Replaying the Tape: An Investigation into the Role of Contingency in Evolution
Tim Taylor
Department of Artificial Intelligence, University of Edinburgh, UK
http://www.dai.ed.ac.uk/daidb/people/homes/timt/
John Hallam
Department of Artificial Intelligence, University of Edinburgh, UK
Abstract
The role of contingency (random events) in an artificial evolutionary
system is investigated by running the system a number of times under
exactly the same conditions except for the seed used to initialize the
random number generator at the beginning of each run. Twelve different
measures were used to track the course of evolution in each run, and
``activity wave diagrams'' were also produced. The results of 19 runs
are presented and analyzed. The performance of every run was compared with
each of the others using a non-parametric test (a randomization version of
the paired-sample t test). When comparing absolute values of the measures
between the runs, some significant differences were found. However,
looking at the *difference* in values between adjacent sample points
for a run, no run was significantly different to any other for any of the
measures. This suggests that the general behaviour is the same in all
runs, but the accumulation of differences results in significantly
different outcomes. The results lead us to propose a rule of thumb for
future experiments with the system: to check whether the outcome of any
particular experiment is robust to contingency in the evolutionary
process, at least nine runs should be conducted using different seeds
for the random number generator, to be confident of seeing a variety
of results. The results are likely to be applicable to other A-Life
platforms of self-replicating computer programs, but at this stage can
probably tell us little about the role of contingency in biological
evolution.