We model some of the crucial properties of biological novelty generation, and abstract these out into minimal requirements for an ALife system that exhibits constant novelty generation (open ended evolution) combined with robustness.
The requirements are an embodied genome that supports runtime metaprogramming (self modifying code), generation of multiple behaviours expressible as interfaces, and specialisation via (implicit or explicit) removal of interfaces.
The main application of self modifying code to date has been top down, in the branch of Artificial Intelligence concerned with learning to learn. However, here we take the bottom up Artificial Life philosophy seriously, and apply the concept to low level behaviours, in order to develop emergent novelty.
full paper : PDF 161K
@inproceedings(SS-ECAL11-23, author = "Simon Hickinbotham and Susan Stepney and Adam Nellis and Tim Clarke and Ed Clark and Mungo Pay and Peter Young", title = "Embodied genomes and metaprogramming", pages = "334-341", crossref = "ECAL11" ) @proceedings(ECAL11, title = "ECAL 2011, Paris, France, August 2011", booktitle = "ECAL 2011, Paris, France, August 2011", publisher = "MIT Press", year = 2011 )