The Evolution of Selection: From Simple Rules to Cognition This paper develops a unified framework for understanding selection as a family of distinct but interacting processes rather than a single, monolithic force. Selection takes different forms — physical, algorithmic, and cognitive — that arose at different stages in the history of life and continue to operate together in modern organisms. Physical forces act directly on bodies. Algorithmic processes shape outcomes through biochemical, physiological, and metabolic computation within organisms. Cognitive processes introduce evaluative decision-making, allowing repeated choices to bias survival, reproduction, and parental investment. These forms of selection are not competing explanations but layered modes that operate simultaneously, and each higher-order mode is built upon the physical implementation established by the modes that came before. A central claim of this paper is that most selection mechanisms are themselves evolved biological traits, not external forces acting on biology from outside. The machinery of algorithmic and cognitive selection — regulatory networks, immune systems, nervous systems making evaluative decisions — is biology doing biological work. Like any other heritable feature of organisms, this machinery varies, imposes costs, and confers differential reproductive consequences on its bearers. It therefore evolves under selection through the same processes of variation, heritability, and differential reproduction that govern any other trait. This makes "the evolution of selection" a literal claim rather than a figurative one. As organisms evolved larger body size, longer generation times, and smaller population sizes, the diversification of selection mechanisms from physical to algorithmic to cognitive forms represents an adaptive evolutionary response to the constraints these life-history changes imposed. To distinguish these forms of selection precisely, the paper offers a formal definition. Selection is the process by which traits or states of one biological system bias the actions of another — or of itself — leading to differentiated effects on the survival, reproduction, or reproductive investment of some organism, accumulated over generational time. This definition keeps three roles structurally distinct: the system whose traits or states structure incentives, the system whose actions are biased, and the organism whose evolutionary persistence is affected. These roles often coincide within a single organism, but distinguishing them prevents the conflation of trait, mechanism, and outcome that conventional formulations permit. The classification of any given selection event follows the upstream causal architecture rather than the proximate physical instrument: lead poisoning from contaminated groundwater is physical selection, while a death from a lead bullet fired in a gunshot is cognitive selection, even though the proximate biochemical injury — lead entering the body — is similar in both cases. Within this framework, sexual selection emerges not as a single force but as a structured system of interacting selection forces, including phenotype-preserving selection, incest avoidance selection, uniqueness selection, ornament exaggeration selection, vigor selection, mimicry selection, coercive sexual selection, intelligence selection, growth termination selection, senescence selection, and parental investment selection. A central problem these forces collectively address is the informational limitation of mate choice. Mate selection occurs at a point in time and cannot directly evaluate many traits that are expressed intermittently, conditionally, or only under rare ecological stress. Phenotype-preserving selection biases reproduction toward self-similar mates, stabilizing phenotype across generations, while uniqueness selection recruits low-frequency heritable traits as reliable markers of shared ecological and evolutionary history. Together they probabilistically preserve latent genetic functions that may not be expressed at the time of mate choice. Ornament exaggeration selection then amplifies discriminability along trait dimensions that uniqueness selection has already recruited, without requiring any change in the chooser's evaluative architecture. Empirical evidence from artificial ornament experiments in birds (Burley, 1981, 1986; Burley et al., 1982) and from the extreme diversification of genital morphology across taxa (Eberhard, 1985, 1996, 2010) demonstrates that novel or arbitrary traits can immediately bias mate choice and then undergo progressive exaggeration. A critical structural argument is that asymmetric mate choice — the fact that choosers evaluate across the entire available pool while candidates compete for individual acceptance — is the mathematical origin of sexual selection's amplifying power, and is fundamentally incompatible with classical assortative mating theory. Under rank-ordered pairing, reproductive success distributes in proportion to pre-existing competitive rank, and selection intensity cannot exceed what rank differences already specify. Under asymmetric evaluative choice, multiple choosers independently applying their criteria to the same pool can concentrate reproductive success in specific candidates far beyond what rank-ordered pairing permits, generating the variance in reproductive success that drives rapid adaptive change. The paper further shows that the named sexual selection forces are not atomic but are composed of combinations of five proto-selection operations — template matching, frequency evaluation, performance integration, temporal projection, and social observation — which combine both simultaneously within individual mate choice decisions and serially across evolutionary time as one force creates the conditions under which the next can operate. This framework supplies a missing mechanism for one of evolutionary biology's longest-standing empirical patterns. The fossil record across many taxa shows extended periods of phenotypic stasis interrupted by relatively rapid speciation events — the pattern formalized as punctuated equilibrium (Eldredge Gould & Eldredge, 1977, 1993). Population-genetic accounts have not adequately explained why stasis is so prolonged in the face of continuous mutation and gene flow. This framework proposes that stasis is not the default outcome of weak selection but the active result of integrated cognitive selection forces preserving lineage identity at high resolution across generations. Punctuation occurs when these stasis-maintaining mechanisms are weakened by reduced population size, isolation, or ecological disruption — and the framework predicts that resulting divergence should be sharply non-random, concentrated in sexually evaluable traits while traits under direct viability selection remain conserved. This is a differential prediction not derivable from drift-based accounts. Cognitive selection itself enables a fundamental shift in the relationship between selection and time. Algorithmic selection, despite its sophistication, operates through stimulus-response computation: the system responds to conditions as they currently exist. The most sophisticated forms of cognitive selection enable forecasting of future states, allowing organisms to act in anticipation of conditions rather than only in response to them. Intelligence selection creates recursive feedback in which cognition selects for itself, progressively increasing the dimensionality of selective discrimination within the constraint of metabolic cost. By treating selection as biased action rather than filtering, by making explicit the mechanisms through which bias is implemented, and by recognizing that selection mechanisms are themselves evolved biological traits, this framework integrates physical, algorithmic, and cognitive forms of selection into a coherent structure. It clarifies how selection mechanisms evolve to enhance adaptive capacity, how sexual selection can simultaneously preserve lineage identity and generate diversity, how stasis can be active rather than passive, and how cognitive selection enables forecasting that fundamentally transcends stimulus-response computation. The framework offers a foundation for understanding not only how traits evolve, but how the mechanisms of selection themselves diversify and elaborate over evolutionary time.
Kevin Brown (Fri,) studied this question.