User:ThomHImself

From Wikipedia, the free encyclopedia

Tom English http://www.BoundedTheoretics.com

[edit] No conflict of interest: Robert J. Marks II and Evolutionary Informatics Lab

In June 2007, someone edited No free lunch in search and optimization to indicate that William A. Dembski and Robert J. Marks II were disseminating at the Evolutionary Informatics Lab website, then hosted by a Baylor University server, technical papers invoking the "no free lunch" (NFL) theorems in arguments for intelligent design (ID). I had long suspected that Marks secretly advocated ID, so this came as no surprise. I immediately scrutinized the website, fully expecting to find ID content. Instead I found novel technical concepts that might find legitimate applications in engineering. It was clear to me that the concepts could also be put to use in arguments for ID, but the fact is that they were not put to that use. I revised the NFL article and gave an explanation on the talk page.

Note not only that my colleagues regard me an authority in evolutionary informatics ("Tom English is almost solely responsible for making the intriguing connections [of NFL] with information theory."[1]), but that I have criticized ID theory in a peer-reviewed chapter, "Intelligent Design and Evolutionary Computation," to appear in an edited volume, Design by Evolution: Advances in Evolutionary Design. If you examine the excerpts from the chapter available at my website, you will see that I give Dembski's latest definition of complex specified information a more rigorous mathematical treatment than Dembski himself does. In short, there are very few people with my combined knowledge of evolutionary informatics and ID.

I examined the Evolutionary Informatics Lab's website a month before Baylor University deleted the site from its web. The key factor in my decision to protest Baylor's action was that my prior scholarly assessment of the site's content was documented at Wikipedia. I sent email notes to Baylor's regents, directing them here for evidence that I was both critical of ID and evenhanded in my response to the website. I also contacted Marks (we had never before communicated) and arranged for him to list me as an affiliate of the lab. In public explanations of this, I have consistently stated that I assessed the site as a Wikipedia editor before the site became controversial. That is, I have tried to persuade the world at large to avoid rushing to judgment of Marks and his lab, and to consider the neutral and cautious perspective of our encyclopedia.

As a computer scientist with scholarly credentials both in evolutionary informatics and (criticism of) intelligent design theory, I can say as a matter of fact that Marks sticks to engineering at the website. When I first wrote that at Wikipedia, it was obvious that I held the interests of encyclopedia paramount. No one questioned the neutrality of my point of view at the time. In fact, the NFL article was promoted to GA status after I changed the discussion of the Dembski-Marks concepts. No one could have predicted that Marks and the website would be covered in the documentary Expelled: No Intelligence Allowed. It is absurd to say now that my point of view is attributable to my past affiliation with the lab. It is the other way around. I joined the lab because I had carefully assessed the website from a WP:NPOV for Wikipedia.

My interests are not in conflict with those of Wikipedia. But my standards of scholarship and encyclopedia writing, not to mention ethics, are in conflict with those of the Defenders of Science committed to exposing the Great ID Conspiracy. I believe that if ID is accurately characterized in neutral terms, drawing only on high-quality sources, the average reader will not miss the fundamental dishonesty of it. ThomHImself (talk) 01:34, 6 May 2008 (UTC)

[edit] How a participant in the Intelligent Design WikiProject "improved" a GA on computer science

Once upon a time, No free lunch in search and optimization, written almost entirely by me, was one of fifteen good articles in the "computer science" category. The subsection below was in the article when it passed GA review. Odd nature, a participant in the Intelligent Design WikiProject, deleted it without discussion, noting in the article history, "Size and credulousness of section gave undue weight to ID proponents views." As you can see here, the size was not excessive in relation to that of the entire article. You can gauge my credulousness for yourself. Odd nature replaced the section with two sentences, one of which was factually incorrect, and bluffed sources to boot:

The no free lunch theorem is often invoked by intelligent design proponents Robert J. Marks II and William Dembski as supporting intelligent design and Dembski's concept of specified complexity which he alleges is evidence of design.^[1] The scientific community has rejected both the notions of specified complexity and that the no free lunch theorem supports intelligent design.^[2][3]

If Odd nature had read the article, s/he would have known that there is not just one NFL theorem. Marks has never embraced specified complexity, and had little to say about NFL until last year. Dembski had quite a bit to say about the NFL theorems in the book cited in the first reference, which he (and not Marks) published in 2002. But he has never claimed that the NFL theorems support his concept of specified complexity. The second reference is to a paper I published in 1996, six years before I first heard about specified complexity and intelligent design! (Odd nature and his partner in reversion, Baegis, have permitted me to remove this embarrassment, but to change nothing else.) As for the third reference, it is a 2002 review of Dembski's book by an author of the earliest NFL theorems. The review says nothing about specified complexity or the response of the scientific community. What do you suppose Odd nature and Baegis have been up to in defending this grotesquerie?

Odd nature's user page says, "Those who can, create. Those who can't, delete." S/he should have stuck to deleting. Here's what I created:

[edit] Specified complexity, active information, and NFL

The central thesis of intelligent design is that only intelligent processes give rise to material entities with high specified complexity (alternatively, high content of complex specified information). This is similar to the notion that only optimization algorithms matched to problems by practitioners are efficacious. Intelligent design advocate William A. Dembski indeed makes various references to the NFL theorems in No Free Lunch: Why Specified Complexity Cannot be Purchased Without Intelligence, indicating that effective optimization requires intelligently designed algorithms.^[1] While Wolpert and Macready indicate that effective optimization requires the use of existing prior information,^[4] Dembski argues that it requires introduction of novel information into the material universe by non-material intelligence.^[1]

The primary objective of the intelligent design movement is to counter mainstream evolutionary theory. Biological evolution is often regarded as optimization, and evolutionary computation, which mimics biological evolution, also commonly takes the form of optimization. A large body of research in evolutionary computation suggests that algorithms regularly yield solutions with high specified complexity. Dembski claims that experimenters, as intelligent agents, "smuggle" specified complexity into their programs.^[1] In other words, only algorithms imbued with prior knowledge of the problems would have succeeded. But, as emphasized above, algorithms are almost universally effective in theory, and no one has established how commonly they are effective in practice.

There have been several arguments that the NFL theorems do not apply to biological evolution. English wrote in 1996,^[2] and Wolpert concurred in 2003,^[3] that in biological evolution the evaluation of "candidate solutions" (genomes) modifies objective functions (known as fitness functions in this context). There is not an NFL theorem that covers such coupling of search and objective functions. Yet in 2005 Wolpert and Macready asserted that "in the typical coevolutionary scenarios encountered in biology, [...] the NFL theorems still hold" [emphasis added] without having established that the theorems ever held.^[5] Häggström argued in 2007 for inapplicability of the original theorems.^[6]

Notions similar to those associated with Dembski appear at the web site of the Evolutionary Informatics Laboratory founded by Professor Robert J. Marks II of Baylor University. "The principal theme of the lab’s research is teasing apart the respective roles of internally generated and externally applied information in the performance of evolutionary systems." As of January 2008, the site provides access to unpublished technical papers coauthored by Marks and Dembski. The papers make no reference to specified complexity, but develop an analogous notion of active information. A major theme is that certain computational simulations of evolution succeed only because researchers have introduced active information. In the following excerpt, the "target" is a subset of candidate solutions with acceptably good values.

Three measures to characterize the information required for successful search are (1) endogenous information, which measures the difficulty of finding a target using random search; (2) exogenous information, which measures the difficulty that remains in finding a target once a search takes advantage of problem-specific information; and (3) active information, which, as the difference between endogenous and exogenous information, measures the contribution of problem-specific information for successfully finding a target.^[7]

Search algorithms, unlike those in this article, are stochastic (randomized). The objective function is not drawn randomly, but is fixed. The active information for algorithm S reduces to the log ratio of the probability p_S that S obtains an acceptable value with n evaluations to the probability p that a random sample of n candidates includes one with an acceptable value, i.e., log(p_S / p). Given that random sampling is the "average" search algorithm,^[4] active information may be thought of as a measure of how much better (or worse) matched S is to the particular objective function than than are algorithms on average.

Although the concept of active information may have legitimate uses, Dembski and Marks have subtly reintroduced the "specification" and the "complexity" of specified complexity. Fixing the objective function and the target is specification. In some of their examples, the target is minuscule in comparison to the space of candidate solutions. This makes it highly improbable that a random sample of feasible size will intersect the target, and in Dembski's past usage "complex" is equivalent to "improbable"^[1] (see specified complexity).

ThomHImself (talk) 04:13, 8 May 2008 (UTC)