-
Pigeons That Read X-Rays: The Experiment That Proved Birds Can Spot Breast Cancer
In 2015, pathologist Richard Levenson at UC Davis and psychologist Edward Wasserman at the University of Iowa put 16 pigeons in individual chambers, each containing a touchscreen displaying digitized breast tissue biopsies. On either side of the image were two colored buttons—one for benign, one for malignant. If the pigeon pecked the correct button, a computer automatically dispensed a 45-milligram food pellet. If it pecked wrong, nothing happened. No humans were visible during training—the entire process was automated to avoid the Clever Hans effect, where animals appear to reason but are actually reading subtle cues from their handlers.
Within 15 days, individual pigeons were identifying cancerous breast tissue at 85 percent accuracy. When the researchers combined the responses of four birds in a “flock-sourcing” approach—taking the majority answer—accuracy climbed to 99 percent. That’s on par with trained human pathologists.
The pigeons weren’t memorizing slides. When shown completely novel images they’d never encountered—different tissue samples, different magnifications, different degrees of image compression, images with and without color—they generalized successfully. They had learned to detect the visual features that distinguish malignant from benign tissue, not to associate specific images with specific rewards. A bird that had never attended medical school, that has no concept of cells or cancer or pathology, was reading histological slides with the diagnostic accuracy of a specialist who trained for a decade.
What they could do and what they couldn’t
The pigeons’ performance wasn’t uniform across all tasks, and the boundaries of their ability tell you as much as their successes.
Histopathology—digitized microscope slides of breast tissue biopsies—was where they excelled. They learned fast, generalized to novel images, and handled variations in magnification (4x, 10x, 20x) and image quality. Wasserman, who had studied pigeon cognition for over 40 years, said they learned to discriminate benign from malignant tissue as fast as pigeons in any other visual discrimination study his lab had ever conducted. The task wasn’t easy for humans—inexperienced human observers require considerable training to reach mastery on the same slides—but the pigeons picked it up in days.
Mammographic microcalcifications—the tiny calcium deposits that, in certain configurations, indicate breast cancer—were a second success. These appear as patterned white specks against a complex background on mammograms, and the researchers hypothesized that detecting small bright targets in visual clutter is precisely the kind of task pigeons evolved to perform. Finding seeds in grass, finding microcalcifications on a mammogram—structurally, the visual problem is similar. The pigeons could detect microcalcifications on novel mammograms they hadn’t seen during training.
Mammographic masses—the suspicious tissue densities that can signal cancer but lack the discrete visual signature of microcalcifications—were where the pigeons hit their ceiling. Human radiologists achieve about 80 percent accuracy on these images, which are genuinely difficult even for trained professionals. The pigeons took weeks instead of days to learn the training set, and when shown novel images, they performed at chance. They had memorized the specific masses in the training images without extracting the generalizable features—the stellate margins, the irregular borders, the density patterns—that correlate with malignancy. They could learn the specific. They couldn’t learn the abstract.
This boundary matters because it reveals the architecture of what the pigeons are doing. They’re not reading X-rays the way a radiologist reads them—constructing a clinical interpretation from visual features informed by anatomical knowledge and diagnostic frameworks. They’re performing pattern recognition at a level that is, for certain categories of visual stimuli, extraordinarily sophisticated, and for other categories, completely absent. The pigeon has no concept of cancer. It has a visual system that, after millions of years of evolutionary optimization for detecting meaningful patterns in complex environments, can be trained to recognize the visual signatures of pathology on a slide faster than a medical student can.
Why pigeons see what they see
Pigeons have tetrachromatic vision—four types of color receptors compared to humans’ three—and their visual acuity, while not as fine-grained as humans’ for detail at a distance, is optimized for detecting patterns, textures, and small differences across complex visual fields. They can discriminate individual human faces, distinguish paintings by Monet from paintings by Picasso, and categorize photographs of objects they’ve never seen into previously learned categories. Their visual cognition is not simple stimulus-response association. It involves genuine perceptual categorization—the extraction of abstract features that define a class and the application of those features to novel instances.
The pigeon brain processes visual information through a pathway called the tectofugal system, which is analogous but not homologous to the mammalian cortical visual pathway. The computational result is similar—pattern extraction, categorization, generalization—but achieved through different neural architecture. This is convergent evolution at the cognitive level: two lineages separated by over 300 million years of evolution arriving at functionally equivalent solutions to the same problem, which is making sense of a visually complicated world.
The cancer detection experiment wasn’t really about cancer. It was about visual cognition. Levenson, Wasserman, and their colleagues were using medical imaging as a standardized, well-characterized visual discrimination task to probe the capabilities and limits of pigeon perception. The fact that the visual stimuli happened to be diagnostically important—that the patterns the pigeons were detecting are the same patterns that determine whether a patient gets a biopsy or goes home—is what made the study irresistible to the public. But the scientific contribution was the demonstration that pigeon visual cognition can be meaningfully compared to human expert performance on the same images, using the same accuracy metrics.
The practical question nobody expected
Levenson was clear that pigeons are not going to replace radiologists. The regulatory implications alone—”What would the FDA think about pigeons?” he said, “I shudder to think”—make clinical deployment a nonstarter. And for the visual tasks where human expertise is most critical—the ambiguous masses, the complex densities, the cases where clinical context determines interpretation—the pigeons failed.
But the practical application isn’t diagnosis. It’s quality assurance. Medical imaging technology is constantly evolving—new display technologies, new compression algorithms, new processing pipelines, new acquisition hardware—and every innovation needs to be validated by trained observers who evaluate whether the new system makes diagnostically important features easier or harder to see. That validation currently requires recruiting clinicians to spend hours or days doing tedious comparisons of image sets, a process that is expensive, slow, and dependent on the availability of people who have better things to do with their medical training.
Pigeons don’t get bored. They don’t get fatigued. They don’t have clinic schedules or grant deadlines. They can evaluate thousands of images without the performance degradation that affects human observers after prolonged sessions. For the subset of visual tasks where pigeon accuracy matches or approaches human accuracy—histopathology slides, microcalcification detection—pigeons could serve as a rapid, cheap, reliable feedback system for the engineers building better imaging tools. Levenson suspects computers will get there first, and given the trajectory of AI-based image analysis since 2015, he’s probably right. But for a decade, the pigeons were competitive.
What it actually tells us
The deeper lesson of the pigeon cancer experiment isn’t about medicine or about pigeons. It’s about what vision is. A pigeon with a brain the size of a walnut, a lifespan during which it will never encounter a microscope or learn what a cell is, can be trained to perform a visual discrimination task that humans require years of specialized education to master. This means the visual features that distinguish malignant from benign tissue are not visible only to minds that understand cancer. They’re visible to any sufficiently powerful pattern recognition system—biological or computational—that can be calibrated against enough examples.
The pigeon doesn’t know what it’s looking at. It doesn’t need to. The visual signal is in the image. The pigeon’s 300-million-year-old visual system just happens to be good enough to find it.
We cover pigeon visual cognition alongside baboon politics, cuttlefish camouflage, and the full landscape of animal intelligence across our Animal Culture & Knowledge course—including why a bird that can’t tell you what cancer is might still be better at spotting it than a first-year medical resident.
-
Ant Colonies as Superorganisms: How Millions of Tiny Brains Make One Giant Decision
In 2022, Daniel Kronauer and Asaf Gal at Rockefeller University built a system to watch an ant colony make a decision. They placed colonies on a heated platform and slowly raised the temperature. Individual ants felt the heat under their feet but carried on as usual—foraging, tending larvae, wandering with the vaguely purposeless energy of someone who forgot why they walked into a room. Then, at a specific temperature, the entire colony reversed course simultaneously. Every ant evacuated. The decision wasn’t made by any individual ant. It was made by the colony.
The expected finding: a colony of 36 workers evacuated reliably at about 34 degrees Celsius. The surprising finding: when Kronauer and Gal increased the colony from 10 to 200 individuals, the temperature required to trigger evacuation went up. Colonies of 200 held out past 36 degrees. No individual ant knows how many ants are in its colony. No ant has a thermometer or a headcount. And yet the group’s decision threshold shifted based on group size—a variable that no single member of the group can perceive.
The colony was behaving like a neural network. Not metaphorically. Structurally.
The superorganism concept
An ant colony is not a collection of individuals cooperating. It’s a single organism made of many bodies. The queen is the reproductive system. The workers are the immune system, the digestive system, the musculoskeletal system. The pheromone trails are the nervous system. No individual ant contains the information necessary to run the colony, just as no individual neuron contains the information necessary to produce a thought. The intelligence—such as it is—exists only at the level of the system.
This isn’t a cute analogy. Researchers at Arizona State University and other institutions have spent decades studying ant colonies using the same experimental methods psychologists use on individual animals—psychophysics, perceptual discrimination tasks, speed-accuracy tradeoff tests, rationality assessments—and finding that colonies exhibit cognitive properties that individual ants don’t possess. Takao Sasaki and Stephen Pratt published a comprehensive review in the Annual Review of Entomology in 2018 documenting the parallels: colonies balance speed against accuracy in decision-making using the same mathematical relationships that govern neural computation in brains. Colonies make better choices than individuals when the discrimination task is hard—a PNAS study demonstrated that colony-level decisions outperformed individual ant decisions on difficult sensory discrimination tasks but not on easy ones, the exact pattern you’d predict if the colony functions as a signal-averaging system that reduces noise through redundancy.
The superorganism concept, as Sasaki and Pratt frame it, isn’t an illustrative metaphor. It’s a research program. If a colony is functionally equivalent to an organism, then the tools developed for studying organisms should work on colonies. They do.
How decisions actually happen
Deborah Gordon, a biologist at Stanford who has studied ant behavior for over 30 years, describes the central puzzle: individual ants are, to put it charitably, not impressive. Watch a single ant trying to find food, and you’ll see an organism that frequently loses the trail, forgets which direction it was heading, and gets confused by a leaf. Gordon says she probably wouldn’t hire one. But thousands of these bumbling individuals collectively locate food sources, mobilize foraging parties, switch flexibly between tasks, defend the nest, and manage waste disposal—all without any centralized control, any chain of command, any ant that knows the plan.
The mechanism is local interaction. An ant doesn’t know what the colony needs. It knows what’s happening in its immediate vicinity—which other ants it’s bumped into recently, what pheromone concentrations it’s detecting, whether the ant it just touched with its antennae was carrying food or returning empty. From these local cues, each ant follows simple behavioral rules. The sophistication emerges from the interaction patterns, not from the individual agents.
Pheromone trails are the most studied example. When a forager finds food, it lays a chemical trail on the way back to the nest. Other ants that encounter the trail follow it to the food source and lay their own pheromone on the return trip. The trail gets stronger. More ants follow it. The trail gets stronger still. This is positive feedback—the same amplification mechanism that drives neural decision-making in brains. When two food sources exist, the colony will usually converge on one, not split evenly between both, because random early variation in ant traffic gets amplified by the feedback loop until one trail dominates. The colony has “decided” which food source to exploit, and no individual ant made that decision.
Nest site selection in Temnothorax ants is the most precisely documented example of collective decision-making. When a colony needs to relocate, scout ants explore candidate sites and assess them individually—cavity size, darkness, entrance width, structural integrity. A scout that finds a promising site recruits other scouts through tandem running, leading them to the site one by one. Once enough scouts accumulate at a site—a quorum threshold—the ants switch from slow tandem recruitment to rapid carrying, physically transporting the rest of the colony to the new home. The quorum threshold is the decision mechanism: it ensures that the colony doesn’t commit to a site until enough independent assessors have confirmed its quality. It’s a voting system that doesn’t require any ant to count votes.
Nigel Franks at the University of Bristol documented the speed-accuracy tradeoff in this system: colonies that use a lower quorum threshold decide faster but make worse choices. Colonies that use a higher threshold are slower but more accurate. The tradeoff is governed by the same mathematical relationships that describe speed-accuracy tradeoffs in primate neural decision-making. The ant colony and the primate brain are implementing the same algorithm using completely different hardware.
Where the analogy breaks
The superorganism framework is powerful but not unlimited. Colonies also encounter performance costs that individual organisms don’t. The same positive feedback that generates consensus can amplify errors—if early scouts happen to find a mediocre nest site first, the pheromone feedback can lock the colony into a suboptimal choice before better alternatives are discovered. Individual organisms can change their minds; colonies, once committed by positive feedback, have a harder time reversing course.
Gordon’s work emphasizes that the ant-colony-as-brain analogy, while productive, can overstate the degree of centralized computation involved. Ant colonies don’t have a dedicated processing center equivalent to a cortex. They operate through what Gordon calls “the ecology of collective behavior”—the interaction between the colony’s behavioral rules and the specific environmental context in which those rules play out. The same colony, using the same rules, produces different behaviors in different environments, just as the same neural architecture produces different outputs depending on sensory input. The intelligence isn’t in the rules. It’s in the fit between the rules and the world.
There are also roughly 14,000 species of ants, and they don’t all work the same way. Army ants conduct nomadic raids without stable nest sites. Leafcutter ants farm fungus in underground gardens. Harvester ants in the American Southwest manage foraging rates using interaction frequencies that Gordon has compared to TCP/IP internet protocols—the rate at which returning foragers contact outgoing foragers determines whether more foragers are sent out, the same feedback mechanism that regulates data transmission rates in computer networks. The superorganism concept applies broadly, but the specific implementations are as varied as the ecosystems ants occupy.
Why neuroscientists care about ants
The deep reason to study ant colonies isn’t entomological. It’s computational. The question that Kronauer’s evacuation experiment, Sasaki and Pratt’s psychophysics research, and Gordon’s decades of fieldwork all converge on is the same question that drives computational neuroscience: how does a system composed of simple, unreliable components produce complex, reliable behavior?
A neuron, like an ant, is not smart. It fires or it doesn’t. It has no concept of the thought it’s participating in. The intelligence of a brain, like the intelligence of an ant colony, is an emergent property of interaction patterns among components that individually can’t do much. The mathematical models that describe how ant colonies reach consensus—positive feedback, quorum thresholds, speed-accuracy tradeoffs, noise reduction through redundancy—are the same models that describe how populations of neurons reach decisions. The hardware is different. The computation is the same.
Kronauer’s evacuating ants couldn’t know how many of them there were, and yet their collective behavior changed as a function of colony size. The mechanism, he suspects, involves pheromone concentration: more ants produce more “stay” pheromone, which raises the temperature threshold for the “leave” signal to override the “stay” signal. It’s a chemical implementation of the same inhibition-excitation balance that governs decision thresholds in neural circuits. The colony isn’t thinking about whether to leave. It’s computing whether to leave, using the bodies and chemical secretions of its members as the processing substrate.
The ant that forgot why it walked into a room isn’t broken. It’s a single neuron in a brain that works just fine.
We cover ant superorganism intelligence alongside baboon politics, cuttlefish camouflage, and the full landscape of animal cognition across our Animal Culture & Knowledge course—including why the best model for understanding your brain might be 200 confused ants on a hot plate.
-
Spinal Cord Stimulation in 2026: How Electrical Implants Are Helping Paralyzed Patients Walk Again
In 2022, a team led by Grégoire Courtine at the Swiss Federal Institute of Technology published a paper in Nature that identified the specific neurons responsible for restoring walking after paralysis. They called them SCVsx2::Hoxa10 neurons—a population of cells in the lumbar spinal cord that, when activated by epidural electrical stimulation combined with rehabilitation, orchestrated the recovery of walking in nine individuals with chronic spinal cord injury. When the researchers optogenetically silenced these neurons in mice, walking stopped instantly. When they reactivated them, walking resumed immediately. The paper didn’t just show that spinal cord stimulation works. It showed which neurons it works through—a mechanistic explanation for a result that had previously looked like something between a miracle and an engineering trick.
That result sits at the center of a field that has moved, in under a decade, from “we got a paralyzed person to twitch their leg” to “we got a completely paralyzed person to walk, cycle, swim, and climb stairs within 24 hours of turning on the stimulator.” The technology isn’t a cure. It doesn’t repair the severed spinal cord. But it’s the closest thing to functional restoration that exists for the roughly 250,000 to 500,000 people worldwide who sustain spinal cord injuries each year.
How the spinal cord works (and what happens when it breaks)
Your brain doesn’t directly control your legs. It sends high-level movement commands—”walk,” “stand,” “step over that obstacle”—down the spinal cord, where local neural circuits called central pattern generators translate those commands into the precise, rhythmic sequences of muscle activation that produce coordinated movement. The spinal cord below the injury site isn’t dead tissue. The pattern generators, the motor neurons, the sensory circuits—they’re all intact. They’re just disconnected from the brain. A spinal cord injury is less like cutting a power cable and more like cutting the communication line between a command center and a factory that’s still fully staffed and operational.
This is why epidural electrical stimulation works. A thin array of electrodes, implanted on the surface of the spinal cord below the injury site, delivers patterned electrical pulses that mimic the signals the brain can no longer send. The electrodes activate the dorsal roots—the sensory nerve fibers that enter the spinal cord from the body—which in turn excite the spinal circuits that coordinate movement. The stimulator doesn’t replace the brain. It substitutes for the broken communication link, providing enough activation to the intact spinal circuitry below the injury to enable the pattern generators to do what they were designed to do.
What patients can actually do
The results have escalated quickly. In 2018, patients with incomplete spinal cord injuries—some residual sensation or movement below the injury—were able to walk and cycle with epidural stimulation. Courtine’s 2022 Nature Medicine paper pushed further: three patients with complete paralysis—no voluntary movement, no sensation below the injury—could take steps on a treadmill within the first day of turning on the stimulator. One climbed stairs. All could swim, cycle, and perform leg presses. A biomedical engineer at the University of Alberta who reviewed the results called it “a big deal.”
A 2025 case study from Italian researchers at the MINE Lab and EPFL documented the first successful application of epidural stimulation in a patient with a lower thoracic spinal cord injury involving the conus medullaris—the tapered lower end of the spinal cord. Previous trials had excluded these patients, who represent over half of thoracic spinal cord injuries, because of concerns about damaging the nerve roots that control the legs. The patient progressed from being unable to walk to covering one kilometer independently with a walker within six months of the implant. A 2025 paper in Science Translational Medicine showed that high-frequency epidural stimulation reduced spasticity—the involuntary muscle contractions that plague roughly 70 percent of spinal cord injury patients—enabling rehabilitation protocols that further improved recovery.
A systematic review published in 2024 analyzed 64 studies encompassing 306 patients and found improvements in motor function, cardiovascular regulation, pulmonary function, bladder and bowel control, and genitourinary function. The effects extend well beyond walking. Spinal cord stimulation is restoring autonomic functions—blood pressure regulation, temperature control, sexual function—that most people don’t associate with paralysis but that profoundly affect quality of life.
The gap between demonstration and daily life
The results are genuine. The caveats are significant.
First, the stimulator must be on for the abilities to exist. Turn it off, and the paralysis returns. Some patients have shown neural adaptation over months or years of stimulation combined with rehabilitation—one participant in a 2017 study progressively recovered voluntary leg movement and standing without stimulation over 3.7 years of training—but this is the exception. For most patients, the device is a permanent prosthetic, not a temporary bridge to recovery.
Second, the current interface is cumbersome. Users select their desired movement on a tablet, which sends Bluetooth commands to a transmitter worn around the waist. The transmitter must be positioned next to a pulse generator implanted in the abdomen, which activates the electrode array on the spine. Courtine’s team is developing a next-generation system, but the current version requires conscious selection of each movement type before execution. You don’t just decide to walk. You tell the tablet you want to walk, the tablet tells the transmitter, the transmitter tells the implant, and the implant activates the pattern that produces walking. The latency and cognitive overhead are nontrivial.
Third, weight support matters. The patients who stepped within 24 hours did so in harnesses that supported more than half their body weight. The one-kilometer walk with a walker at six months is a more representative picture of functional use than the headline-grabbing first-day stepping videos. And patients with paralysis are at elevated risk for osteoporosis and fragility fractures—bones that haven’t borne weight for years break under loads that healthy bones absorb without issue.
Fourth, the technology doesn’t work equally well for everyone. Severity of the original injury, time elapsed since injury, location of the damage, and the individual patient’s residual neural architecture all affect outcomes. A 2025 study of 11 patients with incomplete injuries found that spasticity improved in all 11, sensation recovered in 10, but only 4 of 11 showed improved lower limb strength. The field is still identifying which patients benefit most and why.
Where it’s heading
The next frontier is closed-loop stimulation—systems that read neural signals from the spinal cord or brain in real time and adjust stimulation parameters automatically, rather than relying on preset patterns selected through a tablet. A 2026 case report from Zhejiang University described a closed-loop spinal neural interface combined with rehabilitation for incomplete spinal cord injury, representing early steps toward systems that adapt to the patient’s intent rather than executing fixed programs.
Intraspinal microstimulation—tiny electrodes inserted directly into the spinal cord rather than placed on its surface—offers potentially more precise and selective muscle activation. A 2026 study in Scientific Reports described the first fully implantable intraspinal microstimulation device tested in a large animal model with spine dimensions similar to humans, moving toward clinical translation.
The broader trajectory: epidural stimulation combined with rehabilitation is progressing from proof-of-concept in research labs to clinical feasibility studies to, eventually, regulated medical devices available to the hundreds of thousands of people living with spinal cord injuries worldwide. The International Neuromodulation Society has published consensus guidance on the safety of epidural stimulation, noting it is “a generally safe, minimally invasive procedure.” The infrastructure for clinical translation is being built.
Courtine’s lab identified the neurons. The Italian team proved it works in injuries previously considered untreatable. The spasticity reduction data from Science Translational Medicine demonstrated benefits beyond motor recovery. The systematic reviews have established the evidence base. What remains is the engineering work of making the system smaller, smarter, more responsive, and cheap enough to deploy at the scale the patient population requires—the same progression from laboratory demonstration to clinical reality that every neuroprosthetic technology follows, measured in years, not months.
We cover spinal cord stimulation alongside brain-computer interfaces, retinal implants, and the full landscape of neuroprosthetic technology across our Neuroprosthetics course—including why the spinal cord below an injury isn’t a broken machine waiting for repair. It’s a working machine waiting for a signal.
-
Cuttlefish Camouflage: How a Colorblind Animal Produces the Most Sophisticated Disguise on Earth
A cuttlefish has up to millions of chromatophores in its skin—pigment-filled elastic sacs, each attached to a ring of tiny radial muscles, each muscle controlled by a small number of motor neurons that extend directly from the brain. When those neurons fire, the muscles contract, stretching the chromatophore from an invisible speck roughly a tenth of a millimeter across to a visible disc up to 1.5 millimeters in diameter, displaying the pigment inside. When the neurons stop firing, the elastic sac snaps back to its resting size. The whole process takes less than a second. There are three color classes of chromatophores—red, yellow/orange, and brown/black, depending on the species—arranged in layers across the skin. Beneath them sit iridophores, cells packed with reflective protein platelets that produce metallic blues, greens, and iridescent effects through thin-film interference. Beneath those sit leucophores, white reflecting cells that scatter all incoming wavelengths equally, providing a neutral base canvas.
Three layers. Millions of individually addressable cells. Direct neural control from the brain. The result is the most sophisticated dynamic camouflage system in the animal kingdom—an animal that can transform its color, pattern, and three-dimensional skin texture in a fraction of a second to match virtually any natural substrate it encounters.
The cuttlefish is colorblind. It has a single type of photoreceptor in its eye. It cannot distinguish colors. And yet it produces color matches that fool the color vision of its predators—di- and trichromatic fish that can see wavelengths the cuttlefish itself cannot perceive. This is the central paradox of cuttlefish camouflage, and after decades of research, science has gotten closer to understanding how it works without fully resolving the contradiction.
The hardware
The chromatophore system is, functionally, a neural display. Each chromatophore is a pixel. The brain is the graphics processor. The motor neurons are the data bus. Gilles Laurent at the Max Planck Institute for Brain Research described the research approach as “measuring the output of the brain simply and indirectly by imaging the pixels on the animal’s skin”—because each chromatophore’s expansion state reflects the firing rate of its controlling motor neurons, tracking chromatophores at high resolution is equivalent to tracking neural activity across tens of thousands of neurons simultaneously in a freely behaving animal.
Laurent’s lab developed methods to track individual chromatophores at 60 frames per second, at single-cell resolution, over weeks of continuous observation as the animal breathed, moved, changed appearance, and grew. They could identify each chromatophore like a fingerprint—every animal’s arrangement is unique—and follow it even as new chromatophores appeared daily during development. By analyzing how chromatophores co-fluctuated—which ones expanded together, which ones were independent—they could infer the structure of the motor neuron populations controlling them, and from there, predict the organization of higher-level control circuits deeper in the brain. Reading the skin to reverse-engineer the brain.
What they found overturned the assumption that cuttlefish camouflage patterns were simple. Traditional taxonomy divided cuttlefish patterns into three categories—uniform, mottled, and disruptive—with roughly 30 subcategories. The high-resolution tracking data revealed something far more complex: skin patterns are high-dimensional and dynamic, with the animal meandering through pattern space, accelerating and decelerating, sometimes producing nearly identical overall patterns using entirely different combinations of individual chromatophores. The skin display isn’t selecting from a fixed menu of preset patterns. It’s navigating a continuous space of possible configurations, course-correcting as it goes.
A breakthrough finding reported in 2023 showed that cuttlefish undergo multiple color changes before settling on a camouflage pattern that matches their surroundings—a trial-and-error approach rather than the instantaneous, pre-programmed response the speed of the transformation seems to imply. The camouflage looks instant to the observer because the iterations happen within seconds. But the animal isn’t computing a perfect match and executing it. It’s generating candidates, evaluating them against what it sees, and converging on a solution. The distinction matters: it’s the difference between a lookup table and a search algorithm.
The texture dimension
Color and pattern alone don’t make a convincing disguise. A smooth-skinned animal on a bumpy coral surface still looks wrong, regardless of how well the colors match. Cuttlefish solved this by evolving papillae—muscular hydrostats in the skin that, when activated, produce three-dimensional bumps ranging from subtle texture changes to dramatic protrusions that mimic algae, coral, or rock surfaces. The papillae are controlled by a neural circuit separate from the chromatophore circuit—the two systems can be activated independently—but coordinated through shared brain regions so that color, pattern, and texture match simultaneously.
A cuttlefish resting on a rocky substrate doesn’t just turn the right shade of brown. Its skin erupts into bumps that mimic the surface geometry of the rock. Move it to smooth sand and the papillae flatten, the chromatophores shift to a uniform sandy tone, and the animal becomes a patch of seabed. The transformation—color, pattern, luminance, texture—happens in less than a second.
The colorblind problem
Cuttlefish have a single visual pigment with peak sensitivity around 492 nanometers. One photoreceptor type means no color opponency—the neural comparison between different wavelength channels that enables color perception in animals with two or more photoreceptor types. By every definition used in visual neuroscience, the cuttlefish is monochromatic. It sees the world in shades of a single dimension.
And yet: hyperspectral imaging studies—using cameras that record full-spectrum light data across every wavelength—have shown that cuttlefish camouflage provides high-fidelity color matches to natural substrates when evaluated through the visual systems of their fish predators. The spectral properties of cuttlefish skin and the substrates they match are often similar enough to fool trichromatic vision. The animal can’t see the colors it’s producing, and the colors it produces are right.
How? The honest answer is that the mechanism isn’t fully understood, but several partial explanations have converged. First, the three chromatophore pigment classes and the underlying structural reflectors can, in combination, produce most colors found in marine environments through subtractive and additive mixing, without the animal needing to know what specific color it’s producing. Second, cuttlefish may be matching luminance—brightness—rather than hue, and getting the color right as a byproduct of getting the brightness pattern right. A 2024 study on octopus camouflage (a closely related cephalopod with the same single-photoreceptor constraint) found that they excel at matching background lightness but often miss color saturation, suggesting brightness matching is the primary computation and color match is a statistical bonus.
Third—and this is where it gets genuinely strange—cuttlefish skin contains opsin proteins, the same light-sensitive molecules found in the retina. Researchers discovered opsin transcripts in the fin and ventral skin of the common cuttlefish. The skin opsins are identical to the retinal opsin, which means they can’t provide color discrimination (same single-pigment limitation), but they could provide local light-level sensing that allows the skin itself to contribute to the camouflage computation without routing all information through the eyes and brain. The skin might be sensing its own output and adjusting locally.
In 2025, researchers at Scripps Institution of Oceanography genetically engineered soil bacteria to produce xanthommatin—the primary chromatophore pigment—at industrial scale, a thousandfold improvement over extraction from actual cephalopods. In 2024, scientists developed CHROMAS, a machine learning pipeline that tracks individual chromatophores frame by frame and quantifies how patterns emerge. The tools to finally crack the colorblind camouflage problem are arriving faster than at any point in the field’s history.
Why it matters beyond marine biology
The cuttlefish skin is, from an engineering perspective, a flexible, high-resolution, real-time display that changes color, pattern, and three-dimensional surface texture under direct neural control, powered by biological materials, operating at millisecond timescales, and doing all of this without the organism understanding color theory. Military researchers, materials scientists, and roboticists have been studying cephalopod camouflage for decades as a blueprint for adaptive materials—fabrics that change color, surfaces that alter texture, coatings that respond to their environment.
But the deeper significance is neuroscientific. The cuttlefish skin is a window into the brain. Because each chromatophore is controlled by identified neurons, the skin pattern is a real-time, high-dimensional neural readout of the animal’s perceptual state. When a cuttlefish camouflages, its skin is displaying what its brain thinks the world looks like—a projection of its visual perception onto its own body surface. No other animal provides this kind of direct, externally visible readout of neural computation at the scale of tens of thousands of neurons simultaneously. The cuttlefish isn’t just hiding. It’s showing you what it sees.
We cover cuttlefish camouflage alongside octopus distributed cognition, mirror neurons, and the full landscape of comparative neuroscience across our Neurozoology course—including why the most important display technology in neuroscience isn’t a screen in a lab. It’s the skin of a colorblind animal that paints what its brain perceives.
-
Baboon Politics: Social Hierarchies, Alliances, and Machiavellian Intelligence in Primates
A baboon can do something that most humans find cognitively demanding and many find socially impossible: induce a more powerful individual to attack a third party on its behalf, without the powerful individual realizing it’s being used as a weapon. The maneuver is called a “protected threat.” The baboon appeases the dominant member of its group, positions itself to make a subordinate appear threatening, and maneuvers to prevent the target from doing the same thing in reverse. It’s social tool use—using another organism as an instrument to achieve a goal—and baboons master it at puberty. Chimpanzees, by comparison, don’t learn to use a stone to crack nuts until adulthood. Primates appear to manipulate social objects with more sophistication and at earlier developmental stages than physical tools, which raises an uncomfortable question about what primate brains actually evolved to do.
The answer, according to a hypothesis that has shaped comparative cognition for nearly four decades, is politics.
The Machiavellian intelligence hypothesis
In the 1960s, lemur researcher Alison Jolly noticed something counterintuitive. Lemurs were terrible at manipulating objects—far worse than monkeys at the mechanical problem-solving tasks that laboratories used to measure intelligence. But their social skills were just as sophisticated as monkeys’. Jolly proposed reversing the common assumption: instead of social complexity being a product of intelligence, intelligence might be a product of social complexity. The technical challenges of foraging—finding food, processing it, remembering where it grows—might matter less than the social challenges of living in permanent groups with dozens of individuals who are simultaneously your allies, rivals, mates, competitors, and kin.
Psychologist Nicholas Humphrey extended this in 1976. He’d watched captive monkeys handle laboratory puzzles with impressive skill, but he couldn’t find anything comparably challenging in their natural foraging environment. The hardest problem these animals faced, he argued, wasn’t physical. It was social—navigating a group where every interaction involved weighing cooperation against competition, tracking who owes what to whom, remembering past conflicts and predicting future alliances, and doing all of this with individuals who are simultaneously doing the same calculations about you.
Frans de Waal’s 1982 book Chimpanzee Politics documented the social maneuvering of chimpanzees in terms that read like a dispatch from the Florentine court—coalition formation, strategic alliance shifts, betrayals, reconciliations, and the systematic deployment of social favors as a form of political currency. Andrew Whiten and Richard Byrne formalized the concept in 1988 as the Machiavellian intelligence hypothesis: the pressure to outmaneuver other members of your social group is a primary driver of the evolution of primate intelligence. The brain got bigger not because the environment got harder but because the social group got more complicated.
Robin Dunbar demonstrated a correlation between primate group size and neocortex size—the most recently evolved part of the brain, and the part that expanded most dramatically in the primate lineage compared to other mammals. Larger groups require tracking more relationships, remembering more histories, predicting more behaviors. The cognitive load scales with the number of social connections, not with the complexity of the physical environment. Primates have brains roughly twice as large as expected for mammals of equivalent body size, and the Machiavellian intelligence hypothesis argues that social computation—not tool use, not foraging, not predator avoidance—is the primary reason.
What baboons actually do
Baboon troops are not democracies. They’re hierarchies maintained through a combination of aggression, alliance formation, grooming, and the careful management of social relationships that function as a currency more stable than any physical resource. Male baboons compete for rank through direct confrontation, but rank alone doesn’t determine reproductive success. Males who form alliances—particularly with unrelated males—can collectively outcompete higher-ranking individuals. The alpha male is not always the most reproductively successful male. The most politically connected male sometimes is.
Female baboons form their own hierarchies, typically more stable than male hierarchies and based heavily on kinship. A female’s rank often follows her mother’s, creating lineages of dominant and subordinate families that persist across generations. High-ranking females get better access to food and water, experience lower stress hormone levels, and have offspring with higher survival rates. The fitness consequences of social rank are measurable, heritable, and real.
Grooming is the central social technology. Baboons groom each other for hours daily, and the distribution of grooming is not random. It correlates with alliance patterns, kinship, and—critically—with what the grooming partner can offer in the immediate social marketplace. Research on wild chacma baboons found that female coalitions were not long-term strategic alliances built through reciprocal grooming over months. They were opportunistic, short-term transactions where both parties benefited immediately. Baboons don’t trade favors across time the way the Machiavellian framework originally suggested. They trade in real time, in a social marketplace where the value of a grooming partner fluctuates based on current social conditions.
This finding—published by Silk, Cheney, Seyfarth, and others—complicated the original hypothesis significantly. The Machiavellian framework emphasized long-term strategic planning, deception, and reciprocal exchange. The field data suggested something more like a spot market: baboons assessing the current value of social partners and adjusting their behavior accordingly, not executing multi-step schemes that require remembering who did what three weeks ago.
Tactical deception
Byrne and Whiten documented tactical deception in baboons—behaviors designed to create false impressions in the minds of other individuals. A subordinate baboon feeding on a preferred food item while a dominant individual approaches will sometimes casually move away from the food and adopt a relaxed posture, as if it had finished eating or hadn’t been eating at all. Once the dominant passes, the subordinate returns to the food. The behavior requires, at minimum, an understanding that the dominant’s behavior is influenced by what it believes about the subordinate’s behavior—a rudimentary form of the social cognition that in humans we’d call theory of mind.
Mountain gorillas suppress their copulation vocalizations during secretive matings with subordinate males, conducted out of sight of the dominant silverback. Both the female and the junior male remain silent—a coordinated deception that requires both parties to understand that the dominant male’s response depends on what he perceives. When these matings are discovered, the dominant male invariably attacks the female, adding a punitive dimension to the social calculation: the cost of being caught is asymmetric, falling more heavily on the female, which means the decision to mate secretly involves weighing the reproductive benefit against a gendered risk of punishment.
Dario Maestripieri at the University of Chicago, studying rhesus macaques, found that these monkeys share with humans “strong tendencies for nepotism and political maneuvering.” His conclusion: “Our Machiavellian intelligence is not something we can be proud of, but it may be the secret of our success.” The cognitive machinery that enables a baboon to manipulate a dominant individual into attacking a rival may be the same machinery that, scaled up and elaborated over millions of years, enables a human to navigate corporate politics, negotiate a trade deal, or run for office.
What the critics found
The Machiavellian intelligence hypothesis has generated productive pushback. Barrett and Henzi, studying baboons and other primates in the field, argued that the hypothesis overemphasizes exploitation and deception at the expense of tolerance, coordination, and cooperation. Primate social life, they contended, is not primarily a chess game of strategic manipulation. It’s “an intricate tapestry of competition and cooperation, of aggression and reconciliation, of nonaggressive social alternatives, and of behaviors and relationships that cannot be easily categorized into simple opposites.”
The orangutan problem is frequently cited: orangutans are largely solitary but outperform the highly social baboon on cognitive tests. If social complexity drives intelligence, the most social species should be the smartest. They’re often not. The relationship between sociality and cognition is real but messier than the original hypothesis suggested—group size correlates with neocortex size across the primate order, but individual species frequently violate the pattern.
The current consensus treats the Machiavellian intelligence hypothesis as an important partial explanation rather than a complete theory. Social complexity is a major driver of primate brain evolution, but it’s not the only driver, and the specific form that social cognition takes—long-term strategic planning versus real-time marketplace trading, deceptive manipulation versus cooperative coordination—varies between species in ways the original framework didn’t predict.
Why it matters beyond primatology
The baboon troop is a small-scale version of the problem every human organization faces: how do you maintain a stable group when every member has individual interests that partially conflict with the group’s interests? The baboon’s solution set—hierarchy, coalition, grooming, deception, reconciliation, punishment, nepotism—is recognizable to anyone who has spent time in a corporate office, a political party, or a homeowners association. The specifics differ. The architecture doesn’t.
The deeper implication is about what brains are for. If the Machiavellian intelligence hypothesis is even partially correct, the enormous human neocortex didn’t evolve primarily to solve physics problems or build tools or develop language. It evolved to navigate other humans—to predict what they’ll do, influence what they think, form alliances that advance your interests, and detect when someone is doing the same to you. The math, the engineering, the art, the philosophy—all of it may be a secondary application of cognitive hardware that was built, under evolutionary pressure, for politics.
We cover baboon social intelligence alongside chimpanzee tool traditions, dolphin communication, and the full landscape of animal cognition across our Animal Culture & Knowledge course—including why the most revealing thing about human intelligence might be how much of it we share with a monkey that learned to weaponize its friends.
