|Wesley R. Elsberry
Joined: May 2002
This thread is for discussing the train wreck in progress that is the Robert Marks and Bill Dembski essay trying to critique Tom Schneider's "ev" program. The title here is "Unacknowledged Errors"; part of the reason for this thread is to watch for when, if ever, the errors in the Marks and Dembski paper become acknowledged by the authors.
The Marks and Dembski essay, in PDF format:
How do genetic systems gain information by evolutionary processes? Schneider  asks this question and then purports to answer it by simulating the evolution of nucleotide binding sites with an algorithm called ev. Upon examination, ev's structure is algorithmically equivalent to the inversion of a perceptron neural network. We show that ev is able to evolve binding site locations only because it is prestructured to do so. The difficulty of the problem is 131 bits. The perceptron search structure adds about 122.2 bits of information, leaving only about 8.8 bits to the search procedure. Although the perceptron structure smuggles in an enormous amount of information, ev squanders it with the evolutionary search. The evolutionary algorithm in ev required over 45,000 fitness queries to achieve success. Based on query count, repeated random queries outperform the evolutionary algorithm by over 10,000%. The simulation of evolutionary processes using ev illustrates a larger problem in interpreting the results of evolutionary simulations. Interpretations of simulation results, if they are to be credible and inspire confidence, should (a) identify the inherent difficulty of solving the problem and, (b) measure the amount of information about the solution provided by the search structure and the search procedure.
Tom Schneider responds to criticism:
Considering 439 queries as 3 orders of magnitude, Dembski's estimate is off by about 13 orders of magnitude.
Schneider provides several different methods to arrive at the conclusion that Marks and Dembski are off, way, way off. So, how long will the erroneous claims of Marks and Dembski be headlining "Baylor's Evolutionary Informatics Laboratory"?
"You can't teach an old dogma new tricks." - Dorothy Parker