61 thoughts on “A New View of Irreducible Complexity

  1. 1: I question the analogy of evolution and computation.

    2: You are making an argument about selection. But evolution involves both selection and mutation.

    3: You rule out mutation, by saying that mutation is random, and randomness cannot produce what we see. But you are ignoring possibility that there is a combination of mutation and selection.

  2. Neil Rickert,

    No, the whole thing is based on mutation. Without mutation nothing at all would happen. Mutation *alone* cannot produce it. The point is that natural selection requires a genotype/phenotype map whose selection consistently points in the same direction. However, complexity theory shows that the road to arbitrary features (i.e., features that were not implicit in the system ahead-of-time) necessarily has a chaotic mapping through that configuration space.

    Thus, for such features, you can only stumble upon them by chance. Information theory shows that chance grows exponentially large with the size of the minimum working system.

  3. johnnyb: The point is that natural selection requires a genotype/phenotype map whose selection consistently points in the same direction.

    I doubt that evolution has such a requirement.

    I should note that I’m not a Darwinist. That is, I don’t agree with the idea that evolution proceeds by means of natural selection optimizing the population.

  4. Neil Rickert: I doubt that evolution has such a requirement.

    I should note that I’m not a Darwinist.That is, I don’t agree with the idea that evolution proceeds by means of natural selection optimizing the population.

    But you are an evolutionist, right? That is, common descent of all species or something like it is true and it has its requirements. Why not give a positive account of what you hold to, so that actual discussion can occur? You know, exchange of ideas, stuff like that.

  5. How is the rornado in a junkyard a new argument?

    This still assumes that there is a direction or a goal. It’s the same shit for brains stupidity over and over.

    Take away the target and there is no probability calculation.

  6. 1:50 Irreducible complexity is a empirical definition of “holism”

    I like to regard myself as a holist, so irreducible complexity should go down well for me. So why doesn’t it?

    Wikipedia on holism, “Holism… is the idea that systems (physical, biological, chemical, social, economic, mental, linguistic, etc.) and their properties should be viewed as wholes, not just as a collection of parts.”

    Holism is a view of irreducibility, an anti-reductionist view. Nothing about complexity. Complexity must come from somewhere else.

    The catch is probably this, “Irreducible complexity is a empirical definition…” With its emphasis on the empirical parts of systems, IC somehow sees fundamental complexity. Or maybe it sees mechanisms and interactions of those parts as hard to explain and thus complex. The latter would be just silly, anti-scientific.

    In contrast, holism sees parts as essentially formal (i.e. arbitrary, insubstantial) subdivisions of systems. Parts/subdivisions are parts/subdivisions of something, this something is primary and its parts/subdivisions are secondary, i.e. dependent on it, and the something is ultimately simple, not complex, because all fundamentals are simple – irreducible, yes, but universal or general, applicable to all of its parts/subdivisions, if formulated correctly. Subtle, but not complex.

    This is as far as I got in the video. Maybe I’ll watch more some day.

  7. In other words, holism is not the view that parts can be explained by positing a designer (in the same slide at 1:50), but that parts can be explained by that parts belong to a whole, to a structured system.

  8. TristanM,
    The video seems to consist of presentation slides only, so simply turn audio off. That’s what I did. The opening made me do it.

  9. So anyway, what’s the conclusion of the subject? I mean, if we watch all the slides, and we understand them, what do we end up concluding?

  10. In one of the first slides called “Goal of IC”, it says:

    If a system doesn’t function without one of it’s parts, the a precursor system would have difficulty evolving to it.

    What is “difficulty” evolving? Does it mean it’s sorta slow, and takes long? If so, how long? It’s too vague to make much sense of as it stands.

    Edit: I see now you were criticising that exact point and are at least pretending to offer a more rigorous definition.

  11. I like how you openly state that one of your goals is to show that “Darwinism” is logically impossible. It’s always encouraging when people are open about when they put their conclusion first (and that it’s based on nothing more than your intuition) before getting around to testing it.

  12. In the slide “Universal vs. Special Purpose Machines”, in the last point you write:

    Therefore, if biology is to evolve to environments it isn’t aware of ahead-of-time, then the proper mathematical model is the universal machine

    The proper mathematical model of what? Of evolution? Why? It’s not at all clear why you say this.

  13. Rumraket:
    I like how you openly state that one of your goals is to show that “Darwinism” is logically impossible. It’s always encouraging when people are open about when they put their conclusion first (and that it’s based on nothing more than your intuition) before getting around to testing it.

    Goal is not the same as conclusion.

    The slide “Redefinition of IC” starting at 25:10 posits a “hard problem” for Darwinist natural selection. I guess that’s the core of the video. I don’t think that things that lead up to it hold neatly enough for that thesis to be something very relevant to reality or to Darwinism. It is what it is.

  14. Erik: Goal is not the same as conclusion.

    True but, I don’t believe this guy set out to *test* anything. I’m tired of extending the benefit of the doubt to these charlatans.

  15. I think the whole ID discussion should be about computability.
    So I can’t wait to get into your presentation. I’m very busy but perhaps this weekend.

    Erik: Holism is a view of irreducibility, an anti-reductionist view. Nothing about complexity. Complexity must come from somewhere else.

    I think this is interesting for me “complexity” is about the size of the holistic system it’s not necessarily an empirical measurement.

    Both PI and my phone number are irreducibly complex
    Pi is more complex than my phone number but neither is empirical.

    peace

  16. For those who want a transcript, I don’t have it, but it is based on a paper I wrote a while back. The funny thing is, when I present the paper, everyone wants a video. When I present a video, everyone wants a paper 🙂 Anyway, the paper is here.

    For those who want cliff notes, here is a discussion of the basic principle, a few years prior to the paper if I recall correctly. Here is my announcement of the paper which includes a little bit more of the personal history behind the ideas.

    The funny thing is that my claim, if false, is almost trivially falsifiable. Find me a genetic algorithm that reliably creates open-ended loops as part of its problem-solving.

    As I showed in the video, you can actually use this to detect the designed parts of Avida organisms.

  17. Erik,

    Erik:
    In other words, holism is not the view that parts can be explained by positing a designer (in the same slide at 1:50), but that parts can be explained by that parts belong to a whole, to a structured system.

    You are confusing the meaning of the term with its implication. I agree with you 100% on the meaning of the term (that it is explained as belonging to a whole), but my additional point was that the existence of holism points to a designer (as the explanation of the whole). In other words, the part is explained by the whole, and the whole is explained by the design and teleology of which it is a part.

  18. johnnyb:
    Erik,
    ….the whole is explained by the design and teleology of which it is a part.

    That’s not how it works. In holism, the whole and the part are radically distinct concepts. The whole is not a part of anything. It’s the whole, full stop, and everything else is a subdivision of it.

    You can delimit a structure or area for special analysis. When you do it, you stay strictly on the analysis of that area. The analysis of it does not warrant conclusions outside of the delimited area.

  19. Rumraket: True but, I don’t believe this guy set out to *test* anything. I’m tired of extending the benefit of the doubt to these charlatans.

    Johnnyb gave a convincing argument. Well, at least he managed to convince himself.

  20. Irreducible Complexity generates a lot of noise, but does little to further the cause of Intelligent Design. Consider that a Designer capable of complex creation must also be capable of creating the very simple, down to the level of individual mutations and selections. Such a Designer cannot be falsified.

  21. johnnyb,

    However, complexity theory shows that the road to arbitrary features (i.e., features that were not implicit in the system ahead-of-time) necessarily has a chaotic mapping through that configuration space.

    The name irreducible complexity is a buzz word which I know you are aware of. It was positioned to counter Darwin’s claim of evolution being slow small steps.

    I think your idea has merit but I am not sure the best value is trying to improve IC. Showing that the genome behaves like a class 4 computer is an interesting idea. Perhaps embryo development may be a way to demonstrate this comparison where animal body plans are actually developed from a single cell.

  22. johnnyb:
    …my additional point was that the existence of holism points to a designer…

    If you absolutely need to posit a designer, then from the holistic point of view the designer is part of the system and must be declared, defined, and described as such. Otherwise you are not employing the holistic point of view, just exploiting the good name of holism to deceive people.

  23. Neil Rickert: I should note that I’m not a Darwinist. That is, I don’t agree with the idea that evolution proceeds by means of natural selection optimizing the population.

    Evolution by natural selection is not an optimization process. It’s a sampling process, and fitness is bias in the sampling. (Fitness is the propensity of a type of organism, defined in terms of heritable traits, to leave offspring. If one type is twice as fit as another type, then an organism of the more-fit type tends to leave twice as many offspring as an organism of the less-fit type. The sense in which this constitutes bias in evolutionary sampling is fairly clear, I think.) Under various models, the evolutionary process converges to an equilibrium distribution on fitness. The mean fitness of individuals over the long term is not necessarily close to the maximum. Basically, the higher the mutation rate, the lower the mean fitness over the long term. What I’m saying is old hat to population geneticists. See Joe Felsenstein’s “Wright, Fisher, and the Weasel,” and also the comments.

    Here’s an animation of an evolutionary process (orange) I obtained by fine-tuning a model to speed the occurrence of maximum fitness. The blue process differs from the orange only in initialization. You’ll see that the processes settle into statistical equilibrium. It’s nothing like optimization. I hope to post on this soon. But with me, you never know.

  24. The opening shot is hilarious. Am I the only person who checks formal expressions in photos to see if they serve any purpose but to impress the viewer?

  25. johnnyb,

    A couple times in your presentation you present your intuitions as being those of an engineer. But your personal intuitions about nature don’t come from being an engineer, they come from being a creationist. We engineers don’t include non-material intelligent agents in our models of nature, not any more than scientists include them in their models of nature.

    Your creationist intuitions about nature are not part of the general culture of the engineering profession. Many engineers personally share your intuitions about nature, but as a group we haven’t found them useful in our work. We are like the scientists in this regard.

  26. Excellent points. Adding to IC is welcome although its a famous idea already.
    The whole IC thing does come down to a scientific investigation that demonstrates at the last stage of reducing living things at their working elements STILL is complex beyond chance mutations cooperating in creating them.
    This is the killer point of IC in ID. It does kill bad guys international.

  27. ‘Irreducible complexity’ is pretty much unavoidable for any genome containing interacting parts, because of evolution. Yes, even ‘stepwise’ evolution. To take a simple case of two proteins that may bind each other, assuming that any binding is better than none, initially random association is selected (screw your equivocations, IDists, that’s the word used!). If still closer association is favoured (I said screw your equivocations!), small changes in either component can also be selected. Eventually, we have an interacting system that cannot be simply destroyed without adverse consequence – the association has become embedded, and may well be pinned in place by further interactions outside the initial pair. It’s co-evolution; each element evolves in the presence of the others.

    Other scenarios can be constructed – Muller’s for example, involving loss of parts. Computer programs are a poor analogy for this kind of process.

    So, if one points to a system that is ‘irreducibly complex’, and says it is a problem for ‘Darwinism’, one needs to rule out the kind of irreducible complexity that is unavoidable in evolution, is due entirely to ‘Darwinism’, and whose absence would be the odd thing, not its presence. One is, to my mind, in the position of having to prove the negative.

  28. petrushka: Evolution doesn’t solve problems.

    This is correct even in evolutionary search. There are two components of a search, one of which generates a sample of possible solutions to a problem, and the other of which outputs the best solution it can find in the sample. We refer to the search as evolutionary when the sampling is done by simulation of an evolutionary process, even though the solution-seeking component bears no relation to biological evolution.

  29. Allan Miller: Computer programs are a poor analogy for this kind of process.

    That depends on the model of computation. Post production systems and Markov algorithms are specified by a bunch of little rules. I bring this up because Jonathan Bartlett latches onto whatever suits his creationist preconceptions, and does not seek to refute his own claims.

  30. Allan Miller:
    ‘Irreducible complexity’ is pretty much unavoidable for any genome containing interacting parts, because of evolution. Yes, even ‘stepwise’ evolution.

    Yes, this has been pointed out by many people. Having parts that now, after some process of evolution, are Irreducibly Complex, does not mean that they cannot have arrived in that state by a stepwise evolutionary process. ID advocates need to find a way to rule that out before they can use the presence of IC as evidence that something other than ordinary evolutionary processes was involved.

  31. In the presentation, right near the beginning, johnnyb says

    IC systems have to made holistically — the parts with the end in mind,

    Not so. There are stepwise ways too. So, given that the objective of the talk is to establish a criterion that detects something that must be made “holistically”, we can stop right there and ask for evidence of that assertion.

  32. As Patrick has pointed out, above, a classical genetic algorithm may not be able, most of the time, to evolve computer programs. But there is a variation called Genetic Programming (see here) that is able to do that. The mutations are not single-character changes but swaps of subtrees in a tree-structured program. They are still random with respect to fitness but now produce runnable programs much more frequently.

  33. Looking further into johnnyb’s presentation we find an assertion that seems to be crucial to his argument, but which I find dubious:

    If the set of needed functions is not-known-ahead-of-time one must use a Universal machine.

    Therefore, if biology is to evolve to environments it isn’t aware of ahead-of-time, then the proper mathematical model is the Universal machine.

    Biological evolution does not always succeed in adapting to new environments. Extinction is a real possibility. Adaptation is less than perfect. So I do not see why biological systems must model Universal computation.

  34. I finally looked ahead at VJTorley’s response to johnnyb’s argument. I see that Torley had already fastened on the same two points that I have just made above.

    I agree with Torley’s critique and his view that those two issues are crucial.

  35. Tom English: There are two components of a search, one of which generates a sample of possible solutions to a problem, and the other of which outputs the best solution it can find in the sample.

    In biology, as others have pointed out, there is no problem to solve. Variations occur, and some variants survive in subsequent generations, and some don’t. There is no problem solving behavior. There may be differential survival due to some attribute of a variation, and we may call this natural selection, but it is more like water finding the bottom of the pond than it is like climbing a mountain. All analogies are, at some point overstretched.

    If a change in the environment occurs that requires an adaptive change, the most likely outcome is extinction.

    Gould pointed out that a drunkard’s walk is likely to produce drift away from the point of origin, but it is fallacious to attribute the direction of drift to intention.

  36. petrushka: In biology, as others have pointed out, there is no problem to solve.

    My point was that, even when simulated evolution is employed in a problem-solving procedure, the role of evolution is only to sample the solution space of the problem. A sampling process does not gain information by processing data, e.g., fitnesses, associated with sampled solutions. (It is impossible to deduce the fitnesses of yet-to-be-sampled solutions from the fitnesses of sampled solutions.) The evolutionary informatics strain of ID says precisely the opposite, referring to the fitness function as an “external information source.”

    I don’t want to quibble over fine points. But there is an identifiable problem when the continued existence of a lineage depends on its acquisition of a particular trait. Of course, I am not saying that something sought to produce chloroquine resistance in malarial parasites. I am saying that when there actually is such an existential threat, the lineage very well may come to an end. Evolutionary biologists are not telling us that evolution is good at producing specific adaptations “on demand.” [ETA:

    Joe Felsenstein: Biological evolution does not always succeed in adapting to new environments. Extinction is a real possibility. Adaptation is less than perfect. So I do not see why biological systems must model Universal computation.

    ] Biologists are constantly identifying new sources of variation in offspring, which is relevant to an earlier point of yours:

    petrushka: Take away the target and there is no probability calculation.

    Even when an adaptation like “chloroquine resistance” is targeted (by us), it is difficult to justify a probability calculation, because it is difficult to identify all of the evolutionary pathways to the target. (Joe Thornton and associates discovered multiple pathways to chloroquine resistance, which is to say that the required adaptation wasn’t as specific as supposed by Behe.) It’s sometimes possible to produce a sensible lower bound on the probability that a specific adaptation will arise, but not the upper bound that ID requires to argue that some unseen intelligence has worked a miracle.

  37. Joe Felsenstein,

    The mutations are not single-character changes but swaps of subtrees in a tree-structured program.

    The apparent analogy with reciprocal recombination is interesting, though I may be mistaken in thinking it a valid one.

  38. In the long-term evolution experiment with E coli, an irreducibly complex function evolved without the involvement of natural (or artificial) selection, without it ever having been the goal or intent of the experiment. Only when the function had emerged, did natural selection (the local conditions in a growth flask) set in to preserve and enhance it.

    I’m speaking about the function of aerobic citrate transport. This function requires three criteria to be met and if any one of them is missing, the function fails completely. As such, it is by definition irreducibly complex, as each component on it’s own is nonfunctional (they only function in conjunction with other components), and only when all three are combined in the right place does the particular function, aerobic citrate transport, proceed.

    The three criteria which must be met, are:
    1. A gene coding for a citrate transporter protein.
    2. A promoter controlling the gene.
    3. The promoter must be active when oxygen is present.

    To begin with in the experiment, E coli could not transport citrate into the cell cytoplasm when oxygen was present in the environment, because the citrate transporter gene was under control of a promoter that was inhibited when oxygen was present.
    But at one point over the course of the experiment, a gene duplication of the citrate transporter gene happened into an area downstream of another promoter, this one active under aerobic conditions, created the specific association such that all three criteria were met. Now the bacteria had acquired the function aerobic citrate transport by a single mutation.

    Demonstrably, irreducibly complex functions/systems therefore can and do evolve.

  39. It gets even better. The particular example JohnnyB gives in his video, of a “feedback loop” system that could not evolve (or at least, the challenge is to identify evolving), is one where:

    BIOLOGY
    * An analogous situation in biolog is the multi-step negative feedback loop:
    * X produces Y, Y produces Z, Z inhibits/regulates X
    * This is chaotic because the weak linkage between X and being able to shut off X
    * Small changes will cause dramatic non-linear results because of the failure to turn off X

    Surprise surprise, that exact system has been observed evolving in a laboratory population of E coli. Utilizing deletion-strains of E coli, an experimental population re-evolved a functional Lac-Operon in the laboratory:

    Evolution of a Regulated Operon in the Laboratory.
    Barry G. Hall, Genetics. 1982 Jul; 101(3-4): 335–344. PMCID: PMC1201865

    Abstract
    The evolution of new metabolic functions is being studied in the laboratory using the EBG system of E. coli as a model system. It is demonstrated that the evolution of lactose utilization by lacZ deletion strains requires a series of structural and regulatory gene mutations. Two structural gene mutations act to increase the activity of ebg enzyme toward lactose, and to permit ebg enzyme to convert lactose into allolactose, an inducer of the lac operon. A regulatory mutation increases the sensitivity of the ebg repressor to lactose, and permits sufficient ebg enzyme activity for growth. The resulting fully evolved ebg operon regulates its own expression, and also regulates the synthesis of the lactose permease.

    I guess the question now is, did God come down from the heavens to cause these mutations to happen?

  40. Allan Miller: The apparent analogy with reciprocal recombination is interesting, though I may be mistaken in thinking it a valid one.

    Joe is referring to an approach called genetic programming. I’m pretty sure the original analogy was no more than that of the genome to a program. That analogy was, of course, in vogue with biologists for quite some time. The genetic regulatory network abstraction is more appropriate. A good response to Bartlett might be to point out the difference between a conventional program and a network of interacting elements.

    The reason that I mentioned Post production systems and Markov algorithms above, and not Church’s lambda calculus, as alternative models of computation is that I see the interactions of a bunch of rules as more like a genetic regulatory network than the composition of expressions of functions. (Genetic programming is mostly the manipulation of LISP symbolic expressions. LISP derives from the lambda calculus.) There has been some work in evolution of production systems. I don’t recall the details. What GA people call classifier systems are somewhat similar.

  41. Joe Felsenstein: But there is a variation called Genetic Programming (see here) that is able to do that. The mutations are not single-character changes but swaps of subtrees in a tree-structured program. They are still random with respect to fitness but now produce runnable programs much more frequently.

    There are parse trees for programs in all languages that people actually use. One of the reasons for using LISP in genetic programming is that all nodes are associated with functions. (Leaf nodes are constant functions.) So when you swap subtrees, you are swapping the calculation of one function for the calculation of another. That is the fundamental reason that crossover works well for LISP expression trees. (The probability of obtaining a well defined calculation is increased by exchanging only subtrees that produce the same type of value.)

Leave a Reply