Studying Evolution with Self-Replicating Computer Programs

A critical discussion is presented on the use of self-replicating program systems as tools for the formulation of generalised theories of evolution. Results generated by such systems must be treated with caution, but, if used properly, they can ooer us unprecedented opportunities for empirical , comparative studies. A new system called Cosmos is introduced, which is based upon Ray's Tierra 15]. The major diierence between Cosmos and previous systems is that individual self-replicating programs in Cosmos are modelled (in a very simpliied fashion) on cellular organisms. Previous systems have generally used simpler self-replicators. The hope is that Cosmos may be better able to address questions concerning the sudden emergence of complex multicellular biological organisms during the Cambrian explosion. Results of initial exploratory runs are presented, which are somewhat diierent to those of similar runs on Tierra. These diierences were expected , and indicate the sensitivity of such systems to the precise details of the language in which the self-replicating programs are written. With the strengths and weaknesses of the methodology in mind, some directions for future research with Cosmos are discussed. Within the last decade, computers have become powerful and aaordable enough to enable a number of research groups to study the evolution of life in a new way. Rather than following the traditional approach of trying to capture properties of whole populations in mathematical models, the new approach models a large number of individual self-replicating entities which are competing against each other for resources required for replication. This is achieved by creating a computer which can run a large number of self-replicating programs in parallel 1. 1 In practice, a virtual computer is created (i.e. implemented in software), with parallelism simulated by time-slicing between Tom Ray pioneered this approach with his Tierra system 15, 16]. Since then, a number of other systems have also been developed, including Avida, developed by Chris Adami and Titus Brown 2], Computer Zoo, written by Jakob Skipper 18], and John Koza's system of self-replicating LISP-like programs 8]. Using such a methodology to study evolutionary systems is attractive for a number of reasons. For example, as the self-replicators are being modelled individually rather than as populations, the simulation respects the fact that a gene does not work in isolation. Rather, it is part of a large ensemble of genes which must all work together with some degree of cooperation in order for the individual organism …