How Insects Built Their Own Version of Brains

[00:00:00]There is a structure inside a bee's head that is smaller than a grain of sand. It contains a few thousand neurons arranged in paired lobes, receives signals from every sensory system the bee possesses, [music] and produces from that impossibly small substrate something that functions like memory, like judgment, like the ability to learn from experience [music] and act differently because of it.

[00:00:22]Neuroscientists have a name for it. They call it the mushroom body.

[00:00:26]And for the last 25 years, the more carefully they have studied it, the more difficult it has become to explain what it does without using words we normally reserve for minds.

[00:00:36]We have always drawn a sharp line between insects and cognition. The assumption was almost architectural, that thought requires a certain scale [music] of brain, a cortex, a hippocampus, a prefrontal region capable of holding competing options in tension long enough to choose between [music] them.

[00:00:53]Insects do not have any of those structures. Their nervous systems are compact, distributed, and ancient. And for most of scientific history, that was taken as evidence of their cognitive simplicity. The evidence now says otherwise.

[00:01:07]A landmark review published in the journal Learning and Memory in 2024, drawing on 25 years of experimental research, concluded that the mushroom body is a complete system for sensory encoding, memory formation, and behavioral decision-making.

[00:01:22]Not a precursor to these things, not an approximation, a complete system implemented in a structure that in most insects contains between 2,000 and 3,000 neurons. For context, the human brain contains approximately 86 billion. The bee achieves the same functional categories of cognition with roughly 40 million times fewer neurons.

[00:01:44]A study published in Current Biology in 2024 by researchers at the University of Bristol documented something even harder to dismiss.

[00:01:53]Heliconius butterflies, navigating the dense layered canopy of tropical forest environments have evolved expanded mushroom body circuits that support spatial memory and route planning.

[00:02:04]These butterflies return to the same flower patches across days and weeks constructing what appears to be a cognitive map of their territory.

[00:02:12]Spatial navigation of this kind has historically been assigned to the vertebrate hippocampus, a structure that took hundreds of millions of years to evolve and requires a skull to house it.

[00:02:23]The butterfly solved the same problem with the circuit that fits inside a pinhead.

[00:02:28]The question this raises is not whether insects are intelligent in the way humans are intelligent. They are not and the differences matter.

[00:02:36]The question is more fundamental than that.

[00:02:38]If cognition can be built from 2,000 neurons organized in the right way, then the threshold requirements for a mind are not what we assumed.

[00:02:46]The architecture is what matters, not the scale and the insect brain, specifically the mushroom body at its center, may be the most elegant cognitive architecture that natural selection has ever produced.

[00:02:58]This film follows that architecture from its origins in the simplest nervous systems on Earth to its role recognized in a 2024 paper in Biology Letters as the primary scientific model for understanding how neural circuit evolution drives cognitive innovation across all of animal life.

[00:03:16]We will trace how a fly learns to fear a smell it has never encountered before.

[00:03:21]We will examine how a butterfly builds a map of a forest in a structure smaller than the period at the end of this sentence.

[00:03:28]We will consider what it means that the same architectural solution to the problem of a mind has appeared independently across hundreds of millions of years of evolution in lineages that share no common ancestor with a complex brain.

[00:03:41]By the end, the insect brain will look less like a simplified version of something greater and more like a proof written in neurons and preserved across geological time that intelligence was never the exclusive property of the large.

[00:03:55]Welcome to Omni, where science meets story and curiosity leads the way.

[00:04:00]Every week we explore the hidden patterns of life, the universe, and everything in between.

[00:04:06]If that sounds like your kind of journey, leave a thought in the comments, hit like, and subscribe to keep exploring with us.

[00:04:25]The honeybee brain weighs less than 1 mg. It is smaller than a sesame seed, lighter than a single raindrop, and contains approximately 1 million neurons packed into a structure so compact that it sits comfortably in the space between a thumbnail and the flesh beneath it. By any conventional measure of cognitive capacity, this brain should be incapable of producing anything more sophisticated than reflexive stimulus-driven behavior.

[00:04:51]And for most of scientific history, that was precisely the assumption that governed how researchers thought about insects.

[00:04:58]The human brain, by comparison, contains approximately 86 billion neurons and weighs around 3 lb.

[00:05:05]It is the most metabolically expensive organ in the body, consuming roughly 20% of the body's total [music] energy budget despite accounting for only 2% of its mass.

[00:05:14]The mammalian brain is a monument to biological investment, and the cognitive capacities it produces language, abstract reasoning, long-range planning, emotional memory, social inference have traditionally been explained by reference to its scale and its architecture, the cortex, the hippocampus, the prefrontal regions that sit at the front of the skull and perform the executive functions we associate with deliberate thought.

[00:05:40]Remove those structures, and the theory goes, you remove the capacity for complex cognition.

[00:05:46]Insects have none of those structures.

[00:05:48]Their nervous systems are distributed across ganglia running the length of their bodies.

[00:05:53]Their brains, to the extent the word applies, are not enclosed in bone, but suspended in a head capsule [music] of hardened protein.

[00:06:01]And for most of the 20th century, the scientific consensus held that insects operated on a fundamentally different level from vertebrates.

[00:06:08]Capable of impressive behaviors, yes, but behaviors that were largely pre-programmed, genetically fixed, and cognitively shallow.

[00:06:15]The problem with that consensus is that it never fully accounted for what insects actually do.

[00:06:21]A foraging honeybee navigates several kilometers from its hive to a food source it has never visited before, encodes the location, distance, and direction of that source in a waggle dance that communicates precise vector information to nestmates, and then returns [music] to the same site repeatedly across days, adjusting its behavior based on whether the resource is still productive.

[00:06:43]A desert ant returns to its nest from foraging distances of over 100 m across terrain with no landmarks, using a continuously updated path integration system that accounts for direction and distance simultaneously.

[00:06:55]A parasitic wasp locates a caterpillar host buried beneath the leaf litter it cannot see, assesses whether it has already been parasitized, and makes a decision about whether to lay eggs based on that assessment.

[00:07:07]None of these behaviors are simple. All of them require sensory encoding, memory, and behavioral flexibility.

[00:07:13]And all of them are produced by brains that, in absolute neuron count, are separated from the human brain by a factor of tens of millions.

[00:07:21]The structure at the center of the explanation for how this is possible is called the mushroom body.

[00:07:26]It is a paired brain region found in nearly all insects, named in the 19th century by the French zoologist Félix Dujardin, who first described its distinctive lobe and cup morphology in a paper published in 1850.

[00:07:40]For most of the century that followed, the mushroom body was treated as a sensory processing center of moderate interest.

[00:07:47]What researchers have established over the last quarter century has substantially revised that picture.

[00:07:52]A landmark review published in the journal Learning and Memory in 2024, synthesizing 25 years of experimental research from laboratories across Europe, North America, and Asia, concluded that the mushroom body is not a sensory relay station.

[00:08:08]It is a complete integrated system for sensory encoding, associative memory formation, and behavioral decision-making.

[00:08:16]The review encompasses hundreds of studies conducted in fruit flies, honeybees, moths, locusts, and cockroaches. And its central conclusion is that the cognitive functions previously considered exclusive to structures found only in vertebrate brains are, in insects, implemented within a circuit containing between 2,000 and 3,000 neurons.

[00:08:37]The scale of that disproportion, 86 billion neurons on one side and 3,000 on the other, is not a story about how insects are lesser. It is a story about what brains actually need in order to think.

[00:08:50]What evolution built in the insect head was not an approximation of a mind. It was a different solution to the same problem, arrived at independently, refined across hundreds of millions of years, and preserved with remarkable structural consistency [music] across the full diversity of insect life.

[00:09:08]Understanding how it works requires starting at of anatomy, with the question of what exactly the mushroom body is and how it is put together.

[00:09:17]The mushroom body gets its name from its shape. Viewed under a microscope in transverse section, the structure presents two lobes that curve upward and outward from a central stalk, and above them a paired cup-shaped region called the calyx.

[00:09:32]The resemblance to the fruiting body of a fungus was close enough that Dujardin's original nomenclature held, and it remains the standard term across all subsequent literature.

[00:09:42]But, the shape is not incidental to the function.

[00:09:45]The architecture of the mushroom body is, in a precise biological sense, the explanation for what it does.

[00:09:52]The bulk of the mushroom body is composed of a single cell type, the Kenyon cell, named after the American entomologist Frederick Kenyon, who provided the first detailed cellular analysis of the structure in 1896.

[00:10:05]In the fruit fly Drosophila melanogaster, there are approximately 2,000 Kenyon cells in each mushroom body hemisphere, giving a total of around 4,000 for the full structure.

[00:10:16]In the honeybee, the count is closer to 170,000 per hemisphere, reflecting the substantially greater cognitive demands of a social organism that navigates complex environments, communicates spatial information, and maintains a colony of up to 50,000 individuals.

[00:10:32]In most other insects studied to date, the number falls somewhere between these two values, but the architecture remains the same.

[00:10:39]Kenyon cells are unusual neurons.

[00:10:42]They're extremely small, densely packed, and they receive input through their dendrites in the calyx from projection neurons arriving from the insect's olfactory system, its visual system, and its mechanosensory pathways.

[00:10:55]This convergence of multiple sensory modalities onto a single compact structure is a key feature.

[00:11:02]In vertebrate neuroscience, sensory integration of this kind typically requires the coordinated activity of multiple cortical regions and subcortical relay stations.

[00:11:11]In the insect [music] brain, it happens within a few thousand cells arranged in a structure roughly a tenth of a millimeter across. The calyx, the cup-shaped input region, is where the sensory information arrives.

[00:11:23]The axons of Kenyon cells then extend downward and outward through the peduncle, the stalk of the mushroom, and branch into two sets of lobes.

[00:11:31]The vertical lobes, sometimes called the alpha and alpha prime lobes depending on the cell type involved, and the horizontal lobes, designated as the beta, beta prime, and gamma lobes.

[00:11:41]These lobes are not anatomically identical, and their functional differences are now reasonably well understood.

[00:11:47]The gamma lobe neurons process short-term memory and appetitive associations.

[00:11:52]The alpha prime and beta prime neurons are involved in longer-term storage and in updating memories in light of new information.

[00:11:59]The alpha and beta neurons are most closely associated with the retrieval and expression of consolidated long-term memories.

[00:12:07]Running throughout and around these lobes are two other critical cell populations.

[00:12:11]The dopaminergic neurons, arriving from a cluster of cells called the dorsal paired medial and protocerebral posterior lateral groups, modulate the activity of Kenyon cells by signaling the valence of an experience, whether it was rewarding or punishing.

[00:12:27]And the mushroom body output neurons, which receive signals from Kenyon cell axons in the lobes and project downstream to motor systems, are the cells that actually translate what the mushroom body knows into what the animal does.

[00:12:39]The circuit, in [music] its most distilled form, is this.

[00:12:43]Sensory information arrives at the calyx, activates a sparse and specific pattern of Kenyon cells. Those Kenyon cells send signals down the lobes.

[00:12:52]Dopamine neurons modulate which signals get reinforced, and the output neurons either promote or suppress a behavioral response based on what the current pattern of activity means in light of what has been learned before.

[00:13:04]The entire system is small enough to fit on the head of a pin. It performs what, in vertebrate terms, would require the coordinated interaction of the hippocampus, the prefrontal cortex, the amygdala, and the basal ganglia.

[00:13:18]This compression of function into minimal architecture is not an accident.

[00:13:22]It reflects a fundamental principle of neural circuit design, that the computational operations required for learning and decision-making can be implemented in very different physical substrates, provided those substrates instantiate the right relational structure.

[00:13:38]The mushroom body does not approximate what vertebrate cortex does by being simpler.

[00:13:42]It implements a genuinely comparable set of operations through a circuit that evolved independently under different constraints across a different evolutionary lineage. The fact that the architecture is so consistent across [music] the full breadth of insect diversity, from flies to bees to beetles to moths, argues strongly that evolution found something close to an optimal solution and then kept it.

[00:14:05]Understanding why requires examining what the mushroom body actually does when an insect encounters something in the world and needs to remember it.

[00:14:13]Beginning with the function for which the system is most completely understood, the encoding of smell.

[00:14:19]Smell is the primary sensory modality of most insects, and the mushroom body's most deeply characterized function is the storage and retrieval of olfactory memories.

[00:14:29]The reason olfactory learning has been so [music] extensively studied is partly practical.

[00:14:34]Odors can be precisely controlled by a researcher in a way that visual or mechanosensory stimuli often cannot.

[00:14:41]They can be paired with rewards or punishments with precise timing, and the neural circuits that process them are relatively accessible. But the olfactory memory system of the insect mushroom body also turns out to be a particularly clean model of the general principles by which associative learning works at the cellular level, which is why it has attracted decades of intensive research.

[00:15:01]The model organism for most of this work is Drosophila melanogaster, common fruit fly.

[00:15:07]Drosophila has a number of properties that make it exceptionally useful for neuroscience.

[00:15:11]A fully sequenced genome, a nervous system accessible to genetic manipulation at the level of specific cell types, and a behavioral repertoire that includes robust measurable learning and memory formation.

[00:15:24]The classic paradigm for studying olfactory memory in Drosophila was developed by Seymour Benzer and his colleagues at the California Institute of Technology in the 1970s, and it is elegant in its simplicity.

[00:15:36]A fly is exposed to an odor while simultaneously receiving a mild electric shock.

[00:15:42]After this pairing, the fly develops an aversion to that odor and will avoid it when given the choice.

[00:15:47]This is conditioned odor aversion, a form of associative learning directly analogous to Pavlovian [music] conditioning in mammals.

[00:15:54]What happens in the mushroom body during this process has been worked out in substantial detail over the subsequent decades, particularly through research from Gero Miesenböck's laboratory at the University of Oxford and the laboratory of Josh Dubnau at Cold Spring Harbor.

[00:16:09]When an odor is presented to a fly, olfactory receptor neurons in the antennae activate, and their signals travel to the antennal lobe, where they are processed and relayed by projection neurons to the calyx of the mushroom body.

[00:16:21]There, a specific subset of Kenyon cells is activated, determined by which combination of olfactory channels was stimulated [music] by that particular odor.

[00:16:30]The key feature of this activation pattern is its sparseness. Out of 2,000 Kenyon cells, a given odor typically activates only around 5 to 10% of them, and the specific set of cells activated is highly consistent from one presentation of the same odor to the next.

[00:16:47]This sparse, reproducible coding means that different odors are represented by distinct, non-overlapping patterns of activity, which is exactly what is required for the system to store multiple memories without interference.

[00:17:00]The electric shock in the conditioned aversion paradigm activates dopaminergic neurons that project into the mushroom body lobes.

[00:17:07]These dopamine neurons signal aversive experience. When they are active at the same time as a specific pattern of Kenyon cells is being driven by an odor, the synaptic connections between those Kenyon cells and the downstream mushroom body output neurons are modified.

[00:17:21]Specifically, the synapses that connect the active Kenyon cells to output neurons promoting approach behavior are weakened, and those connecting to output neurons promoting avoidance are strengthened.

[00:17:33]The result is that the next time that odor is encountered, the same pattern of Kenyon cells activates, but now they drive avoidance rather than approach.

[00:17:41]The memory has been written into the synaptic weights of the circuit.

[00:17:45]The discovery that different mushroom body lobes store different types of memories added a further layer of organization to this picture.

[00:17:52]Work published by Scott Waddell's group at the University of Oxford demonstrated that short-term olfactory memories, those that fade within a few hours, are stored primarily in the gamma lobe neurons.

[00:18:04]Longer-term memories that persist across days require synaptic changes in the alpha prime and beta prime lobes, which are driven by a different population of dopamine neurons and depend on protein synthesis in a way that short-term memories do not.

[00:18:17]And the most durable [music] long-term memories, those that can persist across the lifetime of the fly, involve structural changes in the alpha and beta lobe synapses that are stabilized by sleep-dependent consolidation processes.

[00:18:31]This temporal stratification of memory across anatomically distinct mushroom body compartments is not unique to Drosophila.

[00:18:38]Honeybees show the same general organization with short-term and long-term olfactory memories dissociated by pharmacological interventions that target protein synthesis and by the anatomy of their mushroom body lobes.

[00:18:51]The conservation of this multi-compartment memory architecture across insects separated by several hundred million years of evolution argues that it represents a general solution to the problem of storing multiple memories with different time scales and different valences within a compact circuit.

[00:19:07]What this system achieves, in computational terms, is a form of pattern completion and pattern separation operating simultaneously.

[00:19:16]Pattern completion allows the system to recognize a familiar odor even when the sensory signal is noisy or partially degraded. Pattern separation ensures that similar but distinct odors are stored as separate memories and do not interfere with each other.

[00:19:30]Both properties emerge directly from the sparse coding of Kenyon cells and from the connectivity structure of the circuit.

[00:19:37]The mushroom body does not need to be large to accomplish this. It needs to be correctly wired.

[00:19:43]Memory and decision making are not the same process, but in any brain worth the name, they are connected.

[00:19:49]The value of remembering an experience lies precisely in its ability to guide future behavior, to inform the choice that the animal makes when it next encounters a situation it has encountered before.

[00:20:01]The mushroom body does not merely store what happened.

[00:20:04]Through its output neurons, it translates stored experience into behavioral decisions and the logic of how it does so challenges some of the foundational assumptions of vertebrate-centric cognitive science.

[00:20:15]The mushroom body output neurons, a population numbering roughly 34 in Drosophila, are the cells that ultimately determine what the animal does when the mushroom body is active.

[00:20:25]They project from the lobes of the mushroom body to downstream motor and pre-motor regions and their activity either promotes or suppresses specific behavioral states.

[00:20:34]In the context of olfactory memory, an output neuron called MBON-γ1β' one suppresses avoidance and promotes approach when active.

[00:20:44]Another output neuron, MBON-α3, does the opposite. The balance of activity across these output neurons at any given moment reflects the integrated history of what the animal has learned about the current stimulus and that balance is what drives behavioral choice.

[00:20:59]The dopaminergic neurons that modulate this system are not passive reporters of reward and punishment.

[00:21:05]They update the output neuron balance continuously based on ongoing experience.

[00:21:11]When an animal encounters a stimulus it previously associated with reward, and that reward does not materialize, specific dopamine neurons signal the prediction error, the discrepancy between what was expected and what occurred, and they adjust the Kenyon cell to output neuron synapses accordingly.

[00:21:28]This is a functional equivalent of the prediction error signals recorded from dopamine neurons in the mammalian midbrain, the signals that drive reinforcement learning in the vertebrate basal ganglia.

[00:21:39]The insect mushroom body implements this same fundamental algorithm using a fraction of the cellular hardware.

[00:21:46]The research group of Martin Giurfa at the University of Paul Sabatier in Toulouse has spent more than two decades documenting the range of cognitive operations that honeybees can perform using this system. In a series of studies published across the 2000s and 2010s, [music] Giurfa and his colleagues demonstrated that bees trained in a Y-shaped maze can learn to choose a stimulus that differs from a sample they have been shown, a capacity called delayed non-matching to sample, which requires holding a stimulus representation in short-term memory, and making a comparative judgment about it.

[00:22:18]They can learn to apply rules across novel stimulus categories, choosing the odd stimulus out in a set regardless of what the stimuli are, a demonstration of abstract rule learning.

[00:22:29]They can delay an approach response in exchange for a larger reward presented slightly later, a simplified version of delay discounting that mammalian researchers use to study self-control.

[00:22:40]In 2019, a study published in Science by Adrian Dyer at the Royal Melbourne Institute of Technology and colleagues demonstrated that honeybees could be trained to understand a rudimentary representation of the concept of zero as a quantity smaller than one.

[00:22:56]This is a capacity that human children typically acquire around the age of four, and that some non-human primates can be trained to demonstrate with effort.

[00:23:05]The bees learned it within a few training sessions. The brain performing this computation contains 1 million neurons.

[00:23:12]These findings collectively make a specific argument about the relationship between decision-making and the structures assumed to be necessary for it.

[00:23:20]The vertebrate prefrontal cortex is widely considered the neural substrate of flexible, context-dependent decision-making. The region that holds competing options in mind simultaneously, weighs their relative values, and selects among them based on long-term goals.

[00:23:36]Insects do not have a prefrontal cortex.

[00:23:38]They do not have anything anatomically homologous to it.

[00:23:42]But the functional operations that the prefrontal cortex performs, the integration of sensory information with remembered value, the comparison of options, the selection of behavior based on context, these operations are implemented in the insect mushroom body through a circuit that achieves the same output through different means.

[00:24:00]The cortex is one way to solve the decision-making problem. It is not the only way.

[00:24:06]The cognitive functions discussed so far, olfactory memory, associative learning, abstract rule application, are impressive, but they can still be accommodated with some effort within a framework that treats insect cognition as fundamentally reactive, as a system that learns associations between stimuli and stores their values, but does not construct an internal model of the world.

[00:24:29]Spatial navigation of the kind documented in a 2024 study of Heliconius butterflies requires something more than that.

[00:24:37]It requires something that looks functionally like a map.

[00:24:40]Heliconius butterflies are neotropical insects found across the rainforests of Central and South America.

[00:24:46]They are remarkable animals in several respects.

[00:24:49]Unlike most butterflies, which feed on nectar and live for a matter of weeks, Heliconius supplement their diet with pollen, which provides amino acids that enable a lifespan extending to several months.

[00:25:01]They are brightly colored in patterns that signal chemical unpalatability to predators, and those patterns vary across subspecies in ways that have made them a central study system for evolutionary biology.

[00:25:13]But the behavior that made them the subject of the 2024 Current Biology investigation from researchers at the University of Bristol is their foraging strategy.

[00:25:23]Heliconius butterflies do not forage randomly. They maintain what ecologists call a trap line, a fixed route through their forest territory that visits the [music] same set of host plants in the same sequence repeatedly across days and weeks.

[00:25:38]This behavior has been documented in field studies going back to the work of Lawrence Gilbert in the 1970s, but its neural basis remained poorly understood.

[00:25:46]The Bristol study, led by Stephen Montgomery and colleagues, addressed the question of whether the expanded mushroom bodies documented in Heliconius, relative to other butterfly species, were causally connected to this spatial memory capacity.

[00:26:00]The team compared mushroom body volume across 17 species of Heliconius and related butterflies, controlling for overall brain size and body size.

[00:26:09]They found a strong, statistically significant, positive correlation between mushroom body volume and the complexity of the home range in which each species forages.

[00:26:18]Species that maintain stable [music] trap line routes across complex three-dimensional forest environments have larger mushroom bodies relative to brain size than species that forage more opportunistically across simpler, more open habitats.

[00:26:31]The correlation is specific to the mushroom body. Other brain regions do not show the same pattern.

[00:26:36]The researchers also conducted behavioral experiments in a flight [music] cage environment, in which butterflies were trained to find a rewarded location using spatial landmarks, and then tested for their ability to navigate back to that location after a delay.

[00:26:50]Heliconius species with larger mushroom bodies performed significantly better on this task than related species with smaller ones. They navigated more accurately, required fewer trials to learn the location, and retained the learned route over longer delays.

[00:27:05]Spatial navigation using a stable cognitive map. The ability to hold the layout of an environment in memory and use it to plan efficient routes through that environment is a capacity that vertebrate neuroscience is historically attributed to the hippocampus.

[00:27:19]The hippocampus is a structure that in mammals encodes the relationships between spatial [music] landmarks and uses those relationships to construct and navigate an internal representation of space.

[00:27:31]Hippocampal place cells, neurons that fire selectively when an animal occupies a particular location, were discovered by John O'Keefe at University College London in 1971, work that contributed to his Nobel Prize in Physiology or Medicine in 2014.

[00:27:46]The Heliconius data do not prove that insect mushroom bodies contain place cells, or that the neural mechanism of insect spatial memory is identical [music] to that of the vertebrate hippocampus.

[00:27:56]They prove something in some ways more striking.

[00:27:59]That the functional outcome of hippocampal spatial processing, the construction and use of a stable cognitive map of a complex environment, can be achieved by a structure that evolved from an entirely different origin in an animal whose entire brain contains fewer neurons than a single cortical column of a mouse.

[00:28:16]The mushroom body volume correlates with spatial cognitive demand in Heliconius.

[00:28:19][music] For the same reason that hippocampal volume correlates with spatial cognitive demand in birds that cache food. Because the underlying computational problem is the same, and the structure tasked with solving it expands in response to the selection pressure.

[00:28:35]That spatial cognition is implemented in mushroom body circuits also extends the catalog of functions attributed to this structure in ways that were not anticipated by earlier frameworks focused primarily on olfactory learning.

[00:28:48]If the mushroom body can support the construction and use of allocentric spatial representations, then the range of cognitive operations it is capable of is substantially broader than a model of insect cognition built around smell would suggest.

[00:29:01]The question that this raises, the question that the convergent evolution of this structure across the animal kingdom makes unavoidable, is whether the mushroom body is not just an insect solution to cognition, but something closer to a universal solution. One that evolution has arrived at repeatedly because the underlying architecture is simply very good.

[00:29:20]The mushroom body of insects is not the only structure of its kind in the animal kingdom. Paired lobed and cup formations of densely packed small neurons receiving convergent input from multiple sensory systems and projecting to downstream motor regions through output neurons modulated by dopaminergic signals have been identified in a range of invertebrate phyla.

[00:29:41]Polychaete annelid worms possess structures with this organization in their supraesophageal ganglia.

[00:29:46]Several crustacean lineages, including mantis shrimps and certain crab species, have mushroom body analogs in their brains that are in some cases larger relative to total brain [music] size than those found in most insects.

[00:29:59]Some mollusks, particularly the nudibranch and certain predatory gastropods, have structures in their cerebral ganglia that share key [music] architectural features with mushroom bodies, including sparse sensory coding and dopaminergic modulation of associative plasticity.

[00:30:14]These structures did not inherit their organization from a common ancestor that possessed a mushroom body.

[00:30:20]The lineages in which they occur diverged from each other more than 500 million years ago before any of them had developed anything resembling a complex brain.

[00:30:29]The shared architecture is the result of convergent evolution of independent evolutionary lineages arriving at the same structural solution to the same computational problem.

[00:30:39]A 2024 paper published in Biology Letters by Rudy Loitgeb at the University of California San Diego and colleagues formalized this argument and proposed that the mushroom body and its analogs represent what the authors termed a canonical microcircuit for associative learning. A circuit architecture so computationally effective that evolution has discovered it multiple times across animal phylogeny.

[00:31:03]The concept of a canonical microcircuit has precedent in vertebrate neuroscience.

[00:31:08]The cortical column, a repeating unit of approximately 100 neurons organized across six layers that processes incoming sensory information and generates output signals to other regions, is thought to implement a generalized computation that is replicated with variation across the entire neocortex.

[00:31:25]The same basic circuit handles visual processing in the occipital lobe, auditory processing in the temporal lobe, and motor planning in the frontal lobe because the underlying operation, the integration of convergent input with stored patterns to generate context-appropriate output, is the same in each case.

[00:31:44]The mushroom body appears to be the invertebrate equivalent of this canonical circuit, a generalized solution to the integration and association problem that has been deployed across multiple lineages [music] because it works.

[00:31:56]The evolutionary logic of convergence is worth dwelling on. When two lineages independently arrive at the same solution to a problem, it is strong evidence that the solution is close to optimal given the constraints of the problem.

[00:32:08]The wing of a bat and the wing of a bird are convergent structures built from different anatomical materials through different developmental pathways because flight imposes specific aerodynamic requirements that constrain the range of viable solutions.

[00:32:23]The vertebrate eye and the cephalopod eye are famously convergent, having evolved independently to a nearly identical functional design because the physics of image formation on a curved photoreceptor surface >> [music] >> admits only a narrow range of effective implementations.

[00:32:38]The mushroom body's repeated convergent appearance across invertebrate phylogeny argues that associative learning under the constraint of minimal neural resources imposes specific architectural requirements.

[00:32:49]Those requirements, convergent multi-sensory input onto a sparse coding population, dopaminergic modulation of association weights, and behavioral output through a small number of projection neurons are precisely what the mushroom body implements.

[00:33:04]Every lineage that faced strong selection pressure for flexible associative learning in a resource-limited nervous system ended up in roughly the same place.

[00:33:13]This has a specific implication for how researchers should think about the relationship between intelligence and brain size.

[00:33:19]If the mushroom body architecture is a convergent solution to the core problem of learned flexible behavior, then the appropriate measure of a brain's cognitive capacity is not its total neuron count. It is whether the brain contains the right circuit, wired in the right way, with the right modulation.

[00:33:36]Scale matters in that it allows for greater numbers of memory traces and finer discriminations, but the fundamental operations of learning and decision-making do not require scale.

[00:33:46]They require architecture. The insect brain demonstrates this with a precision that no vertebrate brain can match, simply because the compression is so extreme that the architecture is all there is.

[00:33:57]The question of what cognition minimally requires is one of the oldest in philosophy of mind and one of the most practically underexplored in neuroscience. It tends to get displaced by more tractable questions, such as how specific cognitive functions are implemented in the mammalian brain, where the experimental tools are mature and the funding is reliable.

[00:34:16]But the insect mushroom body forces the question with a directness that is difficult to evade, because the numbers are simply too extreme to ignore.

[00:34:24]A bee with 1 million neurons learns to navigate several kilometers of terrain, encodes the location and productivity of food sources, communicates that information to nestmates, recognizes individual bees within its colony, updates its foraging decisions based on changing resource availability, and performs all of this while managing the demands of flight, predator avoidance, and social signaling.

[00:34:48]The question of how much neural substrate is actually required for this set of operations is not rhetorical.

[00:34:55]It is a precise empirical question with a precise empirical answer. And the answer that the mushroom body literature is converging on is far less than anyone assumed.

[00:35:05]The sparse coding of Kenyon cells is a large part of the answer. In the mushroom body, a given stimulus activates only a small fraction of the total Kenyon cell population, typically somewhere between 5 and 10%.

[00:35:19]This sparseness has a specific computational benefit. It maximizes the number of distinct patterns that can be stored without interference.

[00:35:27]A dense code in which most neurons are active for every stimulus saturates quickly, and novel stimuli become indistinguishable from previously stored ones.

[00:35:36]A sparse code in which each stimulus activates a unique low overlap subset of cells can in principle store an enormous number of distinct associations in the same circuit, as long as the total cell population is large enough that the sparse subsets do not randomly overlap.

[00:35:53]The mathematics of sparse coding in the Kenyon cell population has been worked out explicitly by researchers, including Ashok Litwin-Kumar at Columbia University, who published a theoretical analysis in 2017 demonstrating that the mushroom body circuit, with its specific connectivity parameters, can store on the order of 150 distinct olfactory memories before performance begins to degrade.

[00:36:17]For an insect that encounters dozens of relevant odors across its lifetime, each associated with specific locations, food sources, threats, or conspecifics, that storage capacity is more than adequate.

[00:36:30]The system is not limited by the number of neurons, it is limited by the information theoretic constraints on sparse pattern storage, and it operates close to the theoretical optimum.

[00:36:41]This observation has not been lost on the field of artificial intelligence.

[00:36:45]The mushroom body circuit bears a structural resemblance to a class of machine learning architecture called a random feedforward network, in which a large number of inputs are projected through a randomly connected intermediate layer onto a much smaller set of output [music] units, which then learn associations through a supervised or reinforcement based training signal.

[00:37:05]These networks, sometimes called random kitchen sink models or extreme learning machines, are computationally cheap, train quickly, and perform surprisingly well on classification tasks, precisely because the random projection through the intermediate layer creates the kind of sparse, high-dimensional representations that facilitate linear separability of different input patterns.

[00:37:27]The similarity is not coincidental.

[00:37:30]Research from Dmitri Rinberg's group at the New York University School of Medicine, and from Yoshida Tanaka at the RIKEN Brain Science Institute, has examined whether the specific connectivity parameters of the mushroom body, the ratio of projection neurons to Kenyon cells, the degree of random connectivity, and the sparseness of Kenyon cell activation, are close to those that would be predicted by an optimal sparse coding theory. They are.

[00:37:55]The mushroom body appears to have been shaped by selection toward the same computational optimum that machine learning researchers independently discovered through mathematical analysis. This raises a possibility worth taking seriously. If the insect mushroom body operates near the theoretical optimum for associative memory under resource constraints, and if the same architecture has been independently discovered by evolution across multiple lineages, then the question of what makes a cognitive system powerful is not primarily about scale.

[00:38:24]It is about architecture, about the specific wiring pattern that allows a small number of neurons to cover a large representational space efficiently.

[00:38:33]The human brain's 86 billion neurons achieve many things that the insect brain cannot. But the insect brain may be doing something that the human brain, with all its excess capacity, has not been forced by evolutionary [music] pressure to optimize.

[00:38:46]It is running cognition at something close to the minimum viable configuration. And that minimum is higher than we thought.

[00:38:54]The mushroom body literature, focused as it has been on Drosophila and honeybees, represents only one slice of what insects have been observed doing in the field and laboratory.

[00:39:04]When the full catalog of documented insect cognitive feats is assembled, the picture that emerges is one of a class of animals whose behavioral complexity substantially exceeds what the dominant frameworks of 20th century ethology predicted, and whose neural mechanisms are in many cases still poorly understood even at a descriptive [music] level.

[00:39:24]Bumblebees provide some of the most striking recent data.

[00:39:28]A 2017 study from the laboratory of Lars Chittka at Queen Mary University of London, published in Science, documented bumblebees learning to roll a small ball to a target location in order to receive a food reward after observing another bee performing the task.

[00:39:44]This is a form of social learning mediated by observation, a cognitive operation that requires representing the actions of another individual as goal-directed and inferring the relationship between those actions and their outcome.

[00:39:57]The bees not only learned the task from observation, but showed evidence of innovation.

[00:40:02]Some individuals that had been shown a suboptimal solution to the task devised a more efficient approach on their own after solving it through imitation. A follow-up study, published in 2023 in the same laboratory, again in Science, extended this finding to show that bumblebees can transmit behavioral innovations across generations of a colony through social learning chains, demonstrating what the authors described as the rudiments of cultural transmission.

[00:40:28]Ants present a different kind of cognitive challenge.

[00:40:31]The problem-solving capacities of ant colonies are well established, as are the navigational abilities of individual ants. But a 2021 study from Tomer Czaczkes at the University of Regensburg documented something more specific.

[00:40:45]Individual Cataglyphis desert ants, when prevented from using their normal path integration system, spontaneously switch to a landmark-based navigational strategy, demonstrating the ability to hold multiple navigational systems in reserve and deploy the appropriate one based on situational context.

[00:41:04]This kind of flexible [music] strategy selection, adapting the approach based on which information is currently available, is a hallmark of what cognitive scientists call executive function in vertebrates.

[00:41:15]The paper wasp, Polistes fuscatus, [music] provides one of the most counterintuitive data points in the insect cognition literature.

[00:41:22]Research from the Michael Sheehan laboratory at Cornell University, published in Science in 2011 and extended in subsequent work through 2021, documented that Polistes wasps can recognize and remember the individual faces of other members of their species, discriminating between dozens of individuals by their distinctive facial markings.

[00:41:43]Crucially, this capacity appears to be specific to faces within the species.

[00:41:48]The wasps do not show the same learning speed or accuracy for other kinds of visual stimuli, including artificially manipulated or scrambled versions of the same images.

[00:41:58]This suggests face recognition in Polistes is not a byproduct of general visual learning, but a specialized cognitive module, analogous in its functional organization to the face-selective regions of the primate temporal cortex.

[00:42:11]The monarch butterfly's navigational capacity belongs in a separate category of its own.

[00:42:16]Monarch butterflies undertake an annual migration of approximately 4,000 km from their summer breeding grounds in the northeastern United States and Canada to their overwintering sites in the oyamel fir forests of central Mexico, a journey completed by animals that have never made it before and cannot be guided by learned route knowledge.

[00:42:35]The navigation is accomplished using a time-compensated sun compass.

[00:42:40]An internal clock synchronized to a circadian rhythm, combined with a measure of solar azimuth, allows the butterfly to infer [music] a consistent directional heading toward the southwest regardless of the time of day.

[00:42:53]Research from Steven [music] Reppert's laboratory at the University of Massachusetts Medical School, extending from the early 2000s through 2019, mapped the neural circuits involved, demonstrating that time compensation information is processed in the central complex of the insect brain, with the mushroom body contributing input about learned visual landmarks that modify the default migratory heading near the destination.

[00:43:18]Perhaps the most philosophically unsettling finding in the recent insect cognition literature concerns metacognition, the capacity to represent the contents and reliability of one's own knowledge states.

[00:43:29]A 2019 study by Scarlett Howard and colleagues at Monash University and the University of Toulouse [music] demonstrated that honeybees trained on a difficult discrimination task could be conditioned to indicate uncertainty, [music] choosing an exit option that allowed them to avoid the task in exchange for a reduced reward when the discrimination was difficult and choosing to attempt the task when it was easy.

[00:43:52]This is a behavioral signature of uncertainty monitoring, the ability to assess one's own performance and represent [music] the degree of confidence in one's knowledge, which had previously been documented with any reliability only in primates, dolphins, and a limited number of other large-brained vertebrates. Each of these findings was peer-reviewed and published in journals with high methodological standards.

[00:44:16]None of them have been satisfactorily accommodated by a theoretical framework that treats insect behavior as essentially reflexive.

[00:44:23]Taken together, they describe a class of animals that learns from observation, transmits behavior across generations, navigates by multiple [music] independent systems, recognizes individuals, plans long-range routes, and monitors the reliability of its own knowledge.

[00:44:40]What makes this catalog intellectually significant is not that insects are revealed to be secretly equivalent to vertebrates. They are not.

[00:44:48]It is that the cognitive operations documented in these studies require explanation, and the mushroom body is the structure best positioned to provide it.

[00:44:56]In 2024, a paper published in Biology Letters made a claim that would have seemed extravagant even a decade earlier.

[00:45:04]The authors, drawing on the accumulated literature of mushroom body research across species, proposed that the insect mushroom body should be adopted as the primary model system for understanding how neural circuitry evolution drives cognitive innovation across the animal kingdom.

[00:45:20]Not one model among many, the primary model.

[00:45:24]>> [snorts] >> The system through which the general principles of how brains acquire new cognitive capabilities are most clearly visible and most rigorously testable.

[00:45:33]The argument rests on several converging lines of evidence.

[00:45:36]First, the mushroom body circuit architecture is well enough understood that specific predictions can be made about the relationship between its structural parameters >> [music] >> and its cognitive performance.

[00:45:48]Predictions that can be tested comparatively across species with different mushroom body sizes and different cognitive demands.

[00:45:55]The Heliconia study is one example of this kind of test. Second, the genetic tools available in Drosophila allow researchers [music] to manipulate the mushroom body with a precision that is not yet achievable in any vertebrate brain, enabling causal tests of which elements of the circuit [music] are necessary for which cognitive functions.

[00:46:15]Third, the convergent evolution of mushroom body analogs across invertebrate phylogeny [music] provides a set of natural experiments in how the same circuit architecture evolves under different selection pressures, offering comparative evidence that would take decades to accumulate in vertebrate model systems.

[00:46:32]The implications of positioning the mushroom body as a primary model extend well beyond insect neuroscience. [music] For vertebrate neuroscience, the mushroom body offers a simplified system in which the general principles of associative memory, reinforcement learning, and behavioral decision-making can be studied without the confounding complexity of a cortex containing billions of neurons and dozens [music] of anatomically distinct regions.

[00:46:58]Many of the mechanisms first characterized in the insect mushroom body have subsequently been identified in vertebrate brains, including the prediction error function of dopamine neurons, the role of sparse coding in memory storage, and the multi-compartment organization of short-term and long-term memory.

[00:47:16]The insect system is not a precursor to the vertebrate system. It is a clean, tractable instantiation of the same general principles.

[00:47:24]For the field of artificial intelligence and neural engineering, the mushroom body represents an existence proof of something that machine learning researchers have been trying to achieve through design, a system that learns quickly from few examples, stores many memories without interference, and generalizes to novel situations using minimal computational resources.

[00:47:45]The architecture that evolution found for this problem, sparse random projection onto a small output layer modulated by a value signal, maps closely onto architectures that have been independently derived in machine learning, and the specific parameters of the biological circuit are close to those that theoretical analysis predicts should be optimal.

[00:48:05]Building artificial systems that implement mushroom body-like architectures is an active area of research with applications in neuromorphic computing and low-power autonomous navigation.

[00:48:16]For the philosophy of mind, the mushroom body evidence arrives at a moment when the question of what cognitive systems are morally considerable is acquiring practical urgency.

[00:48:25]The question of animal consciousness has traditionally been addressed through reference to brain complexity, with the assumption that greater structural complexity correlates with greater subjective experience, and therefore with greater moral weight.

[00:48:39]The insect brain's demonstration that sophisticated cognitive functions can be implemented in extremely compact circuits undermines the assumption that brain size or structural complexity is the relevant variable.

[00:48:50]If the operations that define cognition, the encoding of experience, the storage of its value, the comparison of options, and the selection of behavior can be implemented in 2,000 neurons, then the threshold for moral consideration may need to be reconsidered, and it may need to be reconsidered in a direction that includes many more living things than our [music] current frameworks acknowledge.

[00:49:12]We began with a structure smaller than a grain of sand sitting inside the head of a honeybee performing operations that neuroscience took most of the 20th century to describe in vertebrates.

[00:49:23]That structure is still there in every bee foraging across a summer field, in every moth navigating by pheromone trails through a dark forest, in every butterfly plotting its route through a canopy that has not changed in a generation.

[00:49:37]It has been there in something close to its current form for at least 350 million years surviving every extinction event that cleared the stage of larger, more famous animals. The insects remain.

[00:49:50]The mushroom body remains.

[00:49:52]What the last quarter century of research has established is that the mushroom body is not a relic, not a simplified precursor to something more sophisticated that evolution was working toward. It is a finished product, a solution to the problem of learning and decision-making that works so well in such a compact space that natural selection has had little reason to change it since before the first vertebrate crawled onto land.

[00:50:17]The architecture that a fruit fly uses to learn which smell predicts food predicts pain is, in its essential organization, the same architecture that a heliconius butterfly uses to plan a route through a rainforest.

[00:50:29]The same arrangement of Kenyon [music] cells, dopaminergic inputs, and output neurons that allows a bee to remember hundreds of flower patches also allows a paper wasp to recognize the faces of its nestmates. The circuit scales with the cognitive demand. The fundamental design does not change. There is something worth sitting with in the specific compression this represents.

[00:50:52]The operations that the mushroom body performs, associative encoding, sparse memory storage, value-based decision-making, the integration of current sensory information with the residue of past experience to produce behavior that is calibrated to both, are the operations that the entire edifice of vertebrate cognitive neuroscience has been attempting to describe and explain for over a century.

[00:51:16]The insect brain performs them in a structure containing a few thousand cells. Not approximately. Not in a reduced or degraded form. Functionally, these are the same operations. The human brain does things the bee's brain cannot. It is not being argued otherwise, but the bee's brain does things on a scale that raises the prior probability that many more animals than we have assumed are doing something that deserves the word cognition. And the prior probability that cognition, wherever it occurs, requires less physical substrate than the size of our own brains has led us to believe.

[00:51:50]The mushroom body is in this sense an argument. It is nature's answer to the question of how much structure a mind needs. And the answer turns out to be far less than the most successful example of a mind would lead you to expect.

[00:52:04]The insect did not fail to build a cortex. It solved the problem a different way.

[00:52:09]It found an architecture that stores memories without a hippocampus, makes decisions without a prefrontal cortex, and navigate space without a cortical map using a circuit so well designed that it has remained essentially unchanged across hundreds of millions of years of a planet that has not stopped changing.

[00:52:27]That circuit now sits at the center of the most productive research program in comparative cognition. Not because insects are surprising, though they are, but because the design [music] is so good that understanding it may be the most direct available path to understanding what a mind, any mind, fundamentally is.

[00:52:47]We began by asking what a structure smaller than a grain of sand could possibly know.

[00:52:52]The answer, written in 350 million years of uninterrupted evolutionary success, is that it knows enough.