Introduction

There is no physical mechanism for translating knowledge acquired during a single lifetime to thegenetic code so that it will be available for genetic propagation via selection, crossover, and mutation. Indeed, the conception of a genotype composed of encoded learned behavior (a conception referred to as Lamarckian evolution) has been discredited in population genetics for nearly 100 years. There is, however, a hypothesis that is consistent with both the denial of Lamarckian evolution and the belief that learned behavior somehow influences the course of evolution. This hypothesis is called the Baldwin effect, after the nineteenth century geneticist Baldwin [2], and it remains a controversial topic even today. The Baldwin effect states that there are phenotypic tendencies rewarded in organisms that learn a certain skill, and that these rewards serve to change the criteria for fitness. Therefore, even though learning cannot directly affect the underlying genotype of an organism, the genetic makeup of the organisms that did the learning will in effect be rewarded. Individuals with these genes will, therefore, be favored for further evolution.

The Baldwin effect has been shown to occur in experiments with artificial life by Hinton and Nowlan [5] and Ackley and Littman [1]. However, showing that it occurs is only the first step in understanding how it operates. Mitchell [8] identifies the study of the Baldwin effect as an important future direction for research in evolutionary programming [8, page 183]. Likewise, Levy [7], Whitley [12], and Fogel [3] indicate that understanding the mechanics of the Baldwin effect should shed light on the importance (if any) of learning in evolution -- artificial or otherwise.