Impugning randomness convincingly?

I haven’t finished reading “Impugning Randomness, Convincingly” [pdf], by Yuri Gurevich and Grant Olney Passmore, but I’ll go ahead and share its remarks about our old pal William A. Dembski:

---

The idea that specified events of small probability do not happen seems to be fundamental to our human experience. And it has been much discussed, applied and misapplied. We don’t — and couldn’t — survey here the ocean of related literature. In §2 we gave already quite a number of references in support of Cournot’s principle. On the topic of misapplication of Cournot’s principle, let us now turn to the work of William Dembski. Dembski is an intelligent design theorist who has written at least two books, that are influential in creationist circles, on applications of “The Law of Small Probability” to proving intelligent design [TDI, NFL].

We single out Dembski because it is the only approach that we know which is, at least on the surface, similar to ours. Both approaches generalize Cournot’s principle and speak of independent specifications. And both approaches use the information complexity of an event as a basis to argue that it was implicitly specified. We discovered Dembski’s books rather late, when this paper was in an advanced stage, and our first impression, mostly from the introductory part of book [TDI], was that he ate our lunch so to speak. But then we realized how different the two approaches really were. And then we found good mathematical examinations of the fundamental flaws of Dembski’s work: [Wein] and [Bradley].

Our approach is much more narrow. In each of our scenarios, there is a particular trial $T$ with well defined set $\Omega_T$ of possible outcomes, a fixed family $\mathcal{F}$ of probability distributions — the innate probability distributions — on $\Omega_T$, and a particular event — the focal event — of sufficiently small probability with respect to every innate probability distribution. The null conjecture is that the trial is governed by one of the innate probability distributions. Here events are subsets of $\Omega_T$, the trial is supposed to be executed only once, and the focal event is supposed to be specified independently from the actual outcome. By impugning randomness we mean impugning the null hypothesis.

Dembski’s introductory examples look similar. In fact we borrowed one of his examples, about “the man with a golden arm” [i.e., Nicholas Caputo]. But Dembski applies his theory to vastly broader scenarios where an event may be e.g. the emergence of life. And he wants to impugn any chance whatsoever. That seems hopeless to us.

Consider the emergence of life case for example. What would the probabilistic trial be in that case? If one takes the creationist point of view then there is no probabilistic trial. Let’s take the mainstream scientific point of view, the one that Dembski intends to impugn. It is not clear at all what the trial is, when it starts and when it is finished, what the possible outcomes are, and what probability distributions need to be rejected.

The most liberal part of our approach is the definition of independent specification. But even in that aspect, our approach is super narrow comparative to Dembski’s. There are other issues with Dembski’s work; see [Wein, Bradley].

---

I’ve changed the reference numbers to tags that are meaningful to many of you. TDI and NFL are Dembski’s The Design Inference and No Free Lunch, respectively. James Bradley wrote “Why Dembski’s Design Inference Doesn’t Work” [pdf] for BioLogos, and Richard Wein wrote Not a Free Lunch But a Box of Chocolates for TalkOrigins.

Join together to report scholarly misconduct?

Edit (9 Oct 2009): Friends and respected acquaintances have persuaded me to rebut the article of Dembski and Marks in the peer-reviewed literature, and not to fuel the "Expelled" propaganda campaign of the intelligent design movement by lodging complaints of scholarly misconduct.

In my opinion, Dembski and Marks engaged in scholarly misconduct in their article Conservation of Information in Search: Measuring the Cost of Success. I will respond by lodging complaints with the academic institutions that employ the authors. I had planned on supplying readers with contact information and encouraging them to send their own letters of complaint. Now I believe that letters with many signatories would command more respect.

At most academic institutions, there are established procedures for responding to credible allegations of scholarly misconduct by faculty members. Allegations are typically reviewed by committees comprised mostly of ordinary faculty members — people who generally want to make fair decisions. Perhaps Dembski is safe. It would be nonetheless interesting to see a "secularized" Baptist university sanction one author, and a Baptist seminary let the other off the hook.

I must emphasize that the issue is not ID per se. Deceptive manipulation of a scholarly forum to advance any socio-political agenda whatsoever is wrong. As some of you know, I actually protested Baylor University's refusal to let Bob Marks display his "Evolutionary Informatics Lab" webpages with the standard disclaimer promulgated by the American Association of University Professors. What animates me to demand academic freedom is precisely what animates me to demand academic integrity.

Technical language goes nowhere with administrators and academic integrity committees. Furthermore, people may have a hard time seeing that Dembski and Marks misconstrued Dawkins. Here are some transgressions that are relatively easy to establish:

1. The authors analyze two closely-related computational methods without giving their conventional names, without indicating that they have been analyzed many times in the literature, and without citing prior analysis. This is egregious in light of the fact that their analyses did not appear in the paper until I explained one of the methods to Marks, provided its name, suggested analyzing it, and indicated that there were many analyses of it in the literature.
2. The authors mathematically formalize a computational method described informally in a popular-science book by Richard Dawkins, and attribute their formal method to Dawkins. That is, they mention neither that they are disambiguating an ambiguous text, nor that their disambiguation is highly controversial. This is egregious in light of the fact that Dembski has offered elsewhere several distinct interpretations of the text, and has emphasized its ambiguity.
3. The authors falsely attribute the term partitioned search, an apt name for their own formalization, to Dawkins. The attribution heightens the impression that they are relating straightforwardly Dawkins' precise meaning.

Speaking to motivation is tricky, but necessary, I think:

1. The authors submitted the article to a broad-scope journal with editors unlikely to recognize methods coming from the field of evolutionary computation. They evidently did not want to draw attention to the fact that parts of the paper needed the scrutiny of specialists in evolutionary computation. One of the analyses is a rehash of old results, and the other has no apparent utility in engineering.
2. Dembski has engaged in what he calls "cultural war" for many years, and the prominent atheist and evolutionary biologist Dawkins is his arch enemy. Dembski's socio-political ends take precedence over academic honesty in the article. He could not score a categorical hit on Dawkins without unequivocally representing Dawkins' work as something he and Marks knew how to analyze. In fact, the dubious interpretation makes for a very simple analysis. Following publication of the article, Dembski revealed his agenda on the Web:

   Our critics will immediately say that this really isn’t a pro-ID [intelligent design] article but that it’s about something else (I’ve seen this line now for over a decade once work on ID started encroaching into peer-review territory). Before you believe this, have a look at the article. In it we critique, for instance, Richard Dawkins METHINKS*IT*IS*LIKE*A*WEASEL (p. 1055). Question: When Dawkins introduced this example, was he arguing pro-Darwinism? Yes he was. In critiquing his example and arguing that information is not created by unguided evolutionary processes, we are indeed making an argument that supports ID.

   But soon after that, Dembski emphasized the ambiguity of Dawkins' description of his computational method, and decided that Dawkins used a method other than the one he and Marks analyzed — a method much harder to analyze. He also reported that he had communicated recently with Dawkins on the matter, and this brings to the fore the question of why he did not ask Dawkins for clarification prior to publication of the article. Evidently getting an unqualified "critique" of Dawkins through peer review was more important than honestly reporting that he and Marks had analyzed a mathematically convenient interpretation of Dawkins.

Some rumination

Dembski and Marks write, "Partitioned search [12] is a 'divide and conquer' procedure best introduced by example." Italicizing the term and placing a reference immediately after it is significant. By convention, this indicates that the term comes literally from the indicated source. There is, of course, no instance of "partitioned search" in reference [12], The Blind Watchmaker. Students might claim plausibly that they did not know the convention, but not a pair of highly experienced scholars.

There is utterly no way to warp Dawkins' description of how his Weasel program operated into D&M's partitioned search. Perhaps I'm underestimating the academic integrity committees. If they saw Dawkins' description of the Weasel program juxtaposed with D&M's description of partitioned search, they might sense that something's rotten in Texas.

How would an honest and responsible scholar go about disambiguating an algorithm in a popular science book? Obviously he would contact the author, if possible. Dembski has communicated with Dawkins plenty of times in the past, and has communicated with him about the Weasel program since publication of the article. Considering the controversy over the Weasel program, there was absolutely no justification for excluding Dawkins from the loop. This lends credence to the claim that Dembski and Marks chose to engage in false attribution because it served an ulterior purpose.

D&M did not know how to analyze the active information of Dawkins' algorithm, so they pinned on Dawkins an algorithm they felt was "close enough," and that they knew how to analyze. (I am sure that they did not know how to analyze Dawkins' algorithm because the article includes an analysis of a restricted form of it.) Whether partitioned search was "close enough" or not is irrelevant. The issue is that D&M had no justification for flat declaration that it was Dawkins' algorithm.

Back to the task

It would be good, I think, to mail a cover letter, a synopsis of the allegations like that I provided above, and somewhat detailed evidence. For instance, I would quote from my email to Marks, as in my last entry. It would be nice if Wesley were to provide a synopsis of Dembski's weaseling on the Weasel. I can give a succinct and simple explanation that D&M analyzed the (1,2)-ES and the (1+1)-ES, as well as a demonstration that they needed to keep reviewers from looking at prior analyses of ES's.

What I have in mind is to prepare the materials, put them on display for comments, revise, and then solicit signatures. I'd like to hear if you think many people would join in if I prepared something along the lines of what you've seen here. Should I just go it alone?

ADDENDUM

See Jeffrey Shallit's remarks on acknowledging priority at Recursivity.

Never look a gift weasel in the mouth

At least ten years ago, William Dembski misunderstood the description of the Weasel program in Richard Dawkins' 1986 book The Blind Watchmaker. (Take it from an experienced teacher of computer science: Some mathematicians cannot think algorithmically.) Many people have tried to persuade him that there is only one reasonable reading in the context of a book on biological evolution. His response has been to ignore context, and to play up ambiguity in his attempts to save face.

Now Dembski reports that "Oxfordensis" emailed him two Weasel programs, claiming that Dawkins ran Weasel1 in preparing the book, and Weasel2 for a demonstration in a 1987 TV show based on the book. The programs accord well with Dembski's misconceptions, so he has posted gelatinous "analysis" leading to the conclusion that "unless further evidence is presented, ... the single-mutation algorithm implemented by WEASEL1 is the one used by Dawkins in TBW."

I will save for another day the details of how Dembski got himself speared by looking a gift weasel in the mouth. Suffice it to say that both programs are easy to model formally as Markov chains, and that I, unlike Dembski and his colleagues at the Evolutionary Informatics Lab, actually bothered to do the math. (Furthermore, I translated the Weasels from Pascal to C++, and cross-checked my models and programs.) Dawkins gives an example run in which the number of correct characters in the parent goes from 3 to 22 in 19 generations. Dembski is perfectly happy to say that this is what we would expect from Weasel1. In fact, the probability that the program improves the parent so quickly (as quickly as it possibly can) is just .037.

It is quite unlikely that Dawkins used Weasel1 in preparing his book. The probability that a Weasel program like that described in Wikipedia (200 offspring per generation, mutation rate of .05) goes from 3 to 12 correct characters in generations 1 to 10, and from 12 to 22 correct characters in generations 10 to 20, is in the range of .05 to .06. (I am working on a Markov model that will yield a precise probability.) The likelihood of the conventional Weasel program, though low, is greater than that of the apocryphal Weasel1.

We cannot regard "Oxfordensis" as a reliable source of information on the program(s) Dawkins used. Thus, even though 32% of Weasel2 runs obtain the target sentence as quickly as the run in the TV show does, we must worry about a hoax. I will have more to say later as to whether a conventional Weasel program is more likely than Weasel2 to yield the run in the TV show.

Technical Details

Markov model source code: main.cpp. Modify constants according to comments, compile, and run. IOU a LaTeX document explaining the math behind the code.

Output files: weasel1.txt, weasel1-TBW.txt, weasel2.txt [8 MB]. The long lines of numbers give the probabilities of having 0 characters correct, 1 character correct, ..., and 28 characters correct after the specified number of transitions. In the "TBW" output, the initial state corresponds to generation 1, not generation 0 as in the other outputs. Thus 9, 19, and 42 transitions yield the probability distributions for generations 10, 20, and 43, respectively.