Abstract

For 100 years it has been recognized that interactions between learning and evolution, such as the Baldwin effect, can be subtle and often counterintuitive. Recently a new effect has been discussed: It is suggested that evolutionary progress toward one specific goal may be assisted by lifetime learning on a different task that may or may not be “uncorrelated.” The phenomenon is reproduced here in a simple scenario, where the tasks are indeed uncorrelated—‘Another New Factor’ does indeed exist. The effect is then explained as being due to recovery from weight perturbations caused by mutation in a neural network. It is a special case of a recently discovered relearning effect: the spontaneous recovery of perturbed associations by learning uncorrelated tasks.

This content is only available as a PDF.