Via the Church of Critical Thinking comes an essay by Baylor University Associate Professor William Dembski, brilliantly entitled "Why Natural Selection Can't Design Anything".

COCT cites it as an early entry into their on-line contest for the quest for great examples of sloppy logic and muddled thinking...and Dembski's essay should place high.

He starts out by classifying biological organisms as "specified" and "complex", though his definitions and examples are on extremely shaky ground. Having just written and presented a paper on complexity in the context of genetic algorithms, I'm obviously interested, and hopefully reasonably well-versed on these subjects.

Dembski starts out by trying to clarify a few concepts, including complexity. His tack exposes some of the difficulties in dealing with such terms:

A long sequence of random letters is complex without being specified (it requires a complicated instruction-set to characterize but conforms to no independently
given pattern).

I've seen this definition of complexity before, and I don't think it's very satisfying or intuitive. Dembski goes on to say that a salt crystal (or, I would suppose, any crystal) is not complex because it is formed by a repetitive lattice structure.

By this view, a rock would be more complex than a diamond, because it would take a longer and more detailed instruction set to explain the composition of the rock (made of hundreds of different minerals arranged in a haphazard manner), while the diamond could be explained as repeating, interlocking layers of single carbon atoms.

But just before that, he defines complexity this way:

Complexity here is used in the Shannon sense and denotes a measure of improbability.

But if we're using probability as a metric for complexity, then surely we'd say that diamonds are rarer than rocks. Unless we want to say that since all diamonds are similar and all rocks are unique, that rocks are more complex. But does this seem intuitive?

Personally, I think it's more useful to discuss the properties of an entity, rather than it's actual composition. Sure, all rocks may be unique, but if you walk around a rocky field in New Mexico, you'll find that many of the rocks have the same properties, even if their internal make-up may be as unique as a snowflake. Thus, we normally don't think of rocks, or globs of mud, or other random accumulations of matter as complex...such a definition doesn't seem very accurate. (Though by all means let me know if you disagree...I think it's an interesting subject.)

In the next section, Dembski goes on to discuss evolutionary algorithms, and he starts out with a quote describing a scene from the Weird Al Yankovic movie UHF (no, I am not making this up), in which a blind man is trying to solve a Rubik's cube by turning it once and asking his sighted friend "Is this it?".

He then goes on to explain that this type of blind search is massively inefficient, and that in practice, evolutionary algorithms constrain the search for solutions significantly. Well, yeah.

But then he asserts:

In fact, evolutionary algorithms are incapable of generating specified complexity de novo.

Which, if you accept his previous definitions, might be true. Let's take an example from stuff Philip and I have worked on. We tried to evolve artificial neural networks that play Tic-Tac-Toe reasonably well. When we begin each run, we start with fully-connected networks, and we randomize their weights. Once they've evolved to play better, they are generally following certain simplified rulesets (e.g., if the center is unoccupied, play there...if you have two in a row, complete three in a row...if your opponent has two in a row, block). Which is more complex behavior? The random or the ruleset? The ruleset emerges as a result of random mutation, variation, and selection for better performing networks. So is our algorithm generating specified complexity, or not?

Dembski's entire argument seems to rest on the notion that the evolutionary engineer is constraining the search by including all sorts of information in the evolutionary algorithm, so that it is not a truly blind search. Well, sure we are. We're trying to set up conditions that are favorable to allow a solution to evolve. The thing is, our simulations are always going to be crude approximations of conditions in the real world, and in many, many ways, we still don't very good ways to constrain searches or set up optimal conditions. That doesn't mean those conditions don't exist naturally.

Dembski's right that Darwinian thought doesn't explain the origin of life. It doesn't claim that it does. The origin of the first replicating molecules, as I've written about here before, is one of the single most difficult and fascinating open questions in science. Dembski falls into the same old straw man fallacy of asserting that science says it knows all the answers...to everything. No, it doesn't. Darwin didn't explain the origin of life...nobody has successfully done so yet. What he did was explain the origin of species, the variation in organisms and biological features that we see today and in the fossil record. And his explanation is very good.

What, by the way is Dembski's alternative explanation?

What, then, besides the Darwinian selection mechanism could account for biological complexity? The answer is, of course, design.

Yeah, of course. Couldn't be a rational, scientific explanation...has to be an omniscient, spectral overseer.

Dembski's writing typifies the new breed of creationists. He couches his unscientific ideas in the rhetorical garb of science, all the while avoided the "g" word (except for one use, in a footnote).

Basically though, his entire argument boils down to what Dawkins calls the argument from personal incredulity. Whether it's eyes or wings or Venus Flytraps or haemoglobin, the creationist credo is basically "Gosh, I don't understand how this works, and scientists haven't explained every last detail, so it must be the work of an ultra-intelligent spirit!"

That's not scientific, and it's not reasonable.