You see something move on the floor. Long, slithery, self-propelled. Before any thought, your heart hammers, your skin prickles, you take a step backward. The whole reaction took about 200 milliseconds. You did not decide to be afraid.
That cascade is an evolved psychological mechanism doing exactly what it was built to do. It noticed a narrow, specific kind of input — slithery elongated motion. It told you which problem you were facing — danger of poisonous bite. It transformed that input through a fast decision rule into output you didn't author — physiology shifts, attention narrows, you step back. The whole machine was waiting in you, fully assembled, before you ever saw a real snake. It was assembled by your ancestors, generation after generation, every time slithery elongated motion correlated with bites that killed people who didn't pull back.
David Buss describes a mechanism like that as a set of processes inside an organism with six specific properties.1 The mechanism's design features fit the adaptive problem the way a key fits a lock — coordinated, not generic. Failure to mesh with the problem meant failure to pass through the selective sieve of evolution.1 That key/lock metaphor matters. It is the difference between an evolutionary psychology that takes the human mind seriously as a piece of designed equipment, and one that treats the mind as a generic processor that culture writes on.
Six things had to be true at once before selection kept this kind of program around in your head, and you can read them off what just happened to you when you saw the slither.
First, you have the mechanism in the form you do because it solved a specific problem of survival or reproduction over and over in the past.1 People whose ancestors didn't have it, or whose version of it was less well-tuned, are not your ancestors. The form fits the problem the way a key fits a lock.
Second, the mechanism takes in a narrow slice of information.2 Your eye is sensitive to a thin band of the electromagnetic spectrum, with cones tuned for specific colors and edges and motion. The snake-fear program is similar — it triggers on slithery, self-propelled, elongated motion, not on everything that moves. It does not get distracted by leaves blowing in the wind or by branches falling.
Third, the input tells you which problem you're facing, and tells you below the threshold of conscious awareness almost every time. You don't smell pizza and think "food selection problem detected." You salivate.3
Fourth, the input runs through decision rules — if/then procedures — to produce output. The simple version: if you see a slithery thing, then step back. The complicated version handles size, speed, distance from you, whether you have a stick, whether you have shoes on, whether the rival shouting at you is bigger than you or smaller. The rules are conditional, and they specify multiple possible responses.4
Fifth, the output comes out as one of three things, often several at once. Physiology shifts — heart, skin, attention. Information passes to other psychological mechanisms — the slither-sighting feeds the freeze/flight/fight calculator, the smell of pizza feeds the food-selection mechanism. And manifest behavior happens — your foot moves backward, your hand reaches for the food.5
Sixth, the output is aimed at solving the problem the input flagged.6 On average. Not every time. Sometimes you do step back and still get bitten. The point is that across thousands of activations over evolutionary time, the program produced the right behavior more often than alternative programs in the population, and that's how it stayed in the gene pool.
Buss calls the whole structure an "evolved psychological mechanism" — EPM for short — and treats it as the unit of human nature.1 Evolutionary psychology, at its core, is the search for these mechanisms.
A few more properties of how EPMs collectively organize the mind matter for downstream pages: they provide a non-arbitrary way to "carve the mind at its natural joints," they tend to be problem-specific, they are numerous, they are complex, and they yield behavioral flexibility precisely because there are so many of them, not despite.7
Children develop fear of snakes, spiders, heights, darkness, and strange men far more readily than fear of cars or electrical outlets, even though cars kill more people now.8 Take a six-month-old who has never seen either. Show them a snake-shape moving on the floor. Show them an outlet in the wall. The baby tracks and reacts to the snake. The baby ignores the outlet. The outlet is more dangerous in modern environments, but the baby's mechanism never met an outlet in the EEA. It met a lot of snakes.
Disgust shows the same pattern at a different scale.9 Across cultures from the Netherlands to West Africa, people find rotting flesh, dirty food, food prepared by someone with dirty hands, and food that has touched feces or insects to be exceptionally repellent. The disgust response is the same. The facial expression is universally recognized — by congenitally blind people who have never seen the face, by deaf people from infancy. Women show stronger disgust sensitivity than men, and they perceive higher disease risk from contaminated objects, because women historically cared for vulnerable infants and lost more reproductive return when an infant died of pathogen exposure.9
Sexual jealousy gives Buss his own worked example.10 Imagine you've gone to a party with your romantic partner and stepped out for a drink. You return. Your partner is standing close to someone, leaning in, eye contact, light touching. The cues act as input — your jealousy mechanism reads "threat of partner loss," not "they are having a conversation." Decision rules evaluate options: ignore them, threaten the rival, become enraged at the partner, reevaluate the relationship. Output flows three ways — physiological arousal, behavioral confronting or threatening, and information passed back to the relationship-evaluation mechanism. The whole program is running on you before you decide what you think.
Westermarck-style incest avoidance shows the input-side specificity even more sharply.11 If you grew up with someone — same household, ages roughly two to six — you tend to find that person sexually unattractive as an adult, regardless of whether you are actually genetically related to them. Kibbutz children raised in the same peer group rarely marry each other. Taiwanese minor marriages, in which a girl is raised in her future husband's household from infancy, have unusually low fertility and high divorce rates. The cue your incest-avoidance mechanism uses is childhood proximity — a reliable indicator of likely kinship in ancestral environments. The output is sexual aversion. Your nervous system never had to detect actual genes; it just had to detect the conditions that statistically tracked them.
Each example follows the same pattern. Narrow input. Function-specific output. The eye does not pass through smell-receptor inputs. Snake fear does not activate on photographs of cars. Sexual jealousy does not activate when your sister has dinner with someone else. Incest aversion targets people you grew up with, not all people. That specificity is the calling card of an EPM.
Run the same fear program with two different inputs and you get freeze in one case and flee in the other. The program stayed the same. The behavior was the output. Behaviorist accounts that catalog human behaviors miss the level at which the explanation actually works — selection picked the program that picks the behavior, not any single behavior.
A bigger fight runs underneath: whether all human mental architecture works this way, or whether some parts of the mind are general-purpose. Tooby and Cosmides describe the mind as a Swiss Army knife — many specialized tools, each tuned to a different problem. Pinker concurs and gives the position its sharpest line. But Chiappe, MacDonald, Geary, and Huffman push back: humans face many novel problems that did not recur often enough in the EEA for specific adaptations to develop, so general-purpose problem-solvers — general intelligence, working memory, classical conditioning, analogical reasoning — must also exist.12 Buss treats the dispute as open, while writing from inside the specificity camp.
A related fight is over information encapsulation. Fodor's classical version of modularity says modules cannot access information from other modules — the visual system cannot read what the language system knows.13 Buss explicitly rejects encapsulation as a defining feature of EPMs.14 Sight, smell, internal hunger, and memory of previous meals all feed into a single decision about whether to eat the thing in front of you. Walk through the woods and run into a hungry lion, a bush of ripe berries, and a beautiful potential mate at the same time, and you don't experience three separate decisions sealed off from each other. You experience a single prioritization — usually flee the lion first — that reflects superordinate regulatory machinery handling input from multiple EPMs at once.14
The on-average-success criterion catches readers who expected something stronger. Snake fear is an EPM even if it sometimes triggers on a stick. The fact that an EPM is biased — sometimes producing false alarms, sometimes missing real threats — does not disqualify it. What matters is that, across many activations over evolutionary time, the bias toward overestimation cost less than the bias toward underestimation would have. This is the substrate that Error Management Theory builds on.
The flexibility paradox runs counter to most people's intuitions. The intuition: more innate equipment means less behavioral flexibility, because innate things are rigid. Buss reverses it.15 More specific tools mean more flexibility — not less — because each tool unlocks a new task the organism couldn't do before. A bird with feet walks. Add wings, it flies. Add a beak, it cracks seeds. The carpenter analogy makes this concrete: a flexible carpenter has many specialized tools, not one general "flexible" tool. The opposite of an EPM-rich mind is not a free mind. The opposite is a mind that can do almost nothing.
Buss and Pinker stand together on the central claim — that domain-specific mechanisms are credible and "domain-general plasticity" is not. Pinker's most concise statement of the case becomes Buss's anchor. "The idea that a single generic substance can see in depth, control the hands, attract a mate, bring up children, elude predators, outsmart prey, and so on, without some degree of specialization, is not credible. Saying that the brain solves these problems because of its 'plasticity' is not much better than saying it solves them by magic."16 Pinker is making the argument rhetorically; Buss is making it methodologically through the six-property definition. Both are leaning on Tooby and Cosmides's underlying point: natural selection cannot have run on something as vague as "general intelligence" for any of the specific functions humans need. Selection runs on specific design features that solve specific problems better than alternative designs at each step, generation after generation. So the case for EPM specificity is not a methodological preference. It is a constraint that follows from how natural selection actually operates.
Where Buss, Pinker, and Tooby/Cosmides diverge is at the edges. Pinker is willing to accept some domain-general mechanisms layered on top of specific ones — working memory, classical conditioning, certain forms of analogical reasoning. Tooby and Cosmides are not. Buss treads carefully here, reporting Chiappe and MacDonald's case for domain-general mechanisms responding to "non-recurrent problems" without endorsing it. The ambiguity is honest; the data don't yet pick a clear winner. The implication for the vault is that "EPM" should not be invoked sloppily for every observed behavior — the six-property test functions as a check, not a flag.
A second convergence sits underneath the encapsulation question. Buss, Tooby, Cosmides, Hagen, and Sperber all converge on EPMs being functionally specialized and content-rich. They all diverge from Fodor on encapsulation. The consequence for vault use: when a page says a behavior is "modular," what's meant is function-specific, not informationally walled off. Adaptations talk to each other. Your fear EPM, food EPM, and mating EPM run simultaneously when you're walking in the woods, and the choice that emerges depends on input from all three plus a superordinate prioritizing layer. Saying "modular" without that qualifier risks Fodor's ghost showing up to confuse the reader.
A profiler watches someone for thirty seconds across a room and tells you something true about that person's state — likely lying, likely angry under the calm, likely under enormous status pressure. The profiler is not reading character. The profiler is reading what evolved psychological mechanisms leak while they're activating — micro-flinches, eye-attention shifts, breath-rate changes, posture asymmetries that the target doesn't know they're showing.
This is what behavioral-mechanics is, looked at from the EPM frame: applied EPM-activation theory. Hughes's Behavior Tracking Element framework reads physiological output (Property 5: physiology is one of the three EPM output channels) of EPMs that are activating in the target during a conversation. The target is running fear, status-assessment, social-comparison, and trust-evaluation EPMs simultaneously. Each one is leaking output through the eye, face, body, and voice clusters Hughes catalogs. The profiler is not learning a personality theory. The profiler is learning to read mechanism activations through the channels mechanisms broadcast in.
Lieberman's fast-pitch interrogation works for the same reason from the other direction. The interrogator presents fragments of input that an EPM was tuned to detect — suggestion of betrayal by someone close, suggestion that a moral violation is about to be exposed, suggestion that group standing is about to drop. The decision rule fires (Property 4). Outputs leak (Property 5). The fast-pitch is structurally identical to ringing a bell that the target's EPM is wired to respond to. It works on someone with a normally configured jealousy or shame EPM. It fails on someone whose EPM is unusually configured — which is one of the diagnostic markers for psychopathy in interrogation contexts.
Naming this connection produces an insight neither domain generates alone: every influence technique can be redescribed as a particular protocol for triggering one or more EPMs in a target. The redescription matters because it tells the operator who a technique will work on (targets whose EPM is normally configured), who it will fail on (targets whose EPM is atypical), and which countermeasures might neutralize it (interventions that change the EPM's input or decision-rule sensitivity). Bag-of-tricks BM catalogs become coherent under this frame. They become targeted instrument design for hitting specific evolved machinery. The implication for the vault: pages that describe BM techniques should — when they're being rigorous — name which EPM they target and which input/output channels they hit.
A different handshake runs from the EPM frame to creative practice. Stories work on you because they activate evolved psychological mechanisms in sequence. You laugh, cry, gasp, root for or against people who do not exist. The conventional account is that stories are emotion-pumps. The EPM frame gives a sharper account. A romance arc activates pair-bonding EPMs — commitment-cue detection, jealousy, romantic love. A thriller arc activates threat-detection EPMs — snake fear, predator avoidance, betrayal detection. A tragedy activates kinship-loss EPMs — grief, parental investment frustration, the parental-protection failure response. A hero's journey activates status-striving EPMs — dominance hierarchy, prestige attainment, recognition. What separates strong storytelling from weak storytelling is partly the precision of the activations the writer triggers — the cues match what the EPM was actually tuned to detect, so the response is automatic, not effortful. Weak writing triggers no EPM at all (you feel nothing) or triggers one shallowly (you feel a flicker that doesn't sustain).
This connects to the Non-Fiction Writing Craft hub through a sharp prediction. Strong non-fiction prose, like strong fiction, activates the most EPMs per paragraph. Wang Dan writing about modern China activates threat-detection (institutional risk), kinship (China as ancestral homeland for diaspora readers), and curiosity-as-status-striving. His prose lands because the activations are dense and specific. The vault's writing-craft pages and the EPM page meet here in a working diagnostic: if a paragraph doesn't activate an EPM, cut it. The reader will not notice it's there. It is not doing the work prose has to do.
The Sharpest Implication.
You do not have a unified self that processes information and decides. You have a federation of evolved psychological mechanisms, each waiting for its narrow input, each running its decision rules independently, each pushing output into the system. Most of the time the federation feels unified because the EPMs are coordinated and the outputs converge. In moments of conflict — fear meeting hunger, jealousy meeting compassion, status-striving meeting kinship obligation — what feels like a struggle of the will is a competition of EPM outputs for behavioral control. Your conscious deliberation is a layer on top of this competition, partly editing it, partly noticing it, mostly downstream of decisions the EPMs have already made.
This means introspection lies. When you ask yourself why you did something, you do not have access to the EPM-level processing that actually drove the behavior. You generate a story about what you did. Sometimes the story is accurate and sometimes not. Accurate stories tend to come from cases where one EPM was loud enough to read — you felt the fear, you felt the desire, you felt the jealousy. Inaccurate stories tend to come from cases where multiple EPMs contributed and the conscious narrator can only credit one of them.
A more disturbing corollary: your moral self is also a federation. The EPMs that produce empathy toward kin are different from the EPMs that produce empathy toward strangers. The EPMs that produce status-striving are not aligned with the EPMs that produce care. Sometimes you act morally because the relevant EPM happened to win. Sometimes you act selfishly for the same reason. The unified moral agent who chooses good over evil is mostly a narrative the conscious layer assembles after the fact.
Generative Questions.
If introspection reports are unreliable for EPM-level processing, what data sources actually let you diagnose which EPM is driving a particular behavior pattern in yourself or someone else? Behavioral consistency across contexts? EPM-specific physiological signatures? Pattern-matching against known EPM templates? The methodology is not obvious, and the EP-clinical-philosophy intersection has not metabolized this gap.
The EPM model predicts that "self-mastery" is not the will overriding evolved drives. It's the cultivated re-weighting of which EPMs win in which contexts. What does training to re-weight EPM outputs look like, in practice, in a non-clinical setting? Contemplative traditions have practical answers — vipassana practice as a method for letting EPM activations rise and fall without converting them to behavior. Cognitive-behavioral therapy has narrower answers. Are these the same intervention pointed at the same machine, or different interventions on different layers? The vault has pages on both that don't yet meet.
If you genuinely believe your moral choices are EPM-federation outputs, does the felt experience of moral responsibility change? Should it? Buss is silent on this. The implication is potentially radical, and the evolutionary-psychology / clinical / philosophy intersection has not yet absorbed it.