Joe Felsenstein analyzes a talk by William Dembski

On August 14, the intelligent-design (ID) advocate William Dembski spoke here at The University of Chicago. He was brought here by the Computation in Science seminar (I believe it was his Ph.D. advisor here who sponsored or invited him), and his topic was “The conservation of information in evolutionary search.” The video of his talk, a bit longer than an hour, is at the bottom. I didn’t go, as I believe I was out of town, but were I here I wouldn’t have gone anyway as I was working hard on the Albatross. I’ll may get around to watching the talk, but, as you can see below, an evolutionist familiar with the material says there’s nothing new in it.  The IDers will of course scream bloody murder because I’m presenting a critique without seeing the talk, but it’s not my  critique. It’s Joe Felsenstein’s. Rest assured that they’ll put up some criticisms of Felsenstein at Uncommon Descent within a day or two.

Over at Panda’s Thumb, Felsenstein, renowned evolutionary geneticist and ID critic, did the yeoman’s work of watching the video and then analyzing it in a post called, “Dembski’s argument in Chicago—New? Persuasive?” Felsenstein’s answer on both counts is “no!” Felsenstein has saved most of us the trouble of watching the video, for he concludes that Dembski’s talk is just a rehash of old arguments that have been debunked some time ago.

I won’t rehash Joe’s post, for, as usual, it’s clearly written and should be intelligible if you know something about evolution and population genetics.  But let me just point out what Felsenstein sees as Dembki’s two major errors (the model is actually by Dembski and Robert Marks; I’ll call it “Dembski’s model” for shorthand).

1. Dembski’s model depends on all possible genotypes of an organism having randomly assigned fitnesses, so that if you change a single nucleotide in a genome, the chance of improving it are exactly the same as if you changed every nucleotide. That’s insane. Changing every nucleotide will completely destroy your fitness (average reproductive capacity); changing one has at least some chance of improving it. Genotypes that are nearby in genotype space will have similar fitnesses, so the “genomic landscape” is not horribly rugged, as Dembski assumes, but smooth. The “ruggedness” assumption is needed to show that natural selection can’t work, but it’s a lousy assumption. Go see how Dembski justifies the assumption that sometimes natural selection can work.

2. Dembski’s model requires that each nucleotide (and trait) in the genome interacts strongly with every other nucleotide (and trait), which simply can’t be the case. The developmental network almost guarantees that you can change some parts of an organism without affecting other traits. For example, changing the coat color of a bear from brown to white (as probably happened during the evolution of the polar bear) almost certainly won’t affect its musculature, shape of its toenails, or hunting behavior. Those things, if they change, would change independently, based on selection on other genes.

This is just a summary of what I see as the two main errors. Joe lists several others.

These criticisms have been made before, and Dembski is, according to the ID playbook, ignoring them. As Felsenstein says, there’s nothing new in the seminar. But I’ll put the video below for those of you who either keep atop these things or are masochists:

Conclusion: Shame on the University of Chicago for giving a platform to religiously-motivated arguments that were discredited a decade ago.  I’m ashamed of my school.


  1. Posted October 4, 2014 at 1:08 pm | Permalink


    • Diane G.
      Posted October 5, 2014 at 12:59 am | Permalink


  2. Golkarian
    Posted October 4, 2014 at 1:16 pm | Permalink

    What’s especially curious (well not really, it’s to be expected), is that his arguments rely on natural selection being effective. This is one glaring problem with attacking evolution to trump up ID, there arguments are full of contradictions because things that would disprove evolution aren’t necessary logically compatible with each other.

    • Golkarian
      Posted October 4, 2014 at 1:21 pm | Permalink

      In addition, if a fitness landscape is created by god, Dembski is implicitly assuming theistic or possibly deistic evolution. He’s basically saying god is needed for evolution, but admitting that evolution can accomplish the task given certain parameters.

  3. Posted October 4, 2014 at 1:30 pm | Permalink

    Thanks for the kind words. I’d emend the description of Dembski’s argument a bit:

    Dembski (and his coauthor Robert Marks) doesn’t assume random association of fitnesses with genotypes. (He did that in his 2001 No Free Lunch argument). Instead he assumes, implicitly, that random association is the default condition. Then, if the fitness surface is not random but is smooth, smooth enough for evolution to succeed in climbing to its peaks, D&M argue that a Designer was needed to choose that fitness surface out of all the others, which are mostly random.

    I dispute that, as nonrandom smoothness would as happen “because physics”. For the rest of that argument see my post.

    And the unbelievably strong interaction between sites in the genome is just a consequence of random association of fitnesses, not a separate assumption.

  4. Posted October 4, 2014 at 1:52 pm | Permalink

    It’s astounding the depths people will sink to to defend childish fantasy.


    • Cathy Anne
      Posted October 4, 2014 at 9:46 pm | Permalink

      So True

  5. Posted October 4, 2014 at 2:16 pm | Permalink

    Leo Kadanoff was Dembski’s advisor? Wow – I remember that when Kenneth Wilson won the 1982 Nobel prize for his work on applying renormalization group techniques to the study of phase transitions, a lot of people thought that Kadanoff and Michael Fisher had been slighted since all three of them won the 1980 Wolf prize for much the same work. All of which is to say that Kadanoff is no slouch – the mentoring of Dembski looks better on Dembski’s cv than Kadanoff’s.
    Jerry, I think you might have asked Kadanoff whether the invitation of a cold fusion aficionado to give a presentation of his or her descredited bunk would be appropriate for you to do. (To be fair, however, at least cold fusion claims were testable.)

  6. Mark Sturtevant
    Posted October 4, 2014 at 2:39 pm | Permalink

    Thanks, Joe for sitting through that and explaining the details for us.
    The lack of accuracy regarding biology and math among the Id’ers is generally breathtaking, as seen in this example and in the other recent dust-up over whether it was mathematically possible for the malaria parasite to accumulate several mutations that then conferred resistance to the anti-malarial drug chloroquinone. That one too was handled by more realistic math and attention to the actual biology, as I know you know.

    Nothing in evolution makes sense except in the light of biology!

  7. Posted October 4, 2014 at 2:41 pm | Permalink

    Re point number 2 (the polar bear’s coat colour evolving independently of it’s toenail’s length etc)…why,a century or so after Mendal proved this point with his pea garden and the independently inherited pea pod characteristics,is this a topic of conversation? Is there some subtle dimension missing in your summary?

    • Mark Sturtevant
      Posted October 4, 2014 at 3:32 pm | Permalink

      The goal is to make up sciencey sounding b.s. that dazzles the audience while you wave your wand and ‘demonstrate’ how rotten is the core of evolution science.
      Of course the audience that I speak of is not the one at the talk in Chicago. It is for the audience – the book buying, blog watching audience who think that God is (not) Dead had a happy ending. They are not going to care that the science nerds dismiss this.

  8. Torbjörn Larsson, OM
    Posted October 4, 2014 at 2:48 pm | Permalink

    Dembski’s new claim cheats a lot. (Well, who suspected that? =D)

    Originally Dembski claimed that genotypes had random fitness. That obviously fails on the redundancy robustness of the genetic code. Perhaps 1/3 of the single nucleotide changes don’t change the amino acid.

    I assume the criticism has made him detract the earlier claims and propose that the genetic code hasn’t evolved. The latest ribosome phylogenies shots that down of course.

    That said, I’m impressed with Joe’s analysis. That the interactions of physics behave so that they smooth the fitness surface is obvious, but only in retrospect, at least for this layman. It is similar to the argument why we see spacetime, else everything would happen at the same point at the same time.

    And if smooth fitness surfaces are possible, one can assume that evolution would tend to use them in order to have efficient selection. In fact, protein evolution follows a smoothing modular principle:

    “The similarity of the amino acid sequences and the existence of an intermediate form confirm a hypothesis proposed by Birte Höcker, head of the Protein Design Working Group at the Max Planck Institute for Developmental Biology, according to which the two folding types developed in the course of evolution from a common ancestor. “We assume that evolutionarily early proteins consisted of only short amino acid chains. Those fragments then joined together as in a construction kit to form new molecules with new functions,” Höcker explains.”

    But more than that organisms and proteins are modular, a practicality that is enforced by the smoothness of nature’s interactions vs spacetime, protein evolution has a smooth phase space that Dembski and Mark is completely silent on. It seems protein folding, the essential non-coded step after the ribosome translation, has been under heavy evolutionary constraint from the first short useful sequences that could fold and onward:

    “The [folding energy landscape] funnel shows that the protein tries things that are mostly positive rather than wasting time with dead ends,” Wolynes said. “That turns out to resolve what was called Levinthal’s paradox.” The paradox said even a relatively short protein of 100 acids, or residues, that tries to fold in every possible way would take longer than the age of the universe to complete the process.

    That may be true for random sequences, but clearly not for evolved proteins, or we wouldn’t be here. “A random sequence would go down a wrong path and have to undo it, go down another wrong path, and have to undo it,” said Wolynes, who in his original paper compared the process to a drunken golfer wandering aimlessly around a golf course. “There would be no overall guidance to the right solution.”

    So the funnel is a useful map of how functional proteins reach their destinations. “The only way to explain the funnel’s existence is to say that sequences are not random, but that they’re the result of evolution. The key idea of the energy landscape (depicted by the funnel) only makes sense in the light of evolution,” he said. …

    “If the selection temperature is too high in the sequence space, the search will give every possible sequence. But most of those wouldn’t fold right. The low selection temperature tells us how important folding has been for evolution.

    “If the selection temperature and the folding temperature were the same, it would tell us that proteins merely have to be thermodynamically stable,” Wolynes said. “But when the selection temperature is lower than the folding temperature, the landscape actually has to be funneled.”

    “If proteins evolved to search for funnel-like sequences, the signature of this evolution will be seen projected on the sequences that we observe,” Onuchic said. The close match between the sequence data and energetic structure analyses clearly show such a signature, he said, “and the importance of that is enormous.””

    [My bold. I’m sorry for the long quote. The press release was hard to pare down!]

    • Mark Sturtevant
      Posted October 4, 2014 at 3:52 pm | Permalink

      That was interesting. Now, I know pretty much squat about protein folding (it never came up…) but I wonder if the models on evolution of protein folding take into account that proteins are not made all at once and then choose how to fold. They are made linearly, and start to fold as they are made. This should limit the # of ways that a protein can fold overall since the 1st end that is made can only interact with itself at first.
      Also, protein folding is not something that all proteins decide for themselves. Many (I do not know how many) proteins are folded by other proteins.

      • Torbjörn Larsson, OM
        Posted October 5, 2014 at 10:16 am | Permalink

        [May be a duplicate:]

        Oh, I’m not very well informed here either.

        “proteins are not made all at once and then choose how to fold. They are made linearly, and start to fold as they are made.”

        Yes. But it seems to me most proteins need to be able to fold in toto, since a) misfolded proteins are unfolded by chaperones and b) membrane proteins can still fold from an unfolded string [ ], despite that they initially start folding as they are made as you say [ ]

        “Also, protein folding is not something that all proteins decide for themselves. Many (I do not know how many) proteins are folded by other proteins.”

        That seems arguable? “Also, some proteins need the assistance of another protein called a chaperone protein to fold properly. It has been suggested that this disproves Anfinsen’s dogma. However, the chaperones do not appear to affect the final state of the protein; they seem to work primarily by preventing aggregation of several protein molecules before the protein is folded.” [ ]

        Anfinsen’s dogma is related to Levinthal’s paradox and so the energy funnel. It was originally about small globular proteins it seems, but in that article they have to list exceptions (amyloid diseases such as prions) so it looks like a more generic rule.

  9. Nick
    Posted October 4, 2014 at 2:55 pm | Permalink

    It appears from the video that 2, maybe as many as 3, dozen people attended this seminar. Rather an embarassment I’d say. I think James Shapiro (an ID sympathizer) was one of them. The Disc Inst touted this heavily on their blog beforehand, but with such a weak turnout I notice they haven’t said anything about it since.

    • Nick
      Posted October 4, 2014 at 2:58 pm | Permalink

      Sorry – ‘blog’ implies to me that they accept comments; the Disc Inst’s website is too precious to allow possibly dissenting views.

  10. Chris
    Posted October 4, 2014 at 3:37 pm | Permalink

    One thing that I’ve always wondered: what, exactly, does Dembski know about genetics/molecular biology/etc as opposed to information theory?

    I’m also wondering what other fields in the physical sciences could “information” could be emergent.

    • Mark Sturtevant
      Posted October 4, 2014 at 3:54 pm | Permalink

      I think that it has been said before, but Dembski and Deepak almost have a lot in common.

    • Posted October 4, 2014 at 4:03 pm | Permalink

      Isn’t “information” already deeply built into the physical sciences under the name “entropy”?

      • compuholio
        Posted October 5, 2014 at 5:16 am | Permalink

        Yes and no. As it is so often the case it depends on what exactly you mean by “information”. If you ask people in different fields what “Entropy” means you’ll get quite different answers – which might at first glance have nothing to do with “information”. But mathematically the concepts all look very similar. The interpretation of the symbols however can be vastly different.

        In information theory it is the average information content. Physicists often interpret it as a measure of the number of microstates in a macrostate and in probability theory it is interpreted as the expected surprisal of an event.

      • Torbjörn Larsson, OM
        Posted October 5, 2014 at 11:04 am | Permalink

        That depends on your definitions of “information” and “entropy”.

        Physical entropy, which ultimately goes back to the availability of microstates that maps to a certain macrostate with energy E, differ from information theoretic (IT) entropy where it is used as one of several ways to define information of a state with probability P (as -PlnP). “In information theory, entropy is the average amount of information contained in each message received.” [ ]

        We can see that they are not the same by looking at a computer memory cell, unless I am mistaken.

        Initially we don’t know the state, so having two states with probability 1/2 means the information theoretic entropy is 2*(-(1/2)*ln(1/2)) = ln2. If we write a state to the memory (say, 0), we decrease the IT entropy to 1*(-1*ln(1)) = 0.

        However physically you have irreversibly erased an (unknown) state, so the entropy of the system has increased.

        Moving on, information in physics is usually taken as the unitarity restriction on quantum parameters. Unitarity means probabilities sum to 1, that we don’t loose information on the system outcomes as it progress through time. “unitarity is a restriction on the allowed evolution of quantum systems that ensures the sum of probabilities of all possible outcomes of any event is always 1. More precisely, the operator which describes the progress of a physical system in time must be a unitary operator.” [ ]

        IT information on the other hand can be defined several ways.

        Common to all these ways is that information as a measure can’t be reified. The definition is always relative to a system. (For physical information, through its operator. For IT entropy, through its states.) Information is a constraint, not a property.

        • Torbjörn Larsson, OM
          Posted October 5, 2014 at 11:07 am | Permalink

          “information theoretic (IT) entropy where it is used as one of several ways to define information of a state with probability P”.

          Here i meant one of several ways to _measure_ information.

    • Posted October 7, 2014 at 3:02 pm | Permalink

      He also doesn’t know much information theory, as folks like Jeff Shallit and so on have pointed out.

  11. Filippo
    Posted October 4, 2014 at 4:39 pm | Permalink


  12. Bob Michaelson
    Posted October 4, 2014 at 4:39 pm | Permalink

    There are questions starting about at 52:52, although most of them are inaudible. One person, presumably a physicist or astrophysicist, asks near the end why Dembski doesn’t apply his logic to star formation, which is much simpler than biological evolution – Dembski hems and haws and then Kadanoff pulls him out by saying that Dembski doesn’t know about stars and he’ll need to think about it.
    At the very end Kandanoff says something like “I think Bill has made his case and the ball is now in the court of those who believe in evolution” – demonstrating that being brilliant in one area doesn’t mean anything about your ability to reason in another area.

  13. Andy
    Posted October 5, 2014 at 5:01 am | Permalink

    So, has anyone else noticed that WEIT gets a mention around 37 minutes in? 🙂

    There were some interesting things in the talk, such as the computational limit of the universe, so it wasn’t quite as annoying as I expected. I think a good computer scientist could make a better criticism, but here are my thoughts:

    From his answer to a question, it seems he makes the usual creationist distinction between micro and macro evolution. I’m also not sure he appreciates the differences that sexual reproduction might make.

    He keeps missing the point of the weasel demonstration, even though he mentions that people have told him this. His answer is basically to quote mine and then carry on making the same mistake.
    I think his strongest point is that imposing conditions on a search puts information in. So, he’s arguing that the fitness function in the weasel example already contains information (in the target). Of course then he’s implicitly admitting that a search can generate information. The fitness function for evolution consists of the environment (and laws of physics) etc, not putting in a specific target.

    He used the Monty Hall 3 door example, but I couldn’t see why that was relevant to his argument.

    I think he misses the point of the no free lunch theorem. If you sum over all possible strings, a random search is as good as any other; but in the real world you are not interested in all strings, only some subset.
    (To put it another way, I think it’s always possible to produce a dataset that a given lossless compression algorithm will
    not be able to compress.)

  14. Posted October 5, 2014 at 8:07 am | Permalink

    Following Ben’s point above it is one of the most remarkable phenomena in the intellectual world whereby a faux scientist believes that he or she have only to prize open the tiniest crack in scientific knowledge for all the gods and saints to tumble to earth! I used to call it ‘One Percenting’, but since doubting one percent of science no longer has credibility, those who pursue that ignoble quest must these days resort to doubting perhaps one billionth of scientific knowledge with the intention of unravelling that close-woven scientific fact called Theory of Evolution.
    Self-awareness in telling porkies (Cockney riming slang; Pork-Pies=Lies) is always missing from the conversations of ID supporters. Again and again we ask, “Where is their shame?”
    Amid the blather of half-truths, deliberate distortions of fact, and mind-boggling assumptions, the Apologists of this world would seem to be trying to jemmy-open the folds of the skies to let their missing gods through. We all know what they are up-to. It is scandalously obvious that they want to justify Stone-Age myths. But for the rest of us it is like watching a police-tape where the murderer simply repeats, ‘ No comment’ to every question, and then has the audacity to finger his wife or his work-colleague for his crime. Is there, – and was there, – ever a time when an Apologist apologises for the lies, and seeks to make himself whole again by tendering the excuse that he felt that his (after) life was on the line?
    Aquinas is said to have written eight million words of theological exegesis and suddenly stopped; I like to think that he stopped at the moment of his deconversion. If only he had told us so!

  15. Posted October 5, 2014 at 10:40 am | Permalink

    I do not know the policy at Univ. Chicago, but some state universities have no choice in RENTING an auditorium to groups such as the DI. For example, we at the University of Oklahoma have had to endure Dembski and others of his ilk several times. The last visit by DI folks involved a presentation at the Oklahoma Museum of Natural History. The Museum director wrote that rental did not show any approval of the DI visit and the Director (Dr. Michael Mares) further indicated that there was no approval of the presentation or their agenda.

    The DI visit was met with lots of opposition and hard questioning. Dembski and others at the DI complained bitterly about their hostile reception at the University of Oklahoma in several blog posts and in one of Dembski’s books. They have also labeled Oklahomans for Excellence in Science Education (http// ‘militant,’ perhaps a correct description when it comes to opposition of intelligent design and other forms of creationism.

    • Posted October 5, 2014 at 7:22 pm | Permalink

      The University of Chicago as a whole probably played no role in the invitation to Dembski. Leo Kadanoff seems to be one of the principals in running the Computations in Science Seminar. Dembski is his former Ph.D. student. Kadanoff is the formal host, and the nominal source of the invitation — though it is not obvious who initiated the invitation, Kadanoff or Dembski.

      Since the topic was not (at least formally) Intelligent Design, larger university policies would not be relevant, and the University of Chicago is a private university anyway.

      • Posted October 5, 2014 at 7:44 pm | Permalink

        Thanks Joe for the explanation of Dembski’s appearance at Chicago.

  16. jay
    Posted October 6, 2014 at 6:30 am | Permalink

    Dembski’s information theory based argument seems to ignore that information IS added to the system __ through differential survival and breeding success.

%d bloggers like this: