|
Author
|
Topic: Nature Refutes ID?: The Evolutionary Origin of Complex Features
|
RBH
Member
Member # 380
|
posted 07. June 2003 16:04
Being surprised by a result (which I was not in this case), and worth doing to ensure that it does in fact happen as one expects are two different issues. I'm not surprised by it, but having it demonstrated rather than merely believed is, to quote a possibly-soon-to-be-felon, a good thing. ![[Smile]](smile.gif)
Now that this baseline study is done, one goes on, as I'm sure people (including me) will, to extend and expand the study and its implications. It won't sit out there all by itself. I said above that the scientific enterprise is a 5-way conversation. This study is a new voice from the research technologies and models participant in that conversation.
Jack wrote quote:
Evolution simply must have somewhere to go.
But that's simply false. It does not have to have somewhere to "go" because it's not "going" anywhere. A system winds up "being" somewhere other than where it was a while ago as a by-product of the operation of the several variables we've identified, but it isn't "going" there from that initial state.
Jack quoted Wagner and Altenberg at length on the representation "problem" in biology. One might note that all the areas that they identify as "problems" are the focus of active research, and that data and corroborated hypotheses are accumulating. There's a lot we don't know, and a lot we know, and the existence of the former does not trivialize the latter.
Further, I'm not as convinced as Wagner and Altenberg seem to be that constraints on representations (gene-phene mappings) are as tight or limiting as they seem to believe. I don't think the "problem" is as bothersome, in an "in principle" way, as they apparently do.
RBH
IP: Logged
|
|
Jack Foster
Member
Member # 79
|
posted 07. June 2003 16:50
quote:
jack: Evolution simply must have somewhere to go.
rbh: But that's simply false. It does not have to have somewhere to "go" because it's not "going" anywhere. . .
Well, perhaps this is all just a semantics argument then, but evolution that does not "go" anywhere is not evolution at all; it is either stasis, or trivial, superficial change.
But I don't think this really is just a semantics argument. It is generally recognized that evolvability is not a feature provided by the simple Darwinian algorithm of replication, variation and selection; there are indeed more conditions than enumerated by Dennet. That recognition leads then directly to Pim's question (though I believe he phrases it as an answer): Is evolvability itself evolvable?
It's a good question.
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 07. June 2003 17:05
Some links to evolvability
Evolution of Evolvability
quote:
This paper shows how evolution tunes the content and frequency of genetic variation to enhance its evolvability. Genetic evolution is not random or entirely blind. Genetic systems are like nervous systems and brains—they have been structured and organised by evolution to enhance their ability to discover effective adaptations.
The Evolution of Evolvability in Genetic Linkage Patterns
quote:
A number of factors have been proposed that may affect the capacity for an evolutionary system to generate adaptation. One that has received little recent attention among biologists is linkage patterns, or the ordering of genes on chromosomes. In this study, a simple model of genetic interactions, implemented in an evolutionary simulation, demonstrates that clustering of epistatically interacting genes increases the rate of adaptation. Moreover, long-term evolution with inversion can reorganize linkage patterns from random gene ordering into this more modular organization, thereby facilitating adaptation. These results are consistent with a large body of biological observations and some mathematical theory. Although linkage patterns are neutral with respect to individual fitness in this model, they are subject to lineage-level selection for evolvability. At least two candidate mechanisms may contribute to improved evolvability under epistatic clustering: clustering may reduce interference between selection on different traits, and it may allow the simultaneous optimization of different recombination rates for gene pairs with additive and epistatic fitness effects.
The Evolution of Evolvability in Genetic Programming
quote:
The notion of “evolvability”--the ability of a population to produce variants fitter than any yet existing--is developed as it applies to genetic algortithms. A theoretical analysis of the dynamics of genetic programming predicts the existence of a novel, emergent selection phenomenon: the evolution of evolvability. This is produced by the proliferation, within programs, of blocks of code that have a higher chance of increasing fitness when added to programs. Selection can then come to mold the variational aspects of the way evolved programs are represented. A model of code proliferation within programs is analyzed to illustrate this effect. The mathematical and conceptual framework includes: the definition of evolvability as a measure of performance for genetic algorithms; application of Price’s Covariance and Selection Theorem to show how the fitness function, representation, and genetic operators must interact to produce evolvability--namely, that genetic operators produce offspring with fitnesses specifically correlated with their parent’s fitnesses; how blocks of code emerge as a new level of replicator, proliferating as a function of their “constructional fitness,” which is distinct from their schema fitness; and how programs may change from innovative code to conservative code as the populations mature. Several new selection techniques and genetic operators are proposed in order to give better control over the evolution of evolvability and improved evolutionary performance.
Increasing Evolvability
Considered as a Large-Scale Trend in Evolution
quote:
Evolvability is the capacity to evolve. This paper introduces a simple computational model of evolvability and demonstrates that, under certain
conditions, evolvability can increase indefinitely, even when there is no direct selection for evolvability. The model shows that increasing evolvability implies an accelerating evolutionary pace. It is suggested that the conditions for indefinitely increasing evolvability are satisfied in biological and cultural evolution. We claim that increasing evolvability is a large-scale trend in evolution. This hypothesis leads to testable predictions about biological and cultural evolution.
and
M. A. Bedau and N. H. Packard, "Evolution of Evolvability via Adaptation of Mutation Rates". To appear in BioSystems, special issue on evolvability.
quote:
We examine a simple form of the evolution of evolvability---the evolution of mutation rates---in a simple model system. The system is composed of many agents moving, reproducing, and dying in a two-dimensional resource-limited world. We first examine various macroscopic quantities (three types of genetic diversity, a measure of population fitness, and a measure of evolutionary activity) as a function of fixed mutation rates. The results suggest that (i) mutation rate is a control parameter that governs a transition between two qualitatively different phases of evolution, an ordered phase characterized by punctuated equilibria of diversity, and a disordered phase of characterized by noisy fluctuations around an equilibrium diversity, and (ii) the ability of evolution to create adaptive structure is maximized when the mutation rate is just below the transition between these two phases of evolution. We hypothesize that this transition occurs when the demands for evolutionary memory and evolutionary novelty are typically balanced. We next allow the mutation rate itself to evolve, and we observe that evolving mutation rates adapt to values at this transition. Furthermore, the mutation rates adapt up (or down) as the evolutionary demands for novelty (or memory) increase, thus supporting the balance hypothesis.
Fascinating topics
IP: Logged
|
|
Roger R
Member
Member # 200
|
posted 07. June 2003 20:27
RBH writes:
quote:
Let us not forget that those functional intermediates, the subprograms that were coopted in the evolution of programs that performed EQU, themselves evolved. Each evolutionary run started with a population that could only replicate. It evolved those intermediate building blocks. They too were generated by the evolutionary process; they were available because they evolved.
Depending, of course, on what one means by "evolved". The first functional intermediates were built essentially by random mutation, and not selection, since there was no prior selection in operation (ignoring the bonus "points" for size). I think that was what was meant upthread by a poster who said something to the effect that the initial state was near the final goal. Because random mutations are likely to produce the building blocks of EQU. And using those intermediate functions, can randomly assemble EQU. But it isn't as easy to randomly build EQU from the basic instruction set. Again, not particularily surprising. The bigger the leap, the more challenging it will be to random assmbly.
quote:
Roger also remarks on his (and other IDists') requirement that we must "demonstrate that such intermediates exist in the biological world."
A point I have to correct. I have never put a requirement on anybody to "demonstrate that such intermediates exist in the biological world." I merely point out that the Avida program doesn't do so, and that such intermediates are an important issue in the debate over IC structures in the biological world. I don't think they are red herrings at all, but rather the crux of the matter.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 07. June 2003 21:18
Commenting on my statement about the evolved availability of intermediates in the Avida simulation, Roger wrote quote:
Depending, of course, on what one means by "evolved". The first functional intermediates were built essentially by random mutation, and not selection, since there was no prior selection in operation (ignoring the bonus "points" for size). I think that was what was meant upthread by a poster who said something to the effect that the initial state was near the final goal. Because random mutations are likely to produce the building blocks of EQU. And using those intermediate functions, can randomly assemble EQU. But it isn't as easy to randomly build EQU from the basic instruction set. Again, not particularily surprising. The bigger the leap, the more challenging it will be to random assmbly.
Roger, selection was operative from the very beginning of the runs. There was initially competition for space, predicated on differential replicational efficiency, and then as simpler logic operations appeared they were selected for because the fitness function defined them as advantageous and they earned SIPs - additional time slices for replicating. Selection was operating from the start and to suggest otherwise misrepresents the study.
I must emphasize also that the bonus "points" for size did not select for longer genomes; read that part carefully, please. What apportioning SIPs in proportion to length did was make length selectively neutral - there was no selective bias either for or against length.
Your locution "The first functional intermediates were built essentially by random mutation, and not selection ..." betrays a misunderstanding of evolution as a process. Selection "builds" nothing; it differentially preserves that which is built by mutations. So yes, the very first logic functions that were selected occurred by chance. So what? And the variants that thereafter were preserved by selection, so they could form a larger and larger proportion of the population and perhaps be coopted for still more complex functions, were also initiated by chance. Again, so what? Mutations are the ultimate source of the heritable variability on which selection operates to differentially preserve variants according to their fitness. Once variants are available in a population, evolution (often via cooption) can take advantage of combinations of those variants, generating more complex structures while by-passing the "chance" bottlenecks IDists seem to see. If A-B-C and D-E-F exist in a population, A-B-C-D-E-F is just one combinatorial step away from A-B-C, not three additions of one letter at a time. Those kinds of combinatorial steps make nonsense of the one-point-mutation-at-a-time probability calculations one sees so many of.
Roger remarked that quote:
A point I have to correct. I have never put a requirement on anybody to "demonstrate that such intermediates exist in the biological world."
Rereading your posting above, that's technically true, though I took it to mean that one must supply all of the biological precursors in order to support the hypothesis that a given complex structure evolved. Perhaps I am misreading it. That is surely Dembski's requirement. In a thread on Brainstorms he said quote:
If evolutionary biologists can discover or construct detailed, testable, indirect Darwinian pathways that account for the emergence of irreducibly and minimally complex biological systems like the bacterial flagellum, then more power to them -- intelligent design will quickly pass into oblivion
Your requirement - that intermediates exist in the biological world - has been met multiple times over. In one of the two threads currently active on this topic (I'm not going to search for it at the moment), yersinia mentioned five biological systems that Behe held out as IC for which at least some intermediates have been identifed.
My remarks on intermediates were to the effect that not all of them are likely to be visible in extant organisms, and I illustrated how they drop out of a lineage during the course of evolution. Dembski's insistence on a "detailed, testable, indirect Darwinian pathway" is reminiscent of Duane Gish's histrionics about transitional fossils: Every transitional fossil that is found merely increases the number of gaps between intermediates.
RBH
[ 07. June 2003, 21:22: Message edited by: RBH ]
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 08. June 2003 08:30
Charlie D. States:
"I sense another definitional can of worms is about to be opened."
This is telling, in that, a definitional can of worms is only a definitional can of worms if you are artificially restricting a concept from going through natural stages of development. It is only a can of worms if you want a static target to IC-bomb, once and for all.
**************************************
BTW, RBH, when you do the experiment, knocking out nand, remember that it is instruction (p) not (q) though knocking out (q) would be far worse (wouldn't it?)
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 08. June 2003 12:32
Jack wrote quote:
It is generally recognized that evolvability is not a feature provided by the simple Darwinian algorithm of replication, variation and selection; there are indeed more conditions than enumerated by Dennet.
Well, in some sense that's correct: for evolution to occur also requires that the fitness landscapes (not fitness "function") have a pretty general property, that at least some of them be locally correlated in one or more of their dimensions. As Richard Wein pointed out some time ago in his review and critique of NFL, that's not a very constraining condition. But you're right: there is that one additional requirement.
Those conditions (with the addition of "heritable" to variation) are necessary and sufficient for evolution to occur in any meaningful sense of the term "evolution." What is "recognized" (for example, in the Wagner and Altenberg paper) is that some meta-properties of evolutionary systems are themselves evolvable. The literature Pim identifies above is directed at the set of questions surrounding the "evolution of evolvability." But that literature in no way implies that the occurrence of evolution, biological or otherwise, requires anything more than Dennett's three conditions plus the local correlation property of fitness landscapes.
RBH
PS: Micah: Yup, gotta mind them p's and q's.
[ 08. June 2003, 12:34: Message edited by: RBH ]
IP: Logged
|
|
GP
Member
Member # 570
|
posted 08. June 2003 13:29
quote:
Those conditions (with the addition of "heritable" to variation) are necessary and sufficient for evolution to occur in any meaningful sense of the term "evolution."
RBH,
To be fair to the IDists, I think the sticking point is the sufficiency claim. I am not entirely sure (and I've argued as much earlier in this thread) how one can empirically establish this claim.
GP
IP: Logged
|
|
Jack Foster
Member
Member # 79
|
posted 08. June 2003 14:53
Hi Pim:
quote:
Fascinating topics
Agreed. Thanks for the links! Good stuff.
Hi RBH:
quote:
But you're right: there is that one additional requirement.
Thanks. (Geez, it's like pulling teeth! ![[Wink]](wink.gif) )
Hi GP:
quote:
To be fair to the IDists, I think the sticking point is the sufficiency claim. I am not entirely sure (and I've argued as much earlier in this thread) how one can empirically establish this claim.
Well, actually I agree with RBH, once he adds the fourth condition . . . evolvability. (I like one of the definitions in one of the Pim links: "the ability of a population to produce variants fitter than any yet existing.") Though clearly, I think the representation problem is more problematic than RBH does. Fitness landscapes (yeah, not fitness functions) are of course determined mostly by the internal properties of the replicator itself. We can easily imagine replicators that can never attain evolvability regardless of how the environment is manipulated.
The discussion of evolvability should probably move to a different thread. I want to read Pim's finds as well. From some of the abstracts, . . . mturner might like these, too! ("Genetic evolution is not random or entirely blind. Genetic systems are like nervous systems and brains . . .")
As far as this thread goes, I notice that nobody has discussed my mousetrap-evolving system analogy. I know that no analogy is perfect, but doesn't it seem to people that that's more or less what's going on in the Lenski system?
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 08. June 2003 15:33
Hey Jack.
I think you are right on with your evolving mousetrap analogy. An evolving system could be designed with a set of building blocks and an environment conducive to building mousetraps. Indeed, evolutionary algorithms are being used for similar projects this very moment (designing better antennas for example). The evolving system could reward a set of intermediate functionality, and often come upon a mousetrap (perhaps better mousetraps?). This would be no surprise (to me at least). It would be an example of constrained evolution or, one might say, flexible design. I actually argue, in my paper on mental realism, that flexibility and generality are features of mental causation. Polanyi's notions of "constraint" and "organizational principles" seem to fit in well here.
[ 08. June 2003, 15:34: Message edited by: Micah Sparacio ]
IP: Logged
|
|
GP
Member
Member # 570
|
posted 08. June 2003 15:50
quote:
Well, actually I agree with RBH, once he adds the fourth condition . . . evolvability. (I like one of the definitions in one of the Pim links: "the ability of a population to produce variants fitter than any yet existing.")
Hi Jack,
Evolvability.
How does one discuss this topic with respect to sufficient conditions on evolution without invoking circular logic? I think this is what I was pointing to. So evolvability is the ability of a population to produce variants fitter than any yet existing. Therefore, evolvability encompasses the sufficient conditions for evolution occur. But that still does not address what those sufficient conditions are.
So what you really seem to be saying is that intelligence is a sufficient condition for evolution. But, this is no better than simply redefining evolvability to be intelligence. It is sufficiency by fiat.
postscript: Of course, in thinking about it just a bit further. We can posit any number of sufficient conditions for evolution to occur. But the natural question is whether they are necessary.
So I'm confused now. Is the IDist thesis concerning "intelligence" one of necessity or sufficiency?
[ 08. June 2003, 15:56: Message edited by: GP ]
IP: Logged
|
|
GP
Member
Member # 570
|
posted 08. June 2003 16:11
quote:
So are you saying that you're surprised by this result? I know that I'm not. I truly can't imagine that this experiment is particularly valuable or compelling for those within your field. I think Nature's interest in this paper is at least partially provided by the anti-ID propaganda value that can be had by it. The inclusion of Robert Pennock as co-author lends some credence to this view.
Jack,
I also want to address this comment of yours. The propaganda value of a work exists independently of its scientific merits. Let's not second-guess the politics of scientists shall we? Your line of reasoning is easily a double-edged sword, that can be applied to any number of popular ID books and articles. And when it comes to propaganda, one can easily make the case that the ID works which directly appeal to the scientifically illiterate have a much larger impact.
But having said that, science experiments often do not go for the surprise-factor. They are often designed with expected goals in mind. Such is the case with the Avida experiment. I cannot imagine how they can put in an objective criteria based on "surprise" to determine the value of their work.
Here's another way to put it. Suppose you were designing the experiment, Jack. Are you telling us that you would be willing to put in a bunch of non-specific initial conditions, and wait paiently, tracking all of the evolved organisms for the first digital organism to surprise you? What would that experiment reveal?
[ 08. June 2003, 16:13: Message edited by: GP ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 08. June 2003 17:09
Jack Foster wrote quote:
(I like one of the definitions in one of the Pim links: "the ability of a population to produce variants fitter than any yet existing.") Though clearly, I think the representation problem is more problematic than RBH does. Fitness landscapes (yeah, not fitness functions) are of course determined mostly by the internal properties of the replicator itself. We can easily imagine replicators that can never attain evolvability regardless of how the environment is manipulated.
This is where we separate, I think. A fitness landscape is a connected graph induced by an evolutionary operator from a fitness function. (See here for an introduction.)
A fitness landscape is not "determined mostly by the internal properties of the replicator ...". In fact, the graph is not a property of of a particular replicator or population of replicators. The present location of a member of a population of replicators on a graph, however, is determined by the properties of the replicator. If replicators vary in fitness in such a way that the phenotypic variability is visible to selection then replicators will be distributed across some neighborhood (cluster of vertices) on the graph. If the landscape is locally correlated (see below) but not absolutely flat, then the requirement that some variants are fitter than others is automatically realized. Further, if mutation is random with respect to location on the landscape, the requirement that a population be able to produce variants that are fitter than any currently in the population is also met: replicators on the highest point of a slope (in any dimension) on which the population is dispersed have a non-trivial probability of producing offspring that are higher on that slope, the value of the probability depending on the exact topography in that neighborhood of the landscape.
The "locally correlated" requirement means that fitness values must not be randomly distributed across vertices on the graph. Metaphorically speaking, a correlated (but not flat) fitness landscape displays hills and valleys, and has slopes across neighboring vertices rather than random values from vertex to vertex.
If replicators display heritable variability (i.e. produce phenotypes with heritable features that are 'visible' to selection), and if there is competition among replicators, then I for one cannot "... imagine replicators that can never attain evolvability regardless of how the environment is manipulated." I ask Jack to describe - list - the specific conditions under which he can imagine that happening, because I can't. What precisely are the conditions under which a population of replicators cannot evolve?
With respect to the evolving mousetrap analogy, I don't think it's at all useful because it is not an accurate model of relevant features of the Avida simulation. It displays a lack of understanding of what the Avida virtual machine is capable of. While the instruction set in Avida obviously contains the necessary raw material (instructions) for a program that performs EQU, Avida's virtual machine is Turing-complete, meaning that it has the necessary 'raw material' to perform any computable function. One can't say that about the mousetrap analogy: the parts list of that analogy is tightly constrained in a way that the Avida instruction set and virtual machine architecture is not. About the only functional 'complex' structure that one can construct (or evolve) using all the mousetrap parts is a mousetrap. However, the only constraints on the Avida system that evolved programs that perform EQU are precisely the constraints imposed by Darwinian evolutionary theory. So the evolving mousetrap analogy fails for lack of alternative capability: all it can evolve is a mousetrap. The Avida simulation in principle could have generated anything but didn't; it evolved what the evolutionary mechanisms permitted. And what it evolved in the Lenski, et al., study are structures - assembly language programs - that meet the operational definition of "irreducibly complex."
RBH
[ 08. June 2003, 17:15: Message edited by: RBH ]
IP: Logged
|
|
Jack Foster
Member
Member # 79
|
posted 08. June 2003 18:38
Hi GP:
quote:
Evolvability.
How does one discuss this topic with respect to sufficient conditions on evolution without invoking circular logic?
LOL! That's a very good point. "Evolvability" as fourth condition for evoltuion is circular, so it really must be viewed as a placeholder. It seems that there is a disagreement between RBH and W&A about what condition or set of conditions produce "evolvability".
Hi RBH:
quote:
This is where we separate, I think. A fitness landscape is a connected graph induced by an evolutionary operator from a fitness function. (See here for an introduction.)
Another semantics argument, I believe; but a valuable one. Your link doesn't work, but I went to the homepage where an abstract was available:
quote:
The use of the term "landscape" is increasing rapidly in the field of evolutionary computation, yet in many cases it remains poorly, if at all, defined. This situation has perhaps developed because every one grasps the imagery immediately, and the questions that would be asked of a less evocative term do not get asked. This paper presents a model of landscapes that is general enough to encompass most of what computer scientists would call search, though the model is not restricted to either the field or the viewpoint. It is particularly relevant to algorithms that employ some form of crossover, and hence to genetic algorithms and other members of the evolutionary computing family. An overview of the consequences and properties of the model establishes a connection with more traditional search algorithms from artificial intelligence, introduces the notion of a crossover landscape, and argues the importance of viewing search as navigation and structure.
It's somewhat ironic that you reference a paper for my benefit regarding the meaning of "landscape" that concedes that the term "landscape", "remains poorly, if at all, defined". When I say fitness landscapes are "determined by the internal properties of the replicator itself", I mean the relevant landscapes. Theoretically, you might be able to speculate regarding, say, the fitness landscape on the feature "motility" for a tomato plant. But since tomato plants do not possess the feature "motility", and are unlikely to develop the feature due to their root system, the landscape in question is not truly a fitness landscape for tomatoes; at least not one that is "in play".
I've got to go to a memorial service, but I have more to say. I'll finish my thougths tonight.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 08. June 2003 19:05
Jack wrote quote:
It's somewhat ironic that you reference a paper for my benefit regarding the meaning of "landscape" that concedes that the term "landscape", "remains poorly, if at all, defined". When I say fitness landscapes are "determined by the internal properties of the replicator itself", I mean the relevant landscapes.
In this domain of discourse a fitness landscape is not "determined by the internal properties of the replicator itself." If you want to use some sort of definition that invokes that notion, feel free, but know that it will only engender confusion since it's a purely idiosyncratic usage. To be blunt, no one will understand what you mean.
I screwed up the reference and therefore the URL. That's an earlier paper by the same author. In that abstract, as you will note, he doesn't merely say that in many cases it remains poorly defined, he sets out to remedy that situation. In the later paper he does further work on remedying it. You'll find that the notion of "landscape" is very well - indeed formally - defined there.
RBH
IP: Logged
|
|
|
Printer-friendly view of this topic
