Comments!

Visitors to Science After Sunclipse have recently left some good comments. I’d like to promote two of them to the top level and discuss a little. Replying to yesterday’s post “Michael Egnor, Reverse Engineering and Genetic Algorithms“, Matt from London said the following:

The trouble is that as far as the Egnors of this world are concerned, GAs — and indeed any artificial system that demonstrates the efficacy of variation+selection — is itself the product of design and therefore cannot possibly constitute evidence in favour of evolution. Whoever wrote the code *obviously* secretly included a complete design of the end product they were seeking. It stands to reason.

Not surprisingly, the TalkOrigins people have a lengthy page discussing genetic algorithms and what they mean. One classic creationist canard is that GAs don’t mean anything for biology and don’t prove evolution works because they have preordained goals. Like most creationist memes, this shows up all over the place; one significant example is young-earther Don Batten‘s essay for Answers in Genesis, “Genetic algorithms — do they show that evolution works?” Not to prolong the suspense, here is what TalkOrigins (in the person of Adam Marczyk) has to say about that:

Like several others, this objection shows that Batten does not fully understand what a genetic algorithm is and how it works. He argues that GAs, unlike evolution, have goals predetermined and specified at the outset, and as an example of this, offers Dr. Richard Dawkins’ “weasel” program.

The WEASEL program, described in The Blind Watchmaker, begins with a random string of letters and employs random copying errors to evolve that string into the Hamlet quotation, “Methinks it is like a weasel” (act 3, scene 2). It’s pretty easy to whip up a version in your language of choice and watch it fly.

However, the weasel program is not a true genetic algorithm, and is not typical of genetic algorithms, for precisely that reason. It was not intended to demonstrate the problem-solving power of evolution. Instead, its only intent was to show the difference between single-step selection (the infamous “tornado blowing through a junkyard producing a 747”) and cumulative, multi-step selection. It did have a specific goal predetermined at the outset. True genetic algorithms, however, do not.

That’s the essential point, but the rest is entertaining:

In a broadly general sense, GAs do have a goal: namely, to find an acceptable solution to a given problem. In this same sense, evolution also has a goal: to produce organisms that are better adapted to their environment and thus experience greater reproductive success. But just as evolution is a process without specific goals, GAs do not specify at the outset how a given problem should be solved. The fitness function is merely set up to evaluate how well a candidate solution performs, without specifying any particular way it should work and without passing judgment on whatever way it does invent. The solution itself then emerges through a process of mutation and selection.

Batten’s next statement shows clearly that he does not understand what a genetic algorithm is. He asserts that “Perhaps if the programmer could come up with a program that allowed anything to happen and then measured the survivability of the ‘organisms’, it might be getting closer to what evolution is supposed to do” – but that is exactly how genetic algorithms work. They randomly generate candidate solutions and randomly mutate them over many generations. No configuration is specified in advance; as Batten puts it, anything is allowed to happen. As John Koza (2003, p. 37) writes, uncannily echoing Batten’s words: “An important feature… is that the selection [in genetic programming] is not greedy. Individuals that are known to be inferior will be selected to a certain degree. The best individual in the population is not guaranteed to be selected. Moreover, the worst individual in the population will not necessarily be excluded. Anything can happen and nothing is guaranteed.” (An earlier section discussed this very point as one of a GA’s strengths.) And yet, by applying a selective filter to these randomly mutating candidates, efficient, complex and powerful solutions to difficult problems arise, solutions that were not designed by any intelligence and that can often equal or outperform solutions that were designed by humans. Batten’s blithe assertion that “Of course that is impossible” is squarely contradicted by reality.

My proposed structure-solving GA differs from the “weasel” program in another way, which one could crudely call the presence of morphology. If we evaluate the “fitness” of a phrase by comparing it letter-by-letter to the string, “METHINKS IT IS LIKE A WEASEL”, then the fitness function is pretty direct. No subtlety there, just subtracting ASCII values. There’s no real distinction between genotype and phenotype, but in the X-ray scattering example, the situation is quite different. There, the “genotype” is the location of molecular units in 3D space, and the “phenotype” is the 2D image (or set of images) which would be produced by X-rays scattering off those molecules in that arrangement. Those images must be calculated, and even in the case of a simple crystalline arrangement, it’s not immediately obvious how that works. The genes must coordinate the “development” of our abstract organism. This mapping is complicated and hard to turn around (which is the whole point of inventing nifty ways to “invert” the problem, anyway).

The only information we put into the program is the X-ray image of the DNA molecule we’re trying to solve — and of course any solution method whatsoever will require that much input, without doubt!

So, while like most creationist arguments the whine about “you specified a target in advance” can serve to stop the brain from thinking, it doesn’t really hold up.

The other comment I’d like to address is a reply to my conclusion:

Incidentally, there’s a good chance that Francis Crick was experimenting with LSD while he and Watson were trying to beat Linus Pauling to the DNA structure. Are we now forced to say that LSD is essential in understanding the “design inference”, that one cannot know the Great Designer without LSD, or perhaps that LSD is God Himself?

Ed Darrell had this to say in reply:

LSD would explain much of the product of the Discovery Institute. Is it in their budget? Have the trustees checked?

I don’t have the budget figures at hand, but I’d hypothesize that the DI wouldn’t be too eager to embrace that interesting molecule. After all, look what it did to the Beatles: could the plain and simple folk of the DI even acknowledge the lifestyle of the people who were bigger than Jesus? Incidentally, if you’re looking for entertainment of the morbid variety, Focus on the Family has a pet magazine called Plugged In which reviews movies, Top 40 music and so forth so that Christian parents can protect their children from the infidel. This is what they had to say about the Beatles Anthology 2, under “pro-social content”:

Nearly a dozen songs espouse upbeat sentiments about love and romance, even in cases where a relationship is struggling. “Help” and “Got to Get You Into My Life” confess the need to rely on others for support. Other tunes value people over material possessions (“And Your Bird Can Sing,” “If You’ve Got Trouble”). Fly-on-the-wall songs such as “Penny Lane” and “Being for the Benefit of Mr. Kite” honor the daily activities of common people.

John openly stated that “Got to Get You Into My Life” was about acid.

Even Paul, who was notorious for bowdlerizing his life stories, went so far as to say it was a pot song. See, these are the sort of fascinating details which you miss when you don’t actually read the Beatles Anthology book — but then again, who worries about teenagers reading books?