An African elephant carries about 257 billion neurons. A human carries about 86 billion. On the crudest possible reading of comparative neuroscience, the argument ends there and the elephant wins by a factor of three, and elephant cognition should be three times whatever ours is.
Then you open the brain and find that 97.5 percent of those neurons are in the cerebellum. Roughly 251 billion of them sit in a structure at the back of the skull, leaving the cerebral cortex, which has twice the mass of ours, holding only 5.6 billion neurons against our 16.3 billion. The elephant hippocampus is slightly larger than a human hippocampus by volume and contains around 37 million neurons where the human hippocampus and amygdala together hold something near 250 million.
That distribution is not a rounding error or a quirk of one specimen. It is the single most informative fact about elephant cognition, and it makes the animal a permanent problem for anyone who wants brain size to mean something simple. Every other mammal examined concentrates most of its neurons in the cerebellum, but never much beyond eighty percent. The elephant is an outlier on its own axis.
The temptation is to read this as a deficit, and that reading has been made. It is more interesting and more accurate to read it as a specification. The elephant did not build a larger version of our brain. It built a different machine, for a different body, solving a different set of problems, and the cerebellar investment is the price of the appendage on the front of its face.
What elephant cognition spends 251 billion cerebellar neurons on
The trunk is the reason, and the numbers on it are as striking as the numbers on the brain.
An elephant trunk is a muscular hydrostat with no bones and no joints, containing on the order of 40,000 muscle units, capable of lifting several hundred kilograms and of picking up a single tortilla chip without breaking it. It functions as a nose, a hand, a snorkel, a hose, a weapon, a social organ, and a low-frequency sound emitter. African elephants have two prehensile fingers at the tip; Asian elephants have one and use a different grasping strategy as a result. Recent biomechanical work has characterized how the trunk manages this, showing that elephants create pseudo-joints, temporarily stiffening segments of the trunk to convert a continuously flexible structure into something with discrete bending points, which reduces the control problem by imposing joints where none exist anatomically. That is a nervous system simplifying its own task by changing the mechanics of the thing it is controlling.
Controlling that is a computational problem of a specific kind. A limb with joints has a small number of parameters: this many degrees of freedom, specify the angles, done. A structure with no joints can bend anywhere along its length in any direction, elongate, shorten, stiffen, and twist, which means the number of parameters a controller would need to specify explodes. The cerebellum is the structure vertebrates use for coordinating and refining movement, and the elephant loaded it accordingly.
The full cellular accounting of the African elephant brain established the numbers by dissolving the tissue and counting nuclei directly rather than estimating from volume, which is why the figures are unusually trustworthy for a single specimen. Anatomical work has found the elephant cerebellum to be not merely large but structurally unusual, with more folding, larger and more numerous Purkinje cells, and an internal organization that suggests specialization rather than simple scaling. Facial motor nuclei are enlarged and the trunk’s representation dominates them, which is the same principle as the outsized hand and lip regions in a human sensorimotor map, applied to an organ with vastly more independent parts.
The trunk also carries a dense array of mechanoreceptors, and the tip has fine tactile discrimination that supports object recognition by touch alone. So the cerebellar investment is not only motor output. It is a sensorimotor loop of enormous bandwidth attached to a single organ, and the animals that solved a comparable jointless-manipulator problem with a segmented nerve cord in each arm arrived at a different architecture for the same computational difficulty.
The two solutions are worth setting side by side because they are the only two examples available. An octopus pushed control outward, distributing two-thirds of its neurons into the arms themselves and letting a small central brain issue high-level intent. An elephant kept control centralized and simply built an enormous dedicated coprocessor at the back of the skull. Same problem, opposite architectures, and the difference plausibly comes down to the number of limbs: eight independent manipulators argue for local autonomy, one argues for concentrated bandwidth. Whichever way it goes, a jointless appendage is expensive, and both lineages paid in the currency of neurons.
The correct summary is that elephant neuron counts do not measure what people want them to measure, which is the point the comparison of cortical neuron counts across species makes in more detail. Total neurons is a body-and-brain number. Cortical neurons is a closer proxy for the kind of processing the question usually intends, and on that measure the elephant sits well below humans and in the range of large primates despite the enormous brain.
Two more anatomical details complicate any simple ranking. Elephant cortex is thin, around 1.5 millimeters against roughly 2.5 in humans, with lower neuron density, which means the cortical mass advantage does not convert into cells. And elephants possess von Economu neurons, the large spindle-shaped cells found in humans, great apes, and cetaceans and once proposed as a substrate for social cognition, in a lineage that acquired them independently. The convergence is real and the functional story attached to it has never been nailed down, which makes them a good example of an anatomical marker that gets more explanatory weight in popular accounts than the evidence supports.
Names, and the 2024 result
The most consequential recent finding in elephant cognition concerns communication, and it is unusually well designed.
Elephants produce rumbles with fundamental frequencies extending into infrasound, below the range of human hearing, which propagate over distances of several kilometers and travel further at dawn and dusk when temperature inversions favor transmission. Some of that energy also travels as seismic waves through the ground, and elephants appear to detect ground-borne vibration through their feet and through bone conduction, which gives the system two channels.
Working with recordings collected in Kenya’s Samburu reserve and Amboseli National Park between 1986 and 2022, researchers applied machine learning to 469 calls involving 101 callers and 117 receivers. A model could predict, better than chance, which individual a given rumble was directed at. Then came the test that matters: they played calls back to elephants. Animals responded more strongly and more quickly to rumbles that had originally been addressed to them than to rumbles addressed to others, even out of the original context.
The finding that African elephants address one another with individually specific name-like calls matters because of a specific structural detail. Dolphins and parrots also address individuals, and they do it by imitating the target’s own signature call, which is a bit like getting someone’s attention by doing an impression of them. The elephant data suggest no such imitation. The label appears to be arbitrary with respect to the receiver, which is how human names work and which had not previously been documented outside our species.
The hedging in the original paper is worth preserving. Names did not appear in all calls. The model’s accuracy, while above chance, left a large majority of calls unclassified. And the authors used the word suggest rather than demonstrate on the arbitrariness point. What is solid is the playback result: elephants distinguish calls meant for them from calls meant for others.
The convergence argument is the part worth carrying. Vocal labeling for individuals has now been reported in bottlenose dolphins, in some parrots, in marmosets, and in elephants, which is four lineages that separated from each other tens of millions of years ago and share a specific profile: long lives, fission-fusion or otherwise fluid social groups where individuals are frequently out of sight of each other, and vocal production learning. Those three conditions together appear to be what generates a naming system, and the cetacean societies whose signature whistles function the same way meet all three. That is a hypothesis with a testable prediction: any animal meeting those conditions should be worth checking, and several have not been.
The matriarch as infrastructure
Elephant social structure is fission-fusion, built around a core family of related females and their offspring led by the oldest female, with bond groups and clans as larger nested units, and with males dispersing at adolescence into a looser and less studied social world of their own. Association patterns within a clan can be mapped as a network, and the resulting structure is genuinely multi-tiered in the way human and cetacean societies are, with individuals maintaining differentiated relationships across hundreds of animals rather than simply belonging to a group.
Playback experiments established what the matriarch actually contributes, and the design is elegant. Play a recording of an unfamiliar elephant’s contact call to a family group and watch how they respond. Families with older matriarchs were better at distinguishing familiar from unfamiliar calls and adjusted defensive bunching accordingly. Groups led by older females also responded more appropriately to playbacks of lion roars, showing stronger defensive behavior toward the roars of male lions, which are considerably more dangerous to elephants than female lions.
That is discrimination the younger matriarchs failed to make, and it means the matriarch is not a figurehead. She is a stored model of the social and physical environment: who is a threat, who is a stranger, which water sources persist through which droughts, which routes are safe. Older matriarchs correlate with higher reproductive success across the family, which is the fitness consequence of holding that model.
The drought evidence is the sharpest version. During a severe East African drought in the early 1990s, family groups whose matriarchs were old enough to remember a comparable drought decades earlier left the park for better conditions, and those groups suffered substantially lower calf mortality than groups led by younger females who stayed. A memory of an event roughly thirty-five years old, held in one animal, determined whether calves in that family lived. There are not many findings in comparative cognition where the fitness consequence of a specific memory can be counted that directly.
Vocal learning adds another dimension that is easy to miss. Elephants are among the small number of mammals capable of producing novel sounds by imitation, with documented cases including an Asian elephant that reproduced Korean words with recognizable formant structure by placing his trunk in his mouth to modify vocal tract shape, and an African elephant that imitated truck sounds. Vocal production learning is rare, appearing in cetaceans, bats, pinnipeds, elephants, and humans among mammals, and it is a prerequisite for anything like an arbitrary naming system.
The corollary is the one with teeth. That knowledge is not distributed and it is not written down. It exists in one animal, and poaching removes the oldest individuals preferentially because they carry the largest tusks. Groups that have experienced culling or heavy poaching show disrupted social knowledge decades later, in animals that were calves at the time.
The populations tracked across the Okavango and the ones studied under the very different pressures of Tsavo demonstrate how much of what an elephant knows is local and learned rather than general and inherited. The same vulnerability shows up in cetacean populations whose foraging traditions live in specific individuals and in migratory birds whose routes had to be re-taught by aircraft after the knowledgeable animals were gone.
Death, bones, and a behavior nobody can explain away
Elephants do something around dead elephants that they do around nothing else, and the observational record is now large enough that it cannot be dismissed as anecdote.
They investigate carcasses of their own species with sustained attention, touching the body with trunk and feet, particularly the face and tusks. The behavior appears across all three species and across sites with no contact between populations, and it appears in individuals encountering a carcass for the first time, which argues against it being a local tradition. They return to sites where individuals died. They show heightened interest in elephant bones and ivory encountered in the landscape, handling and turning them, and controlled presentations found they attend to elephant skulls and ivory considerably more than to the skulls of other large species or to comparable objects. Mothers have been observed remaining with dead calves for days, sometimes carrying them. Individuals have been recorded covering bodies with soil and vegetation.
What is genuinely established: the behavior is specific to conspecifics rather than being general investigation of novel objects, it is directed disproportionately at the parts of the body that carry identity, and it persists over long timescales.
What is not established is anything about what the animal understands. Whether elephants have a concept of death, whether they grieve in any sense that maps onto the human experience, and whether the bone interest reflects recognition of a specific individual are all open, and the honest position is that they may not be answerable with available methods. The comparative literature has the same problem with primates that carry dead infants for weeks, and in both cases the behavior is unmistakable and the interpretation is not.
Mirror self-recognition sits nearby and has its own caveats. Asian elephants were tested with a large mirror, and at least one individual, Happy at the Bronx Zoo, repeatedly touched a mark on her head visible only in the reflection. That is a pass. It is also one animal out of three tested, and the small sample is a real limitation that gets lost in the retelling. The species that pass this test at all form a short and taxonomically scattered list: great apes, elephants, some cetaceans, magpies with a failed replication attached, and a handful of contested cases. Whatever the capacity is, it did not arrive once and get inherited, and the birds whose results on the same test remain disputed illustrate how much weight a single paradigm has been asked to carry.
Infrasound, seismics, and a sensory world tuned low
The communication system deserves treatment as engineering, because elephants are running channels most animals do not have access to.
Vocal production happens in a larynx scaled to the animal, with vocal folds long enough to vibrate at frequencies down into single-digit hertz. Long wavelengths diffract around obstacles and attenuate slowly, so infrasound propagates through forest and across savanna in ways that higher-frequency calls cannot. Under favorable atmospheric conditions the range extends to several kilometers, and elephants appear to time long-distance calling toward dawn when inversions extend it.
The seismic channel is the stranger one. Rumbles couple into the ground and propagate as surface waves at speeds different from airborne sound, and elephants adopt postures consistent with attending to ground vibration, leaning forward and lifting a foot. The proposed reception routes are Pacinian corpuscles in the feet and bone conduction through the forelimb to the middle ear. Whether elephants extract directional or content information from the seismic channel or simply detect it remains under investigation.
There is a practical consequence to running a communication system in a frequency band humans cannot hear, and it shaped the science. For most of the twentieth century, observers watching elephants coordinate movement across kilometers of savanna with no audible signal attributed it to something unexplained, and the infrasonic channel was not identified until the 1980s, when a researcher noticed a throbbing in the air near captive elephants and thought to check below the audible range. An entire communication system was invisible for as long as it was because the instrument used to detect it was a human ear. The animals whose signal repertoires turned out to be far larger than assumed once somebody recorded in the right band are the same story, and it recurs often enough to be a methodological warning rather than an anecdote.
Olfaction runs alongside both and may be the dominant modality. Elephants have the largest number of functional olfactory receptor genes of any mammal sequenced, roughly twice the count in dogs and around five times that in humans. Behavioral work has found they can discriminate human ethnic groups by scent, distinguishing the clothing of Maasai men, who historically speared elephants, from Kamba men, who did not, and responding with fear to the former. They also distinguish human age and sex from voice alone, responding with defensive behavior to recordings of adult Maasai men and not to those of Maasai women or boys, which means the discrimination is a threat assessment rather than a novelty response. Elephant cognition in the wild is heavily organized around categorizing humans, which is a rational allocation of attention for an animal whose primary cause of adult mortality is us.
The dogs whose working deployments depend on olfactory discrimination nobody has fully characterized are running the same sense at lower gene count, and the echolocating animals that generate their own signal to probe an environment are the contrast case: elephants are passive receivers across an enormous frequency range rather than active emitters.
Where the cognitive tests get awkward
Elephants do well on some laboratory tasks and badly on others, and the failures are as informative as the successes because they keep tracking whether the task fits the animal.
The clean successes: Asian elephants pass a cooperative rope-pulling task requiring two individuals to pull simultaneously, waiting for a partner rather than pulling uselessly alone, and they learn to wait up to substantial delays. They appear to show targeted helping, with individuals approaching distressed conspecifics and making contact with the trunk to the mouth while producing reassurance rumbles, which is the operational signature of consolation and had previously been documented mainly in great apes and corvids. They use tools, including branches as fly switches with modification of length, and they solve problems by moving objects to stand on when a reward is out of reach. They show numerical discrimination and appear to use absolute rather than relative quantity in some tests, which is unusual, since most animals including humans show a ratio effect where discrimination degrades as two quantities get closer together.
The famous failure is the mirror-and-food self-awareness task, in which an animal must recognize that its own body standing on a mat is what prevents it from handing over the mat. Elephants largely failed a version of this, and the finding got read as a limit on self-awareness.
The alternative reading is procedural and probably right. The body-as-obstacle task was designed for animals that manipulate the world with limbs they can see. The elephant tested for the puzzle with its trunk, which is the appropriate tool for the job from the elephant’s perspective, and the task did not accommodate that. The same pattern recurred throughout comparative cognition and produced decades of wrong conclusions about apes that failed false belief tests until the measure stopped requiring them to act and about small social mammals whose capacities only became visible through field observation.
There is also a straightforward practical problem. You cannot run large samples of elephants. Studies routinely involve fewer than ten animals, frequently captive, often trained for husbandry in ways that shape performance, and a species that lives sixty years cannot be studied longitudinally by any single researcher. Nearly every strong claim about elephant cognition rests on a small n, and the field says so. Sample sizes in the single digits are the norm, several of the most-cited findings rest on one or two individuals, and replication in this species is close to impossible in practice. That is not a criticism of the researchers, who are working with what exists. It is a reason to hold every specific claim about elephant cognition more loosely than the confident tone of most coverage suggests. The findings that hold up best are the ones from the wild, from decades-long field programs where the sample is a whole population and the observation period is long enough to catch a drought.
Three species, and what the differences mean
The genus distinctions matter more than the popular treatment allows.
African savanna elephants, African forest elephants, and Asian elephants are three species, with the forest elephant recognized as distinct relatively recently on genetic evidence. Divergence between the African species runs to several million years, and Asian elephants are more closely related to the extinct mammoths than to either African species.
The behavioral consequences follow habitat. Savanna elephants live in the largest and most structured social groups, which is what the classic matriarch literature describes. Forest elephants in dense Central African habitat live in much smaller units, often a female and her offspring, aggregating at forest clearings where mineral-rich soil draws animals together and where most of the observation has been done. Asian elephants sit between the two and show more variable social organization than the savanna model predicts.
Which means the standard picture of elephant social cognition is largely a picture of one species in open habitat, generalized to a genus. Forest elephants are considerably harder to study, and the social knowledge that a savanna matriarch carries may be organized entirely differently in an animal whose group is three animals in dense forest.
There is an ecological asymmetry worth noting alongside the social one. Forest elephants are seed dispersers on a scale that shapes the composition of Central African forests, moving large seeds distances no other animal manages, with measurable consequences for carbon storage when they are removed. The savanna species does something structurally analogous by knocking down trees and maintaining grassland. In both cases the animal is a landscape process as much as a species, which is a different kind of significance from the cognitive one and arguably a more consequential one. The large herbivores whose removal restructured entire systems are the general version of the argument.
Sexual dimorphism adds another layer. Adult males spend much of their lives outside family groups, form loose associations, and pass through musth, a periodic state of elevated testosterone and heightened aggression. Older males appear to constrain the behavior of younger ones, and removing them produces well-documented disruption, including the case in South Africa where young males translocated without adults began killing rhinoceroses, and the behavior stopped when older bulls were introduced. That case is cited constantly and deserves its prominence, because it is a rare natural experiment demonstrating that social structure regulates behavior in this species rather than merely correlating with it.
Long lives, slow development, and a cancer problem solved
An elephant’s cognition is inseparable from its life history, and the schedule is extreme even among large mammals.
Gestation runs about twenty-two months, the longest of any mammal. Calves are dependent for years, nurse for several, and remain in the family unit long past weaning. Females reach sexual maturity around ten to twelve and can reproduce into their fifties. Wild lifespans reach sixty to seventy years. That is a schedule that only pays off if the accumulated knowledge is worth the delay, and it is the same bet made by long-lived cetaceans, by great apes, and by large parrots that live for decades on a fraction of the body mass.
Elephants also have menopause-adjacent biology worth noting carefully, because it gets overstated. Unlike killer whales and humans, elephant females do not have a well-established post-reproductive lifespan; reproduction slows with age rather than stopping cleanly. The matriarch’s value as an information store therefore coexists with continued reproduction rather than replacing it, which makes the elephant case different from the standard grandmother-hypothesis examples even though it is frequently cited alongside them.
The cancer finding is the one with real biomedical interest. A body that size, with that many cells dividing over that many decades, should accumulate cancers at rates that would make the animal impossible, and it does not. Elephants carry roughly twenty copies of the tumor suppressor gene TP53, where most mammals including humans carry one, and their cells show unusually aggressive apoptotic responses to DNA damage: rather than attempting repair, damaged cells are eliminated. That resolves what is called Peto’s paradox, the observation that cancer incidence does not scale with body size across species as naive arithmetic predicts. A brain that takes sixty years to fill requires a body that survives sixty years, and the genome had to be rewritten to permit it.
The captivity problem, and what it does to the data
Almost everything known about elephant cognition under controlled conditions comes from captive animals, and the conditions are a variable rather than a neutral background.
The welfare literature is uncomfortable and reasonably clear. Zoo elephants show reduced lifespans relative to wild and working populations, high rates of foot pathology and arthritis attributable to substrate and inactivity, obesity, reproductive dysfunction, and stereotypic behaviors including repetitive swaying and head-bobbing that are generally interpreted as indicators of chronic stress or thwarted motivation. An animal that ranges tens of kilometers daily and lives in a multigenerational family confined to an enclosure with a handful of unrelated individuals is in a situation with no wild analogue.
That matters for the research in two directions. Captive animals are the only ones available for controlled testing, and their performance may under-represent capacity for the same reason performance under stress under-represents capacity generally. But captive animals are also habituated to humans, trained for husbandry procedures, and motivated by food rewards in ways wild animals are not, which can inflate apparent performance on tasks requiring cooperation with an experimenter.
The legal dimension arrived recently. Petitions seeking habeas corpus relief for individual captive elephants, including the Bronx Zoo’s Happy, have been argued before high courts and rejected, with the New York Court of Appeals declining in 2022 to extend habeas to a nonhuman animal while acknowledging elephants as intelligent and autonomous beings. That the arguments were heard at that level at all is a marker of how far the evidence has shifted the conversation, and the reasoning turned on legal personhood rather than on any dispute about the cognitive findings.
The claims that do not hold up
An audit, since elephants attract sentimental reporting.
Elephants never forget is a folk claim resting on real findings. Long-term social recognition and spatial memory across decades are well documented. Elephants also forget, misidentify, and make errors, and the phrase implies a general perfect memory that nothing supports.
Elephant graveyards do not exist. Concentrations of bones occur where old animals die near water during drought, and there is no evidence of animals traveling to a designated dying place. The myth was commercially useful to ivory traders, which is worth knowing about its persistence.
Elephants are afraid of mice has no support and fails basic testing.
Elephants mourn their dead overstates what the evidence carries. The behavior around dead conspecifics is real, specific, and unexplained. Mourning is a claim about internal experience.
Elephants are the smartest animals is a ranking claim assuming a single axis, and this animal is the clearest demonstration that no such axis exists. An elephant vastly exceeds a human in total neurons, sits below large primates in cortical neurons, exceeds nearly everything in olfactory receptor genes, and cannot do things a crow manages in fifteen grams.
Elephants get drunk on fermented marula fruit is a durable myth. The calculation of how much fermented fruit an animal of that mass would need to consume, and the low actual alcohol content, makes it implausible, and the original observations involved captive animals given alcohol directly.
Elephants can hear with their feet is imprecise rather than wrong. They detect seismic vibration and the mechanism plausibly involves both foot mechanoreceptors and bone conduction, but the extent to which they extract information rather than merely detecting is unresolved.
What elephant cognition is actually evidence for
The reason elephants belong in a comparative course is not that they are impressive, though they are. It is that they break the measure.
Comparative cognition has repeatedly tried to find a scalar that predicts capability. Absolute brain size fails, since whales and elephants exceed us. Encephalization quotient fails, since it ranks some small animals implausibly high. Total neuron count fails, and elephant cognition is the case that killed it: three times the human number, in an animal that does not do what humans do. Cortical neuron count survives better than the alternatives and still does not explain birds achieving comparable cognition in a walnut-sized forebrain with no cortex at all.
What the elephant demonstrates is that a nervous system is built to a specification, and the specification is set by the body and the ecological problem rather than by any general drive toward intelligence. A six-ton animal with a boneless manipulator containing 40,000 muscle units, operating across a range where a mistake with the trunk means losing the ability to eat, needs an extraordinary amount of motor computation. It bought that. The cortical investment it did not make is not a failure to become smart. It is capacity allocated somewhere the animal needed it more.
Which reframes the whole question. Asking whether an elephant is smarter than a chimpanzee is like asking whether a crane is faster than a motorcycle. Both are answers to engineering problems and the problems are unrelated. The animals whose cognition was built from entirely non-homologous tissue make the same point from a different direction, and the 24-lecture Neurozoology course runs the tree of life on that basis throughout, alongside the first edition’s survey of nervous systems and the working animals whose capacities were discovered by people who needed something from them. The elephants that hauled teak and moved supply columns through Burma demonstrated capacities their handlers observed daily and nobody thought to write down as data.
An elephant carries three times your neurons and one third of your cortical neurons, hears frequencies you cannot, smells with roughly five times your receptor gene count, addresses its relatives by something functioning like a name, and stores the map of its family’s survival in the head of its oldest female. None of that is a version of what we do. It is what a different machine looks like when it is built well, and the reason it keeps generating confusion is that we insist on grading it against a body plan and a set of problems it never had.
All three species are threatened, with forest elephants critically endangered and both African species in long-term decline from poaching and habitat conversion, and the animals being removed are disproportionately the oldest and largest. Which means the specific thing at risk is not only the population but the stored information: the routes, the water sources, the threat models, the social maps that exist nowhere except inside individuals who are being killed for their teeth. The fisheries whose migratory knowledge vanished with the fish that held it are the version of this that has already run to completion.
The 257 billion neurons were never the interesting number. The 97.5 percent was.
Leave a Reply