An Algorithm With No Foresight

Darwin's argument has a shape that is easy to miss under the weight of the examples, and the shape is the reason it works.

Take any population where three things are true. Individuals differ from one another. Offspring resemble their parents more than they resemble strangers. And some variants leave more offspring than others.

That is all. If those conditions hold, the composition of the population must change over generations, with no further assumption required. It is not a claim about finches or about biology at all; it is an algorithm, and it runs in any medium that meets the conditions (bacteria under antibiotics, viruses in a vaccinated population, software under an evolutionary optimiser, antibodies within a single immune system).

What Darwin supplied on top of the algorithm was the argument that it is sufficient: that a process with no goal and no foresight can produce an eye. That is the part that was, and to some people still is, hard to accept.

Begin with what fitness means, because the usual phrase actively obstructs understanding. "Survival of the fittest" was Herbert Spencer's, and Darwin adopted it without much enthusiasm. Fitness in the technical sense is reproductive success, counted in descendants. Survival matters only as a means: an organism that dies young having bred prolifically is fitter than a long-lived one that never reproduces, and traits that shorten life are readily favoured if they raise reproductive output, which is why salmon spawn once and die, and why a peacock's tail exists at all.

The objection that this is circular (the fittest are whoever leaves most offspring, so the theory says survivors survive) dissolves once fitness is estimated independently. Measure a beak's depth, model the seeds available in a drought, predict which birds will feed successfully, and then count the descendants. Peter and Rosemary Grant did exactly this on Daphne Major for forty years, and watched beak depth shift measurably in a single dry season and shift back when the rains returned. The prediction preceded the count.

Selection is not the only process operating, and treating it as though it were is the commonest error in popular Darwinism. In a small population the frequency of a variant can rise or disappear by chance alone, regardless of any advantage, sampling noise, nothing more. Motoo Kimura argued that most variation at the molecular level is of precisely this kind: neither helpful nor harmful, drifting. A founder effect is drift at its most visible, where a handful of colonists carry an unrepresentative sample of the parent population's variation and their descendants inherit the accident.

Nor does selection optimise. It can only work with the variation that happens to be present, and it can only follow paths where every intermediate step is itself viable, which means it climbs the nearest hill rather than locating the highest one. The evidence is written all over anatomy. The vertebrate retina is installed backwards, with the wiring in front of the photoreceptors and a blind spot where the nerve bundle passes through; the cephalopod eye, evolved separately, has it the sensible way round. The recurrent laryngeal nerve runs from the brain down into the chest, loops under the aorta and returns to the larynx a few centimetres from where it started; in a giraffe that is a detour of several metres. Both are perfectly intelligible as history and indefensible as design, which is exactly what a process without foresight predicts and what a designer would not produce.

The hardest case for the algorithm, and the one that took a century to solve properly, is altruism. Why would selection ever favour an organism that reduces its own reproduction to help another?

The wrong answer, which dominated popular writing well into the 1960s, was that it benefits the species. This fails on its own terms: in a population of self-sacrificing individuals, a variant that accepts help and gives none does better than everyone else and spreads. Group-level selection is not impossible, but it requires conditions (strong competition between groups, limited migration) that are demanding and rarely met.

William Hamilton's answer in 1964 was to change the accounting unit. A gene that causes its bearer to help a relative is helping copies of itself, since relatives share genes with a known probability. Hamilton's rule states the condition compactly: help is favoured when relatedness multiplied by the benefit to the recipient exceeds the cost to the helper. Haldane had got there first as a joke, saying he would lay down his life for two brothers or eight cousins, which is the arithmetic exactly.

This is what "inclusive fitness" means, and it dissolved a set of problems that had looked intractable: the sterile castes of social insects above all, where workers are more closely related to their sisters than they would be to their own daughters.

The place all of this becomes immediately practical is a course of antibiotics. Resistance is not created by the drug: variation in susceptibility already exists in a bacterial population of billions, and the antibiotic sorts it, killing the vulnerable first. Stop early, when symptoms ease, and you have removed the pressure at precisely the moment when the survivors are the least susceptible, leaving them to rebuild the population from that starting point. Honesty requires adding that the optimal duration is now genuinely contested (some evidence suggests shorter courses reduce total selection pressure) but the mechanism is not in doubt, and it is the algorithm running in a few days rather than a few million years.