Selection itself is easy enough: whoever leaves more offspring passes their stuff on. But it rests on a premise that usually gets skipped — there has to be something different to pick from. If everything on the field is identical, no amount of filtering produces any change.
So what really sets how far a species can go is often not how harsh the filter is, but how many different things are on the table. Those things arrive from three sources, differing in speed by orders of magnitude — and the largest stream never passes through selection’s hands at all.
Imagine copying a thick book out by hand. Over a hundred thousand words you will slip a few times. Where the slips land has nothing to do with how good the content is — you do not miscopy a passage because that passage happens to be badly written.
A cell copying that long tape is in the same position. An Icelandic study read the genomes of 78 father-mother-child trios: on average each child carried some sixty changes that neither parent had. The great majority came from the father’s side, and the count rose by roughly two for every year of his age.
The reason is concrete: eggs are essentially laid down before a woman is born, while sperm are produced continuously, and every division means recopying the whole book. More copying, more slips.
Here is the sentence that matters. Calling mutation random does not mean every position is equally likely to be miscopied — they are not. It means exactly one thing: the chance a mutation appears is unrelated to whether it would do its carrier any good. A dry year does not call forth extra drought-tolerant mutations.
That sounds like a philosophical stance; it was in fact settled by experiment. In 1943, Luria and Delbrück wanted to know: when bacteria meet a virus that kills them, are the survivors resistant because the virus provoked the change, or because they were already resistant beforehand? Both accounts predict roughly the same average number of survivors, so averages cannot separate them.
But they predict opposite things about the scatter — and scatter is countable.
They counted, and it was the latter. This trick of reading the scatter rather than the mean later earned a Nobel Prize in 1969, and the whole logic of antibiotic resistance rests on it today: antibiotics do not create resistant bacteria, they merely pick out the few already present and clear away all their competitors. Something starting below one in ten thousand can take over in days.
Sixty changes per person sounds like a lot until you spread it over a text three billion letters long; a species cannot wait for that to accumulate. Fortunately a second route is far faster, and it produces nothing new at all — it only recombines what is already there.
When you make sperm or eggs, the set of chromosomes from your father and the set from your mother line up alongside each other, cut across, swap whole segments, then part company — in humans a few dozen swaps per round. Add that each of the 23 pairs contributes one chromosome at random, which alone gives over eight million combinations, and with the swapping layered on top, essentially every reproductive cell you make is unprecedented.
So the division of labour is clean: mutation adds new material to the pool, slowly; recombination re-deals what the pool already holds, quickly. The differences a population can offer up for filtering right now are overwhelmingly the second kind.
Recombination also does something less obvious with large consequences. In a lineage that never recombines — bacteria splitting in two, say — two good mutations landing in different individuals can never end up in the same descendant; they can only compete until one of them is gone. Switch recombination on and both can be merged. This is one mainstream answer to the question of what sex is for: it is a machine for reassembling variation that is scattered across different individuals.
The third route is fastest of all: import a whole batch of ready-made variation from another population. That is gene flow.
There is a small island in the Galápagos called Daphne Major, where the Grants watched for forty years and numbered every finch. In 1981 an immigrant male landed there — bigger than any bird on the island, singing a song nobody had heard. DNA later showed it had flown in from another island more than a hundred kilometres away, and belonged to a different species.
It paired with a local female. By the third generation, its descendants’ beaks and songs no longer matched any other bird on the island; with nobody else to pair with, they bred among themselves. Reproductive isolation that usually takes hundreds of generations was in place in three. The researchers call this line the Big Birds.
Whether that counts as a new species depends on which species definition you use — an old argument belonging to the topic on speciation. One thing is not in doubt: not one of the variants that suddenly appeared on this island arose locally.
A severe drought around 2003 nearly wiped out the Big Birds. Two were left, a male and a female, brother and sister. When the rains returned they paired, and every Big Bird since descends from them. A lineage that had barely formed had its entire holdings compressed into whatever those two birds happened to carry — and not one of the birds the drought killed died for being insufficiently good. This is not filtering. It is sampling.
Try a more everyday picture: in a village of a few dozen households, a few surnames always vanish over a few generations — not because the people bearing them were more capable, but because of which households happened to leave descendants. Every generation is a sample drawn from the one before, and sampling carries noise of its own. This is what genetic drift means.
The backbone of the whole thing is one quantitative relation: the smaller the population, the louder the sampling noise. In a large population a variant carrying a slight advantage climbs steadily; in a small one it is often sampled away before the climb even starts. So whether selection or luck has the final word depends not on how good the variant is, but on how large the population is.
And drift’s territory is startlingly large. Most mutations change nothing once made: the cell reads that tape three letters at a time with each triplet standing for one amino acid, and swapping the third letter often leaves the same amino acid — not to mention the long stretches never read out at all. Such changes have no measurable effect on whether their carrier lives or breeds, so their fate is settled almost entirely by sampling noise. Motoo Kimura pushed this to its limit in 1968: at the molecular level, the great majority of changes that end up fixed are neutral. Heresy at the time; broadly the consensus now.
Which yields the counter-intuitive corollary: seeing a variant spread through a population is not grounds for concluding it helped. Claiming selection pushed it up requires separate evidence — that it climbed faster than drift can account for, say, or that several unrelated populations each pushed the same variant up independently.
Losses from sampling are also unusually hard to repay. Cheetahs worldwide carry astonishingly little genetic difference — in the 1980s, skin grafts between unrelated cheetahs were not rejected, where any other cat would have thrown them off. Northern elephant seals were hunted down to a few dozen animals by the late nineteenth century; there are a couple of hundred thousand today, yet almost no variation shows up at the positions surveyed. The numbers came back; the variation did not.
There is a clear human example too: in an Amish community in Pennsylvania, an inherited condition involving extra fingers and short stature occurs far more often than anywhere else, and the trail leads back to one couple who arrived in 1744. This says nothing about the community being "worse"; it says how few people came in at the start. Genetic differences between human populations are in large part the residue of migration and sampling history, not a scoreboard of merit — reading them the second way is an error that was refuted over and over in the nineteenth century.