A Capsule Summary of the Status of Dembski’s Explanatory Filter

William Dembski’s “explanatory filter” (EF) has been offered as a “rational reconstruction” of how work gets done in various scientific fields. However, it does not actually comport with such work. A direct refutation can be found in Gary Hurd’s chapter in “Why Intelligent Design Fails” from Rutgers University Press.

Taken at another level, Dembski’s EF has fundamental problems. See the paper by Wilkins and me from 2001, available online here. Dembski’s arguments for making a category of “design” a default choice fail to live up to the “rational” part of rational reconstruction. The problem of limited information is not satisfactorily handled by Dembski: on the one hand, he claimed that sufficient knowledge was in hand in 1998 to analyze the examples provided in biology by Michael Behe via his EF and that the results demonstrated “with the weight of science” that “design” was found, yet more recently he has admitted that even his one attempted explication of applying the EF to a bacterial flagellum was flawed by the problem of obtaining accurate probabilities. One wonders where the ensemble of calculations Dembski implied had already been done in 1998 had disappeared to. Despite the lack of consideration in Dembski’s framework for changes in knowledge sets driving decisions in the EF, Dembski repeated claims of absolute reliability, while inconsistently also claiming that partial function of his EF was only to be expected for a procedure in the natural sciences. Further, the issue of lack of warrant for extrapolating ordinary design inferences to rarefied design inferences has not been adequately addressed by Dembski. In “The Design Revolution”, Dembski manages only a handwave in response, saying why accept the framework in which the criticism of his work is made at all? Yet Dembski has been eager to utilize that inductive framework when he believes that it favors his argument, as in various books and articles where he claims that the successes of various “special sciences” provide support for his “rational reconstruction” via the EF. Applying Dembski’s own words to himself is apropos: “This is known as having your cake and eating it. Polite society frowns on such obvious bad taste.”

It seems obvious that despite the problems in the logic of the EF that there was something of interest in the concepts that Dembski brought up. Humans do go about distinguishing between and eventually favoring particular explanations for phenomena. So what might be at the basis of interesting cases, and how is it that explanations come to be preferred? Jeff Shallit and I took that up in an appendix to an online essay we wrote back in 2002. Therein we described the universal distribution, an application of algorithmic information theory to the problem of inductive inference, and showed how it could be cast in a way that corresponded to the tool that actually provides a rational reconstruction of work done in the sciences to achieve ordinary design inferences. We called it “Specified Anti-Information” or SAI to, so far as possible, utilize the terminology Dembski had provided. SAI differs from the EF in many important ways: it is not based on probability assessments, it is simple to apply, and it is based upon solid work in information theory. Perhaps the most important difference, though, is that the inference that application of SAI leads to is not to an overarching notion of “design”, but rather to the inference that a phenomenon is best explained as the result of a simple computational process. SAI is not burdened with the baggage Dembski loads upon his EF of not merely sorting explanatory categories, but also of standing in for an argument that would lead to an inference of an agency at work. SAI cannot, and does not attempt to, distinguish between a computational process crafted by an agent and one where no originating agent is apparent. This contrasts sharply with Dembski’s long-term fascination with a split between “apparent” and “actual” categories of “complex specified information”. For any phenomenon that might be explained as due to chance or not due to chance, any apparent success of Dembski’s EF can be more parsimoniously explained as a “pre-theoretic” approach to the far more applicable, reliable, and useful rational reconstruction of the SAI.

To summarize, the issues with Dembski’s EF are many and well-documented. Dembski’s EF fails to achieve its claimed status as a “rational reconstruction” of how humans empirically approach the problem of sorting competing explanations for natural phenomena. Better methods exist that serve as descriptions of how humans can “eliminate chance” in preferring alternative explanations for phenomena in the natural sciences.