The evolution of evolvability: how evolution learns to evolve

Abstract

It is hypothesised that one of the main reasons evolution has produced such a tremendous diversity of amazing designs is because evolution has improved its own ability to innovate, a process called the ‘evolution of evolvability’. Rupert Riedl, an early pioneer of evolutionary developmental biology, suggested that evolvability is facilitated by a specific developmental organisation that is itself a product of past selection. However, the construction of a theoretical framework to formalise such ‘evolution of evolvability’ has been continually frustrated by the indisputable fact that natural selection cannot favour structures for benefits they have not yet produced. Here we resolve this seeming paradox. Recent work shows that short-term selective pressures on gene interactions are functionally equivalent to a simple type of associative learning, well-understood in neural network research. This is important for the evolution of evolvability because this
type of learning system can clearly change in a way that improves its performance on unseen, future test cases, without the need for the future to cause the past. Recognising a formal link with the conditions that enable such predictive generalisation in machine learning systems unlocks well-established theory that can be applied to understanding the evolution of evolvability. Here we use this to elucidate, and demonstrate for the first time, conditions where short-term selective pressures alter evolutionary trajectories in a manner that systematically improves long-term evolutionary outcomes.

More information

Identifiers

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

Catalogue record

Export record