Animal & Comparative Psychology: Between Anthropomorphism and Anthropodenial
2026.07.26 · BigCat's Inner World
Comparative psychology appears to study animals. What it actually is, is a method: how to infer a mind in a subject you cannot interrogate. It has two symmetrical failure modes — and its most urgent application today may not involve animals at all.
Animal Emotion: Subcortical Circuits Are Evolutionarily ConservedAffective Neuroscience & Ancient Emotional Systems
Affective Neuroscience · Panksepp
Core Insight
Emotion is not a higher function the cortex evolved. It sits in ancient subcortical circuitry that is highly conserved across mammals. Which means the question "do animals have emotions?" is posed backwards: we inherited these circuits from them.
Evidence Base
In Affective Neuroscience (1998), Jaak Panksepp used the reproducible effects of deep brain stimulation to delineate seven cross-species emotional-motivational systems. The criterion is not behavioural resemblance but the intersection of three things: anatomical homology, consistent pharmacological response, and comparable behaviour from stimulating the same site.
PANKSEPP'S SEVEN SUBCORTICAL SYSTEMS
SEEKINGDopaminergic exploration and anticipation. It encodes the pursuit, not the satisfaction of getting there.
FEAR / RAGEThe two poles of threat response — avoid or confront (amygdala–hypothalamus–midbrain).
PANIC/GRIEFThe separation-distress call system, modulated by endogenous opioids; human social pain shares its substrate.
CARE / LUSTParental and pair-bond circuits, modulated by oxytocin and sex hormones.
PLAYThe rough-and-tumble system, independent of the other six; tickled rats emit 50kHz ultrasonic calls.
Mechanism
The implication is counter-intuitive: the older the feeling, the less it yields to reasoning. Reappraisal (Day 51) operates on the cortex; when PANIC/GRIEF fires, what works faster is bodily and relational input — being held, being answered, hard exercise. Barrett's constructionist account holds that emotion categories like "anger" depend on language and culture — the two are not in conflict: the circuits supply raw valence and arousal, the categories are labels applied later. Hence discussing an animal's valence and arousal is relatively safe; discussing its "jealousy" or "guilt" demands great care.
Applications
SELFPLAY is a separate system, not a childish residue; rest cannot substitute for it — rest switches SEEKING off, play runs on something else.
PARENTINGThe death of a pet activates PANIC/GRIEF, homologous with human bereavement. Don't say "it was only a cat."
TEAMSEEKING encodes anticipation, not satisfaction. Slicing milestones finer sustains engagement better than raising the bonus.
Self-check + common misreading: Take your last strong emotion: was it thwarted seeking, fear of exclusion, or a violated boundary? The three call for entirely different responses. Misreading: taking the dog's tucked-tail "guilty look" for guilt — Horowitz (2009) found the expression tracks whether the owner acts reproachful, not whether the dog actually misbehaved. It is a submission signal aimed at scolding.
Key references · Jaak Panksepp, Affective Neuroscience (1998) · Alexandra Horowitz, "Disambiguating the 'guilty look'" (2009, Behavioural Processes) · Lisa Feldman Barrett, How Emotions Are Made (2017)
THIS WEEKPick one day and schedule 20 minutes of pure PLAY: no goal, no score, no output. Reflection: in which of my roles has play been squeezed out entirely? Was the space taken by external demands, or did I decide I hadn't earned it?
Comparative Cognition: Not a Ladder, a RadiationComparative Cognition & Ecological Rationality
Comparative Cognition · Ecological Rationality
Core Insight
"Which animal is smarter?" is a bad question. Evolution does not produce a general intelligence ranking, only solutions to particular ecological problems. Every species' cognitive profile is jagged: astonishing inside its own niche, unremarkable the moment you change the task.
Evidence Base
Inoue & Matsuzawa (2007, Current Biology): the chimpanzee Ayumu reliably outperforms human adults on a numeral working-memory task. New Caledonian crows manufacture hooked tools matched to the job; Clark's nutcrackers recover thousands of cache sites months later. None of this amounts to "smarter" — only to having been selected on a different exam.
FROM LADDER TO RADIATION
Mechanism
What drives the radiation is ecological rationality: a mechanism is good not insofar as it generalises but insofar as it matches the structure of its environment. The nutcracker doesn't have strong general memory; it has a hippocampus specialised for "many discrete locations, one retrieval season." The octopus is more extreme — roughly two-thirds of its ~500 million neurons sit in the arms, which retain substantial local autonomy, and central commands look more like coarse goals than joint-by-joint instructions: a line of intelligence that evolved independently of vertebrates, on a wholly different architecture. The corollary holds for people too: ranking cognition by a single score flattens a jagged profile into one line, and what gets flattened away is the most useful information of all.
Applications
TEAMStop asking who is stronger; ask what problems shaped this person. Replace the ranking with a profile.
PARENTINGTreating "good at maths, weak at writing" as a gap to level out, rather than as the shape of a profile, leads to two entirely different decades.
SELFState your strength as "the class of problems I'm good at," not "the level I'm at." The former transfers and can be matched to an environment (Day 1).
Self-check + common misreading: Write down the last time you judged someone as not smart, then answer: was that exam from my niche or theirs? Misreading: "bigger brain, smarter animal." Absolute brain mass, brain-to-body ratio, and neuron density give mutually inconsistent rankings; birds achieve tool manufacture and planning with a forebrain built on an entirely different plan — anatomical similarity is not a precondition for cognitive capacity.
Key references · Inoue & Matsuzawa, "Working memory of numerals in chimpanzees" (2007, Current Biology) · Peter Godfrey-Smith, Other Minds (2016) · Nicola Clayton & Nathan Emery on corvid cognition
THIS WEEKSketch a four-axis cognitive profile for yourself or a colleague (abstract modelling / failure intuition / cross-domain translation / long-horizon patience, 1–5 each), then mark which two axes the current role actually tests. The mismatch is where the energy goes.
Two Symmetrical Traps: Anthropomorphism and AnthropodenialAnthropomorphism & Anthropodenial
Methodology · de Waal
Core Insight
Casually mapping human concepts onto animals is an error. But insisting that "subjective experience is unknowable, so we won't discuss it" is also an error — it makes an equally unevidenced strong claim while wearing the costume of caution. Frans de Waal calls the latter anthropodenial: for species very close to us, it is the side assuming discontinuity that carries the burden of proof.
Evidence Base
The methodological landmark is Morgan's Canon (Morgan, 1894): do not attribute to a higher process what a lower one explains — necessary discipline, later stretched by behaviourism into "no internal state may be discussed at all." The classic warning on the other side is Clever Hans: the arithmetic horse was reading the questioner's micro-expressions (Pfungst, 1907); today's double-blind designs descend from that lesson.
SYMMETRICAL ERRORS ON BOTH SIDES
Too much anthropomorphismMapping human-specific categories straight on: the dog's "guilt," the model's "wanting" — manufacturing explanations that cannot be tested.
Critical anthropomorphismTake evolutionary kinship and neural homology as a prior, then test it item by item with falsifiable experiments; the closer the homology, the stronger the prior.
Too much anthropodenialRuling anything experiential unknowable amounts in practice to assuming "there is none" — dressing an ethical question as an epistemic one.
Mechanism
Both sides are easy to fall into, because the mind-attribution system is a hair-trigger default-on device — a stuck robot vacuum has people saying "it's trying to get out." Suppressing it also costs something: it shares machinery with empathy. The right posture is not to switch it off but to treat its output as a hypothesis, not a conclusion. One more overlooked pitfall: testing a mind through the wrong sensory channel. Mirror self-recognition (Gallup, 1970) presupposes visual dominance, whereas Horowitz's (2017) "olfactory mirror" showed dogs investigate their own modified odour markedly longer. Dogs used to fail the test because of the test, not because of the dogs.
Applications
SELF / AIAttributing minds to language models is the contemporary version of the same trap, but the prior runs the other way: animals have neural homology behind them, models don't — human-like output comes from fitting human text, so the resemblance lives at the output layer, not the mechanism.
PARENTINGBoth errors are common: over-attributing intent ("he did it on purpose" — usually an executive-function shortfall, see Day 14), and under-attributing ("what would a child know about grief").
TEAMTesting capability through the wrong channel is the mirror test applied to dogs. Judging thinking quality by whether someone speaks up in meetings filters for channel match, not ability.
Self-check + common misreading: Log three moments when you say "it/he wants…", and mark which had mechanistic evidence and which only behavioural resemblance. Misreading: treating last century's ape-language projects (Washoe, Nim) as settling that animals acquired language — Terrace's reanalysis of his own Nim project found that many "sentences" were imitative collages of the trainer's prompts, lacking spontaneous syntax.
Key references · Frans de Waal, Are We Smart Enough to Know How Smart Animals Are? (2016) and Mama's Last Hug (2019) · Alexandra Horowitz, "Smelling themselves" (2017, Behavioural Processes) · Herbert Terrace, Nim (1979)
THIS WEEKTake a recent judgement you made about someone's mind (or a model's) and write it in three lines: the behaviour I observed / the mental explanation I supplied / one observation that would refute it. Reflection: which of my "cautious non-judgements" are really evasions of responsibility?
Evolutionary Continuity: What Is Actually Ours AloneEvolutionary Continuity & Human Uniqueness
Evolutionary Psychology · Tomasello
Core Insight
The "only humans can" list has been struck through item by item: tool manufacture, teaching, cultural transmission, fairness, planning for the future — all turn up in other species to varying degrees. The two entries that hold up best are not single capacities but cumulative culture and shared intentionality — not "I am smarter," but "we can stack cognition."
Evidence Base
Tomasello (2014): human infants develop early the capacity to form joint goals and joint attention with others, the precondition for cultural learning. Henrich (2016) supplies the complementary half: individual humans aren't especially clever; the advantage comes from a cultural ratchet that loses nothing across generations. And Clayton & Dickinson (1998, Nature) found scrub-jay cache memory carries what-where-when structure, challenging the claim that animals live only in the present.
Mechanism
What matters about a ratchet is not invention but not slipping back. Chimpanzees acquire group-specific techniques (the regional tool differences Whiten documented), but improvements rarely stack across generations; high-fidelity imitation, active teaching, and language make each human generation start where the last one stopped. So the substrate of human cognition is not the individual brain but the cross-generational knowledge network — which also sets the boundary on the "AI-augmented individual": however far individual throughput rises, it stays bounded by how faithfully knowledge transfers between people, and between people and systems.
Applications
SELFYour ceiling isn't inside your head; it's in the network you're plugged into. Add one question to the annual review: how much of my knowledge has left my head in a form others can reuse?
PARENTINGHuman teaching is bidirectional — the teacher adjusts to the learner's feedback. Replace the explanation with "you try first, I'll watch, minimum hint when you're stuck."
TEAMThe real measure of organisational capability is whether the ratchet engages: will a newcomer a year from now still receive the lesson from this incident?
Key distinction: "Uniquely human" is not "superior." A ratchet does not choose a direction; it only guarantees no slipping back — including no slipping back on the way somewhere wrong.
Self-check + common misreading: Which of the things you did this year will still be working after you leave? That is your own ratchet tooth. Misreading: "evolution is progress and humans are on top." Evolution has neither direction nor endpoint; every extant species is the present-day tip of an equally long evolutionary history — bacteria are exactly as "new" as we are.
Key references · Michael Tomasello, A Natural History of Human Thinking (2014) · Joseph Henrich, The Secret of Our Success (2016) · Clayton & Dickinson, "Episodic-like memory during cache recovery by scrub jays" (1998, Nature) · New York Declaration on Animal Consciousness (2024)
THIS WEEKTake one piece of tacit knowledge that only you hold (a judgement heuristic, a debugging route) and spend 30 minutes writing it in a form someone else can follow. Reflection: is the "AI-augmented individual" about augmenting the individual, or about rebuilding how the network transfers? The two demand very different investments.
GOING DEEPER
Is anthropomorphising a language model the same error as anthropomorphising an animal?
The structure is similar; the prior runs the opposite way. For animals, neural homology supplies a genuine prior that similar mechanisms yield similar experience; a model has no such prior — its human-like output comes from fitting human text, so the resemblance concentrates at the output layer. Conversely, refusing all functional description on the grounds that "it's just statistics" will miss real differences in capability. There is exactly one workable criterion: is there mechanistic evidence independent of surface performance?
What can the octopus's distributed nervous system teach distributed systems design?
It offers a trade-off structure, not a solution. Highly autonomous arms with the centre issuing only coarse goals amounts to trading consistency for autonomy and robustness: where communication is expensive and local conditions vary sharply, giving up a globally precise state is a good deal. But the analogy has hard limits — a biological architecture is a feasible solution selected for, not an optimal one. Use it to question the default assumption that you must have a globally consistent view, not as a design template.
Is grounding moral status in the capacity to suffer the same criterion as the Buddhist category of sentient beings?
The resonance is real: both hang moral status on the capacity to suffer rather than on reason or language — precisely the direction of the 2024 New York Declaration on Animal Consciousness (strong evidence for birds and mammals; cephalopods and some insects not to be dismissed), and the reason Buddhist compassion was never confined to humans. The difference is in the grounds: the Buddhist category is metaphysical, bound to rebirth premises, its boundary fixed by doctrine; the ethical criterion is empirical, its boundary moving with the evidence. The conclusions converge; the arguments do not.
If emotion categories are culturally constructed, can animal emotion be studied at all?
Yes, but at a different level. Constructionism targets categories like "anger" and "jealousy," which depend on language and concept learning and cannot be tested in animals. Valence, arousal, approach-avoidance tendencies, and the subcortical circuits can be verified independently through preference tests, conditioning paradigms, and pharmacological manipulation. So the safe programme is: strong claims at the circuit level, weak claims at the category level.