Or, as Glenn R. Morton described it, when Dembski had at a conference repeated his criticism of genetic algorithms only to be met by a sea of hands raised by engineers who used genetic algorithms on a regular basis and knew for a fact that Dembski didn't know what he was pontificating about. Morton's description: "looking like a deer caught in the headlights."
This was the only talk I was unable to record. My new tape recorder refused to turn the tape. It took me an hour to figure out why. :-(( Fortunately, I have found a guy who said he would send me a copy of that talk on tape and on paper. All the biggies were there, Ernan McMullen, Paul Nelson, William Lane Craig, Michael Behe, John Baumgarder, Walter Bradley etc.
His talk was entitled "Can Evolutionary Algorithms Generate Specified Complexity?" In it he examined genetic algorithms and made many mistakes concerning their properties and the way they worked.
Life is complex and specified, Dembski said. A random sequence is complex and unspecified; a sonnet is complex and specified. [Thus he made the same error that Meyer made on day 2--ignoring the reality of spy codes which make a specified sequence appear random and unspecified. Specification is nothing more than an agreement between two individuals--it is semantics or meaning. And the second paragraph of Claude Shannon's seminal paper on information theory clearly states that information theory is incapable of dealing with meaning--grm] But Dembski drives on. Evolutionists haven't explained complex specified information (CSI) at the origin of life or the subsequent increase in CSI. Dembski used Dawkins 'me thinks it is like a weasel' example to claim that genetic algorithms have merely played a shell game in which they shuffle information around. He says that the fact that Dawkins put a target sequence into his program means that the information is already there and is being transferred to the string that is being randomly mutated. There is much truth in this. [assuming Dembski correctly described the program and I can't find my Blind Watchmaker to verify it--grm]. Dembski claims that once a letter in Dawkins program is correct, mutations cease at that location. [If so, it is like a poker hand being drawn. and one is indeed transferring info from the ideal sequence to the evolved sequence.--grm]
However, Dembski then goes on to assume that this is how all genetic algorithms work. He spoke of programs specifying a fitness function over the landscape and then using mathematical properties of the fitness function to find the ideal target. This is cheating because one must specify the info in the fitness function prior to searching for the target fitness. He called it the no free lunch theorem. [Of course genetic algorithms make no such assumption. We don't know the fitness function prior to writing a genetic algorithm and so can't use its mathematical traits to help us with our search. I was appalled at the poor understanding of genetic algorithms.--grm]
He then made a third mistake. He assumed that true fitness functions were flat except for the target which had a high fitness function. In a case like this, he claimed that it would be impossible to find the target because the target occupies too small a region of the sequence space for a random search to have a good shot at finding it. . [This is true but it entails an assumption which is questionable. While any of the laity in the room wouldn't understand the assumption that had just been slipped into them, it was clear to me and a few others. What Dembski is saying by defining such a fitness function is that there is one and only one molecular sequence which will perform a given function. The fact that cow enzymes are different from human which are different from slug enzymes, shows that more than one sequence performs a given function and that the target area in a sequences space is not as small as Dembski says--grm]
His fourth mistake was that he seemed to imply that genetic algorithms used a 20-questions approach to finding the target. The algorithm makes inquiries of the fitness function and gets a response like the game. The fitness function tells the program you are getting warmer, no you are getting colder, etc.[ He claimed that is teleology. It would be if that is what actually occurred with genetic algorithms.--grm]
He claimed that Stuart Kaufman agrees with him. [That would be an interesting thing to check on.--grm] He also claimed that there is only one known generator of specified complexity--an intelligent agent.
In the Q&A I raised the issue of biomolecular companies who use genetic algorithms to search for novel functionality. He claimed that it wasn't important and that such programs could never be used to design anything. Then much to my amazement, John Baumgardner said "'Glenn's point was exactly correct.' I nearly fell on the floor. He told Dembski that they were using genetic algorithms at Los Alamos to design lots of things. Two or three other people said the same thing. [One of these, a man named Eide Trotter, I later learned is a well connected Southern Baptist who goes to First Baptist in Dallas. He sat next to me and the next table on Thursday morning when I had breakfast with Paul Nelson, Mark Kalthoff and Tom Judson(?). After breakfast he said that he liked much of what I had said. Gerald Eichoefer, of Greenville College (don't know where that is) tried to come to Dembski's defense. He said that genetic algorithms were terrible inefficient search methods. He was shot down by a guy in the back who said that genetic algorithms vastly outperform a random search. and indeed a genetic algorithm isn't a random search. Later that day in the last session, Frank turned to me and said that Eichoefer claims to be a prophet of God. I must admit he looked like one which may explain why I took his name down off of his name tag.--grm]Hands were upraised all over the room. Dembski had the deer in headlights look. He turned it over to the next speaker.
I have made this letter longer than usual, because I lack the time to make it short.
I'm a great fan of science you know.