Re: [GP] Evolving agents with a genome and phenome
- Hi Russ,
I think that there are many reasonable ideas for what constitutes a "phenotype," and that a great deal of past work would fit your description in the context of some of these ideas.
For example, a phenotype might be a neural network that is constructed by a genotype program and then assessed (for reproduction) purely on the basis of the performance of the neural network. There is a great deal of work of this kind (and also with evolved electrical circuits, etc.). In some cases the phenotypes are assessed in isolation, while in others interactions among phenotypes (sometimes in a virtual world) determine who can reproduce.
In another context, one might (and many do) consider frameworks like Grammatical Evolution to involve a genotype/phenotype distinction, with the program that is produced by the expression of the genotype (via the grammar) considered to be a phenotype (that is subsequently assessed for the sake of reproduction). Again, there is a lot of work of this sort, sometimes explicitly referencing biological concepts of gene expression.
In yet other contexts, some GP systems evolve code that can modify itself as it runs, involving a sort of continuous unfolding of the phenotype (which might be seen as biological in nature). I've used the phrase "ontogenetic programming" to describe some of the work in this spirit (and others have used phrases involving "self-modifying," etc.), and many have used "developmental genetic programming" for one or more of the different ideas described above.
You can find a bunch of the work that I'm describing by searching for "phenotype", "gene expression", "ontogenetic", or "developmental" at the GP bibliography (http://liinwww.ira.uka.de/bibliography/Ai/genetic.programming.html).
Within all of these contexts, however, I think that your question about systems in which there is a requirement for constructing "an increasingly complex phenome" is a really good one. I would guess that in a lot of the work noted above one might be able to make a case that increasing complexity must and does evolve, but this would depend on exactly what one means by complexity and on how one can measure it.
On Sep 1, 2013, at 2:08 AM, Russ Abbott wrote:
> Does anyone know of work involving the evolution of agents with both a genome and phenome? I'm looking for a model in which survival and reproduction depends on phenome-level interactions and where those interactions must evolve competitively. To do that there must be a genome that constructs an increasingly complex phenome. Any examples where this has been done?
Lee Spector, Professor of Computer Science
Cognitive Science, Hampshire College
893 West Street, Amherst, MA 01002-3359
Phone: 413-559-5352, Fax: 413-559-5438