Erica Cartmill on Animal Minds, Play and Teasing, and Comparative Cognition
Comparative cognition scientist Erica Cartmill argues that human minds differ from animal minds not by a single magic ingredient but by a unique constellation of abilities, and shows how play, teasing, and laughter in great apes open a window on social cognition — one whose methods now bear on whether large language models understand anything at all.
Key ideas
-
There is no ladder of intelligence, only a constellation of abilities. The old scala naturae — species arranged on a rising scale towards human perfection — is wrong: humans are as evolved as every other species, each fitted to its own social and ecological niche. For every single feature of language (syntax, reference, compositionality) at least one other species has that piece, but no other species has them all. What is distinctive about humans is the configuration — the way the pieces lock together and build on each other — not any one part.
-
Cognitive strengths are unequal and niche-shaped, even across the great apes. Chimpanzees and bonobos, the most social apes, excel at reading others’ motivations, alliances, and third-party relationships — the ‘Machiavellian’ skills a large group demands. Orangutans, largely semi-solitary, are weaker at social scheming but stronger at patient material manipulation. Cartmill’s analogy: orangutans are theorists who study a problem before acting; chimpanzees and bonobos are experimenters who grab and shake it.
-
Teasing is a grey zone between aggression and play, and a probe of the social mind. Playful teasing — offering then withdrawing an object, disrupting another’s activity, ‘provocative non-compliance’ — has the structure of a very basic joke: set up an expectation, then violate it. Cartmill reads it not as low-grade aggression but as social hypothesis-testing: a low-stakes way to learn how far a specific relationship can be pushed, and an honest signal of how much a partner values the bond over momentary annoyance.
-
Positive emotion is measurable, and it changes cognition. Apes laugh as humans do, in similar contexts (tickling, play), if less musically. Cartmill’s lab uses thermal imaging (a dip in nasal temperature marks arousal) and a cognitive-bias test — training apes that black boxes are rewarded and white are not, then introducing ambiguous grey boxes — to show that bonobos who have heard laughter approach the grey boxes more readily. Laughter carries no information about the box; it simply induces optimism.
-
Animal-cognition methods should discipline claims about AI understanding. You cannot ask a chimpanzee or a one-year-old why it acted, so comparative and developmental researchers have built indirect, non-verbal ways to probe understanding. An LLM will answer ‘why’, but its verbal report cannot be taken at face value the way human self-report conventionally is — and even human self-report is unreliable. Cartmill argues these are not problems to engineer away but questions demanding systematic, cross-modal probing of what the system actually grasps.
Content
Against the ladder: a unique constellation, not a magic switch
Carroll opens with the question the episode circles: is there one fundamental difference between humans and other animals, or an accumulation of many? Cartmill rejects the linear picture outright. The scala naturae — species climbing towards human perfection — has been dismantled over decades of science; humans are ‘equally as evolved as every other species’, each good at what its niche rewards, so it makes no more sense to rank cognition linearly than to fault humans for not echolocating. Her preferred frame is a unique constellation of abilities: take language apart into syntax, reference, and compositionality and you find each component somewhere else in the animal kingdom, but never the whole assembled in one other species. The human difference is the configuration — pieces that fit together and amplify one another — which is exactly why there is no tidy narrative of human superiority.
The chimpanzee Ayumu makes the point vivid. Trained in Kyoto to touch the numbers one to nine in sequence after they flash and are masked, she outperforms untrained human adults at this working-memory task, to the humbling of zoo visitors in Indianapolis who assume they can beat a chimp. Cartmill is careful: it is an unfair comparison of a trained expert against novices, and trained humans can reach expert level too. The deeper contrast is numerosity. Chimpanzees learn each number as a separate, equally hard fact — learning five is no easier than learning four — whereas human children, after a slow start, grasp the cardinal principle that each number is one more than the last, an ‘aha’ that lets them generalise. Great apes can track quantities and even register that four plus five behind a screen should not yield three, but they do not reach the abstraction of the successor operation.
Reading the social world: hierarchy, cooperation, and pointing
Why the strengths fall where they do is, Cartmill concedes, partly storytelling — hypotheses about language, tool-making, and nested hierarchies that are hard to test directly. What comparative work can do is ask whether living apes appreciate structure: do they distinguish ‘the dog chased the cat’ from ‘the cat chased the dog’, can they order intermediate steps towards a goal? The apes’ capacities cluster by niche. Chimpanzees and bonobos, living in large multi-male, multi-female groups, are intensely attentive to who is doing what with whom — genuinely interested in gossip — and Seyfarth and Cheney’s playback experiments show they track not just their own relationships but third-party ones: splice recordings into a fake fight in which a subordinate beats a dominant, and listeners stare longer and then press their advantage against the newly vulnerable individual.
The pointing discussion draws out a species-level trait of humans. Chimpanzees in the wild rarely point, Cartmill notes, because a resource is better kept than advertised; in human care, behind barriers, they readily gesture at what they want. Humans, by contrast, are ‘fundamentally a cooperative species’ — sharing information freely with strangers even when it costs them, and, crucially, trusting that others speak in good faith. That mutual trust is what makes language, and podcasting, work at all: to talk about anything beyond the here and now, the listener must believe the speaker. Cartmill balances this against the richness of animal communication — song sparrows recognise not just a call type but who made it and from where, responding to a neighbour’s song played from the wrong side as though a stranger had trespassed.
Teasing, jokes, and social hypothesis-testing
The heart of the conversation is humour, which Cartmill studies through playful teasing — a grey area sitting between aggression and play. Past work treated animal teasing as low-level aggression; she insists much human teasing is loving, relationship-building, and that its structure is a proto-joke. The founding observation came during her orangutan dissertation work: an infant dangling above her mother repeatedly offered a stick and snatched it back; then, when the infant tired, the mother picked it up and did it back. That role reversal proved it was not an idiosyncratic routine but a game with structure and interchangeable roles — offer as setup, withdrawal as punchline, the basic architecture of a joke: setting up an expectation and violating it.
Cartmill borrows Vasu Reddy’s catalogue of infant ‘clowning’ behaviours — offer-and-withdraw, disrupting others’ activities, and ‘provocative non-compliance’, of which she declares cats the masters. She resists a firm verdict on whether cats have a sense of humour; the appropriate scientist’s question is whether the animal still performs the act when no one is watching, and whether it is calibrated to elicit a response. Her governing hypothesis is that teasing is social hypothesis-testing: unlike the physical properties of an object, which stay fixed once learned, social relationships shift with mood and history, so they must be probed continually. Teasing offers a low-stakes arena for practising the prediction of another’s reaction — how far can I push you, how far will you be pushed — and functions as an honest signal, since tolerating a tease is a costly display of valuing the relationship over the momentary annoyance. She draws the parallel to Judea Pearl’s causal maps of the physical world: teasing and play build the causal map of the social world.
Games, joy, and the biometrics of positive emotion
Play scales from teasing towards games. There is nothing like chess, but plenty like tag, hide-and-seek, and chase, with genuine turn-taking: a chasing dog will stop, look back, play-bow, and prompt a lagging partner, marking the reciprocal structure of the game. Cartmill is now opening a canine cognition lab at Indiana University, and speculates about private games — dogs that collect and arrange objects, or her own dog Bonnie burying a beloved stone turtle for herself to re-find, self-set goals with no survival payoff beyond the satisfaction of achieving them, akin to a child seeing how long she can stand on one leg.
To get at why animals play, Cartmill’s cross-species project (with philosopher Colin Allen, spanning kea parrots, dolphins, and great apes) builds biometric measures of positive emotion. Thermal cameras track a cold-nose signature of arousal; a cognitive-bias test — anchoring apes on rewarded black boxes and unrewarded white, then introducing ambiguous grey ones — measures optimism. Bonobos who have heard laughter, which carries no information about the boxes, approach the grey boxes more, suggesting the laugh simply lifts the mood. Laughter is also the stimulus in experiments on how hearing another’s laughter shifts memory, attention, and prosociality. On imagination and pretence, Cartmill points to play signals such as the dog’s play-bow — species-typical, unlearned, and interpreted as a frame that says ‘what I am about to do should not be taken seriously’ — the same framing teasing relies on, though short of a child’s full pretend play.
Comparative cognition as a discipline for AI
Carroll closes, as he often does, on artificial intelligence. Cartmill calls the AI question fascinating but tongue-in-cheek irritating, because for most people it collapses to LLMs like ChatGPT. Her point is methodological. For much of the history of the field, language was the line between humans and other animals, and human verbal self-report was trusted as a window on the mind. An LLM will happily give a human-sounding answer — it has ingested essentially all of human language — but that answer should not be taken at its word. The big question, ‘does an LLM understand anything?’, calls for the very tools comparative and developmental researchers have honed for subjects who cannot be simply asked: probe understanding by manipulating how the question is posed, changing the modality, changing the situation. Even human adults, she notes, often cannot report the true reasons for their actions. She doubts the AI labs consult psychologists, anthropologists, and philosophers as much as they should, and predicts these will prove problems you cannot engineer your way out of — which, Carroll agrees, are among the most interesting kind. Cartmill points listeners to her citizen-science project at observinganimals.org, collecting reports of animals teasing, and to the Center for Possible Minds and the Diverse Intelligences Summer Institute.
See also
- Sean Carroll — host
- Erica Cartmill — guest
- Over-Attribution of Intelligence — the caution against reading an LLM’s fluent verbal report as evidence of understanding
- Jagged Intelligence — the uneven profile of strengths and weaknesses, mirrored in the constellation view of animal minds
- Language and Mind — language as a decomposable set of features rather than a single human-defining switch
- Language-Thought Dissociation — the gap between fluent output and underlying understanding, in animals and in LLMs alike
- Large Language Models — the systems whose ‘understanding’ Cartmill argues should be probed with animal-cognition methods
- Consciousness — the broader question of other minds that the comparative approach engages