Reusable cognitive primitives
Variation & selection
When many varying candidates are filtered by a criterion and the survivors are copied, the population adapts to the criterion over rounds — with no designer.
- Charles Darwin — Formulated · 1859
- Alfred Russel Wallace — Independently discovered · 1858
Selection as a mechanism was set out by Darwin and, independently, by Wallace; their papers were read jointly at the Linnean Society in 1858.
Sources: The 1858 Darwin–Wallace paper (Linnean Society of London), Encyclopaedia Britannica
Where it appears
Dependencies & synergies
Derived from the graph’s real structure — 13 concepts across 14 disciplines carry this pattern. Every figure is a count, not a score.
How many concepts carrying this pattern are seen through each discipline.
Other thinking patterns that recur on the same concepts — the more shared concepts, the more often they co-occur.
What kinds of relationships the carrying concepts form — the relational signature of the pattern.
Concepts where this pattern does the most cross-disciplinary work — each reaches disciplines beyond its own.
- Natural selectionreachesAnthropologyArtificial IntelligenceAstrobiologyBioinformaticsBiotechnologyCell BiologyComparative LiteratureComputer ScienceDemographyEarth & Space SciencesEcologyGeneticsGeologyMachine LearningMedicineMicrobiologyMolecular BiologyPaleontologyPublic HealthSociologyStatistics
- Immune systemreachesBiomedical ScienceBiotechnologyEpidemiologyImmunologyMedicineMicrobiologyPediatricsPublic Health
- Gradient descentreachesBusinessEconomicsEngineeringMathematical ModellingStatisticsSystems Engineering
- Antimicrobial resistancereachesBiologyEcologyEnvironmental ScienceEvolutionary BiologyVeterinary Medicine
- SpeciationreachesEcologyGeneticsGeologyPaleontology
- ResistomereachesEcologyGenetics
Carrying concepts that others in the same pattern build on (they depend on or follow from these).
- Fitness — 1 other concept build on it
- Natural selection — 1 other concept build on it
Examples across disciplines
Natural selection: fitter variants leave more offspring.
Markets select firms and products that customers reward.
Training keeps the parameter changes that reduce error.
The immune system amplifies antibodies that bind a pathogen.
How this pattern travels
Interpreted from where the pattern recurs in the atlas — structural transfer and coverage, not a claim it is universally the "best" model.
Recurs across 14 disciplines over 13 carrier concepts — concentrated (69% of carriers in one field).
structural · Structural recurrence in the atlas — a pattern is a reasoning lens, not a law.
5 of its 13 carrier concepts are themselves cross-disciplinary connectors.
structural · Structural recurrence in the atlas — a pattern is a reasoning lens, not a law.
Explicit "where it breaks" notes exist for 10 of 13 annotated assignments.
curated · Curated boundary annotations — absence is a representation gap, not evidence the model has no limits.
Coverage matrix
How these concepts distribute across domains and concept families — real counts, not a score.
Knowledge timeline
Real, sourced key dates of these concepts.
- 1858Natural selection · Discovery · Evolutionary Biology
- 1859Natural selection · Publication · Evolutionary Biology
- 1866Mendelian inheritance · Publication · Biology
The statistical fingerprint
How the 13 concepts that exercise this pattern distribute — from the graph, not a ranking.
Disciplinary fingerprint
Carrier concepts under each illuminating lens.
How settled its carriers are
Epistemic status of the concepts that exercise this pattern.
Where the model breaks
This pattern is a reasoning lens, not a law. Here is where it stops helping:
- Adaptation (mechanism): Selection can only filter variation that already exists — it cannot create the trait it "needs"; drift, mutation, development and constraint co-author every adaptation.
- Antimicrobial resistance (mechanism): Selection enriches pre-existing resistant variants; it does not induce resistance on demand — dosing and horizontal gene transfer change the picture the simple model omits.
- Fitness (reasoning tool): Fitness is realised reproduction in a context, not intrinsic "betterness"; used to predict rather than describe, it becomes circular.
- Gradient descent (analogy): Gradient descent follows an explicit objective's gradient; biological selection has neither an objective nor a gradient. The correspondence is structural (variation + differential retention), not literal.
- Immune system (mechanism): Clonal selection is genuine selection, but the analogy to evolution stops at heredity: immune memory is somatic and dies with the individual.
- Mutation (mechanism): Mutation is the variation source, not the selective step itself — it is random with respect to fitness.
- Natural selection (mechanism): Selection needs heritable variation and differential reproduction; it has no foresight or goal and does not act 'for the good of the species'.
- Reinforcement learning (analogy): RL optimises expected reward for one agent within its lifetime; selection acts across generations with no designer. The shared idea is differential reinforcement, not identity.
- Selective pressure (mechanism): A pressure only selects among heritable, varying, differentially-reproducing traits — absent any of those three conditions, nothing is selected.
- Speciation (mechanism): Selection is one route to speciation; geographic isolation and drift can split lineages with little or no selective difference.
7 of 13 explained assignments cite a source; the rest are editorial interpretations. 2 are flagged for review. None is externally validated.
✕ Selection is forward-looking design ('evolved in order to…'). It only ever filters existing variation after the fact.
You have seen this model in one place. Where else could it apply — and where would the analogy break?
Adaptation →Look for three ingredients: variation, a selection pressure, and inheritance/copying. Together they design without a designer.
- Recurs across 14 disciplines.
- 33 concepts exercise this thinking pattern.
- 10 of them explicitly note where the model breaks down.
Derived from the graph — a pattern is a reasoning lens, not a law.
Are all three present — variation among candidates, a pressure that filters them, and copying of the survivors — so the population adapts with no designer?
Keep this question handy when you meet something new — it helps you notice the pattern, not just name it.