Friday, September 18, 2009

New Peer-Reviewed Article Said to Support Intelligent Design

William Dembski and Robert J. Marks have written an article titled: Conservation of Information in Search: Measuring the Cost of Success, which appears in IEEE Transactions on Systems, Man, and Cybernetics—Part A: Systems and Humans. The Discovery Institute has trumpeted (by none other than Casey Luskin) this article as being a "New Peer-Reviewed Scientific Article From William Dembski and Robert Marks Challenges the Creative Mechanism of Darwinian Evolution." As an aside, I am not sure how he has managed to snag a free, postable copy of this article, because we pay through the nose for our IEEE access. IEEE may have done this because Dembski is a senior member of the IEEE and, of course, a Discovery Institute fellow. Luskin writes:
Darwinian evolution is, at its heart, a search algorithm that uses a trial and error process of random mutation and unguided natural selection to find genotypes (i.e. DNA sequences) that lead to phenotypes (i.e. biomolecules and body plans) that have high fitness (i.e. foster survival and reproduction). Dembski and Marks' article explains that unless you start off with some information indicating where peaks in a fitness landscape may lie, any search — including a Darwinian one — is on average no better than a random search.
Note that Luskin uses the "unguided natural selection" language that he so heavily castigated Kenneth Miller for using. Beyond that, though, this is a peculiar read of how evolutionary theory works. Is this how a search algorithm really proceeds? It certainly is not how evolution proceeds. Luskin writes that:
This paper is in many ways a validation of some of Dembski's core ideas in his 2001 book No Free Lunch: Why Specified Complexity Cannot Be Purchased without Intelligence , which argued that some intelligent input is required to produce novel complex and specified information.
When it was released, No Free Lunch was criticized not so much as being wrong as being irrelevant to biological organisms. Dembski and Marks write:
The no free lunch theorem (NFLT) likewise establishes the need for specific information about the search target to improve the chances of a successful search. “[U]nless you can make prior assumptions about the . . . [problems] you are working on, then no search strategy, no matter how sophisticated, can be expected to perform better than any other”
In this example, he is equating (if I understand it correctly) each genetic variation with a term in a search algorithm. Basically, he is stating that genetic variation, left to its own devices, won't turn up anything complex no matter how many iterations it goes through. In this instance, all fitness functions are averaged. Here's the problem: in no environment are fitness functions averaged. That is like saying that the earth has one climate all over and that it imposes no selection pressures. Mark Chu-Carroll of Good Math, Bad Math also makes this observation:
Search algorithms work because they're not intended to succeed in all possible landscapes - they're designed to work in specific kinds of landscapes. A successful search algorithm is designed to operate in a landscape with a particular kind of structure. The performance of a particular search is tied to how well suited the search algorithm is to the landscape it's searching. A hill-climbing search-algorithm will work well in a continuous landscape with well-defined hills. A partitioning landscape will work well in a landscape that can be evenly partitioned in a well-defined way.
Dembski has, in the past, as in this example, used the Dawkins model of selection in which Dawkins runs through a series of non-random iterations (how evolution would actually proceed) in which a meaningless string slowly becomes METHINKSITISAWEASEL. This model of evolution is echoed by H. Allen Orr, who writes:
First consider the odds of forming this target sequence by blind chance, i.e., with monkeys at word-processors. Draw a random letter from the alphabet for the first position in the phrase; now another for the second position, and so on. The odds that you've spelled out the phrase METHINKS… are essentially nil: in fact, with twenty-six letters plus a blank space, the odds of getting the word METHINKS alone are already less than one in 280 billion. But now consider the following "evolutionary algorithm." Start with a random sequence as before but i) randomly change each character that doesn't match the target sequence; ii) if a resulting character matches the target keep it and in the next round change only those characters that don't match. So, if we start with SATHINKS, at the next step we'll randomly change only the first two letters; and if those changes yield MQTHINKS, then at the next step we'll randomly change only the second letter. This two-step evolutionary algorithm of mutation plus selection arrives at the phrase METHINKS… with surprising speed.
Dembski and Marks argue that there is specified information in this algorithm in that you know the outcome. Without this eventuality, they argue, an evolutionary algorithm will never amount to any new information. As Dawkins has correctly pointed out, though, the METHINKSITISAWEASEL algorithm is not strictly correct—it has a known endpoint—while evolution, as we understand it, does not (a knowable one anyway). Dawkins is trying to illustrate the concept of selection. That he does so in a seemingly teleological fashion does not detract from what the example is trying to illustrate. Further, as Chu-Carroll notes, Dawkins' METHINKS is not a simple lock-in-place example, as Dembski and Marks contend. They intentionally ignores the selection aspect of the illustration to focus on the idea that Dawkins has introduced exogenous/active information (the endpoint is known). Indeed, as Mark Perakh has written:
Evolutionary algorithms may be both targeted and targetless. Biological evolution, however, has no long-term target. Evolution is not directed toward any specific organism. The evolution of a species may continue indefinitely as long as the environment exerts selection pressure on that species. If a population does not show long-term change, it is not because that population has reached a target but because the environment, which coevolves with the species, acquires properties that eliminate its evolutionary pressure on the species. Dawkins’s weasel algorithm, on the other hand, stops when the target phrase has been reached.
For Dembski and Marks, it is an either or proposition. For a search algorithm to be successful, it must have active information. This, as the authors see it, contradicts the "random" nature of evolution. If there is no active information, then the search is no better than random and evolution reaches a dead end. Dembski and Marks conclude:
If any search algorithm is to perform better than random search, active information must be resident. If the active information is inaccurate (negative), the search can perform worse than random. Computers, despite their speed in performing queries, are thus, in the absence of active information, inadequate for resolving even moderately sized search problems. Accordingly, attempts to characterize evolutionary algorithms as creators of novel information are inappropriate.

The problem, as I see it, is that, while Dembski and Marks understand how these theorems work, they have no understanding of how evolution proceeds in the wild. They consistently say that some things cannot happen, when biologists have been observing them for hundreds of years. So, is this an article that supports ID while showing that evolution cannot happen? No, it isn't. Once again Dembski and Marks have applied inappropriate mathematics in attempting to show that evolution doesn't work. The argument fails because neither Dembski or Marks understand how evolution actually works. If the Discovery Institute continues to trumpet the fact that they have a peer-reviewed article that challenges evolution, when in fact, it does no such thing, such posturing will be dishonest.

Now playing: Anthony Phillips - The Geese and the Ghost - (2008 Remaster)
via FoxyTunes


  1. Excellent analysis!

    From Luskin's quote:

    Dembski and Marks' article explains that unless you start off with some information indicating where peaks in a fitness landscape may lie, any search — including a Darwinian one — is on average no better than a random search.

    That's what drives me bats about Luskin's (and Dembski's and Marks') misrepresentation of evolution by random mutations and natural selection. The evolutionary algorithm does not have to know "where peaks in a fitness landscape may lie." It needs only be operating on a (high-dimensioned) fitness landscape that is not random but rather displays local autocorrelations -- it has gradients on some variables. It does not give a damn whether there are peaks: Evolution is the survival of the 'good enough,' not the peak. Most species are not on peaks in all their various traits, but are good enough (relative to competitors) to survive and successfully reproduce.

  2. Its Sewall Wright's adaptive peaks and landscapes all over again. If a species radiates into a landscape that is varied in topography and temperature, the traits that help the regional subpopulations thrive will be positively selected for while those that don't will be selected against. Over time you will get regional variation in the phenotypes of the individuals. If that persists long enough, you get allopatric speciation. You don't have to look far to see examples of that. In Dembski's examples, every generation dies and the whole process starts over again because evolution is "random." He is completely out of touch with not just evolutionary theory, but field biology as well. I wonder, seriously, if he has every heard of Dobzhansky and Mayr's "biological species concept?"