The influence of parsimony and randomness on complexity growth in Tierra

Abstract

The issue of how to create open-ended evolution in an artificial system is one the open problems in artificial life. This paper examines two of the factors that have some bearing on this issue, using the Tierra artificial life system. {\em Parsimony pressure} is a tendency to penalise more complex organisms by the extra cost needed to reproduce longer genotypes, encouraging simplification to happen. In Tierra, parsimony is controlled by the \verb+SlicePow+ parameter. When full parsimony is selected, evolution optimises the ancestral organism to produce extremely simple organisms. With parsimony completely relaxed, organisms grow larger, but not more complex. They fill up with ``junk''. This paper looks at scanning a range of \verb+SlicePow+ from 0.9 to 1 to see if there is an optimal value for generating complexity. Tierra (along with most ALife systems) use pseudo random number generators. Algorithms can never create information, only destroy it. So the total complexity of the Tierra system is bounded by the initial complexity, implying that the individual organism complexity is bounded. Biological systems, however, have plenty of sources of randomness, ultimately dependent on quantum randomness, so do not have this complexity limit. Sources of real random numbers exist for computers called {\em entropy gatherers} -- this paper reports on the effect of changing Tierra's pseudo random number generator for an entropy gatherer.