|
Author
|
Topic: Nature Refutes ID?: The Evolutionary Origin of Complex Features
|
yersinia
Member
Member # 324
|
posted 11. May 2003 15:47
I plinked around a bit yesterday to see if I could run Avida on my windows PC; however, the latest compiled version available is only 1.6 and the study parameter files require 1.7 (IIRC).
However, version 2.0 for Macs (why? Don't ask me...) is already available, as is the source code, so those with compilers could probably run it.
As for computational requirements, my completely uninformed opinion is that it seems like the procedures described in the paper would not take a ridiculuous amount of computer power, and that the fact that it was run on a huge computer bank might just be an artifact of the fact that (say) they run everything on their huge computer bank at Adami's lab.
(my 333 MHz Windows 98 machine, however, might be pushing it...)
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 11. May 2003 16:29
charlie d....
quote:
Micah, despite being a very smart guy and certainly an ID-savvy polemicist
Thanks for the compliment, but I don't consider myself either. I'm certainly not a very smart guy (I notice this especially when I hang around all of you), and I waver back and forth between where I stand on the ID-savvy polemicist range of things.
I've found ID exciting, indeed one of the most intellectually exciting/fun things that I've come across (I'm fairly new to the earth though). I'd like to see it get its fair share of consideration before sending it off to the junk yard.
Having said that, you should realize that I'm a computer scientist whose trying to become a philosopher, and I will forever have doubts wrt the most important element of the discussion: biology. Both sides give me two contradicting stories and I really don't see any way to arbitrate between the two, especially when you factor in the theory ladenness of data interpretation.
Best.
Micah
[ 11. May 2003, 16:37: Message edited by: Micah Sparacio ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 11. May 2003 17:07
On a 2.4 GHz machine with a gene string that size (>50) and a 3600-organism population, it would take 6 to 7 hours per 100,000-update evo run, not counting data summary and analysis time, with no logic being performed, just replication. Add the performance of logic functions and that'd easily double. Hence, for just 50 experimental runs (not counting the control runs and pilot runs to make sure it's all working right), it'd be on the order of 600 hours of CPU time on that machine.
I hadn't noticed you have to have 1.7 to run their parm files. Rats. I'll have to look at editing the parm files or compiling 1.7. I haven't looked to see if the full source is available for 1.7.
RBH
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 11. May 2003 17:56
Nelson: and how much of this ICness can be gleaned from the end result, that is, if EQU is IC, and it was written by intelligent agents, it seems to be moot for ID critics to argue it is IC, since they want to show that an IC system can likely evolve.
And if ICness can be shown to have arisen non-intelligently or at least through mechanisms of evolution (mutation and selection) what would this mean?
I have seen people suggest that since the program was intelligently designed that thus the outcome was intelligently designed which seems to suggest to me the ultimate front loading. Did the designer use RMNS to achieve his purpose? Certainly the fact that it required intelligent designers to write the program does not mean that it required intelligent design for EQU to evolve.
Micah: I will respond to your posting in more detail later but I would like to point out that according to most definitions of IC, the resulting systems surely seems to fit. That removing a component removes original function does not mean of course that all function is removed.
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 11. May 2003 18:12
A few brief comments then I'm back to stuffing myself.
Charlie,
The EQU program was written by the operators. In fact, they started with 3600 identical copies of the program with 15 instructions on how to replicate. This function was intelligently designed from scratch, it was not evolved. The organisms then replicated themselves and evolved and blah blah blah.
The fact that every EQU function acheived replication in different ways, but 3 instructions were conserved, shows that it is those 3 instructions, not surprisingly that were intelligently designed, that had functional constraint, and thus are truly IC. On the other hand, there is plasticity among the other instructions which means they is no specification there. This is equivalent to what we see with eubacterial flagella, eubacterial flagella have 20 parts which are universal and conserved, all eubacterial flagella have the same 20 parts, the rest of the parts vary, in that there is plasticity among the the other parts that are not IC.
3 instructions were preserved because they were essential and thus IC. Those 3 instructions were intelligently designed.
In flagella, 20 instructions are preserved, because we see that they are essential and thus IC. Those 20 instructions were intelligently designed.
The more detail I get from the paper, the more I'm beginning to see it's an argument for ID. Of course, I'm still reading it, and these are my preliminary comments, be back later.
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 11. May 2003 19:26
Nelson: 3 instructions were preserved because they were essential and thus IC. Those 3 instructions were intelligently designed.
And the relevance of this is what? That replication was intelligently designed or that replication function /genes can be reused in other parts?
Certainly these three genes are not sufficient for EQU thus they cannot be IC wrt to EQU function. Perhaps we have a good example though of how IC systems can evolve from simpler IC systems. Surely that by itself is a powerful finding.
I am somewhat confused about ICness though, the definition seem to range from essential to original function to cannot evolve naturally. It may be helpful to first agree on which definition of IC is the 'real' one lest we want to introduce ICBehe versus ICDembski versus ICNelson?
All in all while the paper's focus was not towards ID or IC, it surely shows that evolutionists are doing most of the hard work that could also help resolve many of the open questions raised by the ID proponents. Is Nelson aware of any ID efforts in this area? Also the work shows how there are multiple paths to EQU, complicating Dembski's probability calculations, how hitchhiking can be essential part of evolution, how co-option from simpler (IC?) systems is an essential part of evolution.
It surely seems to show how RMNS are sufficient to generate IC systems, how RMNS are sufficient to generate complexity in the genome, and how epigenesis, hitchhiking, co-option are all very real and relevant processes.
It would be interesting to generate a different replication routine to see if EQU were to arise as well. Given the previous solid results I would not be surprised. But many of these questions could make for interesting ID research? Of course it also makes for even more interesting evolutionary research it seems.
Thus when Nelson states that "In flagella, 20 instructions are preserved, because we see that they are essential and thus IC. Those 20 instructions were intelligently designed. " it seems clear that Nelson has failed to show that for instance these 20 instructions were essential throughout evolution of the flagellum. All we see is an assertion which fails to be based on much concrete evidence, certainly in light of the results of the paper by Lenski et al, one may certainly question Nelson's leap via the fallacy of analogy.
All the paper shows is that these 3 genes are essential for reproduction but not that they are essential for the arrival of the EQU function or that they could not have arisen through RMNS if the replication function genes were not reused by evolution for other purposes showing that evolution marches in unpredicatable and innovative ways.
[ 11. May 2003, 19:30: Message edited by: Pim van Meurs ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 11. May 2003 20:15
quote:
The EQU program was written by the operators. In fact, they started with 3600 identical copies of the program with 15 instructions on how to replicate. This function was intelligently designed from scratch, it was not evolved. The organisms then replicated themselves and evolved and blah blah blah.
You should actually read the paper, Nelson (I guess all the important parts were in the blahblahblah part that you seem to have missed).
The EQU function was nowhere in the starting conditions, none of the initial replicators could perform the EQU function, and all the EQU-capable programs at the end of the successful runs did so in different and unpredictable ways (and one even did it better than the operators thought was possible: 17 instructions instead of 19 - suggesting that random mutation and selection might be more efficient than a designing programmer's brains).
And you are mistaken about the "core" as well. Any one of the final EQU-capable programs required many more than those 3 instructions. Each and every one had many more IC "core" components than 3 (by the functional, knock-out definition of IC); those 3 instructions alone could not generate the EQU function, and thus could not, by definition, represent the "core" of the IC system.
On the other hand, if you argue that because there are many alternative ways to generate the EQU function, the EQU function is not IC, that's fine, but then there are also many ways to achieve bacterial motility, so none of those ways must be IC either.
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 11. May 2003 22:52
Some questions/ideas (preliminary):
1. Why do organisms get compensated(energy in proportion to genome) for longer genomes?
2. Isn't the fitness function defined in terms of nand primitives: rewards based on number of nands required to perform a function?
3. Isn't a fitness function defined in terms of its target inherently teleological?
4. Is it possible that systems exist which can't be achieved in 1 or 2 coordinated mutations from a functional intermediate? Or, are we defining such systems out of existence?
5. It just occurred to me, based on my work with MESA, that it probably wouldn't take long for the EQU function to come together like a tornado in a junkyard ... I'll have to think about this some more.
6. What do we make of the mutation that resulted in a >50% fitness reduction and IN THE NEXT STEP mutated the solution to EQU? This seems to deflate the significance of the entire project to me.
7. Though the genotypic realizations of the phenotypic EQU function may be different (sizes, additional functions, etc.) the phenotypic realization is always the same. An EQU function is always an EQU function (A ^ B) or (~A ^ ~B)
That makes the analogy to the flagellum problematic. Bacterial motility isn't necessarily tied to the existent versions of flagella nor the form of motility used. Variations in method, speed, etc. are all variable.
RBH, I still think the implications of the project are being overdrawn. It just isn't surprising to me at all that, given the setup, you get the result...an EQU function is not a surprising result. For me to be impressed, I'd need to see a system that evolved a function that wasn't practically guaranteed by the initial conditions (nand primitives being moved around to form logical functions...one of which is an EQU function)
[ 11. May 2003, 22:53: Message edited by: Micah Sparacio ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 12. May 2003 01:55
This posting will violate Brainstorms guidelines, in that it will quote each of Micah's questions and provide either an answer or a pointer to an answer. Since Micah's posting implies a Q&A format I'm going to assume that's OK. quote:
1. Why do organisms get compensated(energy in proportion to genome) for longer genomes?
Receiving SIPS in proportion to genome length levels the playing field across lengths. A SIP allows the execution of one instruction, among them replication instructions, and since competition in Avida is on reproductive fecundity the allocation of SIPs in proportion to length makes genome length a selectively neutral feature. Beyond that, organisms earn further SIPS for performing the various logic functions. quote:
2. Isn't the fitness function defined in terms of nand primitives: rewards based on number of nands required to perform a function?
The fitness 'score' for performing a given logical operation is defined in terms of the minimum number of NANDs that operation requires, though just having some NANDs in the genome garners no rewards. The fitness function is graded in terms of the number of NANDs that the various logical functions require, but SIPs are not earned merely by having NANDs in the genome: the logical function must be performed by the sequence of instructions in the genome. Note the language "All other logic functions can be constructed using one or more nand instructions within an integrated framework of other instructions." (p. 140, emphasis added) I should also note that a genome that performs a given logic operation using more than the minimum number of NANDs (which is easily possible, and I'd be amazed if it didn't occur in the lineages) does not earn extra SIPs for the extra NANDs; the reward is contingent on performing the operation, not for having NANDs. quote:
3. Isn't a fitness function defined in terms of its target inherently teleological?
Sure, because this is an experiment. Experiments have goals - that's why one does them. That does not imply that natural processes have goals or are teleological, but only that experimenters do and are. If you want to find out whether a genome that can perform a complex logic operation can evolve, you have to set that as a goal. That in no way implies that the occurrence of such kinds of operations in nature occur because someone(thing) set that as a goal in a meta-experiment. (Well, I know some folks believe that, but it's not a necessary implication of humans having done this experiment.) quote:
Is it possible that systems exist which can't be achieved in 1 or 2 coordinated mutations from a functional intermediate? Or, are we defining such systems out of existence?
I suppose it's possible. But I need to see one where it is demonstrated, not merely asserted, that such systems exist in biology where evolution could not have generated it. Provide an example and run the research to see if it - or a system mapped from it into a simulation - can evolve. One of the fascinating things I've learned in working with GAs for 12 years is that even after all that time, I still cannot predict nor even reliably intuit how a GA will come to a population of solutions. It is still perpetually surprising to me. Hence I don't trust arguments based on "common sense," reason, intuition, or even logic in these matters, because the mapping from those kinds of representations to what is actually happening in the phenomena is untrustworthy. I need now to see the research done, not more IDist formal or logical or mathematical arguments. I don't trust the mapping of those arguments into nature (or hypothetical experiments) any more. quote:
5. It just occurred to me, based on my work with MESA, that it probably wouldn't take long for the EQU function to come together like a tornado in a junkyard ... I'll have to think about this some more.
Do think about it some more, and in thinking about it, consult the control runs they ran, and do the calculations for each of the 23 different genotypes that evolved to perform the EQU function. Then calculate the joint probability that 23 different tornados in 23 different junkyards produced all 23 different instruction sequences that perform EQU, all in just 50 evolutionary runs. Betcha it's lower than the UPB! ![[Smile]](smile.gif) quote:
6. What do we make of the mutation that resulted in a >50% fitness reduction and IN THE NEXT STEP mutated the solution to EQU? This seems to deflate the significance of the entire project to me.
I don't see why it vitiates the experiment. What it demonstrates is that outliers in populations are important contributors to the evolutionary process. Variability is the raw material on which evolution works. Bear in mind also what "step" means in the experiment. It is not comparable to "generation" or "update cycle." It's not intuitively obvious what a "step" is, so one needs to read that carefully. I would be curious (and I'll try to find out) if comparable kinds of things occurred in the other 22 lineages that produced genotypes that perform EQU. quote:
7. Though the genotypic realizations of the phenotypic EQU function may be different (sizes, additional functions, etc.) the phenotypic realization is always the same. An EQU function is always an EQU function (A ^ B) or (~A ^ ~B)
That makes the analogy to the flagellum problematic. Bacterial motility isn't necessarily tied to the existent versions of flagella nor the form of motility used. Variations in method, speed, etc. are all variable.
Which is one of the important points from the study: functionally equivalent outcomes can come from a variety of underlying genotypes and from a variety of lineages. That makes hash of any notion that there is a unique pathway to a given observed phenomenon, and hash of the notion that there is one and only one way to skin a cat. And it surely makes hash of any probability calculation that assumes the observed structure of the flagellum is the "target." quote:
RBH, I still think the implications of the project are being overdrawn. It just isn't surprising to me at all that, given the setup, you get the result...an EQU function is not a surprising result. For me to be impressed, I'd need to see a system that evolved a function that wasn't practically guaranteed by the initial conditions (nand primitives being moved around to form logical functions...one of which is an EQU function)
Then I'm afraid nothing one can do experimentally to demonstrate it will impress you. I need to hear from you in detail how the outcome was "practically guaranteed." What is it that "practically guarantees" the outcome?
As you contemplate the reason(s) you are not surprised, read carefully the structure of the Ancestor genotype with which the program was seeded. (That's the part Nelson hadn't read when he made his "blahblahblah" remark.) And then I'd like to hear reasons that the finding that simple straightforward evolutionary processes generate an irreducibly complex outcome (in two dozen different ways!) from that kind of ancestral population isn't surprising to you. If you do not find the outcome of the study surprising, then you should also not find the claim that evolution produces irreducibly complex outcomes in biological organisms surprising. I'm not surprised the outcome was as described in the paper because I've been confident that the relevant evolutionary mechanisms have the power to produce IC structures. That you are not surprised requires explanation, I think.
RBH
Added in edit. On Question #6, also keep in mind that in the Avida 'world,' there are at any given time a number of different "species" evolving simultaneously, and species with lesser fitness can hang on at low numbers for a fairly long time (number of updates) before finally petering out. So it's not at all surprising that an outlier species (lineage) survived for some number of cycles before the next "step" - a mutational change that moved a lineage toward the final genome that performed EQU. Notice that "steps" must be defined in terms of individual lineages - it is not a population measure. Notice also that one cannot ascertain whether a given "step" at, say, update #41,267 is a stage-setting mutation for a later mutation that produces a genome that performs EQU except in retrospect. When it happens it's merely a mutation that generates some variability. It's only in retrospect that one can say "There's an important mutation." And an interesting outcome is that those 'stage-setting' mutations are frequently deleterious, which means they move the lineage away from the current local optimum. Gee. Evolution is not a monotonic unidirectional hill climb. Interesting, but not particularly surprising: That's been unsurprising for a very long time except to those who haven't worked with evolutionary systems in vivo, vitro, or silico. This is an area where your experience with MESA is liable to lead you astray, too. MESA has a monotonic single-peak fitness landscape, and intuitions informed by that sort of structure are actively misleading with respect to EA behaviors on complex multidimensional landscapes.
[ 12. May 2003, 03:17: Message edited by: RBH ]
IP: Logged
|
|
Roger R
Member
Member # 200
|
posted 12. May 2003 07:08
quote:
Receiving SIPS in proportion to genome length levels the playing field across lengths.
And in the real biological world, where the BWM is selecting without regard to "fairness", who or what is taking it upon him/her/it - self to level the playing field? It sounds a little more like intelligent selection than natural selection.
I believe Dembski had the same criticism of the Tierra simulation, which demonstrated a trend towards minimally complex organisms until larger size itself was selected for.
And as for the issue of whether one should find the claim that evolution can produce IC surprising or not, I still see no claim from the authors of the article that it does, nor any case being made here that anything remotely like IC is being produced. All the authors claim is that GA's can produce more complex organisms, and when you reward size itself, that isn't very surprising.
[ 12. May 2003, 07:20: Message edited by: Roger R ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 12. May 2003 11:23
Two quick comments:
- fitness decreases (Micah's objection): downward fitness variation is of course very common in biological organisms, and even a 50% decrease in fitness does not mean immediate extinction (that would only occur for fitness 0). According to population genetics, the average time (generations) to fixation of a new allele is 2/s*ln(2N), with s the selection coefficient (in this case -0.5) and N the population size. I have not thought through whether this can be applied straight to the AVIDA case, but if it can, a genotype with fitness 0.5 (s=-0.5) in a population with N=3600 would go extinct in 36 generations: quite some time for new mutations to occur. Even more so if the 50% decrease occurs in a low frequency genotype with already high fitness - in that case it may still result in an above-average fitness (>1), and the mutant genotype would actually keep expanding in the population (until the average fitness surpasses it because of the expansion in the higher fitness genotype, that is). Everything is relative.
- Genome length (RR's objection): Clearly program length in the AVIDA simulation makes a big difference in terms of computation cycles, and access to computation cycles is how fitness is rewarded in AVIDA - therefore, a compensation for the loss in efficiency of longer programs was necessary (I guess the alternative would have been to simply give an even bigger reward to more effective programs, but the result would have been the same). In nature, however, very little if any disadvantage is generally associated with increased genome complexity (up to a certain limit at least - but that's pretty high) thus no "leveling of the playing field" for genome complexity is necessary in real-world evolution.
- authors claims re IC (or lack thereof) (also RR): actually, the authors say the following quote:
We ran the functional-genomic analyses on all 23 pivotal genotypes. The number of instructions required for EQU ranged from 17 to 43, with a median of 28 instructions. Notice that one evolved type apparently needed only 17 instructions to perform EQU, whereas our shortest hand-written program used 19 instructions. Further examination showed that this unexpectedly low value occurred because the evolved genome was partially redundant, such that one-at-a-time null mutations did not reveal the full extent of its computational network. These analyses therefore provide a minimum estimate of the number of instructions that a genotype uses to perform a function.
...
Although the complex feature first appeared as the immediate result of only one or two mutations, its function invariably depended on many instructions that had previously evolved to perform other functions, such that their removal would eliminate the new feature.
This is as close as one can get to the IC "functional" definition. The reason IC was not mentioned explicitly, I bet, is a) that the overwhelming majority of Nature readers would have absolutely no clue what it is, and b) that given its definitional ambiguities and inherent contradictions (as seen in this thread), IC is currently of very limited, if any, applicability in biology.
Finally, I would like to add that I find it ironic that so far almost all the criticisms of whether the final EQU programs are IC have dwelled on their evolutionary history (intermediates, selection coefficients, precursors), and not on their intrinsic characteristics; however, unlike CSI, IC can in principle be determined regardless of pre-existing knowledge. Indeed, that's how the flagellum ICness has been dealt with so far.
Thus, to objectively compare this to the flagellum situation, all ID advocates should do first is to look at the EQU programs, play with knock-outs a little to figure out which subcomponents are required and which not, and come up with an answer on whether they are or not IC. Then, and only then, it would be fair to go back to the simulations' history and determine whether the programs' ICness evolved or was "designed". So far, no one in the ID camp has stated convincingly why any of the outcomes of the simulation should not be considered IC - shall we take this as a preliminary concession that in fact they are?
[ 12. May 2003, 11:24: Message edited by: charlie d. ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 12. May 2003 11:55
Roger objects that the experimental procedure in the reported study that made genome length a selectively neutral feature somehow implicates intelligent design in biology. In designing an experiment, one holds a number of variables constant, those one is not interested in for the purposes of the study. In this instance, genome length is one such variable. The question in the study was 'Can evolutionary processes operating on primitive elements generate a complicated outcome?' Hence the experimental design sought to hold constant other variables - genome length, length of runs, composition of the pool of primitives, and so on. And, as charlie d notes, increased genome length in biology is not strongly selectively penalized by selection, so the implication Roger hints at, that making length selectively neutral somehow invalidates the findings, is itself invalid.
Charlie d's remark about ICness being determinable by examining just the outcomes, the 23 evolved programs that perform the EQU logic operation, are apropos. Since the Supplementary Information for the experiment is available on the web at the URL in the OP, it would be real nice to see ID advocates do some actual analyses of those data. Those data form a good test bed for evaluating the claim that IC structures cannot evolve by Darwinian mechanisms. And the very first step an IDist in defense of that claim must take must be to establish whether their notion of ICness applies to those outcome. Given the definitional slipperiness that is manifesting itself, a detailed analysis is required; as I used to hear in classes, "Show your work." That should provide us all with an operational definition of ICness, so any of us can actually make the determination. This research is actually a boon to the study of IC systems, since it provides a set of outcomes - the 23 runs that generated EQU-performing programs - that can serve as a test bed for testing the conception of IC. If ID really is interested in a scientific research program, here's an opportunity to demonstratre that interest on material of known provenance and properties. That doesn't allow fudging, slipping around, or spinning. It requires systematic and detailed analyses with the procedures and outcomes of the analyses published in enough detail to pin down just what was done and what was found. That's science, kids. As Dembski advised you (I presume he was exhorting his own troops), "Do the calculation!"
RBH
[ 12. May 2003, 12:28: Message edited by: RBH ]
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 12. May 2003 12:22
Any lethal mutations? Can an organism have a non-viable offspring?
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 12. May 2003 12:48
Micah asked quote:
Any lethal mutations? Can an organism have a non-viable offspring?
Yes. A mutation that disrupts the replication instructions can produce an offspring unable to replicate itself, and thus it's lethal to that branch of a lineage. Very briefly, (as I recall - I haven't read the manual for several months and don't have time to look it up now) replication of an Avida organism requires first allocating memory for the offspring, then copying the parent into that memory to create an offspring, and then copying the offspring into an adjacent cell in the Avida world matrix. Any of those multi-primitive-instruction operations can be disrupted by mutation.
RBH
P.S. Roger's remark above about organisms being rewarded for genome "size" is false. Genome length was selectively neutral.
[ 12. May 2003, 12:54: Message edited by: RBH ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 12. May 2003 13:32
RBH:
I think Roger was referring to the fact that every organism received a certain number of basic SIPs proportional to its genome length, in addition to whatever "bonus" it accrued based on its performance.
Again, IIUC this was done because otherwise any even small genome expansion would have been immediately significantly disfavored (because larger programs would use up SIPs faster). This, of course, is not the case in living organisms in which small genome expansions are generally neutral - thus the requirement for a compensatory mechanism in the program.
Therefore, you are right, genome size was ultimately selectively neutral in the AVIDA simulation, as it is in biology.
[ 12. May 2003, 13:36: Message edited by: charlie d. ]
IP: Logged
|
|
|
Printer-friendly view of this topic
