Module 6: Cognition, Memory & Grief

An adult African bush elephant (Loxodonta africana) brain weighs approximately 4.8–5.4 kg — more than three times the mass of a human brain and second only to the sperm whale (~7.8 kg) among all animals. Asian elephant (Elephas maximus) brains average 4.3–4.8 kg. The cerebrum dominates the overall volume, with a heavily folded neocortex whose temporal and frontal lobes are particularly enlarged relative to a generic mammalian baseline.

Absolute brain size, however, scales with body size across taxa, so raw mass is a poor predictor of cognitive capacity. The standard correction is Jerison’s encephalisation quotient (EQ), the ratio of observed brain mass to that predicted from body mass under an allometric baseline:

Inserting M_brain ≈ 5 kg and M_body ≈ 4 t gives EQ ≈ 1.88, somewhat greater than that of a domestic cow (EQ ≈ 0.5), chimpanzees (EQ ≈ 2.3), and cetaceans (EQ 1.8–5.3). For comparison, the human EQ is ~7. The point is not that elephants match humans but that their brains are substantially larger than expected for their body size — by a factor close to that of a great ape.

Manger (2006) and Herculano-Houzel et al. (2014, Journal of Comparative Neurology) used the isotropic fractionator method to count neurons directly. The elephant brain contains 257 billion neurons (more than three times the human total of 86 billion), but 97.5% of these are in the cerebellum — only 5.6 billion neurons reside in the cerebral cortex, compared with 16 billion in humans. The discrepancy suggests the elephant’s computational power for fine-grained somatosensory and motor control of the trunk is concentrated in the cerebellum, while neocortical capacity for high-level cognition is somewhat lower than in humans but still comparable in absolute terms to that of great apes.

A particularly striking cytoarchitectural feature of elephant cortex is the presence of von Economo neurons (VENs)— large, spindle-shaped projection neurons with a single thick apical dendrite and a single thick basal dendrite. VENs are concentrated in the anterior cingulate cortex (ACC) and frontoinsular cortex (FI). Hakeem et al. (2009, The Anatomical Record) demonstrated their presence in both African and Asian elephants, in roughly the same morphology as in great apes (chimpanzee, bonobo, gorilla, orangutan) and in cetaceans (humpback, fin whale, sperm whale).

VENs are thought to subserve rapid, intuitive socially relevant processing — including self-monitoring, empathy, and emotional salience — because of their large size (fast axonal conduction) and specialised connectivity. Crucially, elephants, great apes, cetaceans, and corvids form an independently-evolved cluster of large-brained, highly social species that all share VEN-like neurons, strongly supporting the hypothesis that VENs are a convergent adaptation to the demands of complex social cognition rather than a shared-ancestry feature.

VEN numbers drop markedly in frontotemporal dementia in humans, which preferentially ablates social cognition and empathic behaviour — a suggestive loss-of-function argument for VEN involvement in those cognitive domains. No comparable loss-of-function data exist for elephants, but the neuroanatomical parallel is striking.

Beyond VENs, Jacobs et al. (2011, Brain Structure & Function) quantified pyramidal-cell dendritic complexity in elephant cortex and reported extremely extensive basal-dendritic arbors — in some cases exceeding those of great apes. Such morphology is thought to support high-dimensional associative processing, reinforcing the picture of a cortex optimised for integrative social cognition.

Gallup’s mark test is the canonical assay of mirror self-recognition (MSR). After an animal is habituated to a reflective surface, a visible mark is applied to a part of its body it cannot see without the mirror. Mark-directed behaviour — touching, probing, or attempting to remove the mark while facing the mirror — is taken as evidence that the animal understands that the reflection represents itself.

Plotnik, de Waal, and Reiss (2006, Proceedings of the National Academy of Sciences) administered the test to three adult female Asian elephants at the Bronx Zoo (New York). A large 2.44 m × 2.44 m reflective wall was installed so that each animal could see her full reflection. After several days of mirror-directed investigation (examining the reflection, repetitive movements in front of it, peering behind the wall), one animal — Happy — spontaneously began mark-directed touching of a white X painted on her forehead, while ignoring a sham mark visible only through an identical invisible-ink substance. The mark-directed behaviour was restricted to sessions when the mirror was available; it did not occur in no-mirror controls.

The Bayesian analysis in Simulation 2 formalises the statistical inference from such data. Under the self-recognition hypothesis H_selfthe mark-touch rate λ_self is an order of magnitude higher than the chance-exploration rate λ_chance that would obtain in a no-mark or no-mirror control. Observing Happy’s 12 trials yields a log Bayes factor of ~40 in favour of self-recognition, corresponding to a posterior probability > 1 − 10⁻¹⁷.

Happy therefore joined a small clade of confirmed MSR species: great apes (Gallup 1970), bottlenose dolphins (Reiss & Marino 2001), Eurasian magpies (Prior et al. 2008), and — with ongoing debate — cleaner wrasse (Kohda 2019). The phylogenetic spread (placental + avian + teleost) again speaks to convergent evolution of self-monitoring in large-brained highly social species.

McComb et al. (2014, PNAS) performed playback experiments in Amboseli with recorded human voices speaking the same phrase (“Look, look over there, a group of elephants is coming”) in two local languages: Maasai (whose speakers pastorally graze cattle and historically have speared elephants in competition for water) and Kamba (whose speakers practise agriculture and pose little direct threat).

Elephant family groups reliably responded to the Maasai voice with defensive bunching, silent retreat, and vigilant posturing; they responded to the Kamba voice with continued foraging or mild attention. The response asymmetry was robust across adult-male, adult-female, and child voices, and was preserved under randomised presentation. The acoustic discrimination must therefore have been based on purely phonetic/phonological features — formant trajectories, phonemic inventories, prosody — rather than on semantic content (to which the elephants have no access).

Earlier, McComb et al. (2001) had shown an analogous phenomenon within the conspecific-call domain: family groups recognise the calls of ~100 individual elephants by voice alone, and retain recognition of a deceased matriarch’s rumble for at least two years after her death. Together these studies imply an auditory-category-formation faculty capable of partitioning complex acoustic spaces into ethologically meaningful classes with long-term retention — a competence closely paralleling songbird and primate vocal categorisation.

Elephants routinely modify objects in their environment for functional ends:

• Branch-switch fly-swatting (Hart et al. 2001): Asian elephants break twigs from trees, strip off lateral shoots, and swing the resulting flexible implement against the skin to repel flies. The implement is shaped to the needed length and flexibility; the behaviour is widely distributed geographically.
• Waterhole plugging (McComb 2000; Sukumar 2003): In some Asian and Sri Lankan populations, elephants chew bark and clay into a plug and press it into an excavated pit to prevent evaporative loss of accumulated water — a multi-step pre-planned action.
• Lock-picking and door manipulation (Roocroft 2005; anecdotal but well-documented): captive elephants have been observed sliding bolts, lifting latches, and bending gate pins with the distal trunk grip.
• Box stacking / height augmentation (Foerder et al. 2011, PLoS ONE): captive Asian elephants spontaneously move a cube into position to stand on it and reach fruit hung overhead — a clear instance of insightful problem solving.

Each of these behaviours requires a causal model of the external world: branches break when bent, latches lift when displaced, objects can be stacked for height. The elephant’s cognitive apparatus must therefore include representations of object affordances and physical reasoning closely matching those documented in great apes and corvids.

Bates et al. (2008, Journal of Experimental Biology) provided one of the most compelling field demonstrations of elephant empathy. Adult females observed injured or distressed conspecifics and responded with species-typical consolatory behaviours: gentle trunk contact, placement of the trunk in the injured animal’s mouth, protective positioning of the body, and alarm vocalisations that recruited other group members. The same responses occurred whether the distressed individual was a relative or an unrelated companion, ruling out kin-selection as the sole explanation.

Plotnik & de Waal (2014, PeerJ) performed a controlled experiment in which a captive Asian elephant was observed to respond to a conspecific’s distress vocalisation with rapid approach, trunk-touching, and bunching behaviours — classical signatures of consolation, a behaviour previously documented only in great apes, corvids, and canids among non-human mammals.

Cross-species empathy has also been documented: elephants have been reported to respond to human distress calls with vigilance, approach, or defensive posturing, and have been observed carrying dying mahouts in their trunks to a safe location (Siebert 2011; anecdotal but cross-validated). Sharma et al. (2020) reviewed 38 distinct published instances of elephants protecting or assisting humans in distress across Asia and Africa.

At the neural level, the combination of von Economo neurons in ACC/FI (Section 2) and extensive pyramidal dendritic arbors (Jacobs 2011) is hypothesised to underpin these behaviours by analogy with human empathy networks. Definitive functional-imaging demonstrations remain rare because of the technical difficulty of scanning large awake mammals, but early studies in Asian elephants (Dahle 2019; Maseko 2023) are consistent with this conjecture.

Douglas-Hamilton and colleagues (2006, Applied Animal Behaviour Science) published one of the first systematic accounts of elephant responses to death. Family members of a deceased individual were observed to examine the body intensively with the trunk, lift and move bones, vocalise in characteristic low rumbles distinct from routine calls, and return repeatedly to the carcass over hours or days. The behaviour was statistically distinguishable from responses to random-species carcasses (zebra, buffalo): conspecific remains elicited sustained, repeated investigation; heterospecific remains did not.

Goldenberg & Wittemyer (2020, Primates) presented a systematic review across 32 documented cases, confirming the pattern: conspecific bones are handled with the trunk, often lifted, carried short distances, and sometimes sniffed with the vomeronasal organ. Skulls and tusks are frequently the focus of attention. McComb et al. (2006) showed experimentally that family groups preferentially investigate a set of elephant skulls over other wildlife skulls presented at a uniform distance.

Whether such behaviour constitutes grief in the human emotional sense is philosophically contested. It does, however, meet a number of ethological criteria: (i) species-specificity, (ii) persistence over extended time scales, (iii) attenuation of routine activity during the episode, and (iv) the involvement of specific individuals linked to the deceased by social bonds. At minimum, elephants possess a category concept for “conspecific bones” that is behaviourally distinct and emotionally charged.

A central organising idea in elephant cognition is that the matriarch functions as a biological memory store, accumulating over decades the locations of waterholes, dry-season refuges, predator histories, and safe migration routes. McComb et al. (2011, Proceedings of the Royal Society B) played back lion roars of varying age-sex composition to family groups with matriarchs aged 28–70 years. The quality of the groups’ alarm-response — measured as the rate and intensity of defensive bunching around calves — increased with matriarch age, even after controlling for group size and family size.

Crucially, groups led by young matriarchs (< 35 y) were just as likely to bunch in response to the less-dangerous lioness-only roars as to the more-dangerous mixed-sex (or male-dominated) roars — they failed to discriminate. Older matriarchs disproportionately deployed defensive responses to the genuinely dangerous stimulus and no response to the safer one. The pattern directly demonstrates that long-lived matriarchs carry information that translates into enhanced survival for the group.

The model in Simulation 1 formalises this by giving the matriarch a knowledge-capital variable K₀(t) that grows with age and is socially transmitted to juniors at rate η. Group fitness is a Hill function of the group’s mean knowledge. Loss of the matriarch induces a multi-year fitness decrement whose magnitude depends on the juniors’ pre-loss knowledge accumulation.

Long-term demographic data from Amboseli (Douglas-Hamilton, Moss, and the Amboseli Elephant Research Project 1972–present) show that matriarch-led groups typically have lower calf mortality, higher inter-calving-interval efficiency, and more successful navigation during drought years, corroborating the knowledge-capital framework from an empirical direction.

Cultural transmission is rare in non-human mammals but well-documented in elephants. Poole and ElephantVoices (2011 onward) have catalogued population-specific vocal dialects: rumble spectrotemporal signatures differ systematically between Amboseli (Kenya), Samburu (Kenya), Gorongosa (Mozambique), and Udawalawe (Sri Lanka) populations in a manner not explained by genetic divergence. Calves raised in a foster population acquire the foster-group dialect.

Migration-route knowledge is likewise socially transmitted. Foley (2002) and Wittemyer (2007) documented multi-generation retention of seasonal movement patterns in Tarangire (Tanzania); after a period of poaching disrupted matriarch survivorship, adjacent populations recovered migration patterns only slowly across decades. The Selous population recovery after the 1980s-1990s poaching crisis proceeded at a similar sociocultural pace.

The classic evolutionary signature of culture — that knowledge acquired from conspecifics outweighs individually acquired knowledge in determining lifetime behaviour — is readily met in elephants, as young individuals follow matriarch-led routes for years before adopting independent decision-making. Gorongosa National Park (Mozambique) provides a striking example of cultural disruption: after the Mozambican civil war (1977–1992) killed an estimated 90% of the park’s elephants, the surviving cohort was disproportionately juvenile, leading to destabilised social structure and altered behaviour for a generation (Gaynor 2018). See Module 8 for the parallel evolutionary consequences for tusklessness.

The schematic below summarises the elephant’s brain structure and highlights the cortical regions implicated in social cognition. The anterior cingulate cortex (ACC) and frontoinsular cortex (FI) contain the von Economo neurons; the temporal lobe houses the expanded auditory and social-recognition pathways; the cerebellum accounts for the majority of total neuron count and supports fine-grained motor control.

The cognitive richness documented in the preceding sections has direct welfare implications. Clubb & Mason (2003, Nature; 2008, Science) showed that zoo-housed elephants have substantially reduced life expectancy compared with their protected-area conspecifics — median 19 years in European zoos vs 56 years in Amboseli for African elephants — with high rates of stereotypic behaviour(rhythmic swaying, head bobbing, pacing) interpreted as signs of chronic psychological stress.

Stereotypy rates correlate with social-housing instability, social group size below 3–4 individuals, and inadequate enclosure complexity. The interpretive framework is that elephant psychological welfare is intrinsically coupled to the availability of appropriate social partners and an environment that permits expression of their behavioural repertoire — a direct consequence of the high-dimensional social cognition outlined above.

Modern zoo design accordingly emphasises: (i) matrilineal-compatible grouping of 3+ related females, (ii) enclosures exceeding 0.3 ha with complex terrain, and (iii) enrichment regimes explicitly targeting cognitive challenge (puzzle feeders, novel olfactory stimuli). There is emerging behavioural evidence that such reforms reduce stereotypy in long-captive individuals (Williams 2019). Nonetheless, many workers argue that captive housing of elephants remains incompatible with their species-typical cognitive and social needs.

Several central questions remain open:

• Causal vs associative reasoning.Are elephant tool-use and problem-solving behaviours underpinned by causal/counterfactual reasoning, or can they be accounted for by associative learning with large-capacity memory? Controlled experiments analogous to the Aesop’s fable paradigm used in corvids are largely absent.
• Functional mapping of VENs.Pharmacological or neurophysiological perturbation of elephant ACC/FI is not feasible with current methods. Correlative non-invasive techniques (awake fMRI in trained individuals; magnetoencephalography) may bridge the gap.
• Ontogeny of matriarchal expertise.At what age does a female elephant acquire sufficient ecological knowledge to function as a matriarch? Is the transition gradual or threshold-like? Long-term Amboseli cohort data should be suitable to address this within a decade.
• Emotional states and subjective experience. Whether the behaviour of a mother elephant attending a dead calf reflects something that we would reasonably call “grief” is an instance of the broader consciousness-in-non-human-animals question (de Waal 2019). A combination of behavioural, hormonal (cortisol), and neural markers may provide converging evidence.
• Cross-species cognitive comparisons.Systematic comparisons between elephants, cetaceans, great apes, and corvids along a common battery of cognitive tasks (e.g. the Primate Cognition Test Battery of Herrmann 2007 adapted for non-primates) would clarify which cognitive domains are truly convergent vs. phylogenetically constrained.