Genetic Algorithms vs Natural Selection

A recent post by quinton got me thinking about genetic algorithms, and more specifically, about the fact that the algorithm is essentially based on Darwin's theory of natural selection. It's a great technique for solving certain classes of problems, but it troubles me somewhat that the process of evolution (upon which the algorithm is based) is not properly established.

In general, I think the evidence for evolution is very strong. However, the mechanics of how it works still needs proper explanation - at the present time natural selection seems unable to explain things properly. Let me outline a few objections to the theory of natural selection:

1) How the first replicator could arise is not at all explained. (The first replicator is the first entity able to reproduce itself). There are any many suggested theories, none satisfying. Currently, the first replicator mechanism is not explained, not proved and not reproducable.

2) The fossil subset is very poor. Only a tiny fraction of species show intermediate forms, and no species show smooth transitions. Darwin himself thought that the lack of fossil records was the biggest threat to his theory, but was confident that in the fullness of time, sufficient fossil data would come to light. Good hominid fossil records have been found in recent times (presumably because human evolution is more interesting), but that aside, it's arguable that over 100 years after his death, the fossil records are even poorer (as some promising fossil records from his time were proved to be incorrect).

3) Irreducible complexity. This is a popular weapon of creationists, typically they refer to complex components such as the eye. More compelling are some of the metabolic pathways present in organisms - the fact that complex molecules are synthezed in 12 or more steps, with no useful by-products. Hence the whole process would have to have been discovered "at once".

4) Staggeringly variable rates of evolution. Certain species seem to have "forgotten" to evolve despite being subject to the same evolutionary stresses.

5) Problems with DNA itself. Despite the genome mapping project, it is becoming increasingly difficult to see how DNA could contain enough information to define a complex phenotype.

6) The timescales present major difficulties, perhaps even the most significant objection. The timescales to evovle from hominoid to hominid seem too short by many orders of magnitude.

Now, the bulk of these objections can be overcome by either (a) substantially increasing the time available or (b) coming up with a better mechanism than natural selection.

Let me stress again that I have no time for creationists, who for me fall into the same category as astrologists, homeopaths and psychics (i.e. people who believe in things despite the absence of scientific evidence). However, based on the points above, I think the theory of natural selection is currently inadequate to explain evolution properly. And as such, its use as a basis for a computer algorithm is suspect.

Thanks to Alex for his input in structuring this article.