Evolution by natural selection

Foundations

The word that ruins it

The single most common thing people say about evolution is that it is random. It is the opposite of random, and the confusion comes from mixing up two steps that do very different jobs.

Step one is random. Mutations happen without regard to what the organism needs. A bacterium under attack by antibiotics does not decide to mutate a resistance gene. Mutations occur constantly, in all directions, mostly neutral, some harmful, occasionally useful, and the organism gets no say.

Step two is the opposite of random. Whatever happens to reproduce more leaves more copies. Do that for a few thousand generations and the population is systematically reshaped, not by chance but by the most repetitive filter imaginable.

Calling the whole thing random is like calling a sieve random because the sand falls in randomly. The falling is random. The sieve is not.

Three conditions, and that is all

Natural selection is not really a theory in the sense of a guess. It is closer to a piece of logic. If three things are true, it follows automatically.

1. Individuals vary.
2. Some of that variation is inherited.
3. More offspring are produced than can survive, and the inherited variation affects which ones do.

If those three hold, then the next generation is enriched in whatever helped. Nothing else is required. No goal, no plan, no designer, no direction. The mechanism is close to a tautology, which is exactly why it is so powerful: it will run in any system where those conditions hold, including in computer code and in a test tube of molecules.

Darwin worked this out in the 1830s and sat on it for twenty years. What forced him to publish was a letter in 1858 from Alfred Russel Wallace, working in the Malay Archipelago, containing the same idea. Their papers were read together at the Linnean Society that year, and the Origin followed in 1859.

The other word that ruins it

“Survival of the fittest” is Herbert Spencer’s phrase, not Darwin’s, and it misleads two ways.

First, survival is not the point. Reproduction is. An organism that lives a century and leaves no offspring has a fitness of zero. A mayfly that lives a day and lays five hundred eggs is doing better.

Second, fitness is not strength or speed or complexity. It means expected number of surviving offspring in a particular environment, and nothing else. Change the environment and fitness changes with it.

The clearest demonstration is parasites. A tapeworm has no gut, no eyes and a drastically reduced genome compared with its free-living ancestors. By any intuitive measure it is degenerate. By the only measure that counts it is a triumph, because those losses save energy that goes into producing eggs by the million. Evolution has no preference for complexity. It went the other way here, and it was right to.

Not everything is an adaptation

There is a habit of assuming that every feature of an organism is there for a reason. Often it is not.

Populations are finite, so gene frequencies also drift by pure chance, and drift is stronger the smaller the population. A variant can spread because it happened to be in the individuals that survived a flood. Some features are by-products of others. Some are historical accidents, locked in because too much depends on them to change now.

The vertebrate eye has its light sensors pointing backwards, with the wiring in front of them, which requires a hole in the retina for the nerves to escape. That hole is your blind spot. An octopus, which evolved eyes separately, has the wiring behind the sensors and no blind spot. Neither design is a plan. Yours is a good solution built from an awkward starting point that could not be undone.

The evidence, which is not fossils

Fossils are the famous evidence and the weakest of the strong ones, because they are patchy by nature. The stronger evidence is in living organisms.

Homology. Your arm has one bone, then two, then a cluster, then five digits. So does a bat’s wing, a whale’s flipper and a mole’s digging paw. Four completely different jobs, one inherited layout. If each were designed for its task there is no reason for the shared arrangement.

The genetic code. Nearly universal, as lesson 5.1 described, with no chemical requirement that it be this one.

Broken genes. You carry the gene for making vitamin C. It does not work, because one letter is missing. The same gene is broken in the same way in chimpanzees and gorillas, and works fine in most other mammals. An inherited breakage explains that. Independent origins do not.

Viruses in your genome. Retroviruses insert themselves into host DNA, and if they land in a cell that makes eggs or sperm they are inherited. Humans carry thousands of these, and humans and chimpanzees share them at the same positions in the genome. The chance of the same virus landing at the same base pair independently, thousands of times over, is not a number anyone needs to compute.

Common descent has also been tested formally, as a statistical hypothesis against models in which groups arose separately, and universal common ancestry wins by an enormous margin.

Watching it happen

You do not have to infer any of this from the past. Antibiotic resistance evolves in hospitals within months. Pesticide resistance evolves in fields within seasons. Influenza changes fast enough that the vaccine is redesigned every year.

The most careful demonstration has been running since 1988. Richard Lenski’s laboratory has kept twelve populations of E. coli in identical flasks, transferring a sample to fresh food every day, for more than eighty thousand generations. On the human scale that is roughly two million years.

Two things came out of it that a fossil could never show. Because samples are frozen every five hundred generations, the experiment has a physical archive of its own past, and you can restart the past. And in one population, after about thirty-one thousand generations, something appeared that this species is defined by not being able to do: using citrate as food in the presence of oxygen. When the team replayed that population’s history from frozen samples, the ability only re-evolved from later time points. Earlier mutations, useless on their own, had made the innovation reachable.

That is evolution being contingent, and it is the closest anyone has come to running the tape of life twice.

Deeper

Selection, drift and population size

Fitness differences are opposed by drift, whose strength scales inversely with effective population size. A variant with a selective advantage smaller than roughly the reciprocal of the effective population size is effectively invisible to selection and behaves as though neutral. This has a large consequence: species with small populations, including most multicellular eukaryotes, are less efficient at purging mildly deleterious changes than bacteria with enormous populations. Lynch’s argument is that much of the structural complexity of eukaryotic genomes, including introns, mobile elements and expanded intergenic regions, accumulated because reduced population sizes weakened selection against them rather than because that complexity was favoured.

What the long-term experiment measured

Fitness relative to the frozen ancestor rose steeply at first and then decelerated, but genome sequencing showed mutation accumulation continuing at a roughly constant rate for the first 20,000 generations, indicating that most substitutions were beneficial rather than neutral even as fitness gains slowed. Later work established that fitness has continued to improve without an obvious upper bound, following a power law rather than approaching a ceiling. Several populations independently evolved elevated mutation rates, and several later evolved them back down.

Contingency and replay

The citrate-using lineage matters because E. coli is defined in part by its inability to use citrate aerobically, so this was a genuine functional novelty rather than a tuning of existing capacity. Replaying from frozen samples showed the trait re-evolving only from populations after about 20,000 generations, implying a potentiating mutation earlier in that lineage’s history. The result is directly relevant to Gould’s thought experiment about replaying the tape of life: history matters, and the same starting point does not reliably give the same outcome.

Testing common ancestry formally

Theobald’s approach treats universal common ancestry as a hypothesis comparable to alternatives in which major groups arose independently, and evaluates them with model-selection criteria on protein sequence data. The support for a single ancestry is overwhelming, on the order of 10²⁰⁰⁰ times better than the best multiple-origin model tested. The test’s assumptions have been debated, but the qualitative conclusion is not seriously contested by working biologists.

Frontier

How much is selection

The relative contributions of selection, drift and mutational bias to genome evolution remain actively argued. Detecting selection in genomic data is statistically difficult because demographic history produces similar signatures, and methods disagree on what fraction of substitutions between species are adaptive.

Evolvability

Whether the capacity to evolve is itself shaped by selection, and how modularity and robustness arise, is an open area. It is difficult to study because it requires reasoning about long-term consequences of variation structure, which selection cannot see directly.

Transcript and shot list

Target runtime about 8 minutes. Built as the correction of a single word: separate the random half from the non-random half, then show what the non-random half can build. Shot types: ANIM is rendered from code with Manim, AI is a generated shot from Higgsfield with the shared style reference, REAL is agency imagery with credit, CAM is the narrator.

[00:00] S01 · ANIM · A pair of dice tumbling; the word RANDOM across them; the dice are struck out and replaced by a sieve with sand falling through.

The most common thing people say about evolution is that it is random. It is the opposite of random, and the mistake comes from mixing up two steps that do completely different jobs.

[00:20] S02 · ANIM · Step one: mutations appearing across a genome in all directions, most neutral, some harmful, a few useful; a bacterium under antibiotic attack, unable to choose.

Step one is random. Mutations happen without regard to what the organism needs. A bacterium under antibiotic attack does not decide to mutate a resistance gene. Mutations occur constantly, in all directions, mostly neutral, occasionally useful, and the organism gets no say at all.

[00:45] S03 · ANIM · Step two: a population where one variant reproduces slightly more; run forward a thousand generations; the population is transformed.

Step two is the opposite of random. Whatever reproduces more leaves more copies. Run that for a few thousand generations and the population is systematically reshaped, not by chance but by the most repetitive filter imaginable. Calling the whole process random is like calling a sieve random because the sand falls into it randomly. The falling is random. The sieve is not.

[01:15] S04 · ANIM · Three conditions appearing as numbered lines, then an arrow: therefore.

And the mechanism is almost embarrassingly simple. Individuals vary. Some of that variation is inherited. More offspring are produced than survive, and the inherited variation affects which ones do. If those three things are true, the next generation is enriched in whatever helped. That is it. No goal, no plan, no direction. It is closer to a piece of logic than a theory, which is exactly why it is so powerful: it runs in any system where those three hold, including in computer code and in a tube of molecules.

[01:55] S05 · ANIM · “Survival of the fittest” struck through; a hundred-year-old organism with zero offspring beside a mayfly with 500 eggs; then a tapeworm losing gut, eyes and genome while egg output climbs.

Now the other phrase that ruins it. Survival of the fittest is Herbert Spencer’s, not Darwin’s, and it misleads twice. Survival isn’t the point; reproduction is. An organism that lives a century and leaves nothing has a fitness of zero. And fitness is not strength or complexity. It means expected number of surviving offspring, in a particular environment, and nothing else. The clearest case is a parasite. A tapeworm has no gut, no eyes, and a drastically reduced genome. By any intuitive measure that is degeneration. By the only measure that counts it is a triumph, because every one of those losses saves energy that goes into eggs. Evolution has no preference for complexity. Here it went the other way, and it was right to.

[02:50] S06 · ANIM · A vertebrate retina in cross-section: photoreceptors facing backwards, wiring in front, nerves punching a hole; the blind spot demonstrated; then an octopus eye with the wiring behind.

Nor is everything an adaptation. Your eye has its light sensors pointing backwards, with the wiring in front of them, which means the nerves need a hole in the retina to escape. That hole is your blind spot. An octopus, which evolved eyes separately, has the wiring behind and no blind spot. Neither is a plan. Yours is a good solution built on an awkward start that could no longer be undone.

[03:30] S07 · ANIM · Four limbs aligned: human arm, bat wing, whale flipper, mole paw; the same bone sequence highlighted in each.

So what is the evidence? Fossils are the famous answer and the weakest of the strong ones, because they are patchy by nature. The better evidence is in living things. Start with this. One bone, then two, then a cluster, then five digits. A human arm, a bat’s wing, a whale’s flipper, a mole’s digging paw. Four completely different jobs, one inherited layout. If each were built for its task, there is no reason for the shared arrangement.

[04:05] S08 · ANIM · The vitamin C gene shown intact in most mammals, broken at the same position in human, chimpanzee and gorilla; then retroviral insertions at matching genomic coordinates in human and chimp.

Then the broken things. You carry the gene for making vitamin C. It does not work, because a piece is missing. The same gene is broken the same way in chimpanzees and gorillas, and works fine in most other mammals. Inherited breakage explains that. Separate origins do not. And better still: retroviruses insert themselves into host DNA, and if they land in a cell that makes eggs or sperm, they are inherited. You carry thousands. Humans and chimpanzees share them at the same positions in the genome. The odds of the same virus landing on the same base pair independently, thousands of times over, is not a number anyone needs to work out.

[04:55] S09 · REAL · Flasks in a shaking incubator in the Lenski laboratory (research image; confirm credit line). S10 · ANIM · A freezer archive with samples every 500 generations; a counter passing 80,000.

And you don’t have to infer any of it from the past. Antibiotic resistance evolves in hospitals within months. Influenza changes fast enough that the vaccine is redesigned every year. But the most careful demonstration has been running since 1988. Twelve populations of E. coli, identical flasks, a sample moved to fresh food every day, for more than eighty thousand generations. On a human scale, roughly two million years. And because samples are frozen every five hundred generations, this experiment has a physical archive of its own past.

[05:45] S11 · ANIM · At generation 31,000 one population turns cloudy; a citrate molecule; then the replay: restarts from early samples fail, restarts from later samples succeed.

Which lets you do something no fossil allows. After about thirty-one thousand generations, one population did something this species is defined by not doing: using citrate as food in the presence of oxygen. So they replayed it. Restarted that population’s history from frozen samples. The ability only came back from the later time points, which means earlier mutations, useless on their own, had made the innovation reachable. That is contingency, measured. It is the closest anyone has come to running the tape of life twice, and the answer was that history matters.

[06:40] S12 · ANIM · A formal model comparison: universal common ancestry against independent origins; the support figure appearing.

Common descent has also been tested as a formal statistical hypothesis, against models in which the major groups arose separately. Universal common ancestry wins by a margin with two thousand zeros in it.

[07:05] S13 · ANIM · The three conditions again; then a fast montage of what they built: a bacterium, a fern, a beetle, a whale, an eye, a brain.

So: three conditions, one filter, no plan, and four billion years. Mutation supplies the raw material blindly. Selection edits it relentlessly. And drift, chance, and history make sure that what you get is not the best possible answer but a workable one, built out of whatever was lying around.

[07:40] S14 · AI into ANIM · An E. coli colony under a microscope; pull back to the calendar.

Next lesson, the first time life changed the whole planet, and very nearly ended itself doing it.

Shot list

One AI shot, twelve code-rendered shots, one research image. Estimated Higgsfield use: 2 to 4 generations including retries.