Go To the Tierra Home Page. A variety of Tierra programs, or Tierra-like programs are available or linked to at this site, if you are interested in experimenting with Tierra yourself.
Download a description document of Tierra. In Evolutionary Systems we do not run the Tierra synthesis of life; it takes far too long to get meaningful results. However, it is a very important culmination of our look at genetic algorithms, and this document summarizes what we learn from it.
As Tom Ray, the creator of Tierra says, The intent of this work is to synthesize rather than simulate life. As a result this genetic algorithm is in a class by itself, and the results coming from it are in a class by themselves.
Brief Description taken from document above:
Most Genetic Algorithm's are not open-ended. That is, they are designed Top-Down; someone endows each agent in the program with a genome consisting of a set of pre-defined genes, each of which may exist in a pre-defined set of forms. The agents do not contain their own mechanism for self-replication; they are simply copied by the simulator if they survive the selection phase. Furthermore, the simulator program mutates and/or recombines, selects, and replicates the genes according to criteria built into the simulator.
Tierra represents an entirely new class of genetic algorithms. The Tierran ecosystem consists of hand crafted organisms, but ones capable of self-replication and open-ended evolution independent of the designer. Once the initial replicator (the ancestor) is created it is capable of self-evolving its own code to enhance its own survival. In other words, what is evolving here is the computer's code, the instructions usually written by the programmer.
The Tierran agents are based on sequences of machine instructions that live and evolve in the CPU (Central Processing Unit) core of the computer. Machine instructions similar to those used in the Tierra Simulator have been shown to be capable of "universal computation"' and in this sense are like Turing's Universal Computer. This suggests that evolving machine codes should be able to generate any level of complexity. Although the model is limited to the evolution of machine instructions, this may have a potential comparable to evolution based on sequences of organic molecules.
