How evolution learns to evolve: principles of induction in the evolution of adaptive potential

Abstract

Explaining how organisms can exhibit suitable phenotypic variation to rapidly adapt to novel environmental conditions is central in evolutionary biology. Although such variability is crucial for the survival of a lineage and its adaptive potential, it remains poorly understood. Recent theory suggests that organisms can evolve designs that help them generate novel features that are more likely to be beneficial. This is possible when the environments that the organisms are exposed to share common regularities. Selection though cannot favour phenotypes for fitness benefits that have not yet been realised. Such capacity implies that natural selection has a form of foresight, which is inconsistent with the existing evolutionary theory. It is unclear why selection would favour flexible biological structures in the present environments that promote beneficial phenotypic variants in the future, previously unseen environments.

In this thesis, I demonstrate how organisms can systematically evolve designs that enhance their evolutionary potential for future adaptation relying on insights from learning theory. I investigate how organisms can predispose the production of useful phenotypic variation that helps them cope with environmental variability within and across generations, either through genetic mutation or environmental induction. I find that such adaptive capacity can arise as an epiphenomenon of past selection towards target optima in different selective environments without a need for a direct or lineage selection. Specifically, I resolve the tension between canalisation of past selected targets and anticipation of future environments by recognising that induction in learning systems merely requires the ability to represent structural regularities in previously seen situations that are also true in the yet-unseen ones. In learning systems, such generalisation ability is neither mysterious, nor taken for granted. Understanding the evolution of developmental biases as a form of model learning and adaptive plasticity as task learning can provide valuable insights into the mechanistic nature of the evolution of adaptive potential and the evolutionary conditions promoting it.

More information

Identifiers

ORCID for Richard Watson: orcid.org/0000-0002-2521-8255

Catalogue record

Export record