From: Martha Mercaldi (mercaldi@cs.washington.edu)
Date: Wed Dec 01 2004 - 08:38:10 PST
The evolutionary origin of complex features
Richard E. Lenski, Charles Ofria, Robert T. Pennock & Christoph Adami
This paper simulates evolution of organisms, using computer programs to
represent organisms. For the most part it validates properties of
evolution (i.e. initially harmful mutation, if survived, required to
produce beneficial feature) which have been validated by experiments on
other organisms, such as in biology.
Simulating these processes in computers, if we trust that the
environment we contrive is representative of the real world, allows the
researcher to peer deeper in the process, to turn back time, to know
exactly what mutations occur and in what order, and to simulate
hypothetical situations. Biologists in a lab cannot do this, and the
ability to study our model in this way has the potential to reveal
properties of evolution that have not been observable in a
laboratory. Of course it would be wise to devise (if possible) a
biological experiment to test any hypotheses generated from simulation,
Evolution as search:
Evolution can be seen as a search problem. The state space is infinite
(all possible genotypes) and the transitions are mutations. This is
just for asexual reproduction. The transition rules for sexual
reproduction would be more complex. The fitness function and the
environment comprise a heuristic that guides the random walk. The
“frontier” of the search is the current population. The environment
using the fitness function supports replication of organisms with
beneficial mutations first, essentially ordering the queue in a BFS and
encouraging exploration of the best states first.
Evolution not as search:
In deterministic search there is no benefit to re-exploring a node that
has already been visited, however, in this case a genotype could
conceivably evolve back to one that existed in the past. If it survives
it is reasonable to continue exploring it as the way in which it
mutates is likely to be different from how it did previously.
Some questions that this work might explore next. How sensitive is the
evolutionary process to the genetic encoding. The authors here picked
some machine language, but what if the instructions were more or less
dense (CISC v. RISC instructions)? What effect would that have on the
speed and success of evolution? Also, this was mentioned in the
conclusion, but it seems that sexual reproduction would change the
model and some of the conclusions. However in the spirit of “best
first” exploration, there are plenty of organisms (i.e. viri and
bacteria) which replicate asexually which we’d love to better
understand, so this seems less pressing.
This archive was generated by hypermail 2.1.6 : Wed Dec 01 2004 - 08:38:10 PST