Sunday, October 21, 2007

I can’t believe Dembski chose Word Mutagenation!!

Take a look at William Dembski’s latest draft manuscript, Active Information in Evolutionary Search The Information Cost of No Free Lunch.

Dembski: Proponents of intelligent design have argued that the NFLT shows that Darwinian evolution cannot generate the information required for biological complexity from scratch but instead merely reshuffles existing information.

The NFLT (No Free Lunch Theorem, Wolpert and Macready 1997) doesn’t suggest that Darwinian Evolution can't generate information in the sense of searching a fitness landscape. Rather, it merely states that it may or may not be the best conceivable algorithm for finding fitness.

Dembski: Making such assumptions about underlying search structures is not only common but also vital to the success of optimizing searchers (...).

Such assumptions, however, are useless when searching to find a sequence of, say, 7 letters from a 26-letter alphabet to form a word that will pass successfully through a spell checker… With no metric to determine nearness, the search landscape for such searches is binary—either success or failure. There are no sloped hills to climb.

I can’t believe Dembski chose Word Mutagenation!! He couldn't be more wrong.

It’s been three years since I published Word Mutagenation on the web, the result of a year-long thread on the newsgroup Talk Origins with Intelligent Design advocate, Sean Pitman. I suppose it’s time to dust off the old Letter Mutator and take another look.

Turns out we don’t have to make any assumptions about the search space of words. For instance, we could start with the single-letter word “O”, a replicating population representing a veritable Pond of O’s flowing down to the Sea of Beneficence.

In the beginning was the Word.

We then mutate words in our Sea. We might change a random letter, add a random letter, delete a random letter, or randomly recombine words in the Sea. Something like this:



If a mutant sequence is not a valid word, that is, if it fails the spell-checker rule that Dembski established above, then it is ruthlessly eliminated with no issue. If it is a valid word, it enters the population as a new strain. (We might optionally limit our Pond of Words to just a certain number of the longest words.) So, what do we expect?

Quick calculation:
10-letter words in dictionary ~10^4
10-letter sequences possible ~10^14
Ratio of valid 10-letter words to possible sequences ~10^10

A random search would take ~10^10 or several billion mutations. An evolutionary search algorithm takes only ~10^5 mutations or about a hundred thousand times faster—consistently. There is nothing about the structure of words in Word Mutagenation. It works in the simplest of fashions; random mutation and spell-check, over and over again.

The structure is found in the words themselves. Turns out that language itself has evolved and evidence of that history is found in the words we use.

The fault, dear Brutus, is not in our stars,
But in ourselves.

Labels: ,