|
Author
|
Topic: Nature Refutes ID?: The Evolutionary Origin of Complex Features
|
YZ2
Member
Member # 91
|
posted 12. May 2003 16:08
Let me inject some more thoughts to this interesting problem:
As Micah has pointed out, (A EQU B) can be computed from 3 NAND operations. Because of the
logical property here, all 3 operations are
needed to get the output exactly.
[Recall: (A EQU B)=(A^B) or (~A^~B)]
Knowing only one of the 3 will not get the exact output all the time. On the other hand, it does
not mean that knowing the output of some (not all)
of the operations is, shall I say, invaluable in
estimating the output of EQU. (For example,
knowing (A^B)=>1 means that (A = B).)
It is because it will help predicting the output
in some cases. In another words, it will give
partial information of the output.
What I can say then is this:
Despite the rigid requirement of the logic of EQU, it is not 'novel' in this sense, because
information is already reflected in
the original individual NAND operation partially.
If you accept this argument, then this experiment
has not answered where biological complexity with
repect to novelty originates.
Comments?
[ 12. May 2003, 16:31: Message edited by: YZ2 ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 12. May 2003 17:02
YZ2 suggests that because more primitive logical operators like NAND contain what amounts to partial information about the EQU operation, "this experiment has not answered where biological complexity with respect to novelty originates." The only things I can say about that are (a) given whatever "biological complexity with respect to novelty" means, as far as I can tell that's not the question the experiment was designed to address so it's not real amazing that it doesn't address it, and (b) "biological complexity with respect to novelty" is yet another undefined bit of wordplay that has no clear meaning, and thus blurs the issue rather than clarifying it. What is it, that we might know it when we see it?
RBH
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 12. May 2003 18:26
It's like saying that because flagellar proteins share tertiary structure domains essential for their function with a number of other proteins, flagella are not 'novel' in this sense either, because information is already reflected in the original individual protein domains, partially.
Whatevah.
IP: Logged
|
|
John Bracht
Member
Member # 5
|
posted 12. May 2003 22:12
Having spent the past few hours poring over this paper, and trying to get up to speed quickly on Avida programming, I've got a few questions to ask.
1. Does anyone else find that the "mutations" described in the program seem utterly unrealistic? Replacing one entire instruction with another entire instruction seems to be operating at an incredibly high level that is not biologically relevant. I would imagine that the biological equivalent of these instructions would be an entire protein that carries out a particular function. Hence, the program basically swaps out one entire functional protein for another. However, in biology, you have to build those functional proteins before they can be swapped. And you don't (usually) just switch them out--you get point mutations that tend to slightly alter the amino acid sequence and (possibly) the ultimate folding structure, hence the function, of that protein. But many point mutations simply destroy protein function because the mutated protein simply can't fold properly anymore (in the computer simulation that would mean destroying the function, without replacing it with another). By making a "point mutation" mean an instruction-swap, it seems the program greatly inflates the probability of getting functional innovations compared to biological reality. Any biologists want to comment?
2. I've noticed that genetic algorithsms that succeed at generating complicated outcomes often start with the high-level building blocks for those outcomes and just shuffle them around until the desired end product is produced. That's what John Kosa's "evolutionary programming" examples did, and that's what this program does, too. Does anyone else wonder why the building blocks, the instruction sets, are precisely what is needed for making EQU logic functions? And if the answer is "these are the building blocks for any higher-order logic functions", then why don't other, equally complex logic functions get produced by the algorithm? (self-evident reply: because they weren't selected. Sort of demonstrates the vitalness of selecting for precisely what will be output, doesn't it?)
3. Does anyone else wonder about the unrealistically high mutation rate? A mu (mutation rate) of 0.0045 mutations per base per genome (0.225 mutations per genome/50 bases per genome) is quite a bit higher than the estimated 10^-9 mutation/bp realistic in most organisms (I'm not considering viral mutation rates here. Anyone have those mutation rates at their fingertips?). Also, keep in mind that they're using a funky definition of "mutation" whereupon entire functional components are swapped out wholesale. Nothing like having to build up those components base pair by base pair. In fact, it seems that given the limited scope of possible functions (26) and the remarkably high mutation rate, the organisms are virtually guaranteed to explore some extremely useful combinations of parts, especially given that some of the intermediate configurations on the way to EQU are selected for. Perhaps most remarkable is the fact that whthout selection for intermediate steps, the system could not evolve EQU even in spite of all the help from the wholesale swapping of components and high mutation rate. I'd say this study should be more than a tad discouraging for Darwinian models of evolution.
4. What about graph 3a on page 141, showoing "replication efficiency"? I've stared at this graph for awhile, it it seems rather amazing, for the simple fact that organisms never increased their overall replication efficiency--indeed, early organisms were far more "fit" than were the ones that eventually took over the gene pool! What's going on here?
I think the problem is that replication efficiency for this graph is calculated as the ratio of the organism's genome size divided by the SIP's ("Single Instruction Processing" unit of energy) used in its life-cycle. However, the actual fitness criteria used to award more "fit" organisms only looks at the genome size and computational merit (number of logic functions performed by that organism). Notice, there is no compensation for SIP's used in life-cycle when the prgram actually awards new SIPs. Which means--there is no penalty for SIP's that get sucked away from reproduction into the lengthy logic functions that give it computational merit. However, if you separate reproductive efficiency out of this equation and graph it, as they have done, compensating for SIP's lost to computational merit, then suddenly the programs that ultimately win out are actually less fit in terms of reproduction efficiency!
What does this imply? It seems obvious that organisms which spend huge resources on generating complex logic functions will automatically be less fit, because they will be replicating huge amounts of DNA that is not needed for replication alone--and using up precious SIP's to not only copy this large genome, but also to cycle through all the logic functions (eg, EQU) which have no direct bearing on reproduction. The only reason these less fit organisms survive and reproduce and dominate the population is because the authors have introduced an entirely artifical fitness landscape which rewards exactly what they want,--at the expense of reproductive efficiency (ie, fitness)!
In other words, what I'm saying is that this program appears to impose an entirely artifical selective pressure to promote a complex result that would never fly out in the real world. In the real world, the only thing that matters is how many offspring you produce. That's it. There's no extra benefit from doing fancy logical calculations or analysis on your own environment (what the heck is the EQU function supposed to contribute towards fitness in the computer simulation anyway???). The bottom line is, the authors of this study introduced this totally arbitrary selection for "computational merit" which has absolutely no tie-in with real fitness and actually rewards organisms which are less fit, in a reproductive sense, than their peers.
The only way they get complexity is by going against the natural tendency of evolution.
That's the real irony from the paper: it's a prime example of why natural selection won't produce the kinds of complexity that we find interesting (and find in the world around us) but it shows nicely how intelligent input (thought the tweaking of the fitness criterion to include the arbitrary computational merit) can inject information that then gets manifest as a complex and interesting result.
John
[ 13. May 2003, 12:19: Message edited by: John Bracht ]
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 13. May 2003 08:34
John,
You appear to have done an admirable job pointing out how this study adjusted assumptions in order to produce a particular interesting result. To take your analysis a step further, if we can identify which assumptions are unrealistic, would in not also be possible to identify or estimate what constitutes realistic assumptions. It would seem to be possible to set realistic assumptions regarding mutation rates, the realistic sets of likely mutations, realistic genotype to phenotype relationships, realistic assumptions regarding the fitness landscapes, realistic assumptions regarding the rates of change in fitness landscapes, realistic assumptions regarding front loading and realistic assumptions regarding the search techniques used by biological systems.
If it is possible to define and validate a range of realistic assumptions, then it would appear possible to develop realistic measures of the capacity of biological systems to evolve.
This study does or may demonstrate that there are logical/mathematical transformation processes which could ‘evolve’ complex functionality from simple functionality. However, as you point out, the study also points out that there are number of key parameters or assumptions which impact the ability of a system to produce the demonstrated transformation.
If the demonstrations presented are to have any relevance to evolutionary processes in biological systems, then it must be shown that the demonstrated transformation can occur using parameters or assumptions which are realistic for biological systems.
Demonstrating evolutionary change in an artificial mathematical universe is a long ways from modeling, simulating or explaining evolutionary change in real world biological systems. To support a claim that a mathematical demonstration has relevance to biological processes, it is necessary to demonstrate that the assumptions used are realistic. I think you have done a good job of pointing out that the assumptions used in the reported study are not realistic.
IP: Logged
|
|
YZ2
Member
Member # 91
|
posted 13. May 2003 09:21
Probably I have not made myself clear. It appears to me that EQU can be computed by acting on using the original NAND function to different part of the problem subspace, a number of times. One part can be when A=B=1, related to (A^B), the other part can be when A=B=0, related to (~A^~B). The third operation combine the results using another NAND operation. If this is the case, then EQU is just another expression of NAND, acting on transformed and partitioned problem subspaces. There is no novelty in the new EQU function. It is the old NAND function. So rather than showing the power of a Darwinian process, surprisingly the experiment has shown its limit.
[ 13. May 2003, 14:26: Message edited by: YZ2 ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 13. May 2003 10:52
John:
yours are sensible objections, and yet totally irrelevant. This was not a simulation of biological evolution, nor did the authors ever claim so.
This work proves one thing and one thing only: that within a population of simple, inaccurate replicators subject to some selective constraint, mutation, selection (and cooption) can result in the generation of complex features (i.e., for what we are discussing here, systems with the hallmarks of IC) without such features being initially "specified", "smuggled in", etc.
Thus, it does not matter what the original replicators looked like, what the mutation rate is, what the encoding language is, what kind of mutations are allowed, etc etc etc.
Again: this is not meant as a simulation of biological evolution in any way, shape or form, but as a proof of principle that IC systems can evolve spontaneously through evolutionary processes. (As the authors state at the beginning of the article, darwinian processes apply, by default, to any entity endowed with the ability to replicate and mutate in a selective environment.)
Of course, one can object, as you are doing, that the computer simulation is more efficient at this process than living organisms are, and everyone (including the authors and myself) will agree (that's why they used computers and not real live elephants). However, in that case, the issue becomes entirely quantitative and debatable; the qualitative uniqueness of IC, that somehow supposedly placed it beyond the reach of naturalistic evolutionary mechanisms, has evaporated.
So far, from what I can see based on the objections here, that conclusion still stands.
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 13. May 2003 13:11
Charlie wrote:
quote:
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).
Ok read the paper last night. The EQU function was not in the starting condition, but the programmers did design from scratch organisms that could already replicate. And this is actually what I was pointing to in my original post, having read the paper, I simply misnamed the starting programs, which were all identical.
Charlie writes:
quote:
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.
Actually I now see that 8 instructions were actually conserved from the ancestral line. And most likely, only those 8 instructions are required and essential in the IC sense. It looks like 35 instruction-EQU might just be an example of cumalative complexity, with a very small core, that might be easily attainable from NAND. The problem with "essential" comes with the fact that the 35 instruction EQU contains the 3 of the 8 essential instructions for replication, and mutants in those 3 instructions are probably what lead to non-viability, but that probably does not matter.
This goes along with Bracth's discussion of how unrealistic this program is. EQU in this program is done in 35 instructions, but I can remove 16 of these instrutions and still get EQU function.
In fact, I can probably remove many more to get just the EQU function:
quote:
An exhaustive search shows that the minimum number of nand operations to perform EQU is five, which is greater than for any other one- or
two-input logic function. Using the 26 available instructions, we wrote a program of length 19 that performs EQU but does not replicate. This program seems, but has not been proven, to be the shortest one to perform EQU.
In fact, no doubt EQU itself only requires very little instructions, this has no equivalent with any of the IC systems that design theorists study. In fact, the bacterial flagellum cannot be reduced so drastically and still retain function, which explains why EQU is so plastic and the flagellum is not. It is universal among eubacteria.
Charlie writes:
quote:
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.
There are many ways to achieve motility. I get along just fine without flagella, as a matter of fact. But we're not talking about just motility, especially with the bacterial flagellum. We are talking about motility via bi-directional rotor motor-driven propeller, there is no such thing as a simpler way to do that. On the other hand, there is a simpler (as admitted by the authors of the paper)ways to do EQU , and which is why there might be plasticity there, there is no plasticity among the parts that are required to attain the function in bacterial flagella. You need all 20 parts, or there is no motility via bidirectional rotor motor-driven propeller. All the parts are conserved and universal.
[ 13. May 2003, 13:38: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 13. May 2003 14:31
quote:
Nelson:
But we're not talking about just motility, especially with the bacterial flagellum. We are talking about motility via bi-directional rotor motor-driven propeller, there is no such thing as a simpler way to do that. On the other hand, there is a simpler (as admitted by the authors of the paper)ways to do EQU , and which is why there might be plasticity there, there is no plasticity among the parts that are required to attain the function in bacterial flagella. You need all 20 parts, or there is no motility via bidirectional rotor motor-driven propeller. All the parts are conserved and universal.
That's just because you arbitrarily adopt the strictest possible definition for flagella (based on the detailed structure of the example of your choice), and the broadest possible definition for EQU (based on their ultimate, shared functionality). However, one could equally well define the flagellum as "an organelle for motility", and any EQU program as a specific set of instructions, line by line, et voila' the situation would be perfectly reversed.
The fact remains, there are several existing flavors of flagella, even more flavors of cellular motility, and the number of theoretically possible motility organelles is simply too large to even consider. In AVIDA, one has the advantage of looking at and comparing the outcome of many simulations; out here, we're still running Evolutionary Simulation #1 (as far as we know, at least).
You (and others here) are still not taking the right steps in fairly analyzing this paper's findings, though. Instead of frenetically looking for some flaw, any flaw, choose one of the outcome EQU programs (like you chose one kind of flagellum), analyze it functionally and say whether it's IC according to Behe's definition. Then, figure out whether it evolved or was "designed" by some programmer input within the simulation.
[ 13. May 2003, 14:34: Message edited by: charlie d. ]
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 13. May 2003 14:41
Charile wrote:
quote:
That's just because you arbitrarily adopt the strictest possible definition for flagella (based on the detailed structure of the example of your choice), and the broadest possible definition for EQU (based on their ultimate, shared functionality).
Actually thats simply false. The EQU function uses the same components (i.e. registers and simpler logic functions) in the same language as the simpler forms do. This is drastically reducing the EQU function to a simpler form using the same components . Do it for the bacterial flagellum. If you can't, then this program is completely irrelevant, or worse for you, an argument for intelligent design.
EQU function is a precise function and what matters is what components I can use to get that function, obviously not many, the 35 functions probably only build on these. However with abstract function of motility, I need to use completely different components and a completely different model in order to get another form of motility (extracting pili in archael flagellum, cilium in eukaryotic flagellum). But even if you were correct, these are all IC too! I can't drastically reduce these guy's parts either. I can with all the EQU functions.
Note, I don't know if there is really a "right" way to analyze this paper. I am certainly concerned about the fact that you (and others) have stated that EQU is an IC system that spontaneously evolved and yet I can remove parts willy nilly as if EQU was nothing but wrapping paper around a christmas present. The real question is to find the IC system (which is most likely the designed function of replication) and find out that origin. Obviously it was intelligently designed. However, not much of these systems seems terribly complex.
Whats important is finding out what you mean by "an IC system can easily evolve" , by finding this IC system and following the information trail to it's origin.
[ 13. May 2003, 15:11: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 13. May 2003 15:08
Looks like I'm just repeating what Micah and XZ2 are saying. I note that even EQU is not very much more complex than the original replicator program, which was intelligently designed. If this is true:
quote:
Despite the rigid requirement of the logic of EQU, it is not 'novel' in this sense, because
information is already reflected in
the original individual NAND operation partially.
in that very few NAND operations are required for EQU functions (and the authors themselves do allude to this), none of this is reflected in any of the systems Behe et. al. studies, not even if you define bacterial motility in the broadest sense.
[ 13. May 2003, 15:16: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
Kirk Durston
Member
Member # 174
|
posted 13. May 2003 15:23
I notice that charlie d. claims that John's objections are irrelevant. I note, however, that the opening sentence of the abstract of Lenski's et al. paper states, ?A long-standing challenge to evolutionary theory has been whether it can explain the origin of complex organismal features. We
examined this issue using digital organisms ...?. The authors are implying, in fact, that what they did is relevant to evolutionary theory and complex features in organic life.
That being pointed out, however, charlie d.'s contention is that the paper may show that IC can be achieved through natural processes, modelled by the program in question and mentioned in the paper. My response is that I'm not sure that it does. A system is not IC if the informational gap is so small that it can be jumped by random processes. In spite of the fact that the virtual mutations are simulated to be random, both the jump to the intermediate NAND function and then the jump to the EQU function have a probability approaching 1, given the way the simulation is designed and the number of generations executed. One does not need ID to achieve such results, nor do the results appear to qualify as IC for the reason I have just laid out.
IP: Logged
|
|
Argon
Member
Member # 276
|
posted 13. May 2003 16:29
Kirk Durston writes:
quote:
A system is not IC if the informational gap is so small that it can be jumped by random processes.
I think this inverts the logic of the argument. A definition found in DBB by Michael Behe suggests that a system is IC if it loses its function after removing a component of the system.
That IC systems (so defined) are unlikely to be generated by random processes + selection is a claim which needs to be tested. Part of the relative usefulness of the original definition was that the assignment of "IC-ness" could be made directly, on the basis of functionality, without prior knowledge of a system's history. The argument that Behe then proceeded to develop was that such systems are unlikely to have evolved naturally.
In contrast, defining an IC system as one which is inaccessible to natural processes such as random variation + selection turns it into a circular definition with respect to evolvability. It's like saying: "An IC system is an unevolvable system which is unlikely to have evolved." Such a definition also makes it very hard to evaluate the IC-ness of a system because now we have to have a fairly detailed knowledge of the precise history of a system -- Information which is not readily available.
I think what RBH is asking for someone to evaluate the final products of the simulation: Are they IC or not? Here's one suggestion for testing: Knock out a few of the instructions used by the coding sequences to generate the EQU function. What happens?
[ 13. May 2003, 16:58: Message edited by: Argon ]
IP: Logged
|
|
yersinia
Member
Member # 324
|
posted 13. May 2003 16:48
Could we have an IDist give us their definition of "IC"? There is a massive amount of goalpost-moving going on.
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 13. May 2003 17:25
yersina,
This is a good question, and I think it highlights a problem with theories that were originally formulated in one domain and which some try to generalize outside of that domain.
Behe's original definition was certainly developed for biological systems and similar 3D systems with mechanical parts. In other words, mechanical, real-world function. That's why he's thrown in things like "well-matched."
My sense is that the more applicable ID concept here is Dembski's specified complexity.
In any case, forgive (or not) my dodging your quesiton. It just raises a point I've been thinking about a lot lately.
One other thing to note. If we want to apply IC fairly to this experiment, we'll have to come up with some clean analogues to the physical world.
What is a "part" in this experiment? Are the parts made of anything or are they simples (atoms)?
Part of the problem in analyzing this is drawing analogies between ontologies. I'm hoping to look at this more in depth tomorrow while I hang out at my parents house waiting for my car to get fixed.
[ 13. May 2003, 17:27: Message edited by: Micah Sparacio ]
IP: Logged
|
|
|
Printer-friendly view of this topic
