Does Dembski's process for detecting Intelligent Design work in the real world?

Post of the Month: June 2009

by

Subject:    | Is Wilkins ever interested in the facts about ID Theory? Not
Date:       | 10 Jun 2009
Message-ID: | 7ae06f7e-0e86-44f0-8b97-7cb4439c511b@l12g2000yqo.googlegroups.com

John Wilkins wrote (in part):
>> Intelligent design (ID) is perhaps the most widely-discussed non-idea of
>> all time.

Tony Pagano wrote:
> The fact of the matter is that Dembski's ID theory has yet to be seriously
> addressed by any of the secular (read: atheist) community. From almost the
> beginning (1998, Dembski's peer reviewed "The Design Inference") the atheist
> tactic has been to politically smear the ID proponents as anti science and
> mischaracterize the theory as disguised theology so that it "doesn't" have
> to be seriously discussed. And in the rare instances when they have gotten
> beyond the politics of belief the atheists engage in story-telling not
> empirical science. Is Sarkar's new article in the same category?

Burkhard responds beginning with a review of some of the general literature:
Well, first I'd say there have been some quite reasonable discussions. E.g. Branden Fitelson, Christopher Stephens and Elliott Sober: "How Not to Detect Design --- A Review of William Dembski's The Design Inference. Philosophy of Science, 1999, 66: 472-488.

I can't see how they are "driven by atheist politics", arguing that

Then there is the analysis piece by Elliott Sober, Intelligent design and probability reasoning, International Journal for Philosophy of Religion 2002 which is as critical of Dawkin's abuse of statistics as it is of Dembski's.

or Peter Godfrey-Smith: Information and the Argument from Design. R. Pennock (ed.), Intelligent Design Creationism and its Critics: Philosophical, Theological and Scientific Perspectives (MIT Press, 2001, pp. 575-596) who comes to the conclusion that

In other words, Dembski is bad even measured against other criticism of Darwin.

Then there are the articles by Jeffrey O. Shallit, a computer scientists with no professional stake in the evolution discussion. e.g. his Review of "No Free Lunch": Why Specified Complexity Cannot be Purchased Without Intelligence, Rowman & Littlefield Publishers, 2002. BioSystems 66 (2002), 93-99.

More recently, there is Olle Häggström's "Intelligent design and the NFL theorems" Biology and Philosophy 2007. He too is a mathematician without professional stake in the evolution debate.

The list of course continues, Norman Johnson's "Design flaws" in the American Scientist, Ken Miller's book etc etc.

So your contention that there was no critical engagement with the work seems thin on the ground. But then you'd probably say that they are all driven by a "big atheist conspiracy" despite the fact that some of them come from outside biology and look purely at the mathematical aspects, and that some of them are Christians.

Burkhard then discusses the (non)use of Intelligent Designer detection in criminal forensics:
So I'll try another tack: Dembski claims that his method has applications outside biology, for instance in anthropology, history and forensic sciences.

Forensic science in particular carries none of the ideological baggage of the evolution discussion. The "designer" here is not the supernatural mass murderer of the bible, but very natural intelligent agents. So no problematic ontological implications, no ideological commitment to atheism or religion here. Researchers in that field are not biologists but mainly mathematicians, and if they come from other substantive fields, only a few of them (the DNA folks) have any links to biology, the rest are (forensic) material scientists, (forensic) computer scientists, (forensic) linguists etc.

So it is there that we should expect an enthusiastic embrace of his methods, IF they are sound and work. But hey, if you look at the standard books on forensic statistics, e.g. Colin Aitken's book of the same title, or David Lucy's Introduction to statistics for forensic scientists, or the Wiley Encyclopedia of Forensic Science, there is no discussion of Dembski, or indeed of any form of inference that even remotely resembles his design inference ((with one notable exception of which later)

Nor do you find a discussion of his work (or indeed papers by him) in the relevant journals in the field, e.g.

• Law Probability and Risk;
• Science and Justice;
• The international journal of Evidence and Proof;
• International commentary on Evidence;
• Jurimetrics, Forensic Science International;
• International Statistical Review;
• etc etc.

Nor do you find his method used in any of the courses on forensic statistics or forensic investigation, university or police academy based, that I know (and i know many).

Again, there is nothing contentious here, they would have nothing to lose to use his method - but they seem to agree with the Fitelson verdict that there is just nothing of value in there, not even something that is wrong in an interesting way and might be improved by discussion.

Finally Burkhard discusses two specific cases where Dembski like procedures were used:
Now if you look further at e.g. the standard books on forensic statistics, there is as I said above one exception to the general disinterest in a design inference - and that is where mistakes are discussed. In particular,

• the expert witness in the Sally Clark case,
• and the expert witness in the Dutch case against Lucia de B.

(see e.g. Ronald Meester, Marieke Collins, Richard Gill, and Michiel van Lambalgen; On the (ab)use of statistics in the legal case against the nurse Lucia de B. Law Probability and Risk 2006 5: 233-250)

What both cases have in common is that they are examples of a Dembski style design inference (even though the experts did not use his method). In both cases, the experts argued that it was just too unlikely that a series of deaths happened by chance or by regularity, and that therefore it must have been an intentional act of murder.

In both cases, they were probably wrong (we know for sure in the Sally Clark case, Lucia de B is still on appeal). Now that in itself is not that significant, science is fallible after all, shit happens. However, Roy Meadow, the expert in the Sally Clark case, was later put before his professional body and stripped of is license to practice, because the reasoning he used was so far off the normal standards to be expected from a scientist that it amounted to gross incompetence^[1].

If you then look at his mistakes in more detail, you find that they are exactly the mistakes for which Dembski has been taken apart in the above reviews.

• The main inference is unsound, the "therefore" is neither empirically nor conceptually sound. That is, even if he had been able to show that the defence theory of sudden infant death (SID) was flawed, he could not reasonably infer that therefore it must have been an intentional (designed) act. It could have been another natural cause. It could have been murder in one case, SID in another. It could have been two murderers acting independently of each other. It could have been SID, but our present knowledge of SID is insufficient (which was what it turned out to be). Several of the above reviews of Dembski make the same point, Fitelson et al in particular when they say that there are no reasons for the sequence of tests proposed by Dembski, and how he moves from rejection of one theory to the acceptance of the other.
• he never provided a probabilistic assessment of the alternative theory. True, SID is rare, but so are child murders, in particular in the sort of family under suspicion. That is a basic mistake. Relevant for the evaluation of evidence is only the ratio of the conditional probabilities for the two alternative theories. (See e.g. Lucy, Introduction to forensic Statistics, chap 8 and 9.) The same problem is noted in the reviews of Dembski who never gave the conditional probabilities for the designer hypothesis - and he freely admits it, he just refuses to draw the obvious conclusions, but instead waffles on no end.
• he ignored large parts of the evidence, and in particular the absence of evidence that any murderer would have left (e.g. trace evidence of fibres in case of smothering). In other words, he talked mainly about the prior and posterior probabilities, which are in themselves irrelevant. The important things are the likelihood ratios and the conditional probabilities from which they are calculated. We are interested in the probability of discovering certain evidence relative to the competing explanatory theories. Or in other words, math only gets you so far, if you want to learn about the world, you need to observe the world.

  His calculations for instance did not account for the fact (observation) that the parents were not alone with the children in the relevant time that there was a history of SID in the wider family, that there were no signs of physical violence etc etc. Transferred back to evolution, to argue about designers it is not that relevant how probable the mechanism is, but how likely we are to find the evidence that we found, from fossil record to the nested hierarchies, IF the postulated mechanism is assumed (P(E/H)). Dembski knows this of course and on one level acknowledges that what he proposes is an argument scheme rather than an argument - and then acts against this insight by drawing conclusions form the argument scheme (or an only partly filled-in scheme) itself.

[1] Note that he was reinstated on appeal, though Lord Justice Auld, said Meadow "was undoubtedly guilty of some professional misconduct" but that it "fell far short of serious professional misconduct".
see also: http://www.richardwebster.net/cotdeaths.html.
Meadow's own reply is here: http://www.bmj.com/cgi/reprint/324/7328/41.

[Return to the 2009 Posts of the Month]