More Recent Comments

Monday, December 05, 2011

William Dembski Disproves Evolution

 
Bill Dembski is another one of the "big guns" of Intelligent Design Creationism. He has a Ph.D. in mathematics (University of Chicago, 1988), a Ph.D. in philosophy (University of Illinois (Chicago), 1993) and a Master's of Divinity (Princeton Theological Seminary, 1996).

Here's a video explaining how Dembski can mathematically disprove evolution. Somehow this leads to proof of god. It may be difficult to follow the logic but that's probably because you and I don't have Ph.D.'s in mathematics or philosophy.

Keep in mind that this is no amateur. Dembski is among the very best of the best in Intelligent Design Creationism. His speculations have been thoroughly refuted by prominent mathematicians, scientists, and philosophers. Under normal circumstances, this would be enough to cause his supporters to abandon him but he's still a fellow of the Center for Science and Culture and he's still promoted as one of the leading supporters of Intelligent Design Creationism.




43 comments :

DGA said...

Actually the video does not explain "how" Dembski can mathematically prove anything at all, only that he "can" prove something and a simple statement like that is sufficient for his mathematically challenged followers. Of course once one begins to look into the probability calculations and how they are misused be creationists, then it falls apart as does Behe's irreducible complexity. But still the production values, the dice, the homey diner, the music, the Scots accent sounding so dour, implacable and skeptical make me want to believe ...

Frank said...

In a flash of masochism, I tried to read Dembski's The Design Inference. Even I, who haven't taken a math course in over 40 years, could recognize that his conception of probability was naive and uninformed. It would cause him to fail a competently taught introductory course on information theory. What was UC thinking?

Paul King said...

Doesn't Dembski's "mathematical proof" boil down to "Michael Behe says this can't happen" ? (And that isn't even true).

To the best of my knowledge Dembski has never done the probability calculations needed by his own method for any non-trivial example - and neither has anyone else. If all he's got to offer is a method of "design detection" that nobody can sensibly use, he hasn't contributed much of anything.

Joe Felsenstein said...

My usual response: I have a good-if-I-do-say-so-myself paper responding to Dembski's arguments, summarizing the responses of many other people to his work, and adding a couple of relevant arguments myself. It is available (free) here.

For the Design Inference he in effect argues that if you see high fitness due to adaptation, high enough that pure mutation cannot be expected to produce it in the whole existence of the Universe, then that indicates Design. However, that's pure mutation without any natural selection. Of course it can be done by natural selection, which has a huge effect in making the adaptation plausible. (His other theorems are supposed to rule natural selection out, but they are wrong, it turns out). So his famous Inference is only a proof of "Design or Natural Selection".

Dembski has never responded to these (devastating) criticisms.

Anonymous said...

However improbable it is that evolution would result in such-and-such, it seems that it is even more improbable that "Intelligent Design" would.

Unless someone can specify restrictions on what ID can do.

The probability of an event is a ratio of two numbers: the number of favorable outcomes divided by the number of possible outcomes. ID, whatever it means, appears to be a way of increasing the number of possible outcomes - ID can do things that natural causes can't. Increasing the number of possibilities only decreases the probability.

TomS

Joe Felsenstein said...

Paul King: No, Dembski's argument is not dependent on Behe's. As witness the fact that Behe only says the Designer intervenes sometimes, while Dembski's "proofs" (his Law of Conservation of Complex Specified Information and his invocation of the No Free Lunch Theorem) are supposed to work to make essentially all adaptation by natural selection impossible: all characters, in all species, almost all the time.

It is astounding that he thought he could prove this, in the face of what was (then) 90 years of work on the mathematical theory of population genetics. In 1981, which is 30 years ago now, I did a bibliography of theoretical population genetics -- it then had almost 8,000 papers. In light of that, he might have been wise to check his argument more.

Roberto Aguirre Maturana said...

The following link contains an Excel macro wich specifically addresses and debunks the claim that an undirected process (like natural selection + random mutations) cannot generate "specified complexity" with a probability of random occurrence in the order of 10^150:

http://db.tt/iDGliSW

(As far as I know it doesn't contain any malicious code, use at your own risk)

Anonymous said...

"Somehow this leads to proof of god. It may be difficult to follow the logic but that's probably because you and I don't have Ph.D.'s in mathematics or philosophy."

As it happens, I do have a Ph.D. in mathematics. I was probably teaching a class in probability at UIC (then called UICC) at the time that Dembski was an undergraduate there, though I don't remember if he was ever in one of my classes.

In any case, here's the story. Most actual events that occur every day, if you examine them in close detail, turn out to be extremely improbable.

Dembski's argument is based on an elementary error. He is calculating the absolute probability that the event might occur. However, he should be computing the conditional probability, given that the event occurred, did it occur naturally.

Given our knowledge, based on our observations of the frequency in which events occur supernaturally, we can reasonably estimate that the probability of a supernatural occurrence is zero. When we put that into the equation for computing condition probability, it turns out that the conditional probability that life arose naturally is pretty near 1.

I'm pretty sure that Dembski knows all about conditional probability. But he won't use that in his reasoning, because he has a creationist axe to grind.

Anonymous said...

@nwrickert:
the frequency in which events occur supernaturally

But the advocates of ID don't give us any description of what it would be like for an event to occur other than naturally. What difference does ID make in an event? ID does not offer an example of what happened, when or where it happened, when a design event took place. Nor, on the other hand, an example of a non-design event. For all we know, things that occur by ID also have natural causes, and everything occurs by ID.

So the probability of an event occurring by means of ID cannot be estimated.

TomS

Paul King said...

Joe Felsenstein: When Dembski tried to use his method to "prove" that a bacterial flagellum was designed in _No Free Lunch_, he ruled out evolution as a possibility on the ground that Behe had claimed that IC structures could not evolve. (In truth Behe had admitted that IC systems could evolve, by what he called "indirect" routes)

I think that this is sufficient to justify my claim. Once evolution has been eliminated, what plausible option is there for the origin of a complex adaptive structure ? It really is the heart of his argument, and all it is is misrepresenting the opinion of another leading member of the ID movement. Hardly a mathematical proof !

Atheistoclast said...

Joe Felsenstein says:

"In 1981, which is 30 years ago now, I did a bibliography of theoretical population genetics -- it then had almost 8,000 papers. "

Sure. But how do you overcome a very rough fitness landscape where there are few if any intermediate paths available? You need to get rid of the constraining power of natural selection. This has never occurred to you. I am surprised.

Joe Felsenstein said...

nwrickert said:

Dembski's argument is based on an elementary error. He is calculating the absolute probability that the event might occur. However, he should be computing the conditional probability, given that the event occurred, did it occur naturally.

Dembski's "specification" in effect uses a scale on which the outcomes are ordered. His probability is intended as a tail probability of a distribution on this scale (for example, fitness of the genotype). So when he says the outcome is improbable, he means that anything that fit, or more fit, is extremely improbable (to occur by pure mutation).

That gets him out of the fallacy, (except for the fact that he failed to rule out natural selection).

Joe Felsenstein said...

nwrickert said (replying to my assertion that Dembski's argument was different from Behe's):

When Dembski tried to use his method to "prove" that a bacterial flagellum was designed in _No Free Lunch_, he ruled out evolution as a possibility on the ground that Behe had claimed that IC structures could not evolve. (In truth Behe had admitted that IC systems could evolve, by what he called "indirect" routes)

I think that this is sufficient to justify my claim.


It is true that Dembski did try to invoke Behe's IC arguments. However his Law of Conservation of Complex Specified Information argument (connected to his Design Inference) and his No Free Lunch argument are separate from, and precede, that discussion. As I mentioned, they are intended to apply to all adaptations, while Behe's arguments are intended to apply only to some of them (such as the case of the bacterial flagellum).

Joe Felsenstein said...

Sorry, that last comment I made was actually a reply to Paul King, not nwrickert.

Atheistoclast said...

Just for folks who, unlike Joe Felsenstein, do actually understand the meaning of CSI, here is a peptide sequence that exemplifies it: it obeys Dembski's law of the conservation of information:


RRRKRTAYTRYQLLELEKEFLFNRYLTRRRRIELAHSLNLTERHIKIWFQNRRMKWKKEN


You get a cookie if you tell me what the sequence does.

lee_merrill said...

Joe Felsenstein:

> But what we have done is to describe the action of the mechanism that creates specified information — if this acts repeatedly at many places in the gene, specified complexity would arise.

But some motion does not imply a large range of motion is possible by this mechanism--i.e. on the order of 500 bits, within the time we have for evolution to have done this. If we have any amount of time, then yes, evolution can do this, but can we expect such changes in the timeframe we have to work with? This needs to be examined.

> On average they are lower, but the chance that one reaches a sequence that is better is not zero.

But it is indeed zero if you're at a local maximum. Then you need multiple mutations to get to another hill to climb, and the probability of this succeeding gets exponentially smaller, the more simultaneous mutations you need.

And this is the ID argument in a nutshell.

P.S. I agree by the way that specified information can increase by random processes, though not at the rate require to produce say, you, or me.

The Other Jim said...

@Atheistoclast,

Euprymna scolopes seems to do fine with only 70% AA identity. And I didn't bother to go to the second page of the BLAST hits.

Not quite as "specified" as you seem to be implying...

Joe Felsenstein said...

Lee Merrill, after quoting my 2007 paper on how natural selection can create 2 bits of Specified Information by one gene substitution:


But some motion does not imply a large range of motion is possible by this mechanism--i.e. on the order of 500 bits, within the time we have for evolution to have done this. If we have any amount of time, then yes, evolution can do this, but can we expect such changes in the timeframe we have to work with? This needs to be examined.


Dembski's argument would need to establish that, in the whole history of the Universe, there is not enough time for 250 of these gene substitutions, which would be needed for us to get up to 500 bits. (And in fact some of them can even go on simultaneously). In my example only 84 generations were needed to get 2 bits of SI. As long as we get to start the Universe before 4004 BC there seems to be ample time.

Lee Merrill again, quoting another passage where I argue that having some advantageous mutations is not as impossible as the No Free Lunch theorem would imply:

But it is indeed zero if you're at a local maximum. Then you need multiple mutations to get to another hill to climb, and the probability of this succeeding gets exponentially smaller, the more simultaneous mutations you need.

And this is the ID argument in a nutshell.


The No Free Lunch Theorem (which is a valid theorem) in effect describes movement on a randomly constructed adaptive landscape that is infinitely jaggy (a "white noise" adaptive surface). I was pointing out that Dembski's argument assumes an unreal adaptive landscape, and so the NFL Theorem does not rule out achieving increased adaptation in the real world.

Of course on real adaptive landscapes one can get stuck. But they can be very high-dimensional and can change through time.

It is a pleasure to run into a supporter of ID who is actually willing to read my 2007 article and engage its arguments.

OK, are Lee Merrill and I agreed that Dembski's LCCSI and NFL arguments don't work?

SLC said...

At one time, Prof. Dumbski allowed as how common descent and an old earth might possibly be true. In his current position in a fundy school, he now is a young earth creationist who rejects common descent.

Anonymous said...

Atheistoclast said:

RRRKRTAYTRYQLLELEKEFLFNRYLTRRRRIELAHSLNLTERHIKIWFQNRRMKWKKEN

You get a cookie if you tell me what the sequence does.


Oh, here, let me give it a shot. Does it make you break out in a rash of the Virgin Mary with a pulsing "JESUS IS LORD" on your chest? Because it would take something like that to prove the case of where you're ultimately going in praising introns and the like.

You can keep the cookie and the grape juice for Sunday morning.

Pentti S. Varis said...

The complex multicellular animal´s genome does contain many important genes, which has been silent in its one-celled "simple" ancestors perhaps 100 000 000 years. This should be cognized in trying mathematically refute "darwinism"..

http://www.panspermia.org/whatsnew67.htm

The Other Jim said...

@Pentti S. Varis

No extant one-celled "simple" organisms are the ancestors of metazoa. You are misunderstanding, not disproving, evolution.

Jud said...

The complex multicellular animal´s genome does contain many important genes, which has been silent in its one-celled "simple" ancestors perhaps 100 000 000 years.

Oh dear, "front loading" again. The Creator was so incompetent that instead of doing "just in time inventory," which even we humans are smart enough to do, He apparently made all life carry that genetic baggage for 3.5 billion years. (Not 100 million, wherever did you come up with that figure?)

So is that your answer - life was created by an entity so stupid He'd be fired by any competent logistics or manufacturing firm today?

Pentti S. Varis said...

Yes, but it will not take long time before we can agree with that:

"We fooled ourselves into thinking the genome was going to be a transparent blueprint, but it's not," says Mel Greaves, a cell biologist at the Institute of Cancer Research in Sutton, UK.

http://www.nature.com/news/2010/100331/full/464664a.html

Many investigations point towards a radical paradigm shift in genomics and..evolution.

http://www.starcitynews.com/study-finds-that-assault-of-genetic-parasites-triggered-modern-mammalian-pregnancy/8871/#more-8871

http://www.nature.com/nature/journal/v454/n7207/edsumm/e080821-05.html

lee_merrill said...

Joe Felsenstein: ... are Lee Merrill and I agreed that Dembski's LCCSI and NFL arguments don't work?

I agree that random processes can generate specified information, as in the proverbial monkeys typing the works of Shakespeare. I haven't really examined the NFL argument, but it seems unlikely to me that this would provide a limit to what evolution can do--for we are not dealing with an algorithm over all possible input sets, but with an algorithm starting on a local hill in a hilly landscape. But I have Dembski's book! It now behooves me to read it.

But I would be interested in your take on Behe's Edge of Evolution--I wish Larry Moran would discuss that in his class instead of Icons of Evolution.

Joe Felsenstein said...

Lee Merrill said: I agree that random processes can generate specified information, as in the proverbial monkeys typing the works of Shakespeare.

In the example in my paper, which you have read, I showed that natural selection (not "random processes") could put 2 bits of Specified Information into the genome. I thought that you already agreed that this was so. Right?

It's not just typing monkeys, as you have already agreed.

Anonymous said...

I would be interested in your take on Behe's Edge of Evolution

I liked Behe's example of malaria, sickle cell anemia, and the manufacture of drugs against malaria. This shows how evolution is more productive than is intelligent design. While sickle cell anemia still is effective against malaria, malaria has managed to evolve resistance against several intelligently designed drugs.

TomS

lee_merrill said...

Joe Felsenstein: I showed that natural selection (not "random processes") could put 2 bits of Specified Information into the genome.

Yes, I agree, evolution can increase SI, and faster than a random search.

> In my example only 84 generations were needed to get 2 bits of SI. As long as we get to start the Universe before 4004 BC there seems to be ample time.

I would disagree, for your hill is extraordinarily high! Most fitness landscapes do not have hundreds of steps where you don't reach any local maxima.

And I checked the No Free Lunch book, and it seems the argument takes "no algorithm has an advantage in reaching any arbitrary goal" to "no algorithm has an advantage in reaching a particular goal" i.e. a particular subset of the space, such as high fitness regions.

But I'm not seeing the connection here to make this transition, and as you note, evolutionary algorithms and evolution both work better than random searches.

Larry Moran said...

lee merrill says,

But I would be interested in your take on Behe's Edge of Evolution--I wish Larry Moran would discuss that in his class instead of Icons of Evolution.

The Edge of Evolution is much too complicated for second year students who aren't majoring in biochemistry or genetics. That's why I don't use it in class.

Behe presents two main arguments.

The first is an argument about the probability of epistatic mutations. Imagine a fitness benefit when two mutations occur where either one alone is detrimental. Behe argues that the probability of these two mutations happening simultaneously in the same organism is extremely low. Too low to happen in most populations. This defines one "edge of evolution." Behe is basically correct here, given his assumptions.

The second argument concerns protein-protein interactions. Behe develops "The Two-Binding Site Rule" for such interactions. The question he asks is ...

... how difficult would it be for two proteins that initially did not bind to each other to develop a strong, specific interaction by random mutation and natural selection?

He claims that you need three or four mutations in order to get binding where each of them causes trouble if they occur singly. Thus, in order to create a new binding site on a protein you need three or four simultaneous mutations and this is "far beyond the edge."

Again, given his assumptions, Behe is correct. That's not going to happen. We are never going to see such events in the history of life.

There are only two possible conclusions one can draw after reading Behe's book. Either his assumptions are wrong or God exists.

If you know anything about evolution, biochemistry, and genetics, it's trivially easy to show that his assumptions are incorrect. Unfortunately, it's not so easy to explain it to people who are not familiar with those subjects and that's why it's more difficult for undergraduates than Jonathan Wells' book Icons of Evolution.

Pentti S. Varis said...

@Jud:

"The Creator was so incompetent that instead of doing "just in time inventory," which even we humans are smart enough to do, He apparently made all life carry that genetic baggage for 3.5 billion years."

Hundreds of investigations show that this is true, e.g. the genes which cause antibiotics resistance are very old

http://www.nytimes.com/2011/09/01/science/01gene.html?_r=2

The all the time discovered new regulation systems of the genes, that causes e.g. that one gene produces many different proteins, are so complicated, that they are not jet fully understandable. See e.g.

http://vizbi.org/Videos/26206370

Jud said...

Hundreds of investigations show that this is true, e.g. the genes which cause antibiotics resistance are very old

Worse and worse. The story you cite shows not that genes for antibiotics resistance sat silently waiting, but that there were genes conferring resistance to natural antibiotics in the contemporary environment of the bacteria that were studied. Not front loading, but natural selection.

Joe Felsenstein said...

lee_merill said:

Yes, I agree, evolution can increase SI, and faster than a random search.

Good, that's progress towards agreement between us.

and merill also said, about my argument that then only 250 of these searches need to occur to get up to Dembski's 500-bit threshold, which he thinks he has shown is unreachable:

I would disagree, for your hill is extraordinarily high! Most fitness landscapes do not have hundreds of steps where you don't reach any local maxima.

Dembski's argument is supposed to show that it is impossible to get to 500 bits, no matter what the fitness surface. So "my" hill is very relevant.

It disproves Dembski's argument. But how can it do that when he claims to have proven that this is impossible? See my 2007 paper where I argue that he changes the specification in midstream to make his argument, when one should not, if one is modeling increase of a function such as fitness. My argument here does the relevant thing, which is to keep the specification the same, so that his LCCSI theorem does not apply, even if it were totally correct.

OK, so are we in agreement about the invalidity of Dembski's LCCSI argument, and thus of his Design Inference?

(As for the No Free Lunch argument, I will deal with that in a subsequent comment).

Anonymous said...

I have to hand it to them, whoever makes their documentaries are competent, if only them.

Joe Felsenstein said...

lee_merrill commented about Dembski's use of the No Free Lunch Theorem (NFL):

And I checked the No Free Lunch book, and it seems the argument takes "no algorithm has an advantage in reaching any arbitrary goal" to "no algorithm has an advantage in reaching a particular goal" i.e. a particular subset of the space, such as high fitness regions.

But I'm not seeing the connection here to make this transition, and as you note, evolutionary algorithms and evolution both work better than random searches.


I hesitate to support Dembski's argument, but if "any arbitrary goal" includes a set of the highest-fitness points in the space (i.e. if that is the set you arbitrarily pick as the goal), then the NFL does apply.

Where the problem is, is that the NFL does not apply to one fitness surface, it applies to an average behavior over all possible fitness surfaces. With P points in the space amd P different fitness values, there are P! (P-factorial) ways you could associate fitnesses with points (genotypes).

A typical one of these is a random association of points with fitnesses. Yes, that has disastrous consequences for evolution. A one-mutation change brings you to a fitness randomly chosen from the set of all fitneses in the space. In other words:

1. A 1-mutation change basically almost-always kills you, and

2. A 1-mutation change is just as bad for your fitness as changing all sites in the genome.

A moment's consideration will show that real fitness surfaces are smoother than that -- a lot smoother. So evolution has a good chance of succeeding on them, even though it would do horribly on fitness surfaces that randomly associate fitnesses with genotypes.

There is nothing wrong with Wolpert and Macready's NFL Theorem. The problem is that in using it Dembski effectively assumes that an infinitely jaggy fitness surface is typical. Considerations of physical reality work against that.

lee_merrill said...

Joe Felsenstein: ... so are we in agreement about the invalidity of Dembski's LCCSI argument, and thus of his Design Inference?

I do believe (not only SI, but) CSI can increase via any process that allows for all possibilities, this includes random variation, this includes evolution.

Pentti S. Varis said...

"The story you cite shows not that genes for antibiotics resistance sat silently waiting, but that there were genes conferring resistance to natural antibiotics"

Antibiotics resistance was really a bad example. Better is the Trichoplax genome. The ancient genes in Trichoplax, could they perhaps be part of the regulation system too?

http://www.nature.com/nature/journal/v454/n7207/abs/nature07191.html

Joe Felsenstein said...

lee_merill wrote: I do believe (not only SI, but) CSI can increase via any process that allows for all possibilities, this includes random variation, this includes evolution.

Good, I think this constitutes agreement that Dembski's Law of Conservation of Complex Specified Information (LCCSI) does not work to justify his Design Inference (also called Design Detector). For the observation of high fitness, high enough that it could not have come about even once in the history of the Universe if all we had was mutation without natural selection.

Thus the many assertions one hears from ID supporters that CSI is evidence for ID because we know of no natural phenomenon that can produce it, are simply wrong.

I think we now agree about that.

Now what about the No Free Lunch argument? In your view does the NFL Theorem establish that the particular fitness surfaces we find in living systems are so rough that natural selection cannot do any better than search by random mutation? I have argued that the NFL does not establish this because it averages over all possible fitness surfaces. Do you agree?

lee_merrill said...

Joe Felsenstein: I have argued that the NFL does not establish this because it averages over all possible fitness surfaces.

I agree, the fitness landscapes we see are a subset of all possible landscapes, and evolution does much better in these smoother landscapes than random variation would. Evolution would have no overall advantage over all possible landscapes, but that does not constrain it from doing well on a subset.

Joe Felsenstein said...

lee_merrill said:

I agree, the fitness landscapes we see are a subset of all possible landscapes, and evolution does much better in these smoother landscapes than random variation would. Evolution would have no overall advantage over all possible landscapes, but that does not constrain it from doing well on a subset.

It seems that this means you agree that William Dembski's arguments for the ineffectiveness of natural selection lack force.

(His more recent "search for a search" arguments are arguments about information needing to be built into the universe to enable natural selection to work, and that is a different matter as this does not rule out natural selection being effective).

Of course, this does not mean that all arguments for ID are necessarily wrong. But Dembski's arguments that are the subject of the video in Larry's original post are wrong. Are we agreed on that?

lee_merrill said...

Joe Felsenstein: But Dembski's arguments that are the subject of the video in Larry's original post are wrong. Are we agreed on that?

I would agree, though with the caveat that I may be misunderstanding some point about the No Free Lunch argument. I'm would say Bill Dembski is not one to make a simple mistake here, so I would be glad to have him clarify.

Joe Felsenstein said...

lee_merrill responded to my question as to whether he agrees that William Dembski's main arguments are incorrect:

I would agree, though with the caveat that I may be misunderstanding some point about the No Free Lunch argument. I'm would say Bill Dembski is not one to make a simple mistake here, so I would be glad to have him clarify.

The criticism of his No Free Lunch argument that I made was made (by about 7 other people) multiple times since 2002. Dembski has never responded to it, except to point to his more recent papers. However those papers raise a totally different argument (his Dearch For a Search) argument), one which does not rule out the effectiveness of natural selection.

If you can get him to respond and clarify why he has not withdrawn his assertions, you would be doing everyone a favor. Because his No Free Lunch argument is repeatedly cited by ID proponents as an unanswerable criticism of the effectiveness of nstural selection.

In fact, just today, at Uncommon Descent, the ever-astonishing Denyse O'Leary is baiting and imsulting Kenneth Miller for allegedly erecting a strawman of ID rather than dealing with its actual arguments. And how does she start her article? With this sentence:

The threatening part of ID is Dembski’s No Free Lunch hypothesis. In a world looking for secular magic, it is the most deeply threatening idea imaginable.

Joe Felsenstein said...

Oops, that is supposed to have been "Search For a Search", not "Dearch For a Search".

Unknown said...

The evolution camp seems to be unaware that the evolutionary research is actually a kind of creation research: Could the Creator create the known flora and fauna by the mechanisms of Darwinian evolution? (The answer is no as Dembski and others have shown)