From: Ravi Kiran (kiran@cs.washington.edu)
Date: Tue Nov 30 2004 - 23:20:05 PST
"THE EVOLUTIONARY ORGIN OF COMPLEX FEATURES"
- Richard.E.Lenski, Charles Ofria, Robert T.Pencock,
Christoph Adami
One-line summary of the paper:
This paper explores the possibility of using digital organisms --
computer programs that self-replicate, mutate and compete -- thereby
obtaining descendant populations as the consequence of an evolution
mechanism that parallels the biological analogue and through a series of
experiments, show that such organisms provide opportunities to address
important issues in evolutionary biology.
Two most important ideas in the paper and why:
Idea 1: Using a digital organism with a fairly simple set of instructions
can be used to demonstrate the validity of hypothesis, which is also
supported by comparitive and experimental evidence, that complex features
of organic forms evolve by modifying structures and functions. This is
helpful in addressing issues in evolutionary biology for studying problems
that are difficult to study with organic forms (e.g. evolutionary trace of
complex organs such as eye) owing to incomplete information, insufficient
time and impracticality of such experiments.
Idea 2: Choosing the mode of reproduction to be asexual enabled selection
to occur among organisms rather than genes, as in the sexual case, where
there is a distinct possibility of a "dominant" organism enabling a
harmful mutation to persist for a longer time than it would have in the
case of asexual reproduction. This is important because asexuality permits
beneficial combinations of mutations to spread via organisms even though
they are individually harmful while weeding out organisms deemed harmful.
Flaw(s) in the paper:
In the discussion on the Avida system, the authors give an example
using nand and eql. The idea of a no-op instruction ( an instruction which
locks the contents of thp particular register ) modifying something was
confusing. The example could have been a bit more clearly presented.
The authors observe that the space of solutions is only a tiny
fraction of the total genotypic search space. However, there is no mention
of any attempts to incorporate additional randomness into the algorithm
which would assist in increasing the solution pool. For example, simulated
annealing is a local search algorithm which is suitably random in its
behavior and employs a 'cooling' schedule which can be made adaptive.
Evolutionary biology subscribes to the fact that great upheavals ( e.g.:
onset of Ice Age ) provide significant changes in the evolutionary
mechanism. Simulated annealing can be used to of as 'shaking' the table of
evolutionary search space and letting the spheres of fit organisms roll
into appropriate slots. Such algorithms should also be explored to enhance
the effectiveness of genetic algorithms.
Is the problem and solution just another instance of search ? What, if
anything, makes this search problem different from the usual search
problem ?
In some sense, it is a search. There is a start state ( starting
organism) and there is an objective function according to which future
states(descendants) are spawned. The goal states are the organisms that
ultimately survive. However, the generation of successors involves a
mutation from among the existing states. Also, evolution can also involve
moves which are sideways and which regress and yet are still important.
This archive was generated by hypermail 2.1.6 : Tue Nov 30 2004 - 23:20:05 PST