Remembering the Future

In order to be tractable, the number of genes required to specify a phenotype must be orders of magnitude less than the number of structural units composing that phenotype. Nature has shown such representational systems to be possible on an enormous scale. Even with 100 trillion neural connections in the human brain, there are only about 30 thousand active genes in the human genome (2800 million amino acids) [19, 23, 42, 89]. Such representational efficiency is made possible through gene reuse. In an indirect genetic encoding, a single gene may be used multiple times at different stages of development. There are two primary forms of reuse.

First, phenotypic structures can occur in repeating patterns, where the same structural theme, perhaps with some variation, appears over and over again. Each time a pattern repeats, the same gene group can provide the specification. Examples of repeating patterns in biological organisms include the numerous left-right symmetries of vertebrates [65: 302–303], and the numerous receptive fields in the visual cortex [29, 40]. Repetition frequently involves variation on a general theme. For example, each vertebra in the spine is formed similarly to the others, albeit with different incoming and outgoing connections [89: 30–31].

The second primary form of reuse occurs when the same gene product is used to initiate separate developmental pathways. For example, Cohn et al. [17] found that the same gene product, fibroblast growth factor (FGF), induces the appearance of both forelimbs and hindlimbs, depending on the part of the body where the FGF is applied. Thus, the same gene can be used to initiate different structures at different locations.

Natural organisms implement gene reuse through a process of development, or embryogeny.1 The same genes can be used at different points in development for different purposes, and the order in which activations of genes take place determines when and where a particular gene is expressed [65]. Recently, researchers have begun to replicate this process in artificial developmental systems. The hope is that extremely compact codes can evolve to represent immensely complex phenotypes.

Topics: Artificial Life | Representation | Development | Generative Programming