientation, and landmark sequences produce feats that look map-like and are not, and the homing pigeons whose wartime performance made them famous were doing something that a century of research still has not fully resolved, involving some combination of magnetic, olfactory, visual, and possibly infrasound cues. The birds credited with saving units by returning through fire were extraordinary. What they were extraordinary at is still partly an open question.

Where this leaves the internal GPS

The picture that emerges is less tidy than the Nobel citation and considerably more interesting.

There is no single internal GPS. There is a stack of navigation systems of increasing cost and capability, and different lineages sit at different points on it, often running several at once with a preference hierarchy that shifts by context. Path integration is cheap, ancient, and available to animals with under a million neurons. Any account of a cognitive map that cannot say which tier of that stack an animal is operating on is not saying much. Compasses are cheap and heterogeneous, built out of eyes, antennae, clocks, and possibly iron. Maps in the strong sense are expensive and rare, and demonstrating one in the wild took four years and eighteen million position fixes.

Where the mammalian system is unusual is not that it navigates well. Ants navigate well. It is that the coding scheme turned out to be general. The same population geometry that supports getting to a fig tree supports representing a conceptual space, a social hierarchy, a task structure, and a hypothetical route nobody has taken, and it runs those representations offline while the animal sleeps. That is a memory and inference system that happens to have been discovered in a maze.

The elephants whose knowledge of water sources across enormous ranges keeps a family group alive through drought are the case that makes the stakes legible, since that knowledge is held disproportionately by the oldest matriarch and it dies with her. The populations studied under different pressures show how much of that map is individual experience rather than species instinct, and the cooperative hunters covering large territories raise the same question about how much spatial knowledge is distributed across a group rather than held in one head. The sentinel systems that let a foraging group use space it could not safely use alone are a version of the same trade, and the corvids that cache thousands of items and recover them months later are the standing demonstration that spatial memory capacity and brain size have a looser relationship than anyone expected.

Which is where the 24-lecture Neurozoology course puts the emphasis throughout, and it is the reason the ant on stilts is the right place to start rather than the Nobel Prize. Starting with the Nobel Prize teaches you that the brain has a positioning system. Starting with the ant teaches you to ask what any given animal is actually computing, which is the question that survives contact with the next twenty years of results. The nervous system is not a magic positioning device. It is a set of expensive, error-prone, energy-hungry estimators, each one solving a problem some ancestor actually had, layered on top of each other with no plan and no cleanup. The knowledge that visibly moves between animals, the working animals whose capacities were discovered by people who needed them, and the first edition’s survey of nervous systems all run on the same assumption.

Glue bristles to an ant’s legs and it walks past its own front door, having done everything right. That is not a failure of the animal. It is a precise readout of what the animal was actually computing, which is the only kind of answer worth having.

In June 2025 a research team published drone footage of southern resident killer whales biting off lengths of bull kelp stalk, positioning the piece between their own body and a partner’s, and rolling it back and forth for as long as fifteen minutes. It was the first documented case of tool manufacture and use in any marine mammal. It happened in the Salish Sea, to a population of fewer than eighty animals that has been individually photographed, named, catalogued, and followed since the 1970s, by researchers who know these whales by the notches in their dorsal fins.

Sit with the arithmetic on that. Fifty years of continuous observation of the most intensively studied cetaceans on the planet, and a behavior involving manufactured objects and a fifteen-minute duration went unrecorded until somebody put a camera above them instead of beside them. The whales did not learn this in 2024. We learned to look down.

That is the honest shape of almost every recent addition to the animal tool use catalog, and it is why the list keeps getting longer without the animals getting any smarter. The list was never a measure of what animals can do. It has always been a measure of what we have managed to see, filtered through a definition we wrote for our own convenience, and the underlying capacity turns out to be far more widely distributed than the catalog ever suggested. There is a specific neural mechanism behind that distribution, it was worked out on monkeys that barely use tools in the wild, and it is the reason the lecture title says extended self rather than clever animals.

What counts as animal tool use, and who decided

The working definition most of the field still uses descends from Benjamin Beck: the external employment of an unattached or manipulable attached object to alter more efficiently the form, position, or condition of another object, another organism, or the user itself, with the user holding or carrying the tool during or just prior to use.

Read that carefully and notice how much work the qualifiers are doing. The object must be unattached, which is why a chimpanzee using a rock as an anvil sits in a different category from a chimpanzee wielding a rock as a hammer. The user must hold or carry it, which excludes a great deal of interesting behavior. A bearded vulture dropping a bone onto rocks to shatter it is not using a tool under this definition, because the rock is part of the landscape and the bird never held it. Sea otters smashing shellfish against a stone on their chest are sometimes classified as tool users and sometimes demoted to proto-tool users depending on whether the stone or the shellfish is doing the moving.

None of that is arbitrary in the sense of being random. It is arbitrary in the sense of being a human decision about category boundaries, made to keep a research literature tractable, and it has consequences. Every few years someone proposes a revision, most recently a framework distinguishing tooling from mere object use on the basis of whether the animal is dynamically managing a mechanical relationship rather than statically placing something. Under different definitions, different animals join and leave the club.

There is a second layer of gatekeeping underneath the first, and it concerns manufacture. Using a found object is one category. Modifying an object to suit a purpose is another, and modifying it before you can see the situation it will be used in is a third that people treat as the interesting one. The whale biting a length off a kelp stalk crosses into manufacture. So does the crow trimming a twig into a hook. So does the palm cockatoo shaping a drumstick. Whether that boundary tracks anything real in the nervous system, as opposed to tracking our intuitions about foresight, is an open question that the field has largely declined to answer while continuing to use the boundary.

This matters because it means a meaningful fraction of the growth in the animal tool use list is definitional rather than empirical. Behavior that was known and excluded gets reclassified as included. And it means the reflexive question people ask about a new finding, whether it really counts, is usually a question about our filing system rather than about the animal.

Worth noticing what the definition was built for. It was written to discipline a literature that had a real problem with overclaiming, in an era when a single anecdote about a clever animal could circulate for decades without anybody checking it. As a filter against nonsense it earned its keep. As a description of what nervous systems are doing it was never meant to be load-bearing, and it has been quietly carrying that weight for forty years because nothing better arrived.

Where the extended self is actually located

Here is the mechanism, and it is the part that reframes everything else.

In the mid 1990s Atsushi Iriki and colleagues trained Japanese macaques to retrieve distant food with a rake and recorded from bimodal neurons in the caudal postcentral gyrus, cells that respond to both touch on the body and vision near the body. Those neurons carry something like a map of the hand and the space immediately around it. When the monkey used the rake, the visual receptive fields of those cells changed. They stretched to include the length of the rake, or to cover the expanded region the animal could now reach. The subsequent review by Angelo Maravita and Iriki on tools and the body schema laid out the case that a tool in active use gets incorporated into the neural map of the body, as though the effector had been elongated to the tip of the implement.

The tool does not feel like a held object to the nervous system doing the holding. It gets annexed. Peripersonal space, the zone the brain treats as immediately body-adjacent and worth defending, expands to the end of the stick. Related work found tool-use training driving immediate-early gene expression and neurotrophic factor expression in intraparietal cortex, which means this is not a transient perceptual illusion but an actual plastic change in tissue.

Now the detail that makes it important. Japanese macaques do not habitually use tools in the wild. They can be trained to be dexterous with them, and they are not a tool-using species in any natural-history sense. The neural machinery for incorporating an external object into the body schema was sitting there anyway, unused, in an animal whose ecology never called for it.

That is the finding that should govern how anyone reads a tool-use headline. The capacity is not the achievement. Body schema plasticity appears to be a general property of nervous systems that have to coordinate a limb with a visual field, which is most of them, and what varies between species is not whether the machinery exists but whether ecology, anatomy, and opportunity ever conspire to switch it on. Animal tool use is a behavior that gets expressed, not a faculty that gets evolved from scratch each time.

The human side of the same literature runs in parallel and is worth a beat, because it establishes that this is not a monkey curiosity. People using a tool show measurable shifts in how they judge distances and in how visual and tactile events get bound together across the extended reach, with the effects appearing after minutes of practice and decaying after the tool is set down. Patients with parietal damage show tool-related deficits that dissociate from ordinary grasping, which implies at least partly separate circuitry for acting through an object rather than on it. The consistent finding across species and methods is that the body model is a running estimate rather than a fixed inventory, continuously refitted to whatever the organism is currently doing.

Once that is on the table, the comparative question changes shape. Instead of asking which animals are smart enough to use tools, the productive question is which animals have a manipulator worth extending, an ecological problem that extension would solve, and enough tolerance for failure to get through the learning curve. Those are three separate constraints and they are all about circumstances rather than intellect, which is why they can be satisfied in a walnut-sized parrot brain and go unsatisfied in the largest brains on the planet.

The orcas, and the fifty years nobody saw it

Back to the whales, because the case study is unusually instructive.

Between April and July of 2024, researchers flying an unoccupied aerial vehicle over the central Salish Sea recorded roughly thirty instances of what they named allokelping, published in Current Biology as a report on the manufacture and use of allogrooming tools by wild killer whales. The whales detach a complete bull kelp stalk, bite off a short length of the stipe, maneuver it between themselves and a social partner, and roll it along their bodies. The kelp is firm but flexible with a slippery surface, which one researcher compared to a filled garden hose, and the leading hypotheses are skin hygiene, since orca skin accumulates scaly buildup, and social bonding, since grooming in other species is at least as much about relationships as about cleanliness.

The details that make it look like a real cultural behavior rather than an oddity: older whales with more skin sloughing were more likely to participate, participation was biased toward close kin, and specific pairs did it repeatedly, including a twenty-nine-year-old female and her five-year-old daughter, and grandmothers with grandsons. Cetaceans have long been known to drape kelp over themselves, a behavior called kelping. Doing it with a partner, using a length the animal shortened itself, is a different thing.

There is a conservation edge on this that deserves stating. This is a critically endangered population, bull kelp in its habitat is declining with warming water, and the researchers noted that the behavior’s persistence may be at risk. The southern residents are also the population whose vocal dialects made cetacean culture a serious research subject, which means we now have two independent culturally transmitted traditions in the same eighty animals, both of which could be lost with them.

But the methodological point is the one to carry forward. The behavior was invisible from a boat. It required looking straight down, in good light, with enough resolution to see a piece of kelp against a black-and-white animal, over enough hours to catch thirty instances. Every one of those is a technology problem, not a biology problem.

There is also a specific reason cetaceans were a hole in the catalog rather than a genuine absence, and it is anatomical rather than cognitive. Tool use as classically defined requires holding or carrying, and a whale has no hands. What it has is a mouth, a rostrum, and pectoral fins that do not oppose. Any cetacean tool behavior therefore has to route through the mouth or through pressing an object between two body surfaces, which is exactly what allokelping does, and which no primatologist writing a definition in the 1970s would have thought to accommodate. The sponge-carrying and shell-trapping traditions in bottlenose populations had already established that the mouth-and-rostrum route works. The deep-diving species whose social behavior is hardest to observe at all remain the obvious place to look next, and nobody has managed sustained aerial observation of them.

Elephants, hoses, and one elephant turning off another’s shower

The elephant result is more recent and considerably funnier, and the comedy is doing real analytical work.

Researchers observed a female Asian elephant at a zoo using a hose as a flexible shower head, and not simply holding it: adjusting her grip and trunk posture to direct the spray at different parts of her body, switching techniques for different regions, apparently handling the hose as a manipulable object with variable behavior rather than as a fixed water source. That is a tool by any reasonable reading of the definition, deployed on the user’s own body, which is the clause in Beck’s formulation people usually forget is there.

Then the second elephant. A companion animal was observed interfering with the water supply, kinking the hose and disrupting the flow while the first elephant was showering, in a pattern the researchers were careful to describe cautiously and which the coverage immediately and irresistibly called a prank. Whether it represents intentional interference or something less interesting is not settled, and the honest reading is that a single individual’s behavior in a captive setting is a weak base for strong claims.

What is not weak is the general point about elephants and tools, which has an anatomical wrinkle. An elephant trunk is roughly forty thousand muscle units with no bone, capable of grip, suction, precision manipulation, and demolition. It is already the most versatile manipulator in the animal kingdom, which means the ecological pressure to extend it with objects is lower than it would be for an animal with less capable hardware. That the populations studied in the wild still use branches as fly swatters, scratch with sticks, and plug water holes anyway is more interesting given that they hardly need to, and the long-term behavioral records from different ecological contexts keep turning up local variation in what they bother to pick up.

The trunk case generalizes into a principle that explains several gaps in the catalog. An animal with an extremely capable native manipulator has less to gain from an external one, which predicts low tool-use rates in elephants and in cephalopods relative to their cognitive capacity, and both predictions hold reasonably well. Run it the other way and the prediction is that tool use should be concentrated in animals whose native anatomy is almost but not quite sufficient for the job, which is a decent description of a chimpanzee facing a termite mound, a crow facing a beetle larva in a hole, and a sea otter facing a shell it cannot crack with its teeth. Necessity is not the mother of invention here. Near-sufficiency is.

The working-animal record contains a version of this too. Elephants employed in Burmese teak extraction and in wartime logistics learned to handle objects and equipment in ways nobody trained explicitly, and the best-documented individual cases come from handlers whose survival depended on noticing what the animal figured out. Those observations were never collected as tool-use data. They were collected as work notes, which is another way material gets lost.

Cockatoos, and the arrival of the tool set

Parrots have quietly become the most productive experimental system in the field, and Goffin’s cockatoos are the reason.

Goffin’s are not tool users in the wild in any documented systematic way, which makes them the same kind of case as Iriki’s macaques: latent capacity without ecological expression. In the laboratory they innovate tools, and in a task requiring the use of one object to control the movement of a second, a setup the researchers called the Golf Club Task, individuals worked out composite tool use, which is the simultaneous coordinated use of more than one tool and which had been reported in very few non-human animals, mostly specific nut-cracking techniques in chimpanzees and capuchins.

More striking, Goffin’s have been shown to transport tool sets. Given a task solvable only with two different implements, and given a distance to cross, birds carried both tools together rather than making two trips, which implies some representation of the requirements of a job that has not started yet. Tool sets were for a long time a signature of great ape technology, particularly the multi-implement termite and honey extraction kits documented in central African chimpanzee populations.

The wild-parrot side keeps producing too. Sulphur-crested cockatoos in Sydney worked out how to open kerbside waste bins and the technique spread geographically as a social innovation, then the same population was documented operating public drinking fountains. And the palm cockatoo, which manufactures a drumstick from a branch and beats it against a hollow trunk with individually distinctive rhythms, remains the only known non-human case of manufactured instrumental sound production, which is a category with an audience of exactly one species and no competitors.

The kea’s reputation for dismantling anything left unattended belongs in the same conversation, and so does the awkward fact that parrots achieve all of this with a beak and one foot, in a brain the size of a walnut with no cortical layers, which is a fairly direct problem for anyone who wants tool use to be a story about cortex.

The tool-set finding deserves one more paragraph because of what it implies about representation. Carrying two implements across a distance to a job you have not started requires holding something about the structure of the task while the task is not in front of you, which is the kind of claim that used to be reserved for apes and which the experimental design was specifically built to test rather than to assume. Birds also adjusted their transport behavior when the task only required one tool, which is the control that makes the result interesting: they were not simply carrying everything available. Chimpanzee tool sets in central African populations had established the behavior in a primate lineage with a plausible evolutionary story attached. Finding it in a parrot removes the evolutionary story and leaves the capacity.

Corvids, and what is actually new

New Caledonian crows have been the flagship for two decades and the recent work has shifted from whether they use tools to how they think about them. The established repertoire is genuinely impressive: hooked tools manufactured from specific plant species, stepped cuts in pandanus leaves, tool selection by task, and in one much-discussed result the assembly of a functional long tool from separate short components that were individually useless, which is compound tool construction with no obvious template.

What has changed is the framing. The species is now studied less as a curiosity and more as a system for asking about planning, memory for tools, and whether the birds represent a tool’s function independently of the specific object. The Hawaiian crow, extinct in the wild and maintained in captive breeding, turned out to be a habitual tool user as well when anyone finally tested it, which is a reminder that absence of evidence in a poorly studied species means very little.

For ravens, the most cognitively flexible of the widely distributed corvids, the tool-use record in the wild is thinner than their reputation implies, and the laboratory record is strong. That gap between wild behavior and demonstrable capacity is now such a consistent finding across corvids, parrots, and macaques that it has stopped being an anomaly and started being the pattern.

The pattern has an uncomfortable implication for how the catalog gets read. If most tested species turn out to have more capacity than their wild behavior displays, then the documented distribution of animal tool use across the tree of life is not a map of ability. It is a map of ecological opportunity crossed with research attention, and the honest version of any such map would need error bars wide enough to swallow most of its own conclusions. The songbirds whose learning has been characterized in the most detail have never been seriously tested for object manipulation, not because anyone thinks they would fail but because nobody has had a reason to ask.

The primate updates are not about sticks anymore

Chimpanzee tool use has been documented since Jane Goodall, so the interesting recent work has moved to the level of material selection and to categories nobody was filing under tools at all.

At Gombe, analysis of termite-fishing implements found that chimpanzees are not grabbing whatever stem is handy. They preferentially select plant species with mechanical properties suited to the job, favoring materials with the flexibility to navigate a curved termite tunnel, which is a materials-engineering decision embedded in a foraging behavior. The long-term study populations in the Mahale mountains show their own local technological traditions, and the between-population variation in what gets used and how is one of the strongest lines of evidence for chimpanzee material culture.

Then the category that is genuinely new. Chimpanzees have been observed catching insects, applying them to open wounds on themselves and on other individuals, which is either topical medicine or something that looks remarkably like it. In 2024 a wild Sumatran orangutan named Rakus was documented chewing leaves of Fibraurea tinctoria, a plant with known antibacterial and anti-inflammatory compounds, and repeatedly applying the resulting material to a facial wound, which then healed without infection. A single individual is a single individual and nobody should build a theory on one orangutan, but the behavior was targeted, repeated, and directed at a specific injury.

Self-medication with an applied substance sits awkwardly against the classical definition, since a chewed leaf poultice is not exactly an unattached object employed to alter another object. It is also obviously the same underlying competence: using something external to change a physical situation. Which brings us back to the definitions doing more sorting than the animals.

The medicinal cases also arrive with a methodological trap attached. A single wild individual doing something once, observed by researchers who were already watching closely, is exactly the observational situation that generates both genuine discoveries and durable myths, and the two are indistinguishable at the time. The appropriate response is neither dismissal nor a press release, but the field’s incentive structure rewards the press release. Rakus may well turn out to be the first documented instance of a widespread behavior nobody had caught. He may also turn out to be one orangutan who happened to chew a leaf near a wound. Both remain live, and the papers involved were considerably more careful on this point than the coverage.

Underwater, and in animals with no hands at all

The invertebrate and aquatic cases are where the concept gets stress-tested hardest, because the anatomy is wrong for everything our intuitions expect.

The veined octopus collects discarded coconut shell halves, carries them stacked beneath its body in an awkward stilt-walking gait that is slower and more costly than normal locomotion, and later assembles them into a shelter. The carrying is the part that satisfies the definition, since the animal is transporting an object at a cost for delayed future use. Octopuses have also been shown to learn to use mirrors to locate food they cannot see directly, a capacity previously demonstrated only in birds and mammals, and there is a documented and thoroughly enjoyable literature on octopuses propelling debris at each other with jets of water.

Among fish, tuskfish carry bivalves to a specific rock and strike them against it repeatedly to break them open, returning to the same anvil site. Under Beck’s definition the fish is arguably not the tool user, since the rock stays put and the clam does the moving, which is exactly the kind of ruling that makes the definition look like a technicality rather than a biological distinction. The reef fish whose cooperative hunting arrangements with moray eels rewrote assumptions about fish social cognition sit in the same uncomfortable zone: obviously doing something sophisticated, awkwardly served by categories built for primates.

Dolphins in Shark Bay wear marine sponges on their rostrums to probe the seafloor without abrading themselves, a tradition transmitted primarily from mothers to daughters, and separately use empty shells to trap and extract fish in a behavior that spreads through the population horizontally rather than by descent. Bottlenose populations elsewhere show their own local behavioral traditions, and the beluga that spent years working the Norwegian coast demonstrated how quickly a cetacean will incorporate human objects into its own behavior when given the chance.

Insects belong in the list and rarely make it. Certain ant species drop soil particles into liquid food to soak it up and carry the saturated grains back to the nest, which is object use for transport with no other interpretation available. Some wasps use small pebbles to tamp down nest closures. Neither behavior involves anything resembling a brain in the sense the rest of this discussion assumes, and both satisfy the definition as written, which is either a reason to revise the definition or a reason to stop treating the definition as a proxy for cognition. It is probably the latter.

Why the animal tool use list keeps growing

Four things are driving the expansion, and only one of them is about animals.

Instrumentation is first and largest. The orca finding came from a drone. Camera traps have produced tool-use records in species nobody could follow on foot. Biologgers and accelerometers detect stereotyped movement patterns in animals underwater and at night. Higher frame rates and better resolution catch fast manipulations that a human observer registers as a blur. Each new sensor produces a wave of first documented reports, and the wave says more about the sensor than the species.

Observer effort bias is second and it cuts both ways. Tool use gets found where people look, and people look at charismatic, accessible, diurnal animals. The distribution of documented animal tool use across the tree of life correlates disturbingly well with the distribution of research funding and field station locations. The Hawaiian crow case is the cleanest demonstration: a habitual tool user that went undocumented because nobody had run the test.

Captivity is third and it is genuinely double-edged. Goffin’s cockatoos, Iriki’s macaques, and a great deal of the strongest experimental work involves animals with time, safety, and nothing to do, which is a condition that reveals latent capacity and also a condition no wild animal occupies. A laboratory result establishes what a nervous system can do. It does not establish that the behavior is part of the species’ natural repertoire, and conflating the two is the most common error in popular coverage.

Definitional drift is fourth, and it quietly reclassifies old observations as new discoveries without anything being discovered.

There is a fifth factor that belongs on the list even though it is awkward, which is that some of the growth is real behavioral change driven by us. The Sydney cockatoos opening waste bins are exploiting an object that did not exist in their environment a century ago, and the technique spread through the population in a documented geographic wave. Urban animals encountering novel manipulable objects at high density are a genuinely new selective and learning environment, and behavior that emerges there is new behavior rather than newly observed behavior. That is a small share of the catalog and it is the only share that reflects animals actually doing something they were not doing before.

What is not on the list of drivers: animals acquiring new cognitive abilities. On the timescale of the last twenty years of publications, essentially none of the growth in animal tool use records reflects the evolution of new capacity. It reflects epistemics, instrumentation, and in a few urban cases a novel object supply.

What tool use predicts, and what it does not

The folk model treats tool use as an intelligence trophy, a rung on a ladder, and the comparative data will not support that reading.

Brain size does not track it. Cetaceans have the largest brains on Earth and, until 2025, no documented tool manufacture at all. Parrots do sophisticated composite tool work in a brain a few grams in mass. Elephants have three times our brain mass and use tools casually rather than centrally. Meerkats and African wild dogs run intricate cooperative societies with essentially no object technology, and the bowerbird constructing and decorating an elaborate display structure is doing something architecturally sophisticated that the definition mostly excludes.

What does predict it is a mundane trio. Manipulative anatomy: a hand, a beak plus a foot, a trunk, a set of arms with suckers. An extractive foraging niche, meaning food that is embedded, encased, or otherwise not immediately available, which is the ecological problem tools solve. And opportunity, in the form of enough slack in the daily energy budget to fail at something repeatedly without starving.

The macaque troop that famously washes its food is a useful check on the intelligence framing, since the behavior spread socially through the population without any object being employed at all, which means the transmission machinery and the tool machinery are separable. Culture does not require tools and tools do not require culture, even though the two travel together often enough that people assume otherwise.

One more correlation deserves killing. Sociality is often invoked as a driver, on the theory that living in groups creates opportunities for observational learning that accelerate technological accumulation. The theory is reasonable and the data are messy. Octopuses are close to asocial and manage object use. Some highly social primates use almost no tools. The cooperative hunters and pack societies that run the most complex coordination in the mammalian world do it entirely without objects. What sociality plausibly does is speed the spread of an innovation once it appears, which is a claim about transmission rather than invention, and the cases where a behavior demonstrably moved through a population support the transmission half while saying nothing about where the innovation came from.

The extended self, taken literally

Put the mechanism and the catalog together and a cleaner picture emerges than the list-of-clever-animals version.

Nervous systems maintain a model of the body: where the limbs are, what they can reach, which region of space counts as adjacent and worth monitoring. That model is not fixed. It updates continuously, and it will absorb an external object that is being actively used to act on the world, remapping receptive fields to the tip of the implement and expanding the defended zone outward. This appears to be a general property of the relevant parietal machinery rather than a specialization, which is why it shows up in an animal that never uses tools in the wild.

On that account, animal tool use is what happens when an existing plastic body model meets a manipulator, an extractive foraging problem, and some free time. It is not a threshold that gets crossed. It is a capacity that gets recruited, which is why it keeps appearing independently in lineages that separated hundreds of millions of years ago and share almost nothing about their neural organization.

The same principle runs the other direction in a way that ought to be uncomfortable. Human prosthetic and brain-interface work depends on exactly this plasticity, on the nervous system’s willingness to treat a manufactured object as part of the body given adequate sensory feedback and practice, and the engineering succeeds precisely to the degree that it exploits machinery a macaque has too. The attempts to push that integration further are not adding a new human capability. They are leaning on a very old vertebrate one.

Which leaves the self as something less solid than advertised. The boundary between organism and environment is not a fact the brain discovers. It is a hypothesis the brain maintains, revises when a rake is in hand, and revises again when the rake is set down. An orca rolling a length of bull kelp along her daughter’s back has, for those fifteen minutes, a body that includes a piece of seaweed. So does anyone who has ever driven a car into a parking space they could feel the edges of. An orca has a body that ends where her attention says it ends, and so do you. The mechanism does not care about the species, and the 24-lecture Neurozoology course works the tree of life on that assumption throughout, from the animals whose knowledge visibly moves between individuals to the working animals whose capacities were discovered by people who needed something from them and the first edition’s survey of what nervous systems are actually built to do.

The list will keep getting longer. Fifty years of watching eighty whales missed a fifteen-minute grooming ritual until a drone went up, which is a reasonable estimate of how much else is being missed right now, in animals nobody has flown a camera over yet. There is a last thing worth extracting, and it concerns what to do with the next headline. When a new species joins the animal tool use catalog, the productive questions are not whether it counts or how smart that makes the animal. They are: what instrument caught it, and how long had people been watching without seeing it. What manipulator is the animal extending, and what was almost-but-not-quite sufficient about it. Whether the object was found, modified, or made before the problem was visible. And whether anyone has tested the obvious neighboring species, or simply not gotten around to it. Those four questions will tell you more about a finding than any amount of argument about the definition, and they are the questions the course applies to every capacity it examines rather than only this one.

The mechanism was always there. We are just finally in a position to catch it running.

Cross the animal kingdom and almost everything about nervous systems turns out to be optional. Centralization is optional, and a sea star gets along without a brain by running a nerve ring and letting the arms argue it out. Neuron count is wildly optional, spanning from a few hundred in a nematode to eighty-six billion in a person to something north of two hundred billion in a pilot whale. Having neurons at all is optional, and a slime mold will still solve a maze. Vision is optional, hearing is optional, a centralized memory store is optional. The tree of life is a catalog of features that some lineage tried, kept, lost, or never bothered with.

Then there is sleep, which nobody has managed to get out of. Jellyfish sleep. Hydra sleep. Nematodes sleep. Fruit flies sleep, and if you keep them awake they die. Fish sleep, reptiles sleep, birds sleep, and the animals with the most obvious reason to skip it, the ones for whom lying around unresponsive is a straightforwardly fatal proposition, did not skip it. They built workarounds. Expensive, elaborate, structurally bizarre workarounds, which is the single most informative fact in the entire field, and it is the reason animal sleep is a better window into what a nervous system is actually for than almost anything else you could study.

The logic runs like this. If sleep were mainly about conserving energy or staying still while it is dark, selection would have edited it out wherever it became dangerous, the way it edits out eyes in cave fish and wings in island birds. Instead, in exactly those lineages, selection did something much harder. It kept the sleep and redesigned the animal around it.

Animal sleep without a brain to sleep with

Start with the definitional problem, because the field had to solve it before it could go anywhere. You cannot ask a jellyfish for a self-report, and electroencephalography requires a brain organized enough to produce a field potential worth recording. So the working definition of sleep is behavioral, and it has four components: a sustained period of reduced responsiveness to the outside world, rapid reversibility on sufficient stimulation, a species-typical posture or location, and homeostatic rebound, meaning that if you prevent it, the animal does more of it afterward.

That last criterion is the one doing the heavy lifting. Reduced responsiveness alone is just a coma, or a rock. Rebound implies a regulated internal quantity, something being tracked and repaid, which is what separates sleep from mere inactivity and which is why the criterion appears in every serious study of the subject.

Run that test set and the results are unsettling. Cassiopea, the upside-down jellyfish, passes it. Pulsation rate drops during a nightly quiescent period, the animals are slower to respond to being dropped in the water column, they can be roused, and depriving them with pulses of water produces a rebound the following day. Cassiopea has no brain, no central ganglion, nothing but a diffuse nerve net. Hydra passes as well. Caenorhabditis elegans, running about three hundred neurons, has a quiescent state during developmental molts and another triggered by cellular stress, both with sleep-like signatures.

The implication is not subtle. Sleep predates brains. Whatever it is doing, it is being done at the level of cells and small networks rather than requiring a centralized organ, which reframes the whole question. This is not a luxury that complex animals invented once they could afford it. It is closer to a maintenance requirement that came bundled with excitable tissue, and the elaborate architecture in mammals and birds is a later renovation on a much older foundation, in the same way that a modern city’s water system is a series of upgrades wrapped around Roman-era assumptions about gravity.

The fruit fly is where the mechanistic work got traction, because flies are the one sleeping animal you can run genetics on at scale. Drosophila shows consolidated nightly quiescence with elevated arousal threshold, rebound after deprivation, and pharmacological responses that track vertebrate ones: caffeine reduces sleep, antihistamines increase it. Mutations in a potassium channel gene produce flies that sleep a fraction of the normal amount. A specific cluster of neurons in the fly’s dorsal fan-shaped body behaves like a sleep switch, firing more when sleep pressure is high and inducing sleep when artificially activated. Sustained deprivation kills them. That combination, a genetically tractable animal with a conserved sleep phenotype and a lethal deprivation endpoint, is why a great deal of what is now known about the molecular basis of animal sleep came out of an insect rather than a mammal.

Why “rest” was never a good enough answer

The energy-conservation account of sleep has an arithmetic problem. Measure the metabolic savings of a sleeping mammal against quiet wakefulness and you get something in the range of a modest single-digit percentage reduction, which is roughly the caloric equivalent of skipping a slice of bread. Set that against the cost side of the ledger: hours per day of zero foraging, zero mating, zero territorial defense, and substantially degraded predator detection. As a trade it is terrible. Nobody would sign that contract for a five percent discount.

Which means the benefit has to be something that cannot be obtained while awake, and this is where the comparative evidence becomes an argument rather than a catalog. Consider the animals for whom sleep is most expensive and watch what evolution actually did.

A cetacean is a mammal that breathes air and lives in water, so unconsciousness carries a drowning risk that no terrestrial animal faces. A newborn dolphin cannot afford to be unresponsive at all in its first weeks, and neither can its mother. An albatross or a frigatebird on a multi-week foraging flight over open ocean has nowhere to lie down. A migrating songbird crossing the Gulf of Mexico is in the same position. Each of these is a case where the cost of sleep spikes toward lethal, and in each case the lineage did not respond by abolishing sleep. It responded by fragmenting it, halving it, compressing it, or relocating it, at considerable engineering expense.

That is the closest thing the field has to a controlled experiment on the necessity of sleep, and it ran for tens of millions of years across multiple independent lineages. The verdict is consistent. Sleep is not a behavior animals do because they have time. It is a behavior they make time for, and when they cannot make time, they find increasingly baroque ways to take it in installments.

The deprivation evidence points the same direction from the other end. Rats kept awake by sustained forced activity die within weeks, and the cause of death has been difficult to pin to any single organ failure, which is itself informative: something diffuse and systemic goes wrong rather than one subsystem breaking. Flies die. In humans, the fatal familial insomnia prion disease destroys the thalamic circuitry that generates sleep and is uniformly lethal, which is about as close to a controlled demonstration of necessity as ethics will ever permit. Whatever animal sleep is repaying, the debt is not optional and it does not get forgiven.

Sleeping with half a brain at a time

Unihemispheric slow-wave sleep is the marquee workaround and it is stranger than the summary version suggests. In cetaceans, one cerebral hemisphere shows the high-amplitude slow waves of deep sleep while the other shows waking activity, with the corresponding eye typically closed on the sleeping side and open on the waking side. The hemispheres then swap. The animal keeps swimming, keeps surfacing to breathe, keeps some level of vigilance, and still logs slow-wave time in each half of its brain across the day.

Bottlenose dolphins do it, and it is why the persistent claim that dolphins never sleep is precisely wrong rather than approximately wrong. They sleep constantly, just never all at once. The same architecture appears in porpoises, in belugas, and in the sperm whales that also do something entirely different, hanging motionless in vertical formation in the upper water column in what looks like a full-body shutdown, which the observational record on deep-diving cetaceans captured almost by accident when a research vessel drifted into a group that failed to notice it.

The fur seal is the case that makes the mechanism legible, because the same individual switches modes depending on where it is. On land it sleeps bilaterally, both hemispheres in slow-wave sleep, like any ordinary mammal. In water it switches to asymmetric sleep, one hemisphere at a time, with the flipper on the waking side continuing to paddle. Same animal, same brain, two configurations, selected by environment. That is not a fixed adaptation. That is a runtime setting.

The mechanism that makes this possible is worth pausing on because it is not obvious that a brain could do it. Slow-wave activity is a global synchronization phenomenon, and the two hemispheres are connected by a large fiber tract whose entire job is keeping them coordinated. Running deep sleep on one side while the other stays awake means suppressing that coordination selectively, which in cetaceans appears to involve both a reduced interhemispheric connection relative to terrestrial mammals and active regulation of the arousal systems projecting to each side. Asymmetric animal sleep is therefore not simply a matter of letting one half drift off. It requires machinery for keeping the halves apart, and that machinery had to be built.

Birds do a version of it too, and they do it under conditions that reveal the logic. Mallards sleeping at the edge of a group keep the eye facing away from the group open more often than birds in the middle, which is a vigilance allocation problem being solved with hemisphere assignment. The corvids that dominate the cognition literature and the parrots that rival them both show asymmetric eye closure, and so do the long-distance migrants whose entire life history is organized around not being caught out in the open.

The thing worth noticing is what unihemispheric sleep concedes. If sleep could simply be skipped, none of this machinery would exist. Building a brain that can run two incompatible global states simultaneously, with the corpus callosum somehow not smearing them together, is a hard problem. Lineages solved it rather than dropping the requirement.

Sleeping in flight, in units of seconds

The frigatebird result is the one that reset expectations. Great frigatebirds spend weeks continuously airborne over open ocean, and instrumenting them with miniature electroencephalography loggers showed that they do sleep in flight, in both unihemispheric and bilateral bouts, often while circling in rising air. The remarkable part is the quantity. On land the birds slept on the order of twelve hours a day. In flight they slept on the order of forty-five minutes a day, in bouts averaging seconds, and they showed no obvious rebound crash on return.

Take that seriously and it complicates the tidy story. If a bird can operate for weeks on three quarters of an hour of fragmented sleep a day, then either its sleep is dramatically more efficient than ours, or the daily quantity most animals take is substantially above the minimum requirement, or the deficit is being paid down in some way the measurements did not capture. All three possibilities are interesting and the field has not settled which is operating. The measurement problem deserves a flag too: recording animal sleep in a bird the size of a football, mid-ocean, with a logger light enough not to change its flight, means accepting fewer channels and coarser resolution than a laboratory setup, so the possibility that brief or shallow states went undetected is real rather than rhetorical.

Swifts appear to stay aloft for months at a time. Some migratory songbirds shift the architecture of their sleep during migration season, taking many more brief bouts and adding daytime napping, and captive birds in migratory condition show reduced sleep without the cognitive degradation you would predict from equivalent deprivation in a non-migratory period. Sandpipers on Arctic breeding grounds have been found to sleep very little during the competitive mating window, with the least-sleeping males siring the most offspring, which is a fairly direct fitness argument against the idea that sleep quantity is rigidly fixed.

Cross-domain comparison worth making: this looks less like a hard constraint and more like a variable-rate obligation, the way a mortgage can be restructured but not forgiven. The payment schedule flexes enormously. The principal does not go away.

What the flight cases collectively establish is that the daily quantity of animal sleep is far more elastic than the requirement for it. A frigatebird can compress twelve hours into forty-five minutes for weeks. A sandpiper can nearly suspend the whole business for the length of a breeding season. Neither can abolish it, and neither does so permanently. Elasticity within a season and inviolability across a lifetime are two different findings, and conflating them produces most of the bad takes about whether humans really need eight hours.

The animals that barely sleep and the ones that cannot stop

The interspecies range in sleep duration is enormous and the pattern in it is weaker than most summaries admit. Wild African elephants, monitored with implanted actiwatches and collars, slept something like two hours per day, mostly standing, lying down only every few days, and went as long as forty-six hours without sleep after apparent disturbance. Giraffes come in low as well. At the other end, some bats sleep sixteen hours or more, as do certain rodents and armadillos.

The tempting explanation is body size and metabolic rate, and it captures part of it. Larger herbivores need more hours grazing and are more exposed while recumbent, so their sleep gets short and vigilant. Small animals with high mass-specific metabolic rates sleep more. But the correlations are loose, the confounds are severe, and captive measurements have systematically overestimated sleep in exactly the large herbivores where the wild data later came in low, which is a useful reminder that a lot of the older comparative sleep literature was measuring animals in enclosures with no predators and nothing to do.

The elephant cognition literature is instructive on this point, because the same species that reliably shows up in discussions of memory and social knowledge is also running on roughly a quarter of the sleep a human needs, in an animal with a brain three times the mass of ours. Whatever the relationship is between sleep quantity and cognitive capacity, it is not a simple dose-response curve, and the populations studied under different ecological pressures do not converge on a single number.

The developmental cases are stranger still. Bottlenose dolphin and killer whale calves, along with their mothers, show almost no conventional rest in the first weeks after birth, remaining continuously active in a period when terrestrial mammal infants sleep most of the day. Sleep then increases with age, which is the opposite of the mammalian norm. Nobody has fully resolved how a developing cetacean brain gets whatever developing brains normally get from sleep while apparently not sleeping, and the honest position is that the observation is solid and the explanation is not.

Two-stage sleep evolved at least twice, probably more

For decades the alternation between slow-wave sleep and rapid eye movement sleep was treated as a mammal-and-bird arrangement, a signature of endothermy and a big forebrain. That has not survived contact with the last ten years of evidence.

The octopus result is the cleanest demonstration, and it is worth the detail. Octopuses show quiet sleep punctuated roughly every hour by bouts lasting about sixty seconds in which the arms and eyes twitch, breathing quickens, muscle tone changes, and the skin erupts into rapidly shifting color and texture patterns. Work published in Nature on wake-like skin patterning and neural activity during octopus sleep established that these bouts are homeostatically regulated, rapidly reversible, and accompanied by an elevated arousal threshold, which is what qualifies them as a genuine second sleep stage rather than restlessness. The skin patterns during these bouts closely resemble patterns the animals produce while awake. During quiet sleep, the recordings showed waveforms resembling mammalian sleep spindles, localized to brain regions associated with learning and memory.

The evolutionary distance is the point. The lineages leading to octopuses and to vertebrates separated on the order of five hundred and fifty million years ago, and octopus brains are organized nothing like ours, with the majority of neurons distributed into the arms and the central brain wrapped in a doughnut around the esophagus. Two-stage sleep in that architecture is not inheritance. It is convergence, which means the two-stage arrangement is solving a problem that recurs whenever you build a sufficiently complex nervous system, regardless of how you build it.

Cuttlefish show a comparable active stage. The Australian bearded dragon cycles between two states at roughly eighty-second intervals, far faster than mammals, in a forebrain structure that is not a cortex. Zebrafish show two states with signatures analogous to slow-wave and rapid eye movement sleep, in a fish, without a cortex at all. The fish whose behavioral repertoire keeps surprising researchers are running sleep architecture that textbooks reserved for warm-blooded animals a generation ago.

Replay, dreaming, and the line between them

Here is where care is required, because this is the point at which reporting reliably outruns evidence.

In rats, hippocampal place cells that fire in a particular sequence while the animal runs a track fire in compressed versions of that same sequence during subsequent slow-wave sleep, sometimes forward, sometimes reversed. In zebra finches, neurons in the song motor pathway that fire in specific patterns during singing fire in matching patterns during sleep, in a bird that is not singing. Both are robust, replicated, and genuinely remarkable findings about offline neural activity.

Neither is a demonstration of dreaming. Replay is a claim about information processing: patterns of activity recur offline in a manner consistent with memory consolidation and with the strengthening or pruning of specific synapses. Dreaming is a claim about subjective experience, about there being something it is like to be that animal during that activity. The first is measurable with electrodes. The second is not measurable with anything currently available, and the gap between them is not a technical detail to be cleaned up later. It is the central difficulty of the entire subject.

The songbird case is especially clean because the behavior is so well characterized. Young birds learning their species song do something that looks like practice, and the regional song dialects that make sparrow populations distinguishable by ear are the product of a learning process with a sensitive period, a template, and a long refinement phase. Sleep is implicated in that refinement, with song structure degrading overnight and recovering with morning practice in a pattern that suggests offline reorganization. That is a strong mechanistic story about learning. It says nothing about whether the bird experiences anything while it happens.

Keeping those two claims separate is the whole discipline. Almost every overstatement in this field is the result of quietly sliding from the first to the second.

There is a further complication that gets skipped. Replay is not a recording. The reactivated sequences are compressed by an order of magnitude relative to the original experience, they run backward as often as forward, and they include trajectories the animal never actually took, novel paths assembled from fragments of real ones. If you were looking for a neural correlate of something dreamlike, that last detail is the most suggestive one available, since a system generating routes it has not traveled is doing something closer to simulation than to playback. It is also the detail that most resists interpretation, because a planning mechanism and a dreaming mechanism would look identical at the electrode.

So: do animals dream? The question is either easy and uninteresting or interesting and unanswerable, depending on what you mean.

If dreaming means wake-like brain activity occurring during sleep, with sensory and motor patterns recurring offline, the answer is yes and it is thoroughly documented in mammals, birds, at least one lizard, at least one fish, and at least one cephalopod. If dreaming means an experienced narrative that the animal would report if it could, there is no experiment on offer that distinguishes an animal having such an experience from an animal not having it, and pretending otherwise does the field no favors.

What we can do is look for behavioral evidence that constrains the possibilities. Cats with lesions to the brainstem circuitry that normally paralyzes muscles during rapid eye movement sleep act out apparent behaviors while asleep: stalking, pouncing, grooming at nothing. Dogs twitch and vocalize in ways owners find obviously interpretable and which are, at minimum, consistent with motor programs running without inhibition. Sleeping octopuses producing wake-like skin patterns are doing something structurally similar, since those patterns are ordinarily deployed in specific behavioral contexts like camouflage or threat display, and seeing them generated during sleep is seeing a behavioral program run offline.

That is suggestive. It is not proof, and the specific inference that an octopus generating a camouflage pattern in its sleep is dreaming about camouflaging is a leap that the researchers involved were notably more careful about than the headlines that followed. The reasonable position is that the machinery associated with dreaming in humans is present and active in a wide range of animals, that this raises the probability that something experiential accompanies it, and that the probability is not a measurement.

What animal sleep is actually doing down there

Multiple mechanisms are on the table and they are not mutually exclusive, which is worth stating because the field is often presented as a competition with a winner pending.

Synaptic homeostasis holds that waking potentiates synapses broadly, which is metabolically and informationally unsustainable, and that slow-wave sleep globally downscales synaptic strength while preserving relative differences, restoring capacity to learn. The theory makes testable predictions about slow-wave activity tracking prior waking duration, and those predictions have largely held up.

Metabolic clearance holds that sleep facilitates removal of waste products from neural tissue, with work in mice reporting increased interstitial space and enhanced clearance of solutes during sleep. This one deserves an asterisk, because subsequent studies using different methods have reported results pointing the other direction, and the mechanism is genuinely contested rather than settled. Anyone presenting brain-washing-during-sleep as established fact is ahead of the evidence.

Memory consolidation is the best-supported functional account, tied directly to the replay findings, with slow-wave sleep implicated in transferring and stabilizing information and rapid eye movement sleep implicated in integration and in the pruning of weaker associations.

The most interesting recent thread runs through DNA repair. Work in zebrafish found that neurons accumulate DNA double-strand breaks during waking, that the accumulation itself appears to drive sleep pressure, and that sleep permits chromosome dynamics and repair activity that waking suppresses. That is a candidate answer to the question of why sleep cannot be done awake, and it is the kind of cell-level mechanism that would explain why a jellyfish with no brain still needs it. Whether it generalizes is open.

Immune function and thermoregulation both have partial claims as well, and the thermoregulatory account has an interesting comparative angle: rapid eye movement sleep involves a suspension of normal temperature regulation, which is metabolically risky and appears in reduced quantities in animals facing cold stress. None of these accounts is likely to be the single answer, and the reasonable expectation is that animal sleep is doing several unrelated jobs that happen to share a scheduling requirement, the way a maintenance window at a factory gets used for cleaning, calibration, and inventory simultaneously because that is when the line is stopped.

Notice the shape of all four accounts. Each describes a maintenance operation that competes with normal function for the same hardware, which is why it has to be scheduled separately. The nervous system is time-sharing, and sleep is the maintenance window. Anyone who has watched a system get taken offline for patching at three in the morning has the right intuition.

The consciousness ledger, as of now

Sleep research and consciousness research meet because sleep is the one thing that reliably switches consciousness off and on in an intact animal, which makes it the closest thing to an experimental handle the subject has.

The formal state of expert opinion shifted recently and it is worth reading precisely rather than in summary. The 2012 Cambridge Declaration asserted that humans are not unique in possessing the neurological substrates that generate consciousness, naming mammals, birds, and octopuses. In April 2024 a larger interdisciplinary group issued the New York Declaration on Animal Consciousness, which states that there is strong scientific support for attributing conscious experience to mammals and birds, that the empirical evidence indicates at least a realistic possibility of conscious experience in all vertebrates including reptiles, amphibians, and fishes and in many invertebrates including cephalopod mollusks, decapod crustaceans, and insects, and that where such a realistic possibility exists it is irresponsible to ignore it in decisions affecting that animal.

Read the hedging, because the hedging is the content. “Strong scientific support” for mammals and birds is not the same claim as “realistic possibility” for a bee, and the declaration’s own authors were explicit that they were not asserting insects obviously are conscious, only that the probability is high enough to warrant research and precaution. That is a careful, calibrated, deliberately modest document, and a great deal of coverage flattened it into scientists declare insects conscious, which is not what it says.

Underneath the declaration the theoretical situation is unresolved in a way that matters. Integrated information theory and global workspace theory make different predictions about which architectures support experience, an adversarial collaboration between them produced results that neither camp accepted as decisive, and a public letter from a large group of researchers arguing that integrated information theory should be classified as unfalsifiable did more to demonstrate the field’s condition than to resolve it. Meanwhile the anatomical assumptions keep loosening: work on the avian pallium has identified circuit organization comparable in important respects to mammalian cortex, and recordings from crows have identified neural activity correlating with reported sensory awareness in a brain with no cortical layers at all. The engineering side of neuroscience has the same problem from the other direction, and so does every serious attempt to ask whether an artificial system could have experiences: no agreed test, no agreed criteria, strong intuitions on all sides.

The claims that do not survive contact with the evidence

An audit, because the popular version of this subject is unusually contaminated.

Dolphins never sleep is false and instructively so. They sleep in half-brain installments, continuously, which is more remarkable than not sleeping would be. The related claim that sharks never sleep is also poorly supported. Some species must maintain forward motion for ram ventilation, but others pump water over their gills while stationary and show sustained reduced-activity states, and recent metabolic work on at least one species reported the reduced metabolic rate and postural signature consistent with sleep.

Rapid eye movement sleep equals dreaming is wrong in both directions. Human dream reports occur outside that stage, and the presence of the stage in an animal establishes the architecture rather than the experience.

Animals do not dream, the reflexive skeptical position, is no better supported than its opposite. The machinery is present and active. Refusing to draw the inference is a defensible stance; asserting the negative as established is not, and the symmetry here is the part people miss. Anthropomorphism has a mirror-image failure that consists of denying an animal a capacity we would readily grant a human showing identical evidence, and both errors are errors.

The three-second goldfish is the other one worth killing, since it turns up in the same conversations. Goldfish form associations that persist for months, learn to navigate mazes, retain conditioned avoidance across long intervals, and can be trained to press levers at particular times of day. The claim was never based on evidence, and the animals whose spatial and social memory has been documented most carefully happen to be fish, which is a fairly complete inversion of the folk belief.

Sleep is for resting the body fails on its own terms, since muscles recover perfectly well during quiet wakefulness, and the persistence of sleep in animals for which it is dangerous is the evidence against it.

More sleep means more intelligence does not survive the comparative data. Elephants sleep two hours, chimpanzees sleep nine or so, macaques around ten, and armadillos considerably more than any of them. The correlation people expect is not there, and the reason is that sleep quantity is set by ecology and metabolism at least as much as by whatever cognitive maintenance it performs. What does show a relationship, more robustly than total duration, is the proportion of sleep spent in the active stage, which tends to run higher in species with more altricial young and more postnatal brain development. That is a finding about developmental schedules rather than about intelligence, and it is routinely misreported as the latter.

What the sleeping animal is telling us

The most defensible summary of animal sleep is also the least dramatic. Sleep is old, older than brains. It is a cellular and network-level requirement that appears wherever excitable tissue does, it competes with waking function for the same substrate, and it is therefore scheduled rather than continuous. Complex nervous systems elaborated it into staged architecture, and they did so more than once, in lineages that separated before there were vertebrates, which tells us the staging is a solution to a recurring problem rather than a family trait.

The workarounds are the strongest evidence of necessity. Half-brain sleep in dolphins and belugas, mode-switching in fur seals, seconds-long bouts in frigatebirds, restructured architecture in migrating birds, near-elimination in breeding sandpipers and in cetacean calves: every one of these is a lineage paying an enormous engineering cost to keep something it could not discard. Selection had every opportunity to delete sleep in the animals where it hurts most. It did not take the opportunity once.

On dreaming and consciousness the honest ledger is shorter than anyone wants. We can measure offline neural activity and we do, across birds, fish, reptiles, social carnivores, primates, and cephalopods. We can document behavior consistent with motor programs running without inhibition. We can note that expert opinion has moved substantially toward attributing experience more widely, in carefully hedged language. We cannot get inside, and no instrument currently proposed would get us there.

It is worth naming what would move the needle, since a research program that cannot specify its own evidence is not a research program. Convergent behavioral markers would help: an animal that reports, in some trained response, on the presence or absence of a stimulus it was not otherwise trained to discriminate, which is the logic behind the no-report paradigms now used in human consciousness work and behind the crow experiments that adapted them. Sleep-specific versions would help more, and nobody has designed a workable one. Until somebody does, the study of animal sleep will keep producing excellent mechanism and careful silence about experience, and the careful silence is the professional part.

That limitation is not a failure of the science. It is the actual shape of the problem, and treating it as the shape of the problem rather than as a temporary gap is what separates the work that will hold up from the work that will not. The 24-lecture Neurozoology course runs the whole tree of life on that principle, from the ravens and parrots that keep failing to be as simple as advertised to the cetacean dialects that look like culture, the bowerbird building something for reasons of its own, and the animals whose working relationships with people revealed capacities nobody had tested for. It is the same instinct that runs through the study of how knowledge moves between animals and through the first edition’s tour of nervous systems: the mechanism is the marvel, and the mechanism is usually more interesting than the story people tell about it.

There is one more asymmetry worth carrying out of this. Every other capacity in comparative neuroscience is something an animal has: a sense, a memory system, a behavioral repertoire, a neuron count. Animal sleep is the only one that is a thing an animal has to stop doing everything else in order to get, which makes it the only capacity whose cost is measured in foregone life. Elephants pay for it in grazing hours. Frigatebirds pay for it in altitude and attention. Cetacean mothers pay for it by restructuring an entire hemisphere’s worth of neural coordination. That price is the argument, and it is a price paid in every lineage that has ever been examined.

A jellyfish with no brain gets sleepy, and repays the debt the next day. Start there and the question stops being which animals are enough like us to matter, and becomes what kind of thing a nervous system is, that it should need to be taken offline at all.