Entangled Reality
Biochemistry
Biochemistry/ Evolution/ How Selection Works
Evolution / Foundations

How Selection Works

Natural selection is a *filter*, not a generator. It makes some variants more common than others once a system can reproduce, inherit, and vary. This page explains the minimal logic of selection — and why selection cannot “start the story” until earlier constraints are solved.

Variation Inheritance Differential reproduction Fitness Drift Constraints
Reading mode

Choose your level

This page is written in layers. Choose how technical and how long you want the explanations to be. (Nothing is tracked; this only changes what your browser shows.)

Tip: the expandable sections below still work even if you ignore the dropdowns.

1) The minimal algorithm

Selection requires three things: variation (different versions), inheritance (offspring resemble parents), and different success (some versions leave more offspring). When that happens, the “better” versions become more common.

Selection is a statistical rule: if heritable variants differ in reproductive success, the composition of the population shifts. Over time, lineages that replicate more effectively (given constraints) become more common.

Key point: selection does not “aim” at improvements. It *retains* whatever happens to work better in the current environment.

In population-genetic terms, selection changes allele frequencies when there is heritable variance in fitness. A useful mental model is: replication + heritable variation + differential replication. Under many conditions, selection dominates drift when effective population size is large and selection coefficients exceed ~1/Ne; otherwise drift can swamp selection.

(We’ll keep math light here unless you want a separate appendix page.)

2) What “fitness” actually means

Fitness is not strength — it is reproductive success

“Fitness” just means leaving more offspring. A trait can be “fit” even if it looks weak, as long as it helps reproduction in that environment.

Fitness is relative and context-dependent. A variant can be favored in one setting and disfavored in another. Selection therefore tracks environments — it does not produce universal “best” solutions.

Also: selection can favor *tradeoffs* (e.g., speed vs accuracy, robustness vs efficiency).

Fitness can be defined in multiple ways (absolute vs relative; individual vs genotype; short-term vs long-term). In real systems, fitness is shaped by interactions, frequency dependence, and changing environments. Even “simple” selection often operates on coupled traits.

Selection can optimize locally, not globally

Selection tends to improve what’s nearby — it can get stuck on “good enough” solutions.

Selection works by incremental retention. If intermediate steps are not beneficial (or at least not harmful), the path is blocked. This is why “stepwise selectability” is a real constraint, especially for systems where multiple components must cohere.

Fitness landscapes can have local maxima and ruggedness. Recombination, drift, and changing environments can help escape local maxima, but none of these mechanisms remove the basic requirement that populations must remain viable at every step.

3) What selection can and cannot explain (in origins discussions)

What selection is great at
  • Refining traits once reproduction exists.
  • Adapting populations to environments.
  • Building complexity gradually from working parts.
  • Refinement: turning “barely works” into “works well.”
  • Reuse (co-option): repurposing existing components for new functions.
  • Robustness: stabilizing systems against noise, drift, and damage.

Most of evolutionary biology is about what happens after a robust reproductive system already exists.

Selection is a powerful search process when the search space is navigable by viable intermediates, heredity is stable, and populations are large enough for selection to dominate drift. Under those conditions, cumulative selection can produce high degrees of adaptation.

What selection cannot do first

Selection can’t operate without reproduction. If nothing copies itself, there is nothing to select.

In origins debates, selection is often invoked too early. Before you have stable heredity and a way to build functional catalysts reliably, selection has nothing persistent to work on. That’s why the “Phase 2” translation system is a bottleneck: it’s the mechanism that converts code into enzymes at scale.

Ordering constraint: selection explains diversification (Phase 3), but it presupposes a functional reproduction/translation regime (Phase 2).

If heredity is too noisy, you get an error threshold problem: beneficial variants cannot be retained because copying errors erase them. If genotype-to-phenotype mapping is unstable, selection can’t “lock in” improvements. These are not optional details; they are the preconditions for cumulative selection to do work.

4) A compact mental model

Selection is like keeping the best recipes — but only if you can reliably copy and cook them.

Selection is a filter operating on outputs, not a designer of inputs. It improves what reproduction can already produce. When people “explain origins by selection,” the hidden assumption is that reproduction + mapping already exist.

Think in layers: (i) chemistry provides components; (ii) translation/reproduction provides stable heredity + functional expression; (iii) selection refines and diversifies once (ii) is in place. Confusions in origins arguments often come from sliding between (ii) and (iii) without noticing.

Continue

If you want the “selection makes the improbable probable” argument, go to cumulative selection next — and then return to the constraint question about whether selection can build translation machinery at all.