In his very readable book “The Extended Phenotype“, Dawkins defined the “Extended Phenotype” as the effects of a gene when those effects are not regarded as being confined to the individual body in which the gene sits. He argues against the arbitrariness of limiting the applicability of phenotypes only to expressions of an organism’s genes in its own body.
In Evolutionary Computing, the term “phenotype” is usually reserved for a final representation of a genome along the process of transformation pursuant to being assigned fitness. In my paper (on search, neutral evolution and mapping in evolutionary computing), we used a new approach similar to Dawkins’ extension. We relaxed the generally used definition of a phenotype to include intermediate representations of a genome along its transformation to a fitness value. This means a Genotype to Phenotype map can be a map to some intermediate or final form.
The idea of extended phenotypes is controversial in biological circles; however relaxing the definition of a phenotype proves to be very useful for analyzing evolutionary computation because it gives a single general framework for analyzing encodings, representations and problems.
As a result of the relaxation we were able to analyze encodings (e.g. parity coding) and representations (e.g. grammatical evolution) without reference to a specific problem. we also used analyzed specific problems (e.g. OneMax) and problems using a particular encoding or representation (e.g. OneMax using parity coding, and boolean parity using cartesian genetic programming). It is difficult to find any other approach that can deal with encoding, representations and problems within the same framework.