Dembski: “Conservation of Information Made Simple”

At Uncommon Descent, Bill Dembski has announced the posting of a new article, Conservation of Information Made Simple, at Evolution News and Views.  Comments are disabled at ENV, and free and open discussion is not permitted at UD. I’ve therefore created this thread to give critics of ID a place to discuss the article free of censorship.

74 thoughts on “Dembski: “Conservation of Information Made Simple”

  1. Dembski gets off to a bad start, committing Hoyle’s fallacy:

    For example, consider all possible L-amino acid sequences joined by peptide bonds of length 100. This we can take as our reference class or backdrop of possibilities — our search space. Within this class, consider those sequences that fold and thus might form a functioning protein. This, let us say, is the target. This target is not merely a human construct. Nature itself has identified this target as a precondition for life — no living thing that we know can exist without proteins. Moreover, this target admits some probabilistic estimates. Beginning with the work of Robert Sauer, cassette mutagenesis and other experiments of this sort performed over the last three decades suggest that the target has probability no more than 1 in 10^60 (assuming a uniform probability distribution over all amino acid sequences in the reference class).

    Later, he cheerfully admits that the evolution of nylonase represents an increase of information:

    Nylon, for instance, is a synthetic product invented by humans in 1935, and thus was absent from bacteria for most of their history. And yet, bacteria have evolved the ability to digest nylon by developing the enzyme nylonase. Yes, these bacteria are gaining new information, but they are gaining it from their environments, environments that, presumably, need not be subject to intelligent guidance. No experimenter, applying artificial selection, for instance, set out to produce nylonase.

    So much for his fellow IDers who insist that mutation invariably causes a loss of information. Dembski admits that there is a gain of information, but argues that it falls below his 500-bit CSI threshold and therefore does not indicate design.

  2. Good idea.  Although I can’t see anything new in the article (and to be fair it is only meant to be a simplified version of what went before).  The LCI still seems to hang on a dubious analogy between a human searching for something like ace of spades in pack of playing cards and natural mechanisms for creating viable organisms.  It also entails performing unnatural acts with the Principle of Indifference e.g. applying it to search strategies.

  3. Take an Easter egg hunt in which there’s just one egg carefully hidden somewhere in a vast area.

    This is a restatement of the “needle in a haystack” or “islands of functionality” argument put forward by Kairosfocus. As the paper Allan Miller linked to shows, there is no reason to assume one Easter egg, one isolated small island or one needle. The haystack may be stuffed to overflowing with needles. There is no reason to suppose functionality is not common in the set of all possible protein sequences and, as indicated by Allan’s paper and Jack Szostak, every reason to suppose functionality is widespread.

    Moreover these arguments are not arguments in support of “intelligent design”, merely arguments of incredulity directed at evolutionary explanations.

  4. Protein invention appears to be rare. About one new protein every two million years per biosphere, sample size of one. I don’t know how that works out in terms of “trials” per protein, but it’s lots.

    Anyway, most evolution is regulatory rather than invention of new proteins.

    Since “life comes from life,” presumably all protein coding sequences other than the first started from an existing sweet spot, evolving from a duplicated sequence. (This seems to imply a universal common protein ancestor. I just made this up, but I read something along this line in a paper on protein evolution.)

  5. In a discussion with gpuccio I pointed out that his theory of protein invention by intelligent designers requires regular visitations by the designer every two million years or so.

  6. Protein invention appears to be rare. About one new protein every two million years per biosphere, sample size of one. I don’t know how that works out in terms of “trials” per protein, but it’s lots.

    Not sure if my general point – that functional protein sequences may not be at all rare –  conflicts with your comment. And, indeed, one could postulate a nested hierarchy fanning out from LUCA. But given another starting point of a protein based organism in another blank canvas or warm pond, who knows what would turn up?

  7. keiths

    Dembski gets off to a bad start, committing Hoyle’s fallacy

    Yes, not just Hoyle’s! (To undermine the credibility of the Wiki, I wrote a bit of it, in one of my few forays into Wikipedia editing back in 2010 :D)

    As the authors of the E coli paper Alan Fox has relinked above remark, even a rather short 100-residue base-20 sequence would, with just one molecule of each different peptide, fill 6 x 10^23 universes … yet a random(ish) subset of 1.5 * 10^6, a few thousandths of a picogram that I could easily lose down the back of the sofa, contained 18 sequences functionally substitutable between them for c1/7th of all the natural proteins assayed. Which I think is rather cool, even if the deck was somewhat stacked by foldability criteria. Getting the FIRST functional peptide sequence is a different matter of course, but anyone knowing its length and amino acid composition, feel free to calculate its probability.

    Dembski:

    Nature itself has identified this target as a precondition for life — no living thing that we know can exist without proteins.

    An interesting one. No existing living thing doing without proteins is NOT the same as protein (especially 20-acid protein) being a ‘precondition for life’. All modern living things appear to have descended from a protein-synthesising ancestor, but that is hardly conclusive.

    Yet I admit a bit of a symmetry problem – ‘we’ tend to argue that no ‘coding’ intelligence has been observed other than that inhabiting human skulls, yet here I am arguing against a similar case on protein. So sue me!

  8. I see this as related to conflating different notions of information.

    I seem to recall that “conservation of information” is a theorem about Chaitin information.  However, Chaitin’s account is about an idealized mathematical conception of information that exists in a closed axiomatic system.

    If we instead use Shannon information, then (in my opinion) we should be talking about the manufacturing of information.  Digital cameras, for example, produce Shannon information that never existed before there were digital cameras.

    When I look at Dembski’s argument, I don’t see an obvious problem.  He seems to begin with the assumption that everything is search (i.e. search for already existing information).  And then, after a detailed argument, he concludes that you cannot get new information.  Well, Bingo.  But isn’t it trivially obvious that if all you are doing is searching for existing information, then that search won’t yield anything new.  The lengthy argument seems pointless, since it should be trivially obvious that the conclusion is already presumed in the initial assumptions.

    Or did I miss something?

     

  9. Dembski’s nonsensical “Law” has already been refuted by Mark Chu-Carroll and Erik.

    The same flaws they identified exist in this summary.  The most egregious issue, as is often the case in Dembski’s work, is the gross misuse of the No Free Lunch theorems.  He claims that:

    The upshot of the NFL theorems is that no evolutionary search outperforms blind search once the information inherent in fitness (i.e., the fitness landscape) is factored out.

    This is not accurate.  The NFL theorems state that no search algorithm is superior to blind search when averaged over all fitness landscapes.  That’s not a problem in biological evolution because we have only one (continuously changing, which is another problem with using the NFL theorems) landscape.  Even leaving aside the serious problems with modeling evolution as a search, it is clear that Dembski’s attempt to use these theorems comes down to nothing more interesting, and certainly no more compelling, than a variant of the anthropological argument for god.

    This is emphasized by another comment in his summary: 

    …we happened to get lucky and live in a universe hospitable to life.

    Dembski seems to fail to realize that if our universe were not hospitable to life, we wouldn’t be here to observe it.

  10. Dembski: For example, consider all possible L-amino acid sequences joined by peptide bonds of length 100. This we can take as our reference class or backdrop of possibilities — our search space. Within this class, consider those sequences that fold and thus might form a functioning protein. This, let us say, is the target. This target is not merely a human construct. Nature itself has identified this target as a precondition for life — no living thing that we know can exist without proteins. Moreover, this target admits some probabilistic estimates. Beginning with the work of Robert Sauer, cassette mutagenesis and other experiments of this sort performed over the last three decades suggest that the target has probability no more than 1 in 10^60 (assuming a uniform probability distribution over all amino acid sequences in the reference class).

    Starting with the easy stuff. Sauer’s paper on frequency.

    Davidson & Sauer, Folded proteins occur frequently in libraries of random amino acid sequences, Proceedings of the National Academy of Sciences 1994.

    Furthermore,

    Lucrezia et al., Do Natural Proteins Differ from Random Sequences Polypeptides? Natural vs. Random Proteins Classification Using an Evolutionary Neural Network, PLoS ONE 2012: “Altogether, our results suggest that natural proteins are significantly edited from random polypeptides and evolutionary editing can be readily detected analyzing structural features.

    In other words, folded proteins are common in random sequences, and natural proteins appear to be descended from random sequences.

     

  11. Dembski: The origin of nylonase is thus akin to changing the meaning of “therapist” by inserting a space and getting “the rapist.” For the details about the evolution of nylonase, see a piece I did in response to Miller at Uncommon Descent.

    Dembski’s claim is that it doesn’t represent CSI (i.e. >500 bits). However, in the case of “the rapist”, according to Dembski, we determine design “even if nothing is known about how they arose”, so we can ignore its history as irrelevant. So we calculate the number of bits directly for ten letters. For nylonase, ignoring its origin, it has roughly 400 residues, or well over the 500 bits that make for the design conclusion.

    It comes down to you can’t ignore its history.

     

  12. Stuart Kauffman: If mutation, recombination, and selection only work well on certain kinds of fitness landscapes, yet most organisms are sexual, and hence use recombination, and all organisms use mutation as a search mechanism, where did these well-wrought fitness landscapes come from, such that evolution manages to produce the fancy stuff around us?

    Dembski: According to Kauffman, “No one knows.”

    That’s a rather facile dismissal of Kauffman’s position, when Kauffman proposes co-evolution as a means for life to structure its own environment.

    More particularly, the very stuff of the universe is structured by *proximity*, and on a deeper level by *symmetry*. For instance, if there is matter in a given place, there is a higher likelihood of more matter nearby. If atoms are collected together, they will interact due to proximity and form three-dimensional structures that have properties not found in the individual components. In this sort of universe, anything that searches by steps is going to more successful than average.

     

  13. Dembski does Weasel, again.

    Dembski: For the target phrase METHINKS IT IS LIKE A WEASEL, Dawkins bypasses the Shakespeare hypothesis — that would be too obvious and too intelligent-design friendly… Dawkins therefore got rid of Shakespeare as the author of METHINKS IT IS LIKE A WEASEL, only to reintroduce him as the (co)author of the fitness landscape that facilitates the evolution of METHINKS IT IS LIKE A WEASEL.

    The target phrase could be a random sequence, and the algorithm would work just as effectively. In other words, the algorithm is not tailored to the specific fitness landscape.

  14. The shape of a puddle is found by a search of the space of a pothole.  How did the puddle know when its molecules finally found the information about the shape of the pothole so that it could then settle in with an exact fit?

  15. I don’t see anyone commenting on the most important property of Dembski and Marks’s “Search for a Search” (of which this latest article is a popularization).   A number of people have pointed out that this represents a step backwards in Dembski’s argument, to one that does not rule out theistic evolution.

    I wrote two posts at Panda’s Thumb in 2009 (here and here) explaining that Dembski and Marks’s conservation law argues that if specified information shows up in the genome, that it was already lying around out there, in the form of the shape of the fitness surface.   And in that case what transferred this information into the genome?  Natural selection!

    So it’s a case of the universe being set up so as to allow natural selection to work and put adaptive information into the genome.  It’s just that Dembski wants to argue that a Designer set the whole thing up in the first place.  That is at most a theistic evolution argument — Dembski and Marks’s Search For A Search argument does not present any reason to say that a Designer intervened in the process later on.

    (It’s a step back away from Dembski’s Design Inference argument which held that there were theorems showing that 500 bits or more of specified information could not be put into the genome.  Now presumably they are conceding that information can be put in, but it’s just that they want to invoke a Designer for the start of the universe).

    I argued in those PT posts that the smoothness of fitness surfaces is not necessarily the result of a Designer choosing them out of all possible, mostly-very-jagged fitness surfaces.   It could just be the result of the weakness of long-range interactions in physics.  A very jagged “white noise” fitness surface assumes that any little change in the genome causes the whole organism to die.  Everything in the organism is then interacting incredibly strongly with everything else.  Real physics doesn’t work that way.

    I think all the discussion here about needles in haystacks misses the fact that Dembski’s (and Marks’s) argument is not that natural selection doesn’t work but that if it does work, a Designer still did it.  I suspect that in reality physics is actually what did it.

    In any case, even if they could somehow show that our universe was specially designed by a Designer to allow natural selection to work, they would have still conceded that here, in our universe, natural selection can work and can explain adaptation.

     

  16. I hate to keep posting and driving other people’s comments off the page, but that is wonderful.

    Perhaps evilutionists should flip the entropy argument by describing functional sequences as attractors or gravity wells rather than hills.

  17. The argument about “why” the universe supports life and evolution is not as important as whether it does. The first is theology and the second is science.

    Religion gradually accepted astronomy and cosmology, reluctantly accepted geology. Biology remains to be accepted. I suspect physics will always have its mystical interpreters, but they do not really interfere with the math.

  18. Joe-I’m in agreement with you.

    Unless I’ve badly understood the article, I think the only options this article leaves for ID are actually demonstrating the intelligently-designed and optimized evolutionary search algorithm (never going to happen), or claiming the environment is fined tuned to put information into the system while simultaneously reverting to origin-of-life questions. 

    This paragraph, in particular, puzzled me: “Nylon, for instance, is a synthetic product invented by humans in 1935, and thus was absent from bacteria for most of their history. And yet, bacteria have evolved the ability to digest nylon by developing the enzyme nylonase. Yes, these bacteria are gaining new information, but they are gaining it from their environments, environments that, presumably, need not be subject to intelligent guidance. No experimenter, applying artificial selection, for instance, set out to produce nylonase.” 

    The information “smuggled,” as Dembski and Marks have put it, into an evolutionary algorithm, or life itself, is from the environment. Although a lot of words follow, I don’t see an ID-friendly answer to what has been confessed here. How many UD denizens would tell you there are NO gains of information observed, and that the environment cannot convey information?

    The other paragraph that puzzled me is: “The key line in the above quote is, “In real life of course, the criterion for optimisation is not an arbitrarily chosen distant target but SURVIVAL.” Survival is certainly a necessary condition for life to evolve. If you’re not surviving, you’re dead, and if you’re dead, you’re not evolving — period. But to call “survival,” writ large, a criterion for optimization is ludicrous. ”

    Why? The eventual answer involves a Boeing engineer describing penalty functions in his algorithm. So: “So what makes the difference? It’s that the engineer, with knowledge of the problem he’s trying to solve, carefully adapts the penalty function to the problem and thereby raises the probability of successfully finding a solution.” 

    Odd. ‘Penalty functions’ that are analogous to those in nature-grow faster, or at higher temperatures, or in the presence of some new compound, have produced interesting information changes in the lab. But we’re to take Dembski’s word for it that nature, unguided, can’t? Why?

  19. Perhaps evilutionists should flip the entropy argument by describing functional sequences as attractors or gravity wells rather than hills.

    That’s always been my preferred metaphor anyway. It makes more intuitive sense to envisage a space with inaccessible steeps and attractive wells, gravity standing in for selective advantage – the steeper the gradient, the more keenly the the ‘force’ acts.

  20. As usual, Robb poses the interesting questions….

    I’ve been commenting at UncommonDescent for four or five years – and I can’t remember a time when I wasn’t in moderation (and I’m not the insulting type).  Perhaps I should be glad that I’ve never been banned…

    Here my latest comment – perhaps this one will appear (my last one didn’t – but I should know better than to ask KairosFocus to be concise…)

     

    Dear Dr. Dembski – as evolutionnews doesn’t allow comments on this article, I hope that you take the opportunity to discuss your insights on this your former blog!

    BTW: I can’t help to notice that now I’m being mentioned twice in the article “The Search for a Search”! May I expect a bottle of whisky (or some other token of your appreciation) for my contributions to the idea of active information?

  21. Joe Felsenstein writes:

    I don’t see anyone commenting on the most important property of Dembski and Marks’s “Search for a Search” (of which this latest article is a popularization). A number of people have pointed out that this represents a step backwards in Dembski’s argument, to one that does not rule out theistic evolution.

    I challenged Dembski on this point in 2009 (posting as ‘beelzebub’):

    beelzebub       May 3, 2009 at 1:47 pm

    Dr. Dembski,

    As far as I can see, nowhere in the paper do you and Dr. Marks express skepticism regarding the ability of Darwinian evolution to account for the diversity of life. Rather, you seem to grant that Darwin was right that random mutation and natural selection are sufficiently powerful, provided that fitness has a teleological origin. That seems like a huge departure from your former indictment of Darwinian theory as flawed and unsupported by the evidence.

    Have you in fact shifted your position? How has the ID camp reacted to your paper?

    Dembski’s rather unhelpful reply:

    William Dembski       May 3, 2009 at 4:28 pm

    Beelzebub: This paper was written under the supposition that common descent holds and that natural selection is the principal mechanism behind it. Writing under a supposition does not mean accepting it. My own views of the truth of the matter are clearly spelled out in THE DESIGN OF LIFE (http://www.thedesignoflife.com). In particular, I think that irreducible complexity at the molecular level (especially in the origin of DNA and protein synthesis) provides compelling evidence for discontinuity in the history of life.

  22. Thanks for noting your UD comment raising this issue.  A few thoughts:

    1. Of course Dembski is allowed to say that his and Marks’s S4S applies if natural selection works.

    2. Where he is contradicting his previous arguments is that those arguments were supposed to show that natural selection could not put specified information into the genome (in effect that natural selection could not improve adaptation).  Now he is building into his model that natural selection can improve adaptation.

    3. It is also rather remarkable that the argument he cites against evolution is not his own Law of Conservation of Complex Specified Information (which is what is alleged to make his Design Inference work), nor does he cite his No Free Lunch argument.   He cites Michael Behe’s arguments, as if his own had been knocked down and he knew it.  But if he sees that, he has not been willing to acknowledge it, and he even reaffirmed his belief in those arguments a couple of years ago at UD.

  23. So the only actual discontinuity that Dembski can point to is OOL. I wonder what the fallback position will be as OOL becomes increasingly Millerized.

    I think I’ll coin the word Millerization to denote the tendency in all branches of science to replace supernatural agency with regularity. It started in cosmology. Then geology. Then chemistry. It’s beginning in neuroscience. We still have spooks lurking in quantum phenomena. Physics has its OOE discontinuity.

  24. But Behe’s Edge is not an argument from any principle. It is just a guess about sparseness. It is a conjecture about the number and spacing of needles. Something testable.

  25. Dembski: Nylon, for instance, is a synthetic product invented by humans in 1935, and thus was absent from bacteria for most of their history. And yet, bacteria have evolved the ability to digest nylon by developing the enzyme nylonase. Yes, these bacteria are gaining new information, but they are gaining it from their environments, environments that, presumably, need not be subject to intelligent guidance. No experimenter, applying artificial selection, for instance, set out to produce nylonase.

    This is a bit of tangent, but for those of us interested in “semiotics”, I think we could see the evolution of a new semiotic system here. The information recieved by the population group from nylon is not nylonase, nor is it the same as the action of eating nylon. Mutation has created a new protocol for the transfer of recorded information. Therefore, nylon is (or contains) recorded information.

    Irreducibly complex? Clearly not!

    Sorry about the O.T., but Robb’s superbly on topic elsewhere.

  26. KF:

    After years of failed attempts — the last one was to suggest clocks evolving in a simulation from gears, pivots and levers, from someone who did not understand how hard it is to cut a functional gear and to set up stable precision backing plates that support them — it is obvious that there are no good counter examples.

    I think he means this: http://www.youtube.com/watch?v=mcAq9bmCeR0

    I have to say I dislike the tone of the text on this. But this is simply a bog-standard GA taking a set of predefined digital ‘components’, giving them some properties and assaying randomly mutated results for ‘clock-like’ activity. KF’s objection – the fact that in reality the components are quite hard to make – rather misses the point! It is akin to saying that digital ‘organisms’ are not a good analogue of evolution because it’s hard to make real ribs.

  27. I’d also like to see how he applies this to biological systems.  The example you quote at your blog shows Dembski, yet again, assuming a uniform distribution (in this case of hypothetical machines).  He occasionally seems to recognize that biological evolution only “searches” near known good solutions, at which point he switches to his privileged universe spiel, but he always keeps the shells moving.  Wouldn’t want the faithful to suspect that he recognizes that evolutionary mechanisms work in the only universe we have access to, after all.
     

  28. kairosfocus: “Indeed, to promote robust adaptability it would make sense to build in a mechanism to do adaptations by chance incremental variation and niche exploitation. “

    Looks like KF is saying that the “designer” designed “Darwinism” into life.

    I don’t know how the ID side can now claim that “Darwinism” as a mechanism does not exist.

    Joe and Dembski won’t be happy.

     

  29. … there is always this naive concentration on primary sequence too, and on the apparent assumption that any two amino acids are as unalike as any other. Acid properties cluster, and their substitutability depends very much on context, which again is a folding matter.

    Activity is often the inclusion of an approximate motif, rather than every acid being sacred.

    And neighbours can also be much closer than they might seem, in terms of the probability with which a particular transformation may occur – because of frame shift, sense/antisense reversal or other gross switches, apparently very distant places can be probed – all from the comparative security of a functional version, no harm done if it’s a disaster.

    All of this is likely to enhance the interconnectedness of local areas, over a rather ‘flat’ view with function sparsely peppered. Which is important for navigation around it.

  30. Cross-posted from AtBC:

    Okay, here’s my best straight-faced attempt at presenting Dembski’s argument:

    1. A search is a process which attempts to find targets within a larger space of possibilities.

    2. A search can be run over and over. Each time it is run, it “lands on” one point in the space of possibilities. If that point is one of the targets, then the search has succeeded. If the landing point is not a target, then the search has failed.

    3. A blind search is one in which the “landing point” is chosen purely at random out of the space of possibilities, without favoring any points over others.

    4. If the possibility space is huge and the target space is tiny, then a blind search will rarely succeed. The odds of hitting the target are just too low. In other words, the cost of finding the target is high with a blind search.

    5. If we use a better search, we improve the odds of hitting the target. In other words, we can reduce the cost of finding the target by employing a better search.

    6. However, finding a better search is itself a search (“the search for a search”). It has its own cost, which must be factored in.

    7. The total cost of finding a target therefore includes both the cost of the search plus the cost of the “search for a search”.

    8. According to the Law of Conservation of Information, this total cost is always greater than or equal to the cost of finding the target through a blind search. One way or another, you have to pay the piper in order to find the target.

    9. Evolution is a search: it looks for [i]viable[/i] organisms (the targets) within the much larger space of [i]possible[/i] organisms.

    10. Evolution obviously cannot succeed as a blind search, because the target space is too small relative to the possibility space. However, evolution uses the fitness landscape as a source of information to zero in on the target space. (A designer may also inject information at crucial moments.)

    11. The fitness landscape doesn’t come for free. The total cost of the evolutionary search has to include the cost of the information contained in the fitness landscape.

    12. The Law of Conservation of Information tells us that the total cost of the evolutionary search, including the cost of the information contained in the fitness landscape, equals or exceeds the cost of a blind search.

    13. Purely material processes don’t generate information. They merely rearrange information that was already there. Therefore, no material process can “buy” you a fitness landscape.

    14. Thus, the information in the fitness landscape comes from an immaterial intelligence. (And so does any information that is injected along the way.)

    15. Without this information, evolution could not succeed.

    16. One way or another, then, evolution depends for its success on information generated by an immaterial intelligence.

    It’s riddled with holes, but that, to the best of my knowledge, is the argument that Dembski is actually making.

  31. One gaping hole: purely physical systems can learn and adapt. So any theoretical objection to information flow from environment to genome is untenable. It’s back to gaps.

    Meanwhile, Joe has been fun at DiEB’s blog. It appears that to model the search for a Dembski machine you must first search for a computer.

  32. Petrushka writes:

    So any theoretical objection to information flow from environment to genome is untenable.

    What’s remarkable about Dembski’s position is that he concedes that information can flow from the environment to the genome, at least with the right kind of fitness landscape. He just doubts that we actually have that sort of fitness landscape:

    It would actually be quite a remarkable property of nature if fitness across biological configuration space were so distributed that advantages could be cumulated gradually by a Darwinian process. Frankly, I don’t see the evidence for this.

    The core of his argument is that if we do have the right kind of fitness landscape, then the information it contains must have been placed there (directly or indirectly) by an immaterial intelligence:

    If biological evolution proceeds by a gradual accrual of functional advantages, instead of finding itself deadlocked on isolated islands of function surrounded by vast seas of non-function, then the fitness landscape over biological configuration space has to be very special indeed (recall Stuart Kauffman’s comments to that effect earlier in this piece). Conservation of information goes further and says that any information we see coming out of the evolutionary process was already there in this fitness landscape or in some other aspect of the environment or was inserted by an intervening intelligence. What conservation of information guarantees did not happen is that the evolutionary process created this information from scratch.

    And:

    One final question remains, namely, what is the source of information in nature that allows targets to be successfully searched? If blind material forces can only redistribute existing information, then where does the information that allows for successful search, whether in biological evolution or in evolutionary computing or in cosmological fine-tuning or wherever, come from in the first place? The answer will by now be obvious: from intelligence. On materialist principles, intelligence is not real but an epiphenomenon of underlying material processes. But if intelligence is real and has real causal powers, it can do more than merely redistribute information — it can also create it.

  33. It’s a ridiculous argument, of course, but it’s progress of a sort.  Dembski is at least conceding that a Darwinian process can produce biological complexity with the right kind of fitness landscape.

    And of course neither he nor the bloviating KairosFocus has shown that we don’t have the right kind of fitness landscapes, or that the ones we have consist of “isolated islands of function surrounded by vast seas of non-function”.

    It’s pure wishful thinking:  I’ll bet the fitness landscape consists of unconnected islands, and oh, by the way, we need a Designer.  Why?  To bridge the gaps between islands, of course!

  34. I notice that Dembski claims:

    it’s worth noting that none of the critiques of this work has appeared in the peer-reviewed scientific/engineering literature, although a few have appeared in the philosophy of science literature, such as Philosophy and Biology; most of the critiques are Internet diatribes)

    Is anyone interested in working with me to submit a criticism to one of the IT journals?  I am thinking specifically of the journals he published in.  If so, contact me on mtf1c08@soton.ac.uk.

  35. If biological evolution proceeds by a gradual accrual of functional advantages, instead of finding itself deadlocked on isolated islands of function surrounded by vast seas of non-function, then the fitness landscape over biological configuration space has to be very special indeed (recall Stuart Kauffman’s comments to that effect earlier in this piece).   

    Yes, the kind of fitness landscape that allows a puddle to fit its hole, which has an analog in the lock-and-key mechanism of enzymatic activity.  

    But as you say, it’s progress of a kind for Dembski. 

  36. What’s remarkable about Dembski’s position is that he concedes that information canflow from the environment to the genome, at least with the right kind of fitness landscape. He just doubts that we actually have that sort of fitness landscape:

    It would actually be quite a remarkable property of nature if fitness across biological configuration space were so distributed that advantages could be cumulated gradually by a Darwinian process. Frankly, I don’t see the evidence for this.

    I hadn’t seen this before.  This sounds like the standard creationist claim that only “microevolution” is possible, where “microevolution” is defined as “evolution we’ve observed and would look like even bigger idiots for denying.”

    He makes this claim in the same breath as talking about how improbable it is for us to be in a fitness landscape like the one we observe, which is nothing more than the anthropological argument for god.  He seems to be just throwing old creationist arguments at the wall to see which ones still might stick.
     

  37. Thinking of evolution as a search for the highest peak in a fitness landscape is tempting.  But it is a mistake in that there is no necessary failure if the process fails to find that peak.   As long as evolution can continue to make progress in its region of the genotype space, as the fitness landscape gradually changes through time, we can regard it as successful.

    We can be pretty sure we aren’t in the region of the adaptive landscape that leads to the most perfect organisms.  Perhaps they would be able to fly through the air at 500 miles per hour, zooming down occasionally to bore through solid rock while composing sonatas all the while.  Perhaps — we don’t know what such organisms would look like. Instead we’re in a “good-enough” region of the fitness landscape, and evolution makes enough progress.

  38. I think the KariosFocus argument works something like this:

    1. If the fitness landscape had connectable islands of function then evolution would be possible.
    2. We know from First Principles that evolution is impossible.
    3. Therefore islands of function are not connectable.

    I think that models his line of reasoning pretty well.

  39. Joe Felsenstein: As long as evolution can continue to make progress in its region of the genotype space, as the fitness landscape gradually changes through time, we can regard it as successful.

    True, but when it comes to a mathematical assertion of the form Dembski uses, only a single example is sufficient. The environment transfers huge amounts of information to the puddle. The environment is very ‘special’ in that way.

  40. Joe Felsenstein:

    Thinking of evolution as a search for the highest peak in a fitness landscape is tempting.  But it is a mistake in that there is no necessary failure if the process fails to find that peak.

    Further, is it not apt to  visualize the topography of fitness landscapes as undulating in several dimensions over evolutionary timescales, such that what was relatively selectively optimized at time A is less so at time B? Organisms therefore succeed by sustaining themselves on the sides of surmountable inclines, the peaks of which may constantly traverse particular dimensions. Sort of surfing backward (surfers are constantly “descending” without changing altitude, rather than climbing).

  41. Yes, that is nice.

    We might even be able to add that just as the “surfer” moves due to the wave, fitness can “surf” from one “island” to the next as the environment of one island starts to look more like another island’s.

    There’s KF’s “islands” argument down the tube!

     

     

  42. One really nice feature of the surfing metaphor is it combines change and homeostasis. I’m kind of excited about that.

  43. Yes, kairosfocus’s “island” is the “wave” and it moves.

    In this analogy, our surfer could transit to another “wave” if he got close enough to one.

     

  44. Or wipe out.

    I offer a possible blog post title from a childhood book. Surfing The Web Of Life.

  45. Another problem with the ‘islands of function’ metaphor is that it encourages people to think in just three dimensions and to apply the intuitions they’ve acquired from a lifetime of operating in three-dimensional space. This can be very misleading.

    In our three-dimensional world an island only has to be surrounded by water in two dimensions. That is, if you move far enough north, south, east or west, or any combination of these directions, you’ll find yourself underwater.

    In an n-dimensional space an ‘island’ has to be surrounded by water in n-1 dimensions. If there is an isthmus along any one of those n-1 dimensions, or any combination of those n-1 dimensions, then you don’t have an island. Thus ‘islands’ become less likely as the dimensionality increases.

    Real fitness landscapes have thousands of dimensions. How likely is it that each of those dimensions, plus every combination of those dimensions, leads underwater?

    The weaknesses raised in this thread have all been brought to KF’s attention at UD, and more than once. Not surprisingly, he’s chosen each time to ignore the challenge instead of confronting it.

  46. I brought up the thousands of dimensions point with gpuccio when he argued a designer could use targeted selection. He just waved it off. I argued that natural selection sees all dimensions simultaneously. To me this is a major argument against design. Unless the Designer is You Know Who.

  47. Joe:”In Dembski’s scenario we can get the final target of a 6 without securing the “6″ machine. And we would never know if the 6 we got was from the “6″ machine or any of the other 5 machines. “

    Why would Dembski even bring something as mathematically conceptual as this to an argument that concerns itself with physical realities?

    That would be like actually calculating the odds of where in your cup, a half cup of water will end up.

     

  48. Dembski’s description of the problem is completely hosed. It’s interesting Joe would answer on UD the question I asked him on DiEblog. The target is a 6. You never know if you picked the machine rich in 6s, regardless of the outcome. You could, of course, run thousands of trials and figure out how the machines are configured, but that’s not allowed under the rules of this game. If it were allowed, then the search for a search could easily find a rule that would outperform a blind search.

  49. Yes, kairosfocus’s “island” is the “wave” and it moves.

    Indeed. The environment, the designer, is dynamic and modelling attempts that don’t account for the multi-dimensionality of the environment are doomed to failure.

  50. kairosfocus: “And, sooner or later, the dirty objector tactics are going to backfire bigtime. “

    The ID side has taken the role of objector since evolution has a mechanism and Dembski won’t even accept that ID has to lower itself to that level of detail.

    Until ID works out a mechanism, they have nothing to teach.

    As for fine-tuning of the universe, I consider that one of ID’s weakest arguments.

     

  51. petrushka quoth (about an adaptive landscape that changes through time):

    That seriously changes my understanding of punctuated equilibrium.

    Indeed.  This was raised in the debates in the 1980s between advocates of punctuated equilibrium and theoretical population geneticists.  There are rather ordinary population-genetic mechanisms for punctuated patterns of change, so the observation of the pattern does not establish that this requires a new mechanism. There’s a really nice paper on punctuated patterns arising from changes in an adaptive landscape by Mark Kirkpatrick in American Naturalist in 1982

    (petrushka again):

    One really nice feature of the surfing metaphor is it combines change and homeostasis. I’m kind of excited about that.

    Careful.  I suspect you are referring to C.H. Waddington’s image of “canalization” in development as little balls rolling down surfaces.  But that is in developmental time in the development of one individual, and this is in evolutionary time.  Those are different.

    (petrushka):

    Or wipe out.

    OK, my brain has failed.  How does one collapse into the adaptive landscape and darned near drown? Or get hit on the head by your surfboard?

    For anyone who thinks that adaptive landscapes are fanciful analogies of no use to actual evolutionary biologists — not so.  I just finished last month co-teaching a weeklong workshop course at the National Evolutionary Synthesis Center on Evolutionary Quantitative Genetics, and the notion is central to that work.   You can all listen to audio recordings of our lectures here (and the slides are there as PDF or PPT files too).  And you can run the lab exercises too.

  52. Well one difference between this site and UD is that we can disagree without being accusatory, and I can accept correction. Wipe out was a fanciful term for extinction. I suppose I’m wrong about stasis, but I was conjecturing that as long as a body type is adapted there will negative selection for the kinds of somatic changes that would show up in fossils.

    I assume that molecular evolution continues during equillibrium periods.

  53. I started listening. Some lectures seem to be unavailable. And I am obviously in over my head on tbis.

    I just assumed that long stretches with no obvious physical changes indicated some contstraint. I guess I had in the back of my mind the possibly erronious thought that feral dog populations tend to revert to one form, even though they might carry alleles of various breeds. Fossils wouldn’t indicate the hidden potential for variety. My thinking on this is obviouusly unencumbered by actual knowledge.

    I seem to have read that large populations re slower to evolve and founder populations faster. But a large population seems indicative of success. A significant change in environment or predation might reduce the population.

  54. petrushka wrote (about the online postings of audio files and slides for my NESCENT summer course):

    I started listening. Some lectures seem to be unavailable.

    I am embarrassed.  They are almost all missing!  I just wrote to the person who should have put them there and asked that the proper links be made available there.  I hope that his can happen in a day or two.  A couple of the guest lectures (Mackay and Losos) were not recorded, as I had not asked them for permission to record and they contain unpublished research trsults.  Also the computational lab sessions were not recorded as there were no lectures then.  Thanks for pointing this out.

  55. Given that Dembski wisely does not dirty his hands dealing with objectors, perhaps his principal representatives on earth, Joe and Kairosfocus, can address these problems with an ‘islands of function’ approach to the exploration of the phase space of a particular n-length v-variants polymer (total loci v^n)

    1) Define ‘function’. Many amateur theoreticians are convinced that the only functional protein, for example, is one with catalytic activity. This is but one role that proteins play in organisms. Many of those functions – including catalysis itself – are not nearly as rigidly sequence-dependent and non-substitutable as design theorists appear to think.

    2) Overall function richness. If you don’t know how the space of ‘well-formed strings’ is structured wrt function (however defined), you can’t simply declare its structure to be islanded. All attempts to analogise (eg from English strings) are completely redundant. Words don’t fold up into tertiary structures with the capacity to affect thermodynamic potential.

    3) ‘Local’ function richness. Evolutionary exploration is from current established ‘bridgeheads’ of viability, not a series of madman’s leaps in the dark. So the structure of the distribution of function in the local region that functional strings in living organisms tend to occupy is more important than the overall structure, except at OoL (and the probably separate OoP – Origin of Protein).

    4) There is a non-static underlying fitness landscape. It is not a simple case of string changes probing points of fixed fitness. The adaptive landscape undulates and ripples continually, due to both external and internal changes to the ‘environment’ which is the ultimate arbiter of whether a change is beneficial, neutral or deleterious (and includes conspecifics). A shifting landscape can ‘capture’ points currently at sea.

    5) The interconnectedness of functional space. A 2- or 3-dimensional space can easily be mentally divided up into islands surrounded by seas – we inhabit such a space in ’3D-world’. But increasing the dimensionality increases the number of potential pathways from point to point, which affects the availability of directional selective ‘improvement’ or pathways of drift.

  56. Your point 1 there is one which has been reiterated time and time again, but which has never been acknowledged, let alone refuted, by ID “theoreticians”

    Even Axe, although he has in the past published at least one paper pointing out just that “substitutability”, seems just to skate over its import for ID 

  57. I have a dumb question about the size of the search being done by evolution. My ignorant understanding sees, let’s say, 30,000 genes, maybe 2000 bases per gene,  four possible values per base. I see that as 120 million possible point mutations (or 90 considering they already have a value).

    So if I’m off by two orders of magnitude, that could be 9000 million possible changes.

    A large number but not astronomical, and trivial compared to populations of bacteria over millions of years.

    How stupid is this?

    How difficult is it to do a blind, exhaustive search?

  58. Well, there has to be viability of intermediates. Exclusive point mutation might find itself a little stumped by blind alleys. But there are plenty of grosser mechanisms to shake things up – including the ID-er’s favourite, the transposon, whose occasional donation of beneficial sequence they regard as ‘the’ reason over half of our genome is stuffed with their debris.

    Replicating genomes are like a ‘superfluid’. Any viable crack or crevice in ‘search space’ and they can trickle through it. ID insists that function is sealed and watertight. But there is no reason to suppose that it is. And huge leaps in the dark can be made with ease – it does not have to be strictly stepwise across a tottering bridge of “dad to dam to dum to mum” (see – there are answers in Genesis, prog-rock fans!). Most will be detrimental but – as with any other kind of mutation – evolution proceeds via the ones that aren’t.

    It really depends on the structure of the space – strictly, the structure of historic space. Every species on earth may have nowhere left to go, but historic evolution could still have proceeded ‘naturally’ up to this point.

    One shell-game is constructing a 20^n space, because there are 20 amino acids now. But for the organism that first stitched together a peptide bond, there may have been as few as 1. Such a poly-amino-acid could still be ‘functional’, even if it could not catalyse a reaction. Then the first catalytic peptide … what’s the minimum specification for a catalytic peptide? Dembski? KF? Joe? Anyone?  

    Once the protein system is in place, the real turbo charger is recombinant sex. This takes proven viable half-genomes (though really, we are double-genomes), hooks them up for a life then, at the end, produces a whole bunch of gametes from stitched-up amalgams of the parental chromosomes. This gives a much more effective ‘search’, by probing regions that would take many single amendments (which may hit impassable ‘roadblocks’), and by recombining each ‘successful’ gene into the population independently of all the others and likewise dropping the deadweights. Without recombination, a gene’s selective advantage/disadvantage is subsumed within that of the collective it finds itself bound to. 

    Turning crossover on in GAs likewise often speeds up ‘search’ no end.

  59. But leaps into the dark are precisely what feeds ID delusions. I’m rather ignorant of details, but it is my understanding that most nonfatal leaps occur in noncoding sequences, or create redundant sequences through duplication, which are then available for stepwise modification.

    I’m looking for some specific range of numbers to counter the ten to the google search space claimed by ID

  60. OK, but those are going to sound suspiciously like search for a search. I’m looking for the size of the search space, the local vicinity. Something to counter the 10^500 argument. But it would be interesting to know what percentage of successful mutations are other than point mutations. I was under the impression it is low.

  61. Ummm … it’s complicated!

    http://www.genetics.org/content/148/4/1667.full

    I think there’s a danger of getting sucked into a game played by ‘their’ rules. They assert that the haystack is enormous and the number of needles tiny without doing any work to back that up, and we have to scuttle round to refute that assertion.

    It is also frequently hard to determine if they are talking of DNA space, protein space, a fitness landscape, a single sequence, a bacterium, a human …

    The papers linked elsewhere which found significant function in a tiny corner of the overall space by, essentially, massive ‘leaps in the dark’ certainly call the assertion into question – we are no more computationally equipped to assay the structure of the space than they are to make their assertions about it, but we do have empirical sampling data.

    One consideration is the mechanistic basis to the various transformations. This is not a ‘search for a search’, but simply stuff that happens in an imperfect replicative world (including meiotic recombination, which even without errors can be viewed as a kind of mass-mutation, with generally benign consequences).

    A genome will probe its ‘point-mutation neighbourhood’ readily, will hit a damaging fraction but can uncover any local benefit. But the logic of ‘leaps in the dark’ is the same. Each transformation occurs at a particular (if variable) rate, and a particular (if variable) fraction may prove to be beneficial. That fraction may be zero for really wild leaps, such as trisomies in humans (distinct from broken chromosomes, which are quite common). But we can see that such leaps have actually happened (if, that is, one accepts Common Descent! :D).They can shake genomes out of local fitness maxima, in much the same way that drift does at population level. People do much the same with GAs.

    If you look at DNA space, you can see how much closer regions really are, in terms of the probability that the commonest types of error will get there easily from a given point. Many leaps will fail; evolution proceeds courtesy of the ones that don’t. AGTCAGCTT is a simple step away from – for example – AGTGCAGCTT (1 insertion), AAGCTGCACT (antisense complement), and AGTCAAGTCAGCTT (short 5-base duplication). The peptides produced, should the sequence happen to frame-align within a translated region, are Ser-Gln-Leu, Ser-Ala-Ala, Lys-Leu-His and Ser-Gln-Val-Ser respectively. These are a significant distance apart in peptide space, but a short step mechanistically in DNA space. Then there are ‘chemical space’ and ‘folding space’ to consider – primary sequence determines it, but what ultimately matters are the chemical properties of the folded protein – trying to ‘digitise’ an analogue space is really rather misleading.

    If one considers a hypothetical early scenario with few amino acids, the space is fairly small. Every time you add an acid, you increase the space exponentially – but if you already had function in the smaller space, you aren’t going to make it RARER simply by adding an acid to your toolkit! Within 5 or 6 broad categories, the acids aren’t all that different from each other, but the variants increase the subtlety available in fine tuning.

Leave a Reply