Change who the "who" is in "who does this benefit", and ledgers that would not balance suddenly do
2026-08-26 · BigCat
"The selfish gene" sounds like a verdict on human nature. It is a change of accounting method.
The Selfish Gene may be the most thoroughly misread title in science.
It sounds like a claim that people are selfish by nature. It contains no claim about psychology at all. What it changes is the unit of account — the "who" in "who does this benefit" moves from a body to the stretch of information being copied.
Changing the unit changes no facts. But a batch of ledgers that would not balance suddenly do, and a batch of phenomena that were invisible come into view.
// 01
What gets passed on is not you, but the segments that pass through you
After you are gone, what remains? The intuitive answer is "my genes". It does not survive one follow-up question: your genome, as a particular combination, exists for exactly one generation and is never copied a second time.
The problem is sex. When sperm and eggs are made, the two sets you got from your parents are shuffled again — cut, swapped, reassembled. So the set you hand a child is neither your father's nor your mother's but a freshly dealt hand. Your child passes it on and it is shuffled again.
What is copied over and over is therefore never "you" but segments of chromosome.
How long is a segment? The scissors decide. Each shuffle makes roughly one or two cuts per chromosome. Short stretches that escape the cut can travel almost intact for hundreds of generations, while a whole chromosome is in pieces within a few. Which gives an odd but crucial conclusion: the boundary of "a gene" is drawn not by chemistry but by how often the shuffling cuts — short enough to be long-lived.
Across four generations the same chromosome is cut into ever smaller pieces — except the green stretch, which is passed on whole every time. The boundary of a "gene" is drawn by the scissors, not by chemistry: short enough to be long-lived. (Schematic; cut positions are illustrative.)
Williams defined the gene along exactly these lines back in 1966: any stretch of heritable information is a unit so long as it is favoured faster than it is cut apart. There is no molecular biology in that definition — just two rates being compared.
Segment length can be measured directly. In European populations, the variant that lets adults keep digesting milk sits inside an unusually long and unusually uniform stretch of DNA. Only one explanation fits: it was strongly favoured and spread fast, faster than the scissors could chop it up. So "how much of the surrounding stretch is left" becomes a clock, reading off how recently this happened.
None of this makes the body unimportant. The body is precisely the interface selection can see — a bird looks at the moth's colour, not at a piece of DNA. The body simply is not the thing being inherited. Biology calls the first a vehicle and the second a replicator: one takes the hits, the other gets passed on.
// 02
"Selfish" is a method of accounting, not a statement about human nature
Why is the change of unit worth making? Because on the "is this a good deal for this animal" ledger, a whole class of phenomena comes out impossible.
A worker bee never breeds and spends its life raising someone else's young. On its own reproductive ledger that is a total loss and should never have been selected. A ground squirrel that spots a hawk stands up and screams, drawing the predator's attention onto itself — the same losing trade.
On the copy ledger it resolves. Copies of a variant do not sit only in the body carrying it; they also sit in relatives. A sibling has a one-in-two chance of carrying the same one. So "sacrifice yourself to save three brothers" comes out positive. (How that sum actually works is for the piece on kin selection.)
The technical meaning of "selfish" has to be nailed down: it is a statement about quantities — any variant that raises its own copy number will become more common, regardless of what it makes the body do. It carries no psychological content, and it implies nothing about how people ought to behave. In fact the same logic often predicts cooperation.
Does the new ledger predict anything the old one cannot? It does, right where the two diverge.
On mouse chromosome 17 there is a stretch of DNA that males carrying it pass to as many as 99% of their offspring, where half would be normal. The trick is nasty: genes in that region damage sperm motility across the board, and only the sperm carrying the stretch have the antidote to rescue themselves. The price is that mice inheriting two copies are sterile or die outright.
Something that demonstrably harms its own carrier and does the mouse population no good whatsoever spreads anyway. On the "is this good for this mouse" ledger the event is unintelligible; on the "is this stretch of DNA becoming more or less common" ledger there is no mystery at all.
// 03
A gene's influence can grow on somebody else
Is a beaver's dam part of the beaver?
Intuition says no — it is outside the body, made of wood and mud. But selection recognises one thing only: whether variation in this stretch of DNA reproducibly changes the odds of its own copying. Whether that is achieved by remodelling your own body or by remodelling something outside it makes no difference at all. Dam-building ability varies with genetic variation, and how good the dam is directly decides how many offspring that beaver leaves — so the dam, like the incisors, is a product of those genes.
The reach comes in roughly three layers: things built (the beaver's dam), others manipulated (a parasite rewriting its host's behaviour), and action at a distance (a cuckoo chick's call driving foster parents to feed it frantically).
The rings show how far the influence reaches, not where the body ends. As long as the dashed loop closes — the effect lands on someone else, the benefit comes back to you — selection can act on it, however many bodies the influence passes through.
Manipulation is the most convincing case. A fungus that infects carpenter ants makes one leave the nest, climb to a height of a few tens of centimetres, clamp onto the vein of a leaf and die there. The temperature and humidity at that spot happen to be ideal for the fungus to fruit and rain spores onto the ant trails below.
How do we know this is manipulation rather than an ant too sick to know what it is doing? Because the precision can be measured. Field surveys show the height, the orientation and even the time of day of the bite are all far from random — clustered around midday, on the underside of leaves a few tens of centimetres up. Precisely matching what transmission requires is exactly the line between adaptive manipulation and a side effect of being ill.
The counterexample matters too. Toxoplasma making rodents stop avoiding the smell of cat urine was long the field's favourite case. Later work found infected mice were generally less anxious — they also stopped avoiding fox and guinea-pig odours — which looks more like a by-product of neural inflammation than precision steering. The framework holds; individual cases have to be judged one at a time.
🎒 Scenarios · BigCat
① Parenting and education — the behaviours a child repeats, the ones that survive being told off many times, are usually paying for themselves on the child's own ledger: they buy attention, or buy not having to face something, or buy a completely predictable reaction. “Selfish is a method of accounting” says a ledger that will not balance usually means the wrong unit of account. So ask what this behaviour is buying before deciding what to change — suppress it without knowing, and the same need generally returns in another form.
② Writing, and this learning site — “a gene's influence can grow on somebody else” says the boundary of an output is where the causal influence ends, not the file you delivered. So what this site actually produces is not pages but a change in how you look at things afterwards. The metric should not be how many issues got written but which single sentence you can still recall three months later — if none, the effort went into coverage rather than into hooks.
🌀 Crossing over
Buddhism · non-self: the wish to "carry oneself forward" finds almost nowhere to stand against the mechanism in “what gets passed on is not you” — your genome as a combination exists for one generation and is never copied again, and what actually passes through you is a set of segments that belonged to nobody before you and will belong to nobody after. Which yields a harder conclusion than "life is short": "continuing yourself" fails literally, not for lack of time, but because that self was never the unit being transmitted — it is a one-off combination, the vehicle rather than the cargo. But here it breaks down: Buddhist non-self is said in order to dismantle attachment, and it will not even grant that the segments have continuity; evolution insists on precisely that, since its whole account rests on something being copied with high fidelity. You can borrow the blade, not the final stroke.
Parenting · the tug-of-war between mother and fetus: pregnancy is usually imagined as a case of perfectly aligned interests. On the accounting in “selfish is a method of accounting” it does not balance that way — each variant in the fetus has a one-in-two chance of not being in the mother, so "extract a little more from her" is positive for the fetus and negative for her. Some of the physiological pulling seen in blood sugar and blood pressure during pregnancy is the residue of that mismatch: the placenta secretes hormones that raise maternal blood glucose, and the mother answers with more insulin. Hence a counterintuitive conclusion: part of what gets called "pregnancy discomfort" is not a body malfunctioning but two ledgers being kept separately inside one body. But here it breaks down: this explains a physiological mechanism and implies nothing whatever about how a child should be raised or treated — the jump from "is" to "ought" is especially dangerous on this subject.
Institutions · a fair rule for dividing beats a shared goal: the thousands of stretches of DNA in a genome hold their interests in line and rarely feud, not because they share an objective but because one rule cannot be altered by any of them — when sperm and eggs are made, every stretch gets exactly a one-in-two chance of going along. That enforced fairness makes "help the whole" and "help myself" coincide. The mouse DNA in “selfish is a method of accounting” is exploiting a loophole in that rule, raising its own share to 99%, which is why it spreads while demonstrably harming its carriers. From which follows a conclusion you would not otherwise reach: to align interests inside an organisation, checking the allocation rule for holes that let someone unilaterally raise their own share beats hoping everyone endorses the same goal — where the hole is, behaviour that damages the whole grows by itself, no malice required. But here it breaks down: the mouse cheat has its own brake, since two copies mean sterility or death and rising frequency runs it into a wall; human institutions have no built-in brake, so someone has to notice and enforce — and noticing is usually slower than cheating.
One layer deeper
If the boundary of "a gene" is drawn by how often shuffling cuts, what is left of this view where there is no shuffling?
Asexual organisms, the small set of genes inside mitochondria, and the Y chromosome are all places where shuffling has been switched off. The whole block becomes one indivisible unit, and the gene's-eye view collapses into the individual's-eye view — which shows its explanatory power came from recombination, not from the word "gene". The price shows up too: a block that never shuffles cannot separate the harmful from the helpful and has to carry both together, which is how the Y chromosome came to degenerate over time.
Where should extended influence stop? Do artefacts and culture count?
Two criteria, both required: there is a stable association between the influence and some heritable variant, and the benefit returns to that variant's copy number. A beaver dam meets both. A human city meets neither — city-building is transmitted culturally rather than genetically, and the benefit does not flow back to any stretch of DNA, so it needs a different model. The framework is not a licence to include everything; it comes with a fairly strict threshold.
If the body is only a vehicle, why has evolution not made it more durable?
Because durability is not on the ledger. What counts is how many offspring you leave, not how long you last. Once the reproductive period is over, selection pressure on you falls away fast, and problems that only appear then are barely filtered at all. Worse, variants that help early and charge their cost late get selected anyway, because the account was settled at the early end. Ageing is not a design failure; it is a mismatch of payment terms.