Evolutionary Theory Makes Predictions About Cognition Beyond Rational Optimization
Abstract
Evolution by natural selection produces the appearance of design in minds, brains, and behavior. While rational models of cognition are usually considered to be consistent with evolution, they are rarely derived from distinctively evolutionary assumptions. Bayesian modeling, reinforcement learning, rational analysis, and resource rational analysis all postulate optimization objectives of coherence in information use or maximization of expected value, potentially under resource constraints. However, cognition is performed by an individual within a lineage where success is determined by the persistence of the lineage across generations. Here, we present a signal detection model that shows how this distinctly evolutionary objective produces predictions that depart from those of rational models. We find that the source of environmental uncertainty determines cognitive design, with weak priors being favored in environments dominated by change. This ensures offspring vary so that some profit against novel challenges, and is favored even at the expense of accurate decision making. Cognitive structure should reflect stable environmental trends, but in fluctuating environments selection strips structure away to release behavioral diversity. We highlight that evolution maximizes the geometric mean number of offspring over generations, an objective that recurs across optimization problems where outcomes accumulate sequentially, including in finance, control engineering, and information theory.
Related
- FEED your mind: The evolutionary roots of human food cognition
- Improving acetate metabolism of Pseudomonas putida KT2440 by evolutionary and rational engineering
- An eco-evolutionary consumer-resource theory of host-microbe symbioses
- Eco-Evolutionary Graph Theory: Regular Graphs can act as Amplifiers
- Reinforcement learning with evolutionary priors unlocks multi-site enzyme optimization from sparse experimental feedback
BibTeX
@misc{turner2026evolutionary,
title = {{Evolutionary Theory Makes Predictions About Cognition Beyond Rational Optimization}},
author = {Cameron Turner and Thomas L. Griffiths},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/vjrkc_v2},
url = {https://osf.io/preprints/psyarxiv/vjrkc_v2/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Evolutionary Theory Makes Predictions About Cognition Beyond Rational Optimization
AU - Cameron Turner
AU - Thomas L. Griffiths
PY - 2026
DA - 2026/09/21
PB - PsyArXiv
DO - 10.31234/osf.io/vjrkc_v2
UR - https://osf.io/preprints/psyarxiv/vjrkc_v2/
AB - Evolution by natural selection produces the appearance of design in minds, brains, and behavior. While rational models of cognition are usually considered to be consistent with evolution, they are rarely derived from distinctively evolutionary assumptions. Bayesian modeling, reinforcement learning, rational analysis, and resource rational analysis all postulate optimization objectives of coherence in information use or maximization of expected value, potentially under resource constraints. However, cognition is performed by an individual within a lineage where success is determined by the persistence of the lineage across generations. Here, we present a signal detection model that shows how this distinctly evolutionary objective produces predictions that depart from those of rational models. We find that the source of environmental uncertainty determines cognitive design, with weak priors being favored in environments dominated by change. This ensures offspring vary so that some profit against novel challenges, and is favored even at the expense of accurate decision making. Cognitive structure should reflect stable environmental trends, but in fluctuating environments selection strips structure away to release behavioral diversity. We highlight that evolution maximizes the geometric mean number of offspring over generations, an objective that recurs across optimization problems where outcomes accumulate sequentially, including in finance, control engineering, and information theory.
ER -