For most of the twentieth century, calling someone a bird brain was not merely rude. It was a citation.
The reasoning ran through a specific and confidently held piece of comparative anatomy. Mammalian cognition happens in the neocortex, a six-layered sheet of tissue folded over the top of the brain. Birds have no such sheet. Their forebrain is nuclear rather than layered, organized into clusters rather than laminae, and when nineteenth-century anatomists looked at it they concluded that most of it was basal ganglia, the machinery of instinct and reflex. They named it accordingly. The bird forebrain regions were called hyperstriatum, neostriatum, archistriatum, and paleostriatum, and the striatum root in every one of those terms encoded the assumption that a bird was running on autopilot with almost nothing on top.
That vocabulary stayed in the textbooks until 2005, when an international consortium formally renamed the entire avian forebrain because the underlying claim had been wrong for about a century. The regions were not striatum. They were pallium, the same embryonic territory that produces the mammalian cortex, and the animals had been doing cortex-grade work in them the whole time while the naming convention insisted otherwise. Every anatomy student for four generations learned the error as vocabulary, which is a useful reminder that a discipline’s terminology is a theory nobody is required to restate before using.
That is a decent summary of how bird intelligence research has gone. Behavioral work kept producing results that the anatomy said were impossible, the anatomy turned out to be wrong, and then, in 2025, the anatomy turned out to be wrong in a considerably more interesting way than anyone expected. The current answer is not that bird brains are secretly like mammal brains. It is that they are not, and the birds do it anyway.
Bird intelligence and the neuron-packing solution
Start with the obvious objection, which is size. A raven brain weighs roughly fifteen grams. A macaque brain weighs about ninety. If cognition scales with neural hardware, and it does to a first approximation, the corvid should not be competitive.
The resolution came from actually counting. Work published in 2016 dissolved the brains of twenty-eight bird species and counted the nuclei, and found that songbird and parrot brains carry roughly twice as many neurons as primate brains of the same mass. The neurons are smaller, packed more densely, and, critically, concentrated in the forebrain rather than distributed proportionally. The consequence is that large parrots and large corvids carry forebrain neuron counts equal to or greater than monkeys with brains several times larger.
A macaw runs on the order of one and a half to two billion neurons, which is more than a macaque manages in a brain roughly six times heavier. A goffin’s cockatoo, a bird you can hold in one hand, carries more forebrain neurons than a squirrel monkey. Follow-up work found that associative neuron numbers specifically track cognitive performance across corvid species, and that neuron counts predict behavioral innovation rates in the wild better than brain size does. That last result is the one with teeth, because innovation rate is measured from field reports of birds doing novel things rather than from laboratory tasks, which means the relationship holds outside the artificial conditions where most comparative cognition gets tested.
There are engineering advantages to small neurons beyond raw count. Shorter interconnects mean shorter conduction delays, which matters for any computation requiring coordination across a network. Smaller cells have lower metabolic cost per unit. And a compact forebrain avoids the wiring problem that scales badly in large brains, where connecting everything to everything becomes physically impossible and the brain has to modularize.
There is a constraint underneath all of this that mammals never faced, and it explains the design pressure. A flying animal pays for every gram, continuously, on every wingbeat. A brain is dense tissue with a high metabolic rate, and enlarging it costs both mass and fuel in an animal already operating near its power limits. Faced with needing more computation and unable to afford a bigger housing, the avian lineage shrank the components. That is a different optimization target than the mammalian one, and it produced a different machine.
So the first correction to the bird intelligence story is straightforward accounting: the hardware was there all along, in a package nobody had bothered to measure properly until a decade ago. The second correction is about what the hardware is organized into, and it is stranger.
The NCL, a prefrontal cortex nobody inherited
Executive function in mammals is associated with prefrontal cortex: working memory, rule learning, inhibitory control, the ability to hold a goal and act on it across a delay. Birds have no prefrontal cortex, because they have no cortex.
They have the nidopallium caudolaterale, a region in the caudal forebrain that does the same jobs. It is densely innervated by dopamine, as prefrontal cortex is. It receives convergent input from all sensory modalities and projects to motor structures, as prefrontal cortex does. Lesion it and birds fail exactly the tasks that prefrontal damage disrupts in mammals.
And it is in the wrong place. Mammalian prefrontal cortex derives from a different part of the pallium than the avian NCL does, which means the two structures are not the same structure inherited from a common ancestor. Two lineages, separated for something on the order of three hundred and twenty million years, independently built a dopamine-rich association area at the top of the sensory hierarchy and used it for the same class of computation. Reptiles, which sit between the two on the tree, did not build anything comparable, which rules out the simple explanation that the ancestor had one and mammals and birds both kept it.
The NCL also does something the mammalian comparison makes vivid: it sits at the top of a sensory hierarchy that includes a visual system with four cone types rather than three, sensitivity into the ultraviolet, and in some species a magnetic sense. Whatever the association area is integrating, it is integrating a richer sensory world than the primate one, which is a point the architecture of sensory worlds makes repeatedly and which rarely survives into popular accounts of how clever crows are.
Recording from that region has produced the most direct evidence in the field. Crows have neurons that encode abstract rules independent of the specific stimulus or the specific response, firing for the concept of same versus different rather than for any particular pair of images. Other neurons encode number, responding preferentially to a specific quantity of visual items regardless of how those items are arranged or what they are. That single-neuron evidence is the reason bird intelligence stopped being a behavioral curiosity and became a neurophysiology program.
The NCL is not the only convergence in the avian forebrain and the others are worth naming, because a single coincidence is an anecdote and a pattern is an argument. Work published in 2020 identified a repeating circuit motif in the avian sensory pallium with an orthogonal, layer-like organization of fibers, described as a canonical circuit comparable in arrangement to the columnar organization of mammalian cortex, in tissue that is anatomically nuclear rather than laminar. Birds also possess a hippocampal formation that handles spatial memory and that enlarges seasonally in caching species. Three separate functional analogues to three separate mammalian structures, in a forebrain built on a different plan.
What the 2025 papers found about bird intelligence
Two studies published in Science in February 2025 changed the framing again, and the change is subtle enough that most coverage flattened it.
The first used birthdating analysis, single-cell RNA sequencing, and spatial transcriptomics to compare the development of known pallial sensory circuits across chicken, gecko, and mouse. The evolutionary convergence of sensory circuits in the pallium of amniotes turns out to be exactly that: convergence. The neurons occupying equivalent positions in structurally and functionally comparable circuits are generated at different developmental times, from different brain regions, with an early divergence in the transcriptomic trajectory of the excitatory neurons. Same circuit, different construction, and the divergence shows up early rather than as a late cosmetic difference.
The second built a cell type atlas of the avian pallium from adult and developing chicken and compared it against mouse and non-avian reptile data. The developmental origins and evolution of pallial cell types in birds showed that inhibitory neuron types are largely conserved across all amniotes, which is the boring and expected result, while excitatory neuron repertoires diverged substantially outside the hippocampal formation. Only a fraction of cells in the hyperpallium are homologous to neurons in the mammalian isocortex. And there is an extensive developmental convergence of gene expression programs between excitatory populations in the hyperpallium and the nidopallium, occurring late in development, which explains why those cells look similar in adults despite arising from different territory.
Read those together and the conclusion is uncomfortable for both of the older camps. The people who argued for homology on the basis of shared circuitry were right that the circuits are equivalent and wrong that they are inherited. The people who argued that shared circuitry arose convergently were right, and the developmental data now say so at the resolution of individual cell types.
What this means for bird intelligence is the interesting part. It is not a story about a hidden cortex. It is a story about two independent solutions to the problem of building high-order sensory processing and executive control, arrived at from different starting material, converging on comparable circuit architecture because the problem constrains the answer. The layered sheet is one way to do it. The nuclear cluster is another. Neither is the requirement.
It is worth being precise about what convergence means here, because the word gets used loosely. Convergent evolution normally describes independent lineages arriving at similar external form under similar selection: the camera eye in vertebrates and cephalopods, streamlining in sharks and dolphins. What the developmental data describe is convergence at the level of gene expression programs during late development, where cell populations arising from different pallial territory end up transcriptomically similar in adults. That is convergence operating inside the developmental process rather than only on the finished product, and it is a considerably more surprising claim than two animals ending up with similar-looking organs.
Number, and crows counting out loud
The numerical work is the cleanest demonstration available of abstract cognition in a bird, and it got substantially more impressive in 2024.
The background: crows have number-selective neurons in the endbrain that respond to specific quantities, whether the items are presented all at once or in sequence, which means the representation is abstracted away from the format of presentation. Later work found sensorimotor number neurons that translate a perceived quantity into a planned number of actions, which is the bridge between seeing three and doing three.
Then a study published in Science in 2024 pushed it into vocal production. Carrion crows were trained to produce a specific number of caws, from one to four, in response to arbitrary cues, and then to peck a key to indicate they were finished. They did it. More revealing than the success rate is what the acoustic analysis found: the properties of the very first caw in a sequence predicted how many the bird was going to produce in total. The crow had planned the count before it started counting.
That single finding requires several capacities stacked together. Volitional control over vocalization, which is rare in animals and which corvids possess. An abstract representation of number decoupled from any particular sensory display. A motor plan specified in advance for a variable-length sequence. And enough monitoring to know when to stop. When the birds made errors, the errors were structured rather than random, in ways consistent with the same kinds of counting failures humans make.
The vocal control element is easy to skate past and is arguably the rarest capacity in the whole result. Most animals cannot produce a vocalization on command. A dog does not reliably bark when asked, a monkey will not vocalize to a cue, and the great apes are notoriously poor at volitional call production, which is one of the standing puzzles in the evolution of speech. Corvids can be trained to caw on cue and to withhold cawing on a different cue, and recordings from the same forebrain region show neurons carrying the decision to vocalize in advance of the vocalization. A bird sitting between a primate and a human on a capacity central to language is not the arrangement anyone drew up.
The comparison worth drawing is not to human arithmetic, which involves symbols and exact quantities and a great deal of cultural scaffolding. It is to the approximate number system that human infants and many animals share. Crows are running that system with unusual precision and unusual voluntary control over the output channel, in a brain with no layered cortex, which is the recurring theme. Bird intelligence keeps turning out to involve the same computations the mammalian literature describes, implemented in tissue that a mammalian neuroanatomist would not recognize.
The consciousness experiment
In 2020 the same laboratory published the result that made comparative neuroscientists sit up, and it deserves careful statement because it is easy to overclaim.
Carrion crows were trained on a visual detection task using stimuli near the threshold of visibility, the standard paradigm for separating what was physically presented from what the subject reports perceiving. Single neurons in the pallial endbrain were recorded during performance. The activity followed a two-stage temporal pattern: an early component tracking the physical intensity of the stimulus, and a later component that tracked the bird’s subsequent report about whether it had seen anything, even when the stimulus was identical across trials.
That later component is what researchers in human consciousness studies treat as an empirical marker of subjective awareness rather than mere sensory processing. Finding it in a brain with no layered cortex removes one of the standing arguments for why consciousness should be restricted to mammals. One commentator described it as another fallen pillar, which captures the situation: not proof of avian consciousness, but the removal of a specific reason to doubt it.
The caveats are real and should be stated. A neural correlate is a correlate. The task requires the bird to report, so the later activity could reflect decision or response preparation rather than awareness as such, which is precisely the confound that no-report paradigms were invented to address in humans. And a marker validated in one species does not automatically validate in another. What the result establishes is narrower than the headlines and still significant: whatever generates that signature in a primate does not require the anatomy people assumed it required.
A follow-up from the same group added a detail that complicates the tidy version. The subjective experience of a stimulus being absent, not merely the failure to detect one, was also associated with specific neuronal activation rather than with a resting state, which suggests the system is representing seeing nothing as a positive perceptual outcome rather than as an absence of signal. Whether that strengthens or weakens the consciousness interpretation depends on which theory of consciousness you were holding when you walked in, which is a fair summary of the entire field’s current condition.
Caching, deception, and what another bird saw
The behavioral literature that predates all of this remains the strongest evidence for something like social cognition, and food caching is where it lives.
Western scrub jays cache food and recover it later, and the recovery behavior encodes what was cached, where, and how long ago. Offered perishable and non-perishable items, they preferentially recover the perishable ones after short delays and switch to the durable ones after long delays, which requires tracking elapsed time per cache rather than merely remembering locations. That combination of what, where, and when is the operational definition of episodic-like memory, and it was demonstrated in a bird before it was demonstrated in most mammals.
The cache protection work is better. A jay that was observed by another bird while caching will return later, alone, and re-cache the items elsewhere. It does this selectively: not when it cached in private, and not when the observer was a partner rather than a competitor. The finding that made the literature famous is that only jays which had themselves stolen from other birds’ caches showed the re-caching behavior. Birds with no thieving history did not bother, which is a within-species difference driven by individual experience rather than a species-level capacity, and that specificity is what made the result hard to explain away as a general caution response. The interpretation, which remains contested and which the original authors framed carefully, is that the experienced thief can project its own past behavior onto the observer.
Ravens extend it. In experiments using a peephole arrangement, ravens cached more cautiously when they could hear a competitor and knew, from prior experience with the apparatus, that a peephole was open, even with no bird visible. The relevant variable was not seeing a competitor but understanding that being seen was possible. The ravens whose behavior in the wild keeps outrunning what the laboratory predicts are among the few non-primates for which a serious case has been made about attributing perceptual states to others.
Clark’s nutcracker sits at the other extreme of the caching problem, storing tens of thousands of pine seeds across thousands of separate sites in a season and recovering them months later under snow, which is a spatial memory feat that no primate approaches and which is achieved with a hippocampus that enlarges seasonally to support it.
The comparison across taxa on this specific capacity is instructive about how badly single-axis rankings perform. On cache recovery a nutcracker beats every ape, every cetacean whose spatial behavior has been characterized, and every human. On social manipulation the scrub jay result sits close to the primate literature. On tool manufacture the cockatoos and crows are competitive with apes. These are not the same bird, and no single species does all of it, which is the argument against treating bird intelligence as a scalar quantity that species can be ranked on.
Planning for a future you are not in
Whether animals can plan for a future motivational state, rather than acting on present drives, was one of the durable open questions in comparative cognition, and ravens produced the result that moved it.
Ravens were trained to use a specific tool to extract a reward from an apparatus, and separately to exchange a specific token with a human for food. They were then offered, at a time when the apparatus was absent, a choice between the correct tool and several distractors including an immediately consumable treat. They selected the tool, held it, and used it successfully when the apparatus returned up to seventeen hours later. They did the same with the bartering token. Performance was comparable to and in some conditions better than apes tested on similar protocols.
The self-control component matters as much as the planning. Choosing a tool over an immediately available food item requires overriding a present drive in favor of a future one, and the ravens did it at rates that get reported as delay-of-gratification results in primates.
Skeptics have pushed back on whether this demonstrates mental time travel or a well-learned association between object and eventual reward, and the objection is legitimate: an animal that has learned tool equals food later does not obviously need to imagine the future to pick the tool. The counterargument runs through the flexibility of the transfer across contexts and object types. This is not settled, and the honest position is that ravens perform at the level that would be taken as evidence of planning if they were primates, which is at minimum a problem for anyone applying different standards across taxa. That asymmetry in evidentiary bars is worth watching for generally. A behavior observed in a chimpanzee gets described in mentalistic language by default; the same behavior in a bird gets an associative explanation first, and the difference is frequently about the observer’s priors rather than about the data.
The parrots that dismantle whatever is left unattended contributed a separate landmark: kea were shown to make true statistical inferences, judging the likely contents of a sample drawn from populations with different ratios of rewarded to unrewarded tokens, and integrating physical constraints and social information into those judgments. Before that result, integrated statistical inference of that kind had been demonstrated only in great apes and human infants.
Culture, and what actually spreads between birds
Individual capacity is one question. What a population accumulates is another, and the transmission side has produced results that the cognition literature sometimes underweights.
Sulphur-crested cockatoos in suburban Sydney worked out how to lift the lids of kerbside waste bins, and researchers tracking reports across dozens of suburbs found the behavior spreading outward geographically in a pattern consistent with social learning rather than independent invention, with birds in different areas converging on locally distinct opening techniques. That is a behavioral tradition with regional variants, emerging within a decade, in an urban environment that did not exist when the species evolved.
Parrots go further on the vocal side. Green-rumped parrotlets have been shown to learn individually distinctive signature contact calls, with parents producing distinct calls toward each nestling and the young converging on modified versions of those calls, which is a naming system in the loose but defensible sense that individuals are identified by a learned vocal label. Wild parrot populations maintain regional call dialects with boundaries that can be mapped, and birds moving between areas modify their calls toward the local variant. That is the same structural phenomenon documented in cetacean pods whose call types mark group membership and in bottlenose populations with location-specific behavioral repertoires, arrived at independently in an animal with an entirely different vocal apparatus.
Crows transmit information about specific threats. The Seattle mask work found that birds which had never been captured learned to scold the dangerous face, acquiring it from the behavior of birds that had, and the response persisted across years and spread through the population. That is social transmission of a specific, arbitrary, individually identified piece of information, which is a demanding definition of cultural learning to meet.
The songbirds whose dialect boundaries shifted measurably as a city changed around them demonstrate the same machinery on a longer timescale, and the cranes that lost a migratory route and had to be taught one show what the failure mode looks like. The study of how knowledge moves between animals generally keeps returning to birds for the same reason the cognition literature does: the behavior is visible, the populations are trackable, and the transmission can be watched happening.
Parrots, vocal learning, and the Alex problem
Vocal learning, meaning the ability to acquire novel vocalizations by imitation rather than producing an innate repertoire, has evolved in a small handful of lineages: songbirds, parrots, hummingbirds, some cetaceans, bats, elephants, and us. Most animals cannot do it at all.
Parrots do it with an unusual neural arrangement. Where songbirds have a single set of song nuclei, parrots have a duplicated system, with a core set of regions surrounded by an outer shell, and the shell appears to be a parrot innovation associated with their exceptional vocal flexibility. The songbirds whose local dialects can be mapped from neighborhood to neighborhood run the core system alone and still produce culturally transmitted regional variation, which suggests the shell buys flexibility rather than learning as such.
Then there is Alex, the African grey parrot studied for thirty years, who could label dozens of objects, colors, shapes, and materials, answer questions about which of two objects differed and in what respect, and who was reported to use the word none to indicate the absence of a difference, which is a zero-adjacent concept that took human mathematics several thousand years to formalize.
Alex deserves both the credit and the asterisks. The credit: the work was conducted with an explicit protocol designed against the Clever Hans problem, using a model-rival training method and testing by people who did not know the target answer. The asterisks: it is a single bird, studied by a researcher deeply invested in the outcome, over decades, with a great deal of the evidence taking the form of reported episodes rather than controlled trials. Single-subject research is not worthless, but it cannot bear the weight that popular accounts put on it, and the appropriate posture is that Alex established what was worth testing systematically rather than settling it. Subsequent work with other grey parrots using controlled designs has replicated parts of the repertoire, including inference by exclusion, where a bird shown that a reward is not in one container selects the other without further information. That capacity has since been demonstrated in several parrot species and in corvids, which is how a single-subject claim is supposed to graduate.
Longevity is the underappreciated parrot variable. Large parrots live fifty to eighty years and some cockatoos considerably longer, which is extraordinary for an animal of that mass and which correlates with relative brain size across parrot species. A long life changes the cognitive calculus entirely: it makes learning worth the investment, it allows accumulated knowledge to pay off, and it creates the generational overlap that social transmission requires. The cockatoo species that manufactures a drumstick and beats out individually distinctive rhythms has the decades available to develop a personal style, which a short-lived animal does not. The contrast with the cephalopods that build comparable problem-solving capacity and then die on schedule after a single reproductive event is close to a controlled comparison on what longevity buys, and what it buys is the possibility of anything accumulating at all.
The claims that do not hold up
An audit, because this field has generated real results and a great deal of adjacent nonsense.
Birds have primitive brains is the original error and it is dead, but the replacement claim that bird brains are basically the same as mammal brains is also wrong, and the 2025 developmental work is what kills it. The circuits are functionally equivalent. The cells are not homologous. Both halves matter.
The magpie mirror test is the weakest widely cited result in the corvid literature. A 2008 study reported that magpies with colored marks on their throat feathers engaged in mark-directed behavior in front of a mirror, which was taken as the first demonstration of mirror self-recognition outside mammals. A subsequent large replication attempt failed to reproduce it. That does not prove magpies lack the capacity, but the honest current status is unreplicated, and it continues to be cited as established.
Crows hold grudges is, unusually, true. Work in Seattle using masks found that crows learned to recognize and scold specific human faces associated with capture, that the response persisted for years, and that it spread to birds which had never been captured, meaning the information was transmitted socially rather than learned individually. This one survived scrutiny.
Parrots have the intelligence of a five-year-old child is a framing that should be retired regardless of which species it is applied to. Human developmental stages are a package of interlocking capacities, and an animal that matches a five-year-old on one task will be nowhere near on another and superhuman on a third. Clark’s nutcracker beats every five-year-old ever born at cache recovery and cannot do anything else a five-year-old does. The same objection applies to every version of this comparison, including the ones made about great apes and about elephants, and it is a formulation that survives only because it is easy to write headlines around. The bowerbirds that build and decorate structures with a sophistication nobody expected would score terribly on any test designed around primate capacities and are doing something no primate does.
Talking parrots understand what they say conflates mimicry with reference. Most vocal production by companion parrots is contextual association at best. The Alex work was specifically designed to distinguish referential use from mimicry, which is precisely why it required thirty years and a purpose-built protocol.
And bird intelligence is not one thing. Corvids and parrots diverged from each other well over sixty million years ago and are not close relatives, which means the two most cognitively impressive bird groups represent two separate elaborations, not one. Lumping them is convenient and slightly misleading. It also obscures a real difference in style: corvids dominate the caching, planning, and social-cognition literature, while parrots dominate vocal learning, longevity, and object manipulation. Two elaborations, two emphases.
Bigger brains mean smarter animals also fails on the bird data specifically. Ostriches have large brains with low neuron density and are not notable problem solvers. Ravens have small brains with high forebrain neuron counts and are. Mass was the wrong variable and it took a decade of tedious cell counting to establish that, which is a good argument for tedious cell counting.
Two constructions, three hundred and twenty million years apart
What the last five years have produced is a cleaner comparison than the field had before, and the cleanliness comes from the anatomy being different rather than the same.
If bird cognition ran on a hidden cortex, it would be a story about one solution appearing twice through shared inheritance, which is interesting but limited. What the developmental data show instead is that the last common ancestor of birds and mammals, an amniote living something like three hundred and twenty million years ago, did not hand down the machinery. Both lineages built high-order association areas out of different pallial territory, from neurons born at different times through divergent transcriptomic programs, and arrived at circuits that do the same jobs.
That makes birds the best available test of which features of cognition are contingent and which are forced. Layered cortex: contingent. A dopamine-rich association area sitting downstream of sensory convergence: apparently forced, since both lineages built one. Abstract rule representation, numerical representation, and a two-stage neural signature during perceptual report: present in both, from different tissue.
There is a further reason to care about a second construction beyond comparative zoology. Any claim about what cognition requires, including claims about whether it can be built in systems with no evolutionary history at all, rests on a sample. Until recently that sample was effectively one architecture, described in mammals and generalized from there. Birds provide an independent draw, and the fact that the independent draw converged on dopamine-modulated association areas, abstract rule coding, and a two-stage perceptual signature is evidence about which features are load-bearing rather than incidental. A sample of one cannot distinguish necessity from accident. A sample of two starts to.
It also puts a specific number on the hardware question. Roughly two billion neurons in a macaw, densely packed, forebrain-weighted, with short conduction paths, in a brain that has to be light enough to fly. Flight imposes a mass constraint that mammals never faced, and the avian solution to needing more computation without more weight was to shrink the components rather than expand the housing.
That constraint-driven miniaturization is the mechanism, and it is more informative than any amount of admiration. The great apes we habitually use as the cognitive reference point built their capabilities with no weight budget at all, the elephants running the largest terrestrial brains had even less pressure to economize, and the cetaceans carrying the biggest brains on the planet are neutrally buoyant and effectively free of the constraint. Only the birds had to be clever in fifteen grams while remaining airworthy. The social carnivores coordinating hunts across open country and the sentinel systems that let a small mammal forage in the open solved their cognitive problems with no mass ceiling whatsoever, and it shows in the hardware rather than in the behavior.
Which is the emphasis the 24-lecture Neurozoology course carries throughout, running the first edition’s tour of nervous systems forward with the constraints in front and the superlatives thrown out, alongside the study of how knowledge actually moves between animals and the working animals whose capacities got discovered by people who needed something from them. The birds that carried messages through artillery fire were running navigation systems nobody had characterized, and the ones credited with saving a village were doing it on the same fifteen-gram budget.
An anatomist in 1900 looked at a crow forebrain, saw no layers, and wrote down striatum. The bird was, at that moment, capable of representing an abstract rule, tracking who was watching it, and planning a sequence of vocalizations before beginning to produce them. Bird intelligence was never hiding. It was sitting in plain view, in a forebrain that had been given the wrong name because the people naming it already knew what they expected to find. The anatomy was not wrong about what it saw. It was wrong about what layers are for.