Biology 101 · Chapter 6

Evolution by natural selection

The answer is the most consequential idea in all of biology, and it's almost embarrassingly simple. Before the drug, the colony already varied — billions of cells, a few carrying a chance mutation that happened to blunt the antibiotic. The drug didn't create resistance; it filtered for it, wiping out the susceptible and leaving the rare resistant few. Those few reproduced, and within days the colony was rebuilt almost entirely from their resistant descendants. No foresight, no striving — just a filter applied to variation, then copied forward.

Three ingredients, and that's all it takes

The whole of Darwin's mechanism reduces to three conditions, each of which you've already met earlier in this track:

• Variation — individuals differ. The source is mutation: copying errors in DNA (Chapter 2) introduced during replication (Chapter 5). Most are neutral or harmful; occasionally one helps.
• Heredity — offspring resemble parents, because DNA is faithfully copied and passed on. The variation has to be transmissible to count.
• Differential reproduction — some variants leave more surviving offspring than others. The environment does the choosing; nothing intends it.

Run that loop across generations and the population's makeup shifts toward whatever reproduces best. Crucially, the filter is cumulative: each generation builds on the gains of the last, which is how blind tweaking, given enough time, assembles structures as intricate as an eye.

Random and not-random, at the same time

The single most common misunderstanding is "evolution is just chance". Half right. Mutation — the source of new variation — is random: errors fall where they fall, blind to what the organism needs. But selection is the very opposite of random: it's a consistent, directional filter that systematically keeps what works and discards what doesn't. Evolution is randomness fed through a relentless sieve. That combination is why it can do something pure chance never could — accumulate.

One tree, one code

Wind the loop back far enough and lineages merge. All living things appear to descend from a single common ancestor — and the smoking gun is sitting in Chapter 2. Every organism, from a bacterium to a blue whale, uses essentially the same genetic code: the same codons spell the same amino acids. That near-universal code is overwhelming evidence of shared ancestry — like finding the same arbitrary encoding scheme in every program ever written. Add the fossil record, the nested similarities in anatomy, and evolution we watch happen in real time (resistant bacteria, drug-resistant viruses, finch beaks shifting in a single drought), and common descent is about as settled as science gets.

Work one, then finish one

Worked: Selection is fast because advantage compounds. Suppose a beneficial variant starts as 1 in 1,000 (0.1%) of a population and, thanks to its edge, roughly doubles its share each generation. Its fraction is 0.1% × 2ⁿ after n generations. To reach about half the population we need 2ⁿ ≈ 500, and since 2⁹ = 512, that's only about 9 generations. For fast breeders like bacteria, that can be a single day. (Real selection slows near fixation — this is a deliberately simple model to show the order of magnitude.)

Your turn: Same setup. Roughly how many generations to reach about 25% of the population? (Answer: 0.1% × 2ⁿ ≈ 25% means 2ⁿ ≈ 250, so n ≈ 8 generations.)

Why this earns a place in your toolkit

Natural selection is substrate-neutral: it doesn't care whether the things copying are cells or code. That's exactly why it became an algorithm. Genetic algorithms and evolutionary computation breed solutions by mutating candidates, scoring them against a goal, and copying the winners — used to design antennas, schedules, and even neural-network architectures (neuroevolution). The same logic underpins population-based training in modern ML. And evolution is the lens for everything else in this track: DNA, proteins, mitochondria, the cell itself are evolved, not engineered — tinkered together by a blind editor, which is why biology is so robust and so gloriously messy. Reading it backwards, evolution in real time is now a forecasting problem: phylogenetics plus machine learning is how we track which flu and COVID variants will spread, and update vaccines before they do.