Hi, ASA Listees -
Here is a new topic to work on, at least I havenít seen it discussed
in this forum.
A few weeks ago I attended the weekly seminar of the Bioinformatics
group, an initiative
of Computer Science and Biology on our campus. The speaker was
Darrel Whitely, a
mathematician of Colorado State.
His topic was ěNo Free Lunch Theorems in Genetic Computationî.
This of course immediately reminded me of Bill Dembskiís recent
book, ěNo Free Lunchî.
I expected to see a vitriolic denunciation of Dembski and his book,
but he didnít mention
Dembski. He went through the process of
Showing that in general for each type of genetic algorithmic
computation one can make,
there is a corresponding NFL theorem. I ended up not being sure
whether he knew
Dembskiís book existed.
His lecture showed that any special search algorithm that is used to
optimize a given
quality (such as a fitness function) can on the average be no more
efficient than a
RANDOM search. And a random search through all of genome space for
an animal gets you
nowhere, for a reasonable population size, and in a time of 10^14
years. So the NFL
principle implies that one should not expect to find Darwinnian
processes that add
appreciable useful information to a species genome.
Whitely did not spell out this last inference for the audience, but
he did not need to,
for most of them. He was not lynched, he was an honored guest.
For those who are not into mathematical analysis of genetic
evolutionary processes, and
NFL theorems, let me explain a bit.
Genetic algorithms are computer programs that model the standard
process of Evolution, which assumes that the increasing adaptedness
of animals is due to
random mutations within a population, followed by Natural selection
of better adapted
The No Free Lunch conclusion says that one should not, averaged over
time, and over
all genetic algorithms, expect purely natural physical processes to contribute
appreciable information content to an animalís genome.
This is basically what Dembski says in his ěNFLî book.
Blessings, Larry Johnston
This archive was generated by hypermail 2b29 : Mon Jul 15 2002 - 14:43:59 EDT