Back in October, one of us (JF) commented at Panda's Thumb on William Dembski's seminar presentation at the University of Chicago, Conservation of Information in Evolutionary Search. In his reply at the Discovery Institute's Evolution News and Views blog, Dembski pointed out that he had referred to three of his own papers, and that Joe had mentioned only two. He generously characterized Joe's post as an "argument by misdirection", the sort of thing magicians do when they are deliberately trying to fool you. (Thanks, how kind).
Dembski is right that Joe did not cite his most recent paper, and that he should have. The paper, "A General Theory of Information Cost Incurred by Successful Search", by Dembski, Winston Ewert, and Robert J. Marks II (henceforth DEM), defines search differently than do the other papers. However, it does not jibe with the "Seven Components of Search" slide of the presentation (details here). One of us (TE) asked Dembski for technical clarification. He responded only that he simplified for the talk, and stands by the approach of DEM.
Whatever our skills at prestidigitation, we will not try to untangle the differences between the talk and the DEM paper. Rather than guess how Dembski simplified, we will regard the DEM paper as his authoritative source. Studying that paper, we found that:
They address "search" in a space of points. To make this less abstract, and to have an example for discussing evolution, we assume a space of possible genotypes. For example, we may have a stretch of 1000 bases of DNA in a haploid organism, so that the points in the space are all 41000 possible sequences.
A "search" generates a sequence of genotypes, and then chooses one of them as the final result. The process is random to some degree, so each genotype has a probability of being the outcome. DEM ultimately describe the search in terms of its results, as a probability distribution on the space of genotypes.
A set of genotypes is designated the "target". A "search" is said to succeed when its outcome is in the target. Because the outcome is random, the search has some probability of success.
DEM assume that there is a baseline "search" that does not favor any particular "target". For our space of genotypes, the baseline search generates all outcomes with equal probability. DEM in fact note that on average over all possible searches, the probability of success is the same as if we simply drew randomly (uniformly) from the space of genotypes.
They calculate the "active information" of a "search" by taking the ratio of its probability of success to that of the baseline search, and then taking the logarithm of the ratio. The logarithm is not essential to their argument.
Contrary to what Joe said in his previous post, DEM do not explicitly consider all possible fitness surfaces. He was certainly wrong about that. But as we will show, the situation is even worse than he thought. There are "searches" that go downhill on the fitness surface, ones that go sideways, and ones that pay no attention at all to fitnesses.
If we make a simplified model of a "greedy" uphill-climbing algorithm that looks at the neighboring genotypes in the space, and which prefers to move to a nearby genotype if that genotype has higher fitness than the current one, its search will do a lot better than the baseline search, and thus a lot better than the average over all possible searches. Such processes will be in an extremely small fraction of all of DEM's possible searches, the small fraction that does a lot better than picking a genotype at random.
So just by having genotypes that have different fitnesses, evolutionary processes will do considerably better than random choice, and will be considered by DEM to use substantial values of Active Information. That is simply a result of having fitnesses, and does not require that a Designer choose the fitness surface. This shows that even a search which is evolution on a white-noise fitness surface is very special by DEM's standards.
Searches that are like real evolutionary processes do have fitness surfaces. Furthermore, these fitness surfaces are smoother than white-noise surfaces "because physics". That too increases the probability of success, and by a large amount.
Arguing whether a Designer has acted by setting up the laws of physics themselves is an argument one should have with cosmologists, not with biologists. Evolutionary biologists are concerned with how an evolving system will behave in our present universe, with the laws of physics that we have now. These predispose to fitness surfaces substantially smoother than white-noise surfaces.
Although moving uphill on a fitness surface is helpful to the organism, evolution is not actually a search for a particular small set of target genotypes; it is not only successful when it finds the absolutely most-fit genotypes in the space. We almost certainly do not reach optimal genotypes or phenotypes, and that's OK. Evolution may not have made us optimal, but it has at least made us fit enough to survive and flourish, and smart enough to be capable of evaluating DEM's arguments, and seeing that they do not make a case that evolution is a search actively chosen by a Designer.
This is the essence of our argument. It is a lot to consider, so let's explain this in more detail below:
As usual I will pa-troll the comments, and send off-topic stuff by our usual trolls and replies to their off-topic stuff to the Bathroom WallThe target
DEM have a "target" for which the search is searching. Except that they don't actually require that the "search" actually search for something that makes sense. The target can be any set of points. If each point is a genotype and each of them has a fitness, the target can be genotypes with unusually high fitnesses, with unusually low fitnesses, mediocre fitnesses, or any mixture of them. They do not have to be points that are "specified" by fitness or by any other criterion. DEM do not require that the "search" even consider the fitnesses. They calculate the fraction of all M points that are in the target. If |T| is the size of the target, for this fraction If we divide that by the number of points in the space, N, we get p = |T|/|N|. This of course is also the probability that a random point drawn uniformly from the space hits the target.
Searches as distributions on the space of points
DEM consider the probability distribution of all outcomes of a search. Different instances of the search can find different results, either because they choose different starting points, or because of random processes later during the search. They assume very little about the machinery of the search -- they simply identify the search with the distribution of results that it gets. Suppose that two searches lead to the same distribution of outcomes, say a probability 0.6 of coming up with point x1, probability 0.4 of being coming up with x12, and probability 0 of everything else. They consider these two processes to be the same identical search. They don't consider what intermediate steps the searches go through. Correspondingly, two searches that lead to different probability distributions of outcomes are considered to be different searches. All distributions that you can make can apparently be found by one or another of DEM's search processes. From this point on they talk about the set of possible distributions, which to them represent the set of possible searches.
Note that this means that they are including "searches" that might either fail to be influenced by the fitnesses of the genotypes, and even ones that deliberately move away from highly fit genotypes, and seek out worse ones. Anything that gets results is a "search", no matter how badly it performs.
Are "searches" search algorithms?
Mathematicians and computer scientists working on optimization are accustomed to investigating the properties of algorithms that try to maximize a function. Once an algorithm is given, its behavior on different functions can be studied mathematically or numerically. DEM do not make this separation between the algorithm and the function. Their definition of a "search" includes both the algorithm and the function it encounters. As an evolutionary algorithm may have different results on different fitness surfaces, in their argument the same evolutionary model can be two different "searches" if it encounters two different fitness surfaces. As we have noted, even "searches" that do not try to maximize the fitness are included in their space.
DEM's "Search For a Search"
A probability distribution on a set of N points simply assigns probabilities to each of them. These probabilities can be positive or zero, but not negative, and they must add up to 1. So DEM consider the N probabilities a1, a2, ..., aN. The conditions that they be nonnegative and add up to 1 forces them to lie in a region of N-dimensional space called a simplex. For example, if N is 3, the numbers must lie in an equilateral triangle in a 3-dimensional space of points (x,y,z), where x+y+z = 1, with its corners on the points (1,0,0), (0,1,0), and (0,0,1). For that small case, each probability distribution would have three probabilities, and be a point in the triangle such as (0.2344, 0.6815, 0.0841).
Now DEM discuss the distribution of searches -- that is, the distribution of probability distributions. Since each probability distribution is a point in the simplex, the distribution of searches is a distribution on that simplex. This is the probability distribution from which the search is said to be chosen. They go to a fair amount of effort, in this paper and in earlier papers by Dembski and Marks and by Dembski, to argue that a uniform distribution of searches on the simplex is a natural starting point from which the searches can be regarded as chosen. They also consider, in the DEM paper, initial distributions that are nonuniform. That does not make much difference for the argument made here. We're not going to argue with the details of their mathematics, but instead concentrate on what in evolutionary biology corresponds to such a choice of a search.
Their theorem
When one draws a probability distribution, which is one of the points in the simplex, one might get one that assigns a higher probability to the target, or one that assigns a lower probability of the target. On average, they argue, one gets one that has the probability p of hitting the target. DEM show that, in the original uniform distribution of searches, at most a fraction p/q of them will have a probability of finding the target as large as, or larger than q.
They then calculate a quantity that they call "active information" by taking the negative logarithm of this ratio and conclude that this is the amount of information that is built in by the choice of that search. In their argument it is implied that the improved success is due to some Designer having made choices that built that information in.
Mostly not using the fitness.
In Joe's earlier post, he argued that Dembski and Marks were examining the choice of a fitness surfaces from among all possible fitness surfaces. He was wrong. In fact, most of the searches in their distribution of searches cannot involve going uphill on any fitness surface. One is already in a very small portion of their distribution of searches as soon as the process is doing that. In that case one has an evolutionary search, and that is drawn from a very small fraction of all of their searches. Here is how we can see that.
A simple "greedy" search algorithm
Evolutionary processes occur in populations of organisms that have genotypes and fitnesses. Will a situation like that do as badly as a randomly-chosen search, where the probability of hitting the target is the same as it would be for random draws from the space? We can make a simple model, which easily shows that it is not the same.
Consider a space of DNA sequences, say all possible sequences of a stretch of 1000 nucleotides. The organism has one of these DNA sequences. In each generation it looks at all of its neighbor DNA sequences that have just one of these 1000 bases changed from the present sequence. There are 3000 of these, since each of the 1000 bases has one of the four bases A, C, G, and T and this means that there are 3 others possible at that site. Each DNA sequence has a fitness. Let's assume that the organism has just one DNA sequence, so it is located at one point in the genotype space. If the most fit of these 3000 neighbors has a higher fitness than the present DNA sequence, let's assume that the organism changes its DNA sequence to that DNA sequence. Otherwise it stays the same. It goes through m-1 generations of this.
This of course is a very simpleminded model of an evolving population, one that looks only at the neighbors of one genotype, but which also responds perfectly to any fitness differences. The question is not whether this is fully realistic, but whether this simple biasing by natural selection has a major effect on the probability of hitting the target. Let's call this beast a Greedy Uphill Climber "bug". We introduce it because it is easy to see what it will do.
Searching for a small target
To make the case even simpler, let's assume that all the genotypes have different fitness values -- there are no ties. There is then only one genotype that has the highest fitness. For our test case, let's define that one as the target T. In DEM's argument, the target can be defined in any way you want. It could even be a set of genotypes of unusually low fitness. But as the issue for evolution is whether natural selection can find highly-fit adaptations, it does not make sense to have a target that has unusually low fitness, especially since natural selection will actively move away from it.
Let's also simplify things by choosing the starting genotype at random from among all possibilities. Our GUC Bug then makes m steps, each time to the most fit of the 3001 sequences that consist of its own genotype, plus the genotypes of its 3000 current neighbors.
Probability of the GUC Bug finding the target
Remember that if we drew at random from a distribution (a "search") which itself was randomly chosen from the simplex of all probability distributions, we would have only a probability p of hitting the target. That is the same as if we just drew the outcome randomly from the set of possible DNA sequences. In the case of our GUC Bug, we start out with a randomly sampled genotype, and if that were all we did, we would have that small probability of hitting the target.
But if we let the bug do just one more step, so m = 2, it will move to the fittest of the 3001 immediate neighbors. This mimics the effect of natural selection, and that makes us much more likely to hit the target. The GUC Bug will find the target if it starts with the genotype which is the target, or if it starts with any genotype that is an immediate neighbor of the target. As there are 3000 neighbors of each of these DNA sequences, the probability of hitting the target will be about 3001 times greater than p.
If we take more steps, it is not clear how much larger is the set of starting points that will allow us to arrive at the target. It depends on how smooth the fitness surface is. At its smoothest, the fitness surface has no local peaks. For each genotype outside the target, there is a best neighbor of higher fitness, so the GUC Bug will move to that neighbor. If m = 50, there will be a great many neighbor genotypes that are less than 50 steps away from the target. In fact, there will be 1.211Ã10107 of those neighbors in all. That's a lot. All of those genotypes are starting points that will lead to T in 49 steps or less. So the probability of a GUC Bug reaching the target is not just p, in the most favorable case it is vastly larger than that.
Behavior on a "white noise" fitness surface
One of us (TE) has carried out computer simulations of this case. He considered 1000-base nucleotide sequences and a GUC Bug started at a random sequence. Running the bug until it reached a local peak of the fitness surface, where no immediate neighbor is more fit, he found that these peaks were typically higher than 99.98% of all points. So even on one of the worst possible fitness surfaces, a GUC Bug does far better than choosing a DNA sequence at random.
Can DEM's "searches" all be carried out by a greedy search bug?
This immediately establishes that most of the searches in DEM's space of searches are much worse at finding the target T than any search that has a GUC Bug and a fitness surface. In our case the average chance of success of one of their searches is only p, which is more than 3000 times lower than the average for a GUC Bug that looks at neighbors on a fitness surface once. So a GUC Bug moving on a fitness surface must be far more successful than a random one of DEM's searches. This is true no matter what the fitness surface is. Simply by having a process that moves to more fit neighbors, we immediately narrow down DEM's searches to a tiny fraction of all possible searches.
But what about more realistic models of evolution?
These have the same property. In the GUC Bug model, we had only one DNA sequence in the species. If instead there is a population of sequences, then the genotypes of the species have multiple DNA sequences, and by multiple mutations and recombination parts of the space further afield can be reached. On the other hand the GUC Bug is more efficient in moving uphill to more fit genotypes than actual evolutionary processes are. So more realistic models of evolution might be either better or worse at climbing the fitness surface. But all of them move to the target from some reasonably large set of points in the neighborhood of the target. All such models will end up at the target far more often that a blind search will, and that immediately signals that these processes are far different from most of the searches in DEM's space of searches.
What causes smooth fitness surfaces?
We can see that evolutionary processes are not typical members of DEM's space of searches, because all of them, no matter what the shape of their fitness surface, do much better than blind search. Within the class of evolutionary processes those that have smoother fitness surfaces do better yet -- enormously better. DEM acknowledge this but do not discuss what makes fitness surfaces smooth. As one of us (JF) argued in his previous posts (here, here, here, and here), the ordinary laws of physics, with their weakness of long-range interactions, lead to fitness surfaces much smoother than white-noise fitness surfaces.
In the white-noise surfaces, changing one base in the DNA brings us to a fitness that is in effect randomly chosen from all possible fitnesses. In fact, it brings us to a fitness that is just as bad as if all bases in the DNA were changed simultaneously. That is not like actual biology. Furthermore in a white-noise fitness surface interactions among changes in different sites in the DNA are ubiquitous and incredibly strong. Changing one base leads to a randomly-different fitness. So does changing another. Changing both of those leads to a fitness that is also randomly-chosen, without regard to what the effects of the two earlier changes were. Combining two deleterious changes will then make no prediction that the result will be even more deleterious. Similarly, combining two advantageous changes will make no prediction that the result will be even more advantageous. But with real physics, those predictions can often be made.
Thus we can see that simply having genotypes with different fitnesses leads to results much better than most of the searches in DEM's space. Considering that "because physics" the fitness surfaces will be nonrandomly smooth brings us to an even tinier fraction of all possible searches, ones that are even more successful. Dembski and Marks would consider these smooth fitness surfaces to have large amounts of "active information", because they lead to much greater success at reaching any target which includes the genotypes of highest fitness. So these two effects do not require any intervention of a Designer, just the presence of genotypes that have fitnesses, and the action of ordinary laws of physics. Some, quite possibly all, of Dembski and Marks's "active information" is present as soon as we have genotypes that have different fitnesses, and genotypes whose phenotypes are determined using the ordinary laws of physics.
Is evolution a search?
The modeling of evolutionary processes as searches is of limited help. It is generally not best to regard evolutionary processes as carrying out a search for a target which is an optimal organism.
Evolution does not withhold its approval until it sees whether the single most-fit possible phenotype is found. Whether a species goes extinct depends on its fitnesses along the way, and a species can be quite successful without ever finding the most-fit genotypes. It is almost certain that we are not as fit as the best organism possible anywhere in in our space of genotypes. Requiring that evolution find that optimum result is unreasonable; we may always be stuck in some isolated region of genome space, and all of our wonderful adaptations may be the ones found there. But that is good enough for us to have developed remarkable abilities, including being capable of analyzing arguments about the evolutionary process, and seeing whether they imply the existence of the intervention of a Designer in the evolutionary process. Or whether they do not.
167 Comments
Joe Felsenstein · 29 March 2015
For those who were viewing PT in the first hour after I posted this, apologies for the chaos, which was due to my own mishandling of the editing, plus WordPress's wierdness.
Now it seems to be as Tom and I wanted it. It is a big, long, somewhat tedious argument, but it is our fairly-serious evaluation of Dembski, Ewert. and Marks's recent arguments, so we think it is appropriate here.
Joe Felsenstein · 29 March 2015
Let me abstract the whole post quickly, for those readers who are busy:
Dembski, Ewert and Marks have presented a general theory of "search" that has a theorem that, averaged over all possible searches, one does not do better than uninformed guessing (choosing a genotype at random, say). The implication is that one needs a Designer who chooses a search in order to have an evolutionary process that succeeds in finding genotypes of improved fitness.
But there are two things wrong with that argument:
1. Their space of "searches" includes all sorts of crazy searches that do not prefer to go to genotypes of higher fitness -- most of them may prefer genotypes of lower fitness or just ignore fitness when searching. Once you require that there be genotypes that have different fitnesses, so that fitness affects their reproduction, you have narrowed down their "searches" to ones that have a much higher probability of finding genotypes that have higher fitness.
2. In addition, the laws of physics will mandate that small changes in genotype will usually not cause huge changes in fitness. This is true because the weakness of action at a distance means that many genes will not interact strongly with each other. So the fitness surface is smoother than a random assignment of fitnesses to genotypes. That makes it much more possible to find genotypes that have higher fitness.
Taking these two considerations into account -- that an evolutionary search has genotypes whose fitnesses affect their reproduction, and that the laws of physics militate against strong interactions being typical -- we see that Dembski, Ewert, and Marks's argument does not show that Design is needed to have an evolutionary system that can improve fitness.
DS · 29 March 2015
It is ironic that Dembski seems to demand that evolution work as though it is an intelligent designer, searching for an optimal solution. It is not. Why must he continue to display his ignorance of basic biological principles? Why must he continue to ignore those who ;point out his errors? What does he hope to gain by continued obfuscation? Does he really think that he is fooling anyone? Why must he misrepresent the way in which natural selection works? Is it because he knows that an accurate representation would eviscerate his argument? Does he really think that assuming your conclusions and trying desperately to develop some twisted mathematical mumbo jumbo to vindicate your preconceptions is productive in any way? Give it up Bill, you has been outed!
Tom English · 29 March 2015
I'm responsible for several months' delay in this post. Joe was ready to go in mid-December. There's been no substantive change since then.
Mike Elzinga · 29 March 2015
Anything that Dembski, Ewert, and Marks can come up with is excruciatingly boring, amateurish, and totally irrelevant compared with the really interesting things that are going on in chemistry, biology, and physics.
ID/creationists always start with sectarian dogma as an implicit, if not explicit, goal of their "science;" everything else is bent and broken to fit dogma. Atoms and molecules are modeled by inert objects such as ASCII characters, dice, junkyard and battleship parts, and coin flips. It never seems to occur to any of them that atoms and molecules have electric charge and interact strongly according to quantum mechanical rules. It never seems to occur to them that biological organisms interact strongly with environments that constrain what they can become.
"Advanced mathematics" to an ID/creationist is high school logarithms to base two; with the result labeled as "information" to obscure the fact that a simple multiplication is taking place. Taking log2 of the product Np of the number of trials, N, by a probability per trial, p, and calling it "information" doesn't suddenly change the product or the concept into something "advanced and profound" and "impossible." It's an amateurish bastardization of and misrepresentation of other work going on in the areas of computer science.
In the real world of science, calculations start with the well-studied properties of atoms and molecules, or with the well-studied properties of complex organisms interacting with their environments. The fact that supercomputers are required to do these kinds of calculations at the level of chemistry and physics should be a reminder of the vast differences between what scientists actually do and what people like Dembski, Ewert, and Marks are doing.
harold · 29 March 2015
Kevin · 29 March 2015
I wrote this a while back: http://www.skepticink.com/smilodonsretreat/2015/01/21/winning-vs-not-losing/
Basically, that last paragraph. The optimum gene doesn't necessarily matter... it's just an allele that is good enough so that the organism doesn't die and can reproduce.
Joe Felsenstein · 29 March 2015
Joe Felsenstein · 29 March 2015
An interesting distinction is between DEM's "searches" and the more conventional notion of a "search algorithm". I can define the latter as a series of operations we do to move in a space of points, each of which has numbers that arise from some function evaluated at those points. Given a search algorithm, we can ask what its behavior is on a given function (a given fitness surface, for example).
But DEM's "searches" have the unusual property that they are the result of applying a given search algorithm to a given surface. Thus when we consider a different surface, we may have to call the result a different "search". So the "searches" encompass all possible algorithms used on all possible fitness surfaces. They describe the "search" only in terms of the distribution of results, so it is also possible that two different search algorithms, on two different surfaces, get the same distribution of results and thus are the same "search".
It is important to keep this in mind when trying to understand their set of possibilities and when trying to relate it to more conventional search algorithms.
fnxtr · 29 March 2015
TomS · 29 March 2015
Is there any hint as to what sort of algorithm would work better in the world of life on Earth?
I gather that DEM's investigations are about searches which are limited by obeying the laws of nature. If they were to include in their analysis searches with more freedom than the fine-tuned parameters of nature allow, would there be a different result?
Mike Elzinga · 29 March 2015
Mike Elzinga · 29 March 2015
eric · 29 March 2015
Joe Felsenstein · 29 March 2015
Keep in mind that having an adequate fitness is not what natural selection favors. A genotype for having a fitness, say, 10% more than that will be favored by natural selection, even if the former genotype was in some sense good enough.
However DEM in their papers have a target (T) for evolution and count whether or not that target is reached. An organism can do quite well without finding the best possible genotype. Which is fortunate for us, since we're almost certainly not that optimal genotype.
TomS · 30 March 2015
Frank J · 30 March 2015
Joe Felsenstein · 30 March 2015
Nick Matzke · 30 March 2015
typo in here somewhere: "This shows that even a search which has is evolution"
Joe Felsenstein · 30 March 2015
TomS · 30 March 2015
Joe Felsenstein · 30 March 2015
TomS · 30 March 2015
Joe Felsenstein · 30 March 2015
harold · 30 March 2015
TomS · 30 March 2015
Mike Elzinga · 30 March 2015
callahanpb · 30 March 2015
Flint · 30 March 2015
I may be missing all the nuances here, but what I'm reading boils down to "let's assume evolution works according to some silly and unrealistic assumptions. Then let's show that it can't work that way. Then let's conclude that the Designer must have done it."
And what this says (as has been pointed out), untangles as "Let's assume the Designer did it. Then let's confect a hopelessly misguided model of how it might work without the Designer. Then let's show that this stupid model must be wrong."
I personally suspect that few normal creationists would bother trying to make any sense of DEM or anything similar. Their approach is probably tougher to counter: "Goddidit, I believe it, go away."
callahanpb · 30 March 2015
bigdakine · 30 March 2015
Matt Young · 30 March 2015
Carl W · 30 March 2015
It seems strange to say that you don't get white-noise fitness surfaces because of physics -- white-noise fitness surfaces are definitely possible with our physics (but with a very different biology). For instance, instead of ribosomes, there could be a system that takes the CRC32 of a gene and then uses the resulting number as a DES key to decrypt the gene; then any mutation anywhere in the gene would effectively change the gene entirely. (With this variant biology, evolution would be essentially impossible, of course.)
So I would say that smooth fitness surfaces are because of biology (in particular, how ribosomes work, and the fact that proteins with similar sequences will often have similar function), rather than because of physics.
Also, if I were Designing life on earth from scratch, it seems like it would be a good idea to use this kind of system and disable evolution entirely... why would a hypothetical Designer set things up so that random chance would degrade a (hypothetically) initially-perfect Design?
Mike Elzinga · 30 March 2015
TomS · 30 March 2015
Mike Elzinga · 30 March 2015
Mike Elzinga · 30 March 2015
For the image and signal processing geeks out there, here is another slant on the smoothing of potential wells.
Consider the so-called "shift theorem" of Fourier transforms.
That phase factor that shows up in front of the Fourier transform as a result of a shift in the function being transformed is dependent on the frequencies in the function; higher frequencies are shifted more than lower frequencies.
The concept of "dithering" is used in signal processing as a type of "poor man's filter" to smear out high frequencies and preserve the lower frequencies in a function. Shifting the function back and forth randomly within a small interval "washes out" the higher spatial frequencies in the function. Then the inverse transform retrains the lower frequencies and the high frequencies that represent "sharp edges" are gone and so are the sharp edges.
Collections of molecules and their mutual potential wells are always in thermal motion.
keiths · 30 March 2015
eric · 30 March 2015
Tom English · 30 March 2015
Mike Elzinga · 30 March 2015
Matt Young · 30 March 2015
The equation is (a + bn)/n = x. I have never seen this anecdote before, so I looked it up. Jeff Shallit says it is bunk, as does the Wikipedia entry for Euler. Shallit, however, attributes the anecdote to an author who Wikipedia thinks merely embellished it. The equation, I assume, is meaningless; is that so?
phhht · 30 March 2015
callahanpb · 30 March 2015
Mike Elzinga · 31 March 2015
Joe Felsenstein · 31 March 2015
DS · 31 March 2015
Joe Felsenstein · 31 March 2015
TomS · 31 March 2015
DS · 31 March 2015
Joe Felsenstein · 31 March 2015
TomS · 31 March 2015
Matt Young · 31 March 2015
DavidK · 31 March 2015
Many kudos to those who understand the bilge that Dembski, et. al are putting forth. It's just far too deep for my hip boots to wade through the muck.
callahanpb · 31 March 2015
Mike Elzinga · 31 March 2015
TomS · 31 March 2015
Carl W · 31 March 2015
Joe Felsenstein · 31 March 2015
callahanpb · 31 March 2015
TomS · 31 March 2015
Mike Elzinga · 31 March 2015
Carl W · 31 March 2015
Mike Elzinga · 31 March 2015
phhht · 31 March 2015
TomS · 31 March 2015
Carl W · 31 March 2015
eric · 31 March 2015
Henry J · 31 March 2015
So evolution uses fuzzy logic?
Tom English · 31 March 2015
Mike Elzinga · 31 March 2015
Steve Schaffner · 31 March 2015
Mike Elzinga · 31 March 2015
Joe Felsenstein · 31 March 2015
callahanpb · 31 March 2015
Mike Elzinga · 31 March 2015
harold · 1 April 2015
I certainly hope that everyone here absolutely understands that all Dembski has done, is to have deliberately built a computer model that doesn't model evolution correctly, and then to have claimed that real biological evolution can't explain the diversity and relatedness of the biosphere without adding magic, because his bad model doesn't do that.
I agree with Joe Felsenstein that unconscious compulsion and authoritarian thinking, rather than conscious deceptiveness, are Dembski's probable motivations.
There's a lot of great stuff here about search algorithms and related topics, but an especially easy way to look at the whole thing is to note that we have a great idea of how evolution fundamentally works, and good models can be and are built.
Within cells or cell-like environments, which means within a certain narrow range of physical conditions/energy levels, and yes it's narrow even if you include extremophiles and you won't be finding any of them on Mercury, DNA replicates in a certain way and codes for proteins in a certain way. This type of replication leads to a certain level of variation in "offspring" sequences, relative to parent sequences. This nucleic acid variation, along with vagaries of development and whatnot, leads to phenotypic variation. The phenotypic variation may lead to selection within an environment. Genetic variation can and will spread through populations in a random, stochastic way, but alleles associated with a phenotype that is selected for or against in a given environment may change in frequency more rapidly than would be the case due to random variation alone. The primary end effect that we see in the modern biosphere is extensive niche adaptation, although a few highly adaptable generalist species, mainly us and some species we complain about, are also widespread.
That's how it works, and that can be modeled. I could write a model in pseudocode, learn an easy language, and set up a half decent computer model of biological evolution with a few months, while working full time at my job. And so could you. Some people reading this could do it a lot faster than I could (caveat: don't go Dunning Kruger; knowing about the biomedical stuff and learning to program usually makes for a better model than knowing how to program and trying to model something you haven't fully studied). Some people reading this do construct excellent computer models of evolution on a regular basis.
Dembski has some motivation to build a bad model and make false claims, and to do it in a verbose way that forces detailed critique, but in the end, he's just building a bad model even though anybody can see how to build a good model. Because of his biases.
TomS · 1 April 2015
Joe Felsenstein · 1 April 2015
Harold, I did not say anything about "authoritarian thinking", and my comments about "unconscious motivation" did not extend beyond saying that the main person who Dembski had fooled was himself.
I'd also point out that Dembski (and Ewert and Marks) have not put forward a "computer model", but have presented mathematical theory of how effective typical "searches" will be. The argument in the original post of this thread was that simply having organisms that reproduce and have fitnesses gives us searches that are far more likely to result in higher fitness than are the randomly chosen searches of DEM's theory. And that typical fitness surfaces are smoothed by physics, which makes strong interaction among the products of all genes in the genome unlikely, and thereby makes the fitness surface easier to climb.
The discussion here about optimization and evolution has been interesting. I would add one consideration. Just because one or a few individuals in a population have found a new peak in the fitness surface does not mean that the population will go there. These individuals have to mate with other members of their population, and the offspring may not be on or near the new peak. Such considerations are central to Sewall Wright's 1932 Shifting Balance Theory (readers might enjoy reading his paper here). About 33 years ago I sent Sewall Wright a copy of a paper in which I discussed a result of his. He sent me a reprint of his famous 1932 paper in return. I was astonished -- surely that was the last available reprint of that famous paper. A few years later I got suspicious -- I heard of someone else later on also getting a 1932 reprint from Wright. I later heard from Wright's scientific executor, Will Provine, that Wright ordered new reprints of that paper repeatedly, and left a stack of them among his papers when he died in 1988.
If anyone here wants to discuss the original post by Tom and I, that is possible too.
Joe Felsenstein · 1 April 2015
TomS, it is true that Dembski's models of evolution (in this case with Ewert and Marks) don't work, in the sense that they fail to achieve substantially higher fitnesses. The issue that should concern us is that they are presented as having behavior typical of evolving populations.
The point that Tom and I have made in the original post of this thread is that the set of "searches" that DEM have considered is far too broad. It includes processes that descend the fitness surface, and lots of processes that ignore the fitness surface or go sideways on it instead of up. But as soon as one requires that there be organisms that have genotypes and have fitnesses that affect their survival and reproduction, one has narrowed down the set of searches considerably, and they are now much more successful than found by DEM's theory.
Since DEM's "Active Information" measures how much more successful the search is than typical members of their set of searches, just having organisms with fitnesses gains us a lot of Active Information. And in fact, it gains that Active Information on all possible fitness surfaces, not just on one particular one.
A similar issue arose when Dembski published No Free Lunch in 2002. His use of the NFL theorems implicitly assumed that fitnesses of neighboring genotypes were uncorrelated, so that the fitness surface was a "White Noise" surface. Evolution can improve fitness on such a surface, as Tom's simulation study described in our post showed. But it cannot effectively find the highest peaks. Once one allows the ordinary laws of physics, the fitness surface becomes much smoother and easier for the population to climb. This was immediately pointed out by Dembski's critics (the point was first made in 2002 by Richard Wein and by Jason Rosenhouse).
The lack of success of evolution in Dembski's models is central to his argument (and DEM's argument). Their arguments fail, not because the models do not evolve higher fitnesses, but because they are not typical of the behavior of reproducing populations in a world that has the rules of ordinary physics.
Tom English · 1 April 2015
It's unlikely that I'll ever contribute as much as you have, Mike. But I am doing what I can. Dembski and Marks have linked me to "conservation of information" several times. I was wrong to use the term. And so are they, for similar reasons. I'll have a strong opening paragraph for an amicus brief, should I be around for the next trial. Joe has taught me some valuable lessons in plain-language exposition.
callahanpb · 1 April 2015
gdavidson418 · 1 April 2015
Mike Elzinga · 1 April 2015
harold · 1 April 2015
Mike Elzinga · 1 April 2015
bigdakine · 1 April 2015
bigdakine · 1 April 2015
eric · 2 April 2015
eric · 2 April 2015
keiths · 2 April 2015
It's worth emphasizing that the active information arguments are a huge step back for Dembski. His earlier CSI arguments were that evolution is inadequate to explain the diversity and complexity of terrestrial life. Now, he's merely saying that if unguided evolution is responsible, then it succeeds by virtue of the active information already inherent in nature -- information that must have been put there by You Know Who.
It's perilously close to what theistics evolutionists believe. I'm surprised that Dembski doesn't get more flak from his fellow IDers on this score.
Just Bob · 2 April 2015
Mike Elzinga · 2 April 2015
Just Bob · 2 April 2015
... and he's a rill perfessor! At a college and all!
Mike Elzinga · 2 April 2015
TomS · 2 April 2015
keiths · 3 April 2015
Tom English · 3 April 2015
keiths · 3 April 2015
And all of it based on probabilities he can't even begin to estimate for real-life evolutionary scenarios.
Tom English · 3 April 2015
callahanpb · 3 April 2015
I want to add one general thought about Dembski's work that has always been the kicker for me, the reason it's obvious to me as a computer scientist (and not a biologist) that he's full of it.
He's essentially claiming an intractability proof. He is saying that whatever it is evolution does, this is insufficient to produce the results we see. And he claims to have proven this mathematically.
One insight from over a half century of theoretical computer science has been that intractability proofs are notoriously difficult (as in the most brilliant researchers have been banging their heads for decades to little avail). Even problems that sound like they ought to be intractable, such as traveling salesman, are only conjectured to be intractable. "Obvious" lower bounds on time complexity are almost never provable, and most really come down to the trivial point that you have to scan the entire input (linear time lower bound). Even a slightly more sophisticated lower bound like Omega(n log n) for sorting, using a decision tree argument, is not too far from saying you would need to look at the entire input if it was encoded in bits (we can quibble over whether that's a good general characterization, but it covers a useful subset).
So if Dembski was really the "Isaac Newton of Information Theory" maybe he could shine a little bit of his brilliance on questions like whether P=NP. Unlike evolution, the problem is well-defined mathematically, and subject to little political or religious controversy. I cannot imagine why anyone would imagine Dembski had the tools to claim rigorous results about what is impossible in the real world when he has never published an intractability proof in a well-defined domain.
Mike Elzinga · 3 April 2015
TomS · 3 April 2015
The indescribable as the explanation where there is nothing needing an explanation.
Paul Burnett · 3 April 2015
Tom English · 3 April 2015
Mike Elzinga · 3 April 2015
Mike Elzinga · 3 April 2015
Henry J · 4 April 2015
That sounds like one disgruntled pig!
Just Bob · 4 April 2015
Mike Elzinga · 4 April 2015
Maybe pudding is the process of staying pud.
gdavidson418 · 4 April 2015
Frank J · 6 April 2015
eric · 6 April 2015
TomS · 6 April 2015
Just Bob · 6 April 2015
Frank J · 6 April 2015
Mike Elzinga · 6 April 2015
I keep going back to my previous assertions that Dembski's - and all ID/creationist's - problem is related to the incredible, mush-headed "discussions" about coins and the second law of thermodynamics that are going on between Uncommon Descent and The Skeptical Zone for the last few weeks. Not one of these characters - including most of those arguing with them - has a clue about the underlying physics, chemistry, and biology.
Ever since Morris and Gish, ID/creationists have been absolutely sure that there is something about the second law that precludes evolution and the origins of life. At the moment, it appears that the current crop of ID/creationists think it has something to do with statistics and the "Law of Large Numbers." None of them seems to have a clue.
When Dembski thinks that finding a "solution" to a complex molecular assembly involves some kind of Easter egg hunt; and then, of all things, an Easter egg hunt for a strategy for doing an Easter egg hunt for a strategy for doing an Easter egg hunt for a strategy for doing an Easter egg hunt for â¦, etc. ad infinitum, he has plunged himself into a hopeless abyss of self-referential, and self-imposed ignorance. He is avoiding the science and is giving his followers "reasons" to mud wrestle endlessly over totally irrelevant "philosophical and semantic issues."
It is this endless mud-wrestling with "The Enemy" that is providing the political "rationale" and motivation for the sectarian grass-roots pressure on legislators and school boards. It maintains the appearance of a substantive debate doing on that is being suppressed by the scientific community.
In reality, very few scientists, if any, are taking ID/creationists seriously any longer. ID/creationists are a bunch of loons being debated by a few amateurs who simply like to mud-wrestle with them for the fun of it. But there is no science being elucidated or learned in any of it. And it is this kind of mud-wrestling that would shut down public science education if it ever got into the classrooms.
stevaroni · 6 April 2015
Just a few hours ago Duke beat Wisconsin to win the NCAA championship, and, once and for all, finalized "The Bracket" for this year.
Now, of course, this happens every year, but here's the really freaky part of this year's championship - there are 2^63 possibilities for the possible winners in a 64-team NCAA bracket.
And Duke won with this exact combination!
Using Dembski-math, that means that Duke beat 9.2 quintillion to 1 odds to win this championship!
Good job Blue Devils, I didn't realize right until this very moment what an incredible underdog you guys were.
Mike Elzinga · 6 April 2015
TomS · 7 April 2015
What it looks to me is like this:
Someone proposes a law of nature. He points out that nature does not obey that law. And then says that because nature does not obey that law, there must be explanation.
Frank J · 7 April 2015
harold · 7 April 2015
eric · 7 April 2015
Mike Elzinga · 7 April 2015
Mike Elzinga · 7 April 2015
scienceavenger · 7 April 2015
Whenever I see a paper like this, I imagine a man sitting in an airport revising his proof that heavier-than-air flight is impossible.
scienceavenger · 7 April 2015
bigdakine · 7 April 2015
callahanpb · 7 April 2015
TomS · 7 April 2015
callahanpb · 7 April 2015
Henry J · 7 April 2015
But that "heat death of the universe" thing is over a time span a good bit larger than the current age of the universe, isn't it? Is it hundreds of billions, or is it trillions of years? So that has nothing to do with things that occur on shorter time scales.
callahanpb · 7 April 2015
eric · 7 April 2015
Mike Elzinga · 8 April 2015
eric · 8 April 2015
Richard B. Hoppe · 8 April 2015
Coming late to this discussion, I'll merely repeat what I've been saying for years now: Treating biological evolution as a formal search process is a snare and a deception. DEM are both snared and (self-)deceived.
Mike Elzinga · 8 April 2015
callahanpb · 8 April 2015
Mike Elzinga · 9 April 2015
TomS · 9 April 2015
I've tried to point out that many of the arguments against evolution apply at least as well against reproduction.
That doesn't seem to work, probably because it is taken as merely ridicule. But it is true. In the Wikipedia article on "Irreducible complexity" it is pointed out that the same language was used in the 1700s precisely to argue against reproduction. No joke.
How can one take Intelligent Design seriously when they are using concepts which are paradigms of rejected ideas: elan vital, homunculus, caloric, ...
callahanpb · 9 April 2015
Mike Elzinga · 9 April 2015
Even more devastating for the ID/creationist is the fact that, for many species, development within an egg or seed is temperature dependent. For example, the sex of a developing embryo in some species is determined by temperature.
But ID/creationists have no clue about what temperature is and what it has to do with the processes that are taking place within the cells of developing organisms. You won't find temperature anywhere within an ID/creationist's "calculations" of the "probabilities" of molecular assemblies. That concept has gone over their heads for the last 50 years; and they still don't get it when it is pointed out to them.
Henry J · 9 April 2015
So what you're saying is that they can't stand the heat?
Mike Elzinga · 9 April 2015
prongs · 9 April 2015
callahanpb · 9 April 2015
I want to add that I never meant to slight Asimov or The Last Question. What bothers me is the use of the 2nd law as shorthand for something like the inexorable march of decay. This use of "entropy" is so commonplace it was even mocked in a mainstream movie like The Four Seasons (1981). One of the middle aged characters: "I have shifted into a state of entropy that's progressing geometrically." In that case, it's actually just the use of metaphor that was already established rather than an explicit reference to the 2nd law. But the term itself was coined purely as a thermodynamic concept (by Rudolph Clausius in 1865 if I can trust Wiktionary).
There is a certain strain of sophomoric wisdom that holds that because the universe itself must have some end (or assuming it does) then everything we do is pointless. I know this firsthand (from my sophomore days of course). I am also familiar with the kind of false-knowing reference to the "2nd law of thermodynamics" as some kind of explanation for why things often seem to be in a state of decay.
And I don't deny that there is constant decay going on around us, though I would counter that there is also constant renewal. It may even all come to an end one day, and the explanation might even have something to do with thermodynamics (not sold on that one). However, this is absolutely irrelevant to the important question of how we decide to live today.
So while I don't blame Asimov, who certainly understood the implications of the potential "heat death" of the universe, I do blame people who take up the 2nd law as a kind of blanket excuse for apathy. It's of course done in a facetious way, but it still plays into the same narrative used by creationists to claim that the 2nd law proves evolution is impossible. I think the 2nd law is probably useful in working out limits to the efficiency of steam engines (but it's a little outside my expertise). Anyway, it is not a useful metaphor for most of the other things it has been applied to, and I wish we could call a halt to all references to the 2nd law by anyone--SF fan or creationist--unless the topic is actually thermodynamics.
anagrammatt2.wordpress.com · 9 April 2015
Sorry I will use an example!
I do not understand much of myself...!
How can I understand origins of life or how wrong in logic Evolution is!
Can anybody please make life in a Laboratory, then we can talk all it took!
If the pathways and roads to life are too complex or not known, we are merely speculating terribly in Graduate and Post Graduate Evolution studies! In which ignorant students are really rusted minds!
anagrammatt2.wordpress.com · 9 April 2015
Sorry I will use an example!
I do not understand much of myself...!
How can I understand origins of life or how wrong in logic Evolution is?
Can anybody please make life in a Laboratory, then we can talk all it took!
If the pathways and roads to life are too complex or not known, we are merely speculating terribly in Graduate and Post Graduate Evolution studies! In which ignorant students are really rusted minds!
Think of analphabetism as a big end road in Evolution!
phhht · 9 April 2015
Tom English · 9 April 2015
Mike Elzinga · 10 April 2015
Joe Felsenstein · 10 April 2015
bplurt · 12 April 2015
Mike Elzinga · 29 April 2015
Mike Elzinga · 29 April 2015
Timothy Horton · 29 April 2015
It gets worse. a little later Ewert drops this dog turd:
"Ultimately, the fact that birds exist has to be explained in terms of the initial configuration of the universe. The universe must have begun with a large amount of active information with respect to the target of birds"
How in the world did he decide that extant birds were one of evolution's "targets"? His logic means that every of the 50 million or so extant species, and the 100X that number now extinct must have also been "targets" predefined by his "active information". Not only that, the "active information" for extant species had to account for all the major mass extinction events including the Earth getting clobbered by a honkin' big asteroid 66 MYA. That's one heck of a lot of "active information". I wonder where it all resided for the last 4.5 billion years?
Mike Elzinga · 29 April 2015
Mike Elzinga · 29 April 2015
Richard B. Hoppe · 29 April 2015
eric · 30 April 2015
TomS · 30 April 2015
Timothy Horton · 30 April 2015
Mike Elzinga · 30 April 2015
Even now - after something like 50 years of watching ID/creationists - I still am amazed at how resistant ID/creationists are to getting a proper education in science. It isn't just that they are ignorant of basic scientific concepts and facts; it's that the crap they believe is dead wrong at even the most basic level. And they work at getting it wrong; one can actually watch the process of bending and breaking concepts in order to support sectarian beliefs.
It may seem humorous in some sense; but in fact, these people are serious about wrecking science education through political action and repeated harassment of school boards and state boards of education. It is no coincidence that people like Casey Luskin show up frequently at school boards, state boards of education, state legislatures, and other places to hawk ID/creationist wares.
ID/creationists portray themselves as persecuted, beleaguered martyrs bucking an evil and corrupt cabal of scientists who are keeping them from achieving the fame and adulation they crave. It's the typical crackpot persecution complex.
Andreas Wagner, author of Arrival of the Fittest had a rather succinct way of describing them, especially the YECs; "Half literate and wholly ignorant."
I think Wagner is being a bit too kind. I don't see how people who routinely scan for quote mines and who distort well-known, basic concepts as they go can be considered literate at any level. Literacy is more than just mouthing words; any computer text reader can do that. Literacy is not about selective scanning for debating points.
TomS · 30 April 2015
paulc_mv · 30 April 2015
(disclosure: was posting as callahanpb until Google dropped OpenId 2.0)
I have often wondered that creationists (particulary the YE variety) manage to dress themselves each morning let alone hold down paying employment. Now, inspired by "Tornado in a Junkyard", I realize that I have a mathematical proof that they cannot.
Let's start by considering socks. A sock should go on a foot. It doesn't matter if it is the left or right foot, but it has a uniform probability of being somewhere entirely different, such as the creationist's ear or nose. For that matter, could hang from any one of ten fingers, and so on. It might simply fall to the floor or hover in the air. It might not even be a sock. Thus, the probability of one sock actually being on the creationist's foot is already quite low. The probability of each sock being on a distinct foot is vanishingly small. Extend this to other articles of clothing, and after pages of additional analysis, taking logs and adding them to make it much more scientific than mere multiplication, we determine that the CSI of a properly dressed creationist cannot be explained by chance. Indeed, only by divine will could any one of them make it out of the house each morning (granting that some may stay at home and type randomly on the nearest available keyboard).
For my next trick, I will explain how it is impossible for light to travel in a straight line.
Henry J · 1 May 2015
Isn't that only if it drank too much?