What is natural selection: definitions, boundaries, and conceptual frameworks that define the entire debate
Natural selection is traditionally defined as the process by which organisms with traits better adapted to their environment survive and reproduce more successfully, passing these traits to offspring. This definition traces back to Darwin and remains the foundation of evolutionary biology (S013).
The philosophical problem arises when clarifying ontological status: is selection a causal mechanism that actively produces changes, or a descriptive phenomenon capturing statistical regularities at the population level?
- Mechanism
- A system of entities and activities organized to produce regular changes (S014, S017). For selection: individuals, populations, genes interact through variation, inheritance, differential reproduction, creating a predictable outcome—change in trait frequencies.
- Phenomenon
- An observable pattern of change in trait frequencies within populations, separated from the underlying mechanisms (differential survival, reproductive success, genetic inheritance) (S012).
D. Benjamin Barros (2008) developed the concept of stochastic mechanisms, recognizing that evolutionary processes include probabilistic elements but can still be considered mechanisms (S018).
Wei (2024) proposes distinguishing selection as a phenomenon from its underlying mechanisms. According to this position, calling selection itself a mechanism is a category error: we observe the result (population change) but attribute causal power to an abstraction rather than concrete biological processes. More details in the Thermodynamics section.
This distinction is not merely semantic—it affects how researchers formulate hypotheses and interpret experimental data.
Population as entity: the central point of disagreement
The key dispute concerns the role of populations. Skipper and Millstein (2005), along with Barros (2008), argued that populations should be considered one of the constitutive entities of the natural selection mechanism (S004).
| Position | Argument | Consequence |
|---|---|---|
| Population as entity | Populations have properties (genetic structure, demography) that actively participate in evolutionary dynamics | Selection is a population-level mechanism |
| Population as context | Population is not an entity but rather a level of description at which selection effects manifest | Selection is a phenomenon; mechanisms lie at the individual level |
Pérez-González (2025) defends the first position, that populations are indeed biological entities (S001, S003). This debate reflects a deeper problem: how individual and population levels of explanation relate in biology.
Related materials: irreducible complexity and intelligent design, creationism versus evolution.
Steel Version of the Arguments: Seven Strongest Cases for the Mechanistic Interpretation of Natural Selection
Before critiquing the mechanistic interpretation, it's necessary to present it in its most convincing form. Defenders of this position advance several powerful arguments that cannot be ignored (S001).
🔬 The Argument from Experimental Manipulation: We Can Intervene in Selection
If natural selection is merely a descriptive phenomenon, how do we explain that biologists can experimentally manipulate its components and predict outcomes? Artificial selection in breeding, experiments with bacterial populations under controlled conditions, studies of adaptation to new environments—all demonstrate that we can intervene in causal processes, not simply observe statistical patterns. More details in the Climate and Geology section.
Mechanisms by definition permit such intervention: we alter components (selection intensity, sources of variation) and observe predictable changes in outcomes.
- Artificial selection in breeding of crops and livestock
- Controlled experiments with bacterial populations
- Studies of organism adaptation to new habitats
- Manipulation of selective pressure intensity under laboratory conditions
🧬 The Argument from Integration with Molecular Biology: Genes as Real Entities
Modern evolutionary biology is tightly integrated with genetics and molecular biology. We identify specific genes undergoing selection, track changes in their frequencies, link genotypes with phenotypes and fitness (S001).
This integration works precisely because natural selection is treated as a mechanism operating through concrete biological entities (alleles, genotypes, phenotypes), not as an abstract statistical phenomenon. If selection were only a phenomenon, such integration would be conceptually problematic.
📊 The Argument from Predictive Power: Mechanistic Models Work
Population genetic models based on mechanistic understanding of selection possess impressive predictive power. Fisher's equations, Wright's models, quantitative genetics theory—all treat selection as a causal process with identifiable parameters (selection coefficients, fitness, heritability).
These models successfully predict evolutionary trajectories in natural and laboratory populations. Fisher's fundamental theorem works precisely within the mechanistic framework (S001).
🧠 The Argument from Causal Explanation: Selection Answers "Why?"
Biologists use natural selection for causal explanations: why do giraffes have long necks? Why do bacteria develop antibiotic resistance? Why do peacocks have bright tails?
The answers appeal to selection as a cause, not simply as a description of a pattern. If selection is only a phenomenon, these explanations lose causal force and become mere redescriptions of observed changes. The mechanistic interpretation preserves the explanatory power of evolutionary theory (S001).
⚙️ The Argument from Stochastic Mechanisms: Probability Doesn't Exclude Causality
Critics sometimes point to the stochastic nature of evolutionary processes as a problem for mechanistic interpretation. However, probabilistic processes can be mechanisms if they involve identifiable entities and activities producing regular (though probabilistic) outcomes (S018).
Quantum mechanics, radioactive decay, many biological processes—all are stochastic, yet no one denies their mechanistic nature.
🔁 The Argument from Multilevel Selection: Mechanisms at Different Levels of Organization
Multilevel selection theory shows that selection can operate on genes, organisms, groups, even species. This hierarchical structure fits naturally into a mechanistic framework: at each level there are entities (genes, individuals, groups) and activities (replication, reproduction, extinction) organized into selection mechanisms (S004).
The phenomenological interpretation struggles to explain how the same "phenomenon" can manifest at such different levels of biological organization.
🧰 The Argument from Research Practice: Biologists Think Mechanistically
Analysis of actual research practice shows that evolutionary biologists formulate hypotheses, design experiments, and interpret results while implicitly assuming the mechanistic nature of selection. They search for "mechanisms of adaptation," "selective pressures," "sources of variation"—all these terms reflect mechanistic thinking (S001).
If selection were simply a phenomenon, research practice would look completely different. The connection between theory and practice indicates that mechanistic interpretation is not just philosophical convenience, but a reflection of how biologists actually understand evolution.
Evidence Base: Detailed Analysis of Empirical Data and Philosophical Arguments from 2024-2025 Sources
Systematic analysis of evidence from recent publications shows that the debate about the nature of natural selection rests on incompatible interpretations of the same data. Wei (2024) and Pérez-González (2025) provide material for testing both positions. More details in the Physics and Meta-Analysis section.
📊 Wei's Position: Selection as a Phenomenon Requiring Mechanistic Explanation
Wei (2024) distinguishes between explanandum (what requires explanation) and explanans (what explains). Natural selection is a phenomenon: an observable pattern of change in trait frequencies in populations (S012). Mechanisms—specific biological processes (differential survival, reproductive success, genetic inheritance)—produce this phenomenon.
Wei's central problem: if a population is an entity in a mechanism, what activities does it perform? Populations don't "do" anything in the sense that organisms reproduce or genes replicate (S004). Populations are a level of description at which the effects of individual processes are observed.
🧪 Pérez-González's Counterarguments: Populations as Real Biological Entities
Pérez-González (2025) argues that populations have causal properties: genetic structure (allele frequencies, heterozygosity), demographic characteristics (size, age structure, growth rates), ecological relationships (competition, predation, symbiosis) (S001). These properties are emergent and not reducible to the sum of individual properties.
Fitness only makes sense at the population level. Individual fitness is not an intrinsic property of an organism, but a relative measure of reproductive success in the context of a population and environment (S001). This makes the population a constitutive part of the mechanism, not merely a context.
If fitness is a statistical quantity, how can it be part of a causal mechanism? Answer: statistical quantities are causally relevant if they reflect real differences in biological properties.
🧾 Statistical Interpretation of Fitness
Fulda (2017) analyzes the tension between mechanistic and statistical interpretations (S014). The statistical interpretation treats fitness as a population parameter describing average reproductive success, rather than as a causal property of individuals.
Reconciliation is possible: statistical quantities are causally relevant if based on real differences in biological properties. Average genotype fitness is a statistic, but it's based on causal differences in survival and fecundity that have mechanistic explanations (physiology, behavior, morphology) (S014). Statistical and mechanistic interpretations complement each other at different levels of analysis.
🔎 Fisher's Fundamental Theorem: Mechanism Through Mathematics
Okasha (2008) showed that Fisher's fundamental theorem—the rate of increase in mean fitness equals the genetic variance in fitness—has deep mechanistic significance (S010). It connects population-level changes (phenomenon) with genetic variation and inheritance (mechanisms).
The theorem works only under certain assumptions: absence of mutation, migration, random drift; additive genetic variance. This emphasizes that statistical patterns in evolution depend on specific biological mechanisms, rather than being autonomous (S010).
| Interpretation | What It Explains | Level of Analysis | Problem |
|---|---|---|---|
| Phenomenological (Wei) | Patterns of change in trait frequencies | Population | Doesn't explain how populations produce effects |
| Mechanistic (Pérez-González) | Specific biological processes | Organismal + population | Requires defining population causality |
| Statistical (Fulda) | Mathematical regularities | Abstract | Connection between statistics and causation |
🧬 Integration with Developmental Biology: Adaptive Developmental Bias
Natural selection creates adaptive developmental bias—the tendency of developmental systems to produce certain phenotypic variations more frequently than others (S008). This blurs the classical Darwinian distinction between random variation and directed selection: selection shapes not only the distribution of existing variants, but also the probabilities of new ones appearing.
This integration supports the mechanistic interpretation: selection acts through specific biological processes (genetic networks, epigenetic mechanisms, developmental processes) that can be studied experimentally (S008). The phenomenological interpretation struggles to explain such integration.
⚙️ Self-Organization and Natural Selection: Interaction of Mechanisms
Batten et al. propose the formula: "self-organization proposes, natural selection disposes" (S007). Self-organization creates structures through physicochemical processes independent of selection. Selection then "chooses" among these patterns those that increase fitness.
This shows that natural selection doesn't explain all evolutionary change. Some patterns arise through self-organization. However, this doesn't refute the mechanistic interpretation; rather, it shows that evolution involves multiple mechanisms interacting with each other (S007). The phenomenological interpretation struggles to explain such interactions.
- Selection is not the only evolutionary mechanism; self-organization creates initial variations.
- The mechanistic interpretation explains how selection interacts with other processes.
- The phenomenological interpretation remains at the level of describing patterns without explaining their origin.
- The statistical interpretation connects population parameters with biological causes.
Related materials: irreducible complexity and intelligent design, Lamarckism and epigenetics.
Mechanism or Correlation: Distinguishing Causality from Statistical Association in Evolutionary Processes
The central problem in debates about the nature of natural selection is distinguishing causality from correlation. More details in the section Thinking Tools.
🔁 Criteria for Causality in Biology: Manipulation, Mechanism, Counterfactuals
Philosophers of science identify several criteria for establishing causality: (1) manipulative criterion — can we change the proposed cause and observe a change in the effect? (2) mechanistic criterion — can we identify the physical process linking cause and effect? (3) counterfactual criterion — would the effect change if the cause were different? (S008).
Natural selection satisfies all three criteria: we can experimentally manipulate components of selection (changing environment, sources of variation, selection intensity) and observe predictable changes. We identify specific biological processes (differential survival, reproduction, inheritance). Counterfactual scenarios show that selection is necessary for observed evolutionary changes.
If selection is merely a phenomenon (an observed pattern of change), how do we distinguish it from other phenomena? The mechanistic interpretation offers a clear answer: identify specific causal processes and distinguish selection mechanisms from drift or migration mechanisms.
🧩 Confounders in Evolutionary Research: Drift, Migration, Mutation
Evolutionary changes can occur not only through selection. Genetic drift (random changes in allele frequencies), migration (gene flow between populations), mutation (emergence of new variants) — all are confounders that create patterns resembling selection outcomes (S008).
Distinguishing these processes requires careful experimental design and statistical analysis. This creates a problem for phenomenological interpretation: how do we distinguish selection from other phenomena if they all produce visible changes in populations?
| Process | Mechanism | Statistical Pattern | Distinction from Selection |
|---|---|---|---|
| Natural selection | Differential survival and reproduction | Systematic change in adaptive allele frequencies | Directional, predictable by phenotype |
| Genetic drift | Random fluctuations in small populations | Random walk of frequencies, loss of variation | Non-directional, independent of fitness |
| Migration | Gene flow between populations | Equalization of allele frequencies between groups | Homogenizing, unrelated to local environment |
| Mutation | Emergence of new genetic variants | Low frequencies of new alleles, random distribution | Source of variation, not factor in frequency change |
📊 Statistical Signatures of Selection: How to Recognize Its Action
Population geneticists have developed numerous statistical methods for detecting signatures of selection in genomic data: tests for excess rare alleles, linkage disequilibrium patterns, ratios of synonymous to non-synonymous substitutions. These methods are based on predictions from mechanistic models of selection: selection creates specific patterns of genetic variation distinct from neutral evolution (S008).
The success of these methods supports the mechanistic interpretation: we can predict what statistical patterns selection will create because we understand it as a causal mechanism. If selection were merely a phenomenon without mechanistic content, such predictions would be impossible.
- Determine which traits vary in the population and whether they are heritable.
- Measure differences in survival and reproduction between carriers of different variants.
- Test whether these differences correlate with changes in trait frequencies in subsequent generations.
- Exclude alternative explanations (drift, migration, mutation) through statistical analysis and experimental control.
- Identify the mechanism through which the trait affects survival or reproduction (e.g., physiological, behavioral, ecological).
The connection between arguments about the complexity of biological systems and the mechanism of selection becomes clear: selection is not merely a statistical pattern, but a causal process that explains how complexity can arise without external design. Mechanistic understanding of selection allows us to predict evolutionary changes and distinguish them from other processes, which is impossible if we view selection as a phenomenon without causal content.
Conflicts and Uncertainties: Where Sources Diverge and Why Consensus Remains Elusive
A century and a half of research has not produced consensus on the nature of natural selection. Analysis of sources from 2024–2025 reveals several key points of disagreement where even authoritative voices diverge fundamentally. For more details, see the section on Logical Fallacies.
🕳️ Ontological Status of Populations: Entity, Context, or Abstraction?
The sharpest dispute concerns populations. Wei argues that populations are not entities of the selection mechanism in the strict sense—they do not perform activities, lack clear boundaries, and possess no causal powers independent of the individuals composing them (S004).
Pérez-González objects: populations have emergent properties (genetic structure, demography) that are causally relevant and not reducible to individual properties (S001).
This disagreement reflects a fundamental question in philosophy of biology: how do different levels of organization relate? Reductionists see populations as convenient abstractions; anti-reductionists insist on the causal autonomy of the population level.
🧬 Fitness: Property of Individuals or Population Parameter?
The second disagreement concerns the interpretation of fitness. Three main approaches:
- Propensity interpretation—fitness as an individual's disposition to reproduce in a specific environment;
- Statistical interpretation—fitness as the average reproductive success of a class of organisms;
- Contextual interpretation—fitness as a relationship between phenotype and environment, not existing outside a specific population and moment in time.
Each approach has implications for understanding selection causality. If fitness is a property of individuals, then selection acts at the organism level. If it's a population parameter, selection becomes a statistical phenomenon rather than a mechanism in the classical sense.
⚙️ Mechanism or Description: Can Selection Be Both?
The third conflict concerns the very definition of mechanism. Okasha (S008) proposes distinguishing between mechanism in the narrow sense (a system of components with clear causal interactions) and mechanism in the broad sense (any regular process explaining a phenomenon).
| Position | Selection as mechanism? | Consequence |
|---|---|---|
| Narrow interpretation | No—selection is a description of statistical patterns | Selection does not explain but reformulates observations |
| Broad interpretation | Yes—selection is a regular causal process | Selection has explanatory power at the population level |
| Hybrid position | Yes, but with caveats—mechanism at population level, description at individual level | Selection works as a mechanism only under certain conditions |
🔄 Causality and Selection: Does Selection Act or Only Describe?
The fourth conflict concerns causality. Some authors argue that selection is not a cause but a filter: the environment selects, organisms do not choose. Selection does not act actively but passively excludes unfit variants.
Others object: selection is a causal process where differential reproduction of organisms with different traits leads to changes in allele frequencies. Without selection, populations would not change predictably.
Paradox: if selection is only a filter, why is evolution directional? If selection is a cause, why does it not act at the individual level but only at the population level?
📊 Empirical Testability: How to Distinguish Selection from Drift?
The fifth conflict is practical. How do we distinguish natural selection from genetic drift in real populations? Both processes change allele frequencies, but selection is directional, drift is random.
Problem: in small populations drift dominates, in large ones—selection. But population size itself depends on ecological conditions that create selection. How do we separate causes?
- Criterion 1: Repeatability
- If trait change repeats in independent populations—likely selection. If random—drift. But populations are rarely independent.
- Criterion 2: Directionality
- If change is directed toward adaptation—selection. But adaptation is defined post hoc, creating circularity.
- Criterion 3: Molecular Signatures
- Selection leaves signatures in the genome (low variability in coding regions). But signatures may come from other processes.
🌍 Universality of Selection: Does the Mechanism Work Everywhere?
The sixth conflict concerns boundaries of applicability. Selection works in biology, but does it work in economics (S001), culture, technology? If yes—selection is a universal principle. If no—selection is specific to biology.
Problem: in economics and culture there are no clear analogs of genes, replication, inheritance. Applying selection to these domains may be metaphor rather than mechanism.
Consensus is unattainable because each position is logically consistent within its own premises. The choice between them is a choice of philosophical framework, not an empirical question.
Related questions about the boundaries of evolutionary theory can be found in articles on intelligent design and complexity, creationism, and Lamarckism with epigenetics.
