Conservation of arguments

William Dembski has recently made two posts (on January 10 and on January 20) at Evolution News, the advocacy site of the Discovery Institute’s Center for Science and Culture. He describes them as sections of a paper he submitted to the DI’s house journal BIO-Complexity. The paper was not immediately accepted, he said, and in the meantime he wanted to post the sections at EN.

As Dembski declares in the preface to the first of these posts, their argument in these posts is, in effect, that

Conservation of information is a big result of the intelligent design literature, even if to date it hasn’t gotten the attention it deserves. It quantifies the amount of information needed to increase the probability of finding a needle in a haystack so that the needle can actually be found. The upshot of conservation of information is that the information needed to find a needle in a haystack in turn requires finding another needle in a haystack, implying there is no free lunch in search.

In Dembski’s account, when evolutionary biologists argue that processes such as natural selection can put adaptive information into the genome, they are failing to explain where the information comes from. They are building it into their evolutionary algorithms as a detailed goal, and not informing the reader that they have done that. The biologists’ argument has thus simply displaced the question, not answered it. Dembski’s first post is illustrated by an illustration of the Shell Game, to label the argument of evolutionary biologists as a dishonest trick.

Actually, the two posts leave Dembski’s argument no further than it was some years ago. We can deal with it without going into great detail, because we have seen these arguments before. Let’s ask, and answer, some brief questions as to what was accomplished:

Do these posts clarify how his Algorithmically Specified Complexity argument deals with natural selection?

No, not at all. That argument, which was the centerpiece of Dembski and Ewert’s recent book: The Design Inference: Eliminating Chance through Small Probabilities, 2nd edition, which I discussed here recently, was left hanging by them, with the essential questions unanswered. And Dembski’s recent two EN posts do not discuss ASC at all.

Have we seen the arguments of the recent EN posts before?

Yes. Applied to evolution, the conservation of information argument is the Active Information argument of Dembski and Marks, which has been discussed here and here. It basically says that if natural selection can find genotypes of better fitness by climbing uphill on a fitness surface, that this is an increase of information in the genome. But, they argued, it is not creating new information, because the information was already there in the structure of the fitness surface.

Does the Active Information argument rule out any role for natural selection?

Not at all, as you can see.

So Dembski (and Ewert and Marks) accept that natural selection explains adaptations?

No, they are arguing that if natural selection can increase specfied information in the genome, this is not new information. But they have always been very resistant to acknowledging that natural selection does much in evolution. In the first of the two EN posts that we are discussing, Dembski grapples with Richard Dawkins’s book Climbing Mount Improbable. Dawkins gives examples of natural selection achieving major adaptations by gradual small changes. Dembski calls Dawkins’s argument “displacement”, arguing this way:

It could be that Mount Improbable is sheer on all sides and getting to the top via baby-steps is effectively impossible. Consequently, it is not enough to presuppose that a fitness-increasing sequence of baby steps always connects biological systems.

So Dembski argues that Dawkins’s examples are a misleading reflection of how well (or badly) natural selection will typically be able to do.

Connection to possible, and likely, fitness surfaces

Dembski argues that fitness surfaces that allow natural selection to work are unlikely, and their existence needs explanation:

Mountains, after all, do not magically materialize — they have to be formed by some process of mountain formation. Of all the different ways Mount Improbable might have emerged, how many are sheer so that no gradual path to the summit exists? And how many do allow a gradual path to the summit? A Mount Improbable with gradual paths to the top may itself be improbable.

Improbable in what sense? We’ll see that only a tiny fraction of all possible fitness surfaces will be smooth – but this is true only if all possible fitness surfaces are equally probable. Which they definitely are not. Smooth ones are vastly more probable.

Connection to the No Free Lunch argument

The issue has come up before, in Dembski’s 2001 book “No Free Lunch: Why Specified Complexity Cannot Be Purchased without Intelligence”. There he invokes a mathematical theorem about the success of search algorithms on all possible problems. This is the No Free Lunch theorem of Wolpert and Macready, which shows that averaged over all possible problems all search algorithms do equally well, and are no more successful than examining a random set of points and choosing the best.

Wolpert and Macready did not prove that all search algorithms will do that badly in finding large values of, say, fitness on all fitness surfaces. As Dembski’s title says, he cites Wolpert and Macready’s theorem as showing that we need to invoke intelligence, as we cannot expect that natural selection on fitness surfaces to find genomes of better fitness.

Was Dembski’s 2001 No Free Lunch argument refuted?

Yes, almost immediately, by many authors, including Richard Wein, Jason Rosenhouse, the late Mark Perakh, by Wesley Elsberry and Jeffrey Shallit (here and here), by Erik Tellgren and by Ole Häggström.

NFL and smoothness of fitness surface

When the original Wolpert-MacReady No Free Lunch theorem considers all possible problems, it has a space, such as all possible genomes. Each is assigned a value (such as a fitness), and all possible assignments are considered. In effect, the fitnesses are scrambled among the genotypes in all possible ways, all equally probable. So when we see a fitness of a genotype, what can we say about the fitness of a nearby genotype, say one which differs by a single change of a single base? Nothing. Every possible DNA sequence has its fitness assigned at random, including the nearest neighbors.

This leads to vast numbers of infinitely rough fitness surfaces, of “white noise” fitness surfaces. They have no smoothness whatsoever. Dembski’s original NFL argument is that these do not allow for much evolution. If they were typical of fitness surfaces in nature, this would be a big obstacle to the effectiveness of natural selection.

Empirical evidence on mutational effects

Such fitness surfaces are very, very atypical in real biology. Look at it this way: the effect of a single base mutation on such a surface is to carry us to a genotype whose fitness is just like that of a totally random sequence, one which differs by being mutated at 75 percent of the sites! If this were a human genome, it would be mutated at about 2,250,000,000 sites simultaneously. The idea that the effect of one mutation is the same as the effect of billions of mutations is bizarre and unreal.

So when Dembski considers “all the different ways” that a fitness surface could have emerged, overwhelmingly those are unreal fitness surfaces that would never emerge in real biological cases. Real ones are much smoother.

Tight Interaction

Not only does the equidistribution of all possible fitness surfaces imply unreal effects of mutations, it also implies incredibly tight interactions between all mutations. Consider the effect of a mutation at one site, and the effect of a mutation at another site, millions of sites away. We’ve seen that in the NFL distribution of fitnesses, these two mutations are each probably disastrous. But what about the interaction between them? What if we make both of these changes? Even if each of the two mutations happened to have a mild effect on fitness, what would be the effect if both mutations occurred? We cannot predict that the effect would also be mild. Because the fitness would, once again, be drawn totally at random, with no predictability from the effects of single mutations. The result is an incredibly strong, and unpredictable, interaction between mutations at these two sites.

But not just these two sites – at all pairs of sites, and at all combinations of sites.

No biological system works like this. Mutational changes in the genes affecting the production of earwax generally have little effect on the development of our kidneys. And for a very simple reason – these are processes separated in time and space, and in the real world it is hard for them to interact strongly. In the NFL world, every gene, by default, interacts incredibly strongly with everything else.

The Anthrop(ogenet)ic Principle

Natural selection could achieve very little if fitness surfaces were rough, white-noise surfaces. We can conclude that they aren’t those, because we have in fact evolved. This is, in a sense, like the Anthropic Principle in cosmology. We’re here, so we must have been able to evolve. Perhaps it should be called the Anthrop(ogenet)ic Principle. Of course this argument would not convince an “ID theorist” who does not believe that natural selection played a major role in our evolution.

Choice of region of the fitness surface

When different parts of a fitnes surface have different properties, the Anthropogenetic Principle argues that life evolved into a region that had sufficient smoothness. That is, it sought out evolvability. Computer simulations by Lee Altenberg and Kenneth Kinnear and Günter Wagner have shown evolution doing this,

And what about physics itself?

Even if the properties of the fitness surface have not evolved, it seems reasonable that physics itself predisposes fitness surfaces to be smooth. After all, the forces of physics tend to be local in time and space and to lose influence with greater distance. If I gesture with my fingers, there is hardly any effect on objects at a distance. The floor of our room does not collapse, nor does the ceiling cave in. Different objects are scarcely affected by each other.

Conclusion

We may summarize briefly that William Dembski’s conservation- of-information argument is not new, and has been convincingly dismissed in the past.

References

• Altenberg L 1994. Kinnear, Kenneth (ed.). “The evolution of evolvability in genetic programming”. Advances in Genetic Programming, pp. 47–74.
• Altenberg L. 1995. “Genome growth and the evolution of the genotype–phenotype map”. Evolution and Biocomputation. Lecture Notes in Computer Science. Vol. 899. pp. 205–259. doi:10.1007/3-540-59046-3_11.
• Wagner G.P., and Altenberg L. 1996. “Perspective: Complex adaptations and the evolution of evolvability”. Evolution 50 (3): 967–976.