|
Author
|
Topic: Joshua A. Smart: On the Application of Irreducible Complexity
|
Pim van Meurs
Member
Member # 541
|
posted 09. June 2003 23:04
Jerry
quote:
2) The ICS must demonstrate molecular punctuated equilibrium.
How does ICS demonstrate molecular punctuated equilibrium. What do you envision?
quote:
4) The ICS must demonstrate specificity and complexity specificity must be over 500 bits in the formula: Cs = ln(W), where Cs is complexity specification and W is the component parts of the system.
Not many systems would be CSI then unless Jerry has a different definition of component parts.
IP: Logged
|
|
Jerry D. Bauer
Member
Member # 756
|
posted 10. June 2003 01:16
Hi Pim:
We dissected this a bit on ARN and I must admit that this is the tenet of Julie’s work that I had the roughest time understanding, but I left it in anyhow.
I’m thinking that her meaning here is similar to Gould’s punk eek in that it just appears quite suddenly into existence rather than arriving uniformitarianistically in the mannerism posed by Darwin through the eyes of his colleague, Charles Lyell.
As to your second question, certainly organisms, cells and organelles would be CSI. But I must admit that ‘Cs’ is a bit shaky for my comfort and I’m hoping that if people care to, we can brainstorm on it a bit and drive it on home.
Many of you are more familiar with Dembski’s work than me. Has he ever dissected CSI in a manner that separately calculates specificity from complexity? That is what is needed in that section of the methodology.
I can envision that an object might be low in complexity, yet high in specificity or vise versa. This is the way that should work mathematically, IMHO.
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 10. June 2003 01:28
Jerry:
quote:
I’m thinking that her meaning here is similar to Gould’s punk eek in that it just appears quite suddenly into existence rather than arriving uniformitarianistically in the mannerism posed by Darwin through the eyes of his colleague, Charles Lyell.
I think that your understanding of Punk Eek may be erroneous. Certainly Punk Eek and 'uniformitarianism' are not mututally exclusive, that is Punk Eek can still happen gradually.
Jerry:
quote:
As to your second question, certainly organisms, cells and organelles would be CSI.
Why? How did you determine this? Are you saying that cells etc have > 3e+150 components? Fascinating. Or should I have used base-e since you used ln instead of the more customary log which would make it 1.e+217?
Jerry
quote:
Many of you are more familiar with Dembski’s work than me. Has he ever dissected CSI in a manner that separately calculates specificity from complexity?
What is specificity?
IP: Logged
|
|
Mike Gene
Member
Member # 149
|
posted 10. June 2003 19:17
As I explained above, according to Behe, the first step in an IC analysis is to identify the function and parts of a system. And I am becoming increasingly convinced that this first step has not been taken. In fact, I’m starting to suspect that people are trying to stretch Behe’s concept to forcefully squeeze the EQU into it.
FUNCTION
According to RBH, the function of EQU is that is a “logical function” – " Please don't conflate "EQU" (logical function) and "programs that perform EQU" (what evolved). "
Okay, but that still seems awfully vague to me. It’s hard for me to comment on the function because there are so many unanswered questions.
1. Is a “logical function” a very broad category?
2. Are we defining function to mean “it does something?” For example, RBH writes: The principal function of the programs is to perform the logic operation EQU. Thus, the function is to perform?
3. How many EQU’s evolved? 23 or 124?
4. Were these 23 (or 124) different ways to arrange the "parts" to get the same function?
5. Why didn't any other functions evolve with EQU-like complexity?
6. As for the simpler functions, how many parts were they composed of?
PARTS
Attempts to define the “parts” of the EQU have led to confusion. First, when I used the instruction itself as a part, RBH corrects me as follows:
quote:
As to parts, in constructing your list of 14 you have conflated "part" and "kind of part." 'q' is a "kind of part;" 'q as instruction #11' is a "part." . It's like a brick pillar made of a slew of bricks piled up one on top of another. All the bricks are identical - all are the same "kind of part" - but pulling out the brick on the bottom of the pillar has a considerably different effect than pulling out the brick on top. They are different "parts" in the context of the structure. I take "part" in the program above to be coterminous with "Instruction #."
But this means that none of the EQUs share parts, as the context is different for all of them. As I explained, if we define a part to include its position, then we can’t disassemble the system into parts. This is because disassembly eliminates the position. And if position is entailed in “part”, the part too disappears. Yet, when I made this point to RBH, clarity did not follow.
I asked a simple question: do any of the 23 independently derived EQUs share parts?
RBH replied: All 23 programs used the same set of 26 primitive instructions. Each used them in a different sequence.
Okay, and I read that as “yes,” the parts are the 26 primitive instructions. So, I tried to clarify this by asking if the 26 primitive instructions were the parts.
RBH then replied: The program I posted on page 2 of this thread is 60 instructions long, and has 60 parts in the definition I deem appropriate: "instruction in sequential context."
So the answer is no, none of the EQUs share parts because the part exists only when a sequential context exists (and none of the EQUs have the same sequential context).
But then, as I explained, this type of “part” is different from a machine part. With machines, parts can be disassembled and still exist. Yes, they must be put in the right position to function (highlighting the importance of assembly), but they can still exist as a “part” apart from the assembly. This is not a trivial distinction, because it means the “parts” of the EQU are not modular, thus they are not machine-like or cooptable. They are not cooptable because they don’t exist as parts apart from the whole.
It would greatly help if the community of IC critics could reach a consensus and define the “parts” of the EQU. Is the primitive instruction the part? Or is the instruction + position the part? The former seems to be more consistent with the concept of a “part.” But I can't continue without some working consensus.
The confusion in defining the EQU’s parts has then led to further confusion about the proper biological analog. What, in biology, does the EQU part map to?
For example, Rex first says: It's worth pointing out that Avida's instructions are more like amino acids than proteins, in that each instruction is identical, and that position is important.
But, in another context, writes:
Rex: However, an Avida instruction isn't a great analogy for either an amino acid or a gene.
Yet Yersinia thinks the instructions map to the genes. And Argon speculates about functional domains.
None of this confusion invalidates the Lenski paper as they did not, after all, seek to rebut IC. The problem comes from those trying to use the Lenski paper as a “last word” rebuttal of IC. As such, it is their burden to take the first step of the IC analysis. As it stands now, and as I mentioned above, there is this distinct impression that people are trying to force the EQU into the IC category. And there is independent evidence that supports this. In trying to explain why the EQU is IC, RBH uses two very suspicious analogies.
First, he appeals to a brick pillar. This should be a red flag, as whether or not a brick pillar should be considered an example of Behe’s formulation of IC is questionable, in the least. Furthermore, without having to spell it out (now), hopefully people can appreciate that brick pillars and motorized propellers are different in some rather significant ways.
Secondly, RBH most recently appealed to the sequence of nucleotides in DNA. This is another red flag, as no disciple of Behe (that I know of) has argued that the nucleotide sequence of DNA is an example of IC. When scoring the parts of the flagellum, I don't tabulate all the individual nucleotides + position to determine the parts. Again, the different between a string of characters and a motorized propeller should be obvious.
SUMMARY
Thus far, I don’t see much evidence that the first step of the IC analysis has been taken. Function has been identified, but many relevant questions remain unanswered, making it hard to probe the assertion. Confusion also remains concerning what constitutes a part and even more confusion remains concerning what aspect of biology these EQU parts map to. Finally, analogies used to render the EQU IC appeal to things such as brick pillars and a sequence of nucleotides.
Without further clarification and some attempt to reach consensus (i.e., not simply a rehash of the same arguments), I don’t see how this work is relevant to Behe’s concept of IC.
[ 10. June 2003, 19:18: Message edited by: Mike Gene ]
IP: Logged
|
|
Argon
Member
Member # 276
|
posted 10. June 2003 20:47
Mike Gene writes: "And Argon speculates about functional domains."
I speculate about "functional domains" because some of the other, multi-instruction logic functions compose the EQU function. Thus some of the "parts" of the derived EQU functions may be grouped as separate assemblages, not just single instructions.
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 10. June 2003 22:11
Mike, I'm having trouble understanding why your concerns are relevant to Avida's implications for IC. Avida is not biology, and it doesn't pretend to be. However, I think it does address the logical structure of IC.
Let me quote Behe (Darwin's Black Box, pp. 39-40):
quote:
What type of biological system could not be formed by "numerous, successive, slight modifications"?
Well, for starters, a system that is irreducibly complex. By irreduciby complex I mean a single system composed of several well-matched, interacting parts that contribute to the basic function, wherein the removal of any one of the parts causes the system to effectively cease functioning. And irreducibly complex system cannot be produced directly (that is, by continuously improving the inital function, which continues to work by the same mechanism) by slight, successive modifications of a precursor system, because any precursor to an irreducibly comkplex system that is missing a part is by definition nonfunctional. An irreducibly complex biological system, if there is such a thing, would be a powerful challenge to Darwinian evolution. Since natural selection can only choose systems that are already working, then if a biological system cannot be produced gradually it would have to arise as an integrated unit, in one fell swoop, for natural selection to have anything to act on.
Even if a system is irreducibly complex (and thus cannot have been produced directly), however, one can not definitively rule out the possibility of an indirect, circuitous route. As the complexity of an interacting system increases, though, the likelihood of such an indirect route drops precipitously.
(Italics as in original, bold mine.)
Note that the core logic doesn't depend on the exact definition of a part, or on the fact that genes are encoded by DNA, or whether there are other complex functions, and so on. Those are interesting additional questions, but the answers are of limited relevance when we are assessing the validity of Behe's core logic.
The claim is as follows, as near as I can tell: if a system is composed of parts, and removing any one part abolishes function, that system is IC; if a system is IC, then it cannot have been produced directly except in one fell swoop; since indirect routes are improbable, and one-fell-swoop solutions are improbable, IC systems do not evolve.
In analyzing the Avida system, it's helpful to use one of Dembski's methods of analyzing the flagellum, presented on pp. 291 of No Free Lunch. There, the probability of coming up with a discrete combinatorial object, p_dco, is p_orig * p_local * p_config. Proteins cannot be neatly divided into origination, localization, and configuration probabilities, since primary sequence plays a large role in determining all three. However, it assists in performing the calculation that demonstrates just why an IC system is apparently so troublesome for gradual evolution. Even if you are in a junkyard that contains all the parts required for a 747 (p_orig = 1), the chance that you'll bring them all together in a sensible way (p_local*p_config) is vainishingly small. Likewise, all the instructions are there in Avida (p_orig = 1), bringing the parts together is a challenge.
When we look at a system like the flagellum, where many components appear to be similar to other proteins (other "parts"), one major question is, "how did that functional unit get *here*, working together with all these other functional units from *there* and *there* and *there*?" It is this question that Avida directly addresses.
First, let's consider the "goal" of EQU. Is this even complex? The authors claim that their best attempt to write an EQU function uses 19 instructions out of 26 possibilities. I don't see a shorter way to do it, personally, although there are a few different ways to do it at 19 instructions. The probability of getting one of these is 26^-19, or about 10^-27. This is definitely not below the universal probability bound, but getting 23 different versions has a probability of about (10^-27)^23 or on the order of 10^-620. So you should *never* see multiple instances of EQU coming up.
Does EQU contain well-matched interacting parts? In the human-created version, it does. EQU is a logic function, which returns True if inputs A and B are equal, and False if they are not. Unfortunately, the program only gets to use the NAND operation in its basic instruction set. It's known from logic that NAND can be used to construct any logic function, but EQU is the hardest, requiring 5 separate NANDs (and many more instructions to make sure you're NANDing the proper things together). A working EQU, built from NANDs, is thus a well-matched set of interacting instructions.
(Specifically, with inputs a and b it can be implemented as (((a#b)#b)#((a#b)#a))#(((a#b)#b)#((a#b)#a)), or with variables c and d available, c = a#b ; d = (c#b)#(c#a) ; EQU = d#d, where # is the nand operation (true exactly when a and b are not both true).
Now, to make a long story short, EQU does evolve. It doesn't evolve directly--if you select for EQU only, by giving the digital organisms extra "energy" (permission to execute more instructions and therefore replicate faster), it doesn't appear. And, well, it had better not, given that it's well below the universal probability bound.
However, when you give energy rewards (increasing with complexity) for performing other logic functions, EQU does evolve. Why? Apparently, simpler logic functions provide an indirect route for the evolution of EQU. Well, okay, this is maybe not such a huge surprise. However, it is a bit more surprising when you consider the following: EQU does not evolve as a logician would create it from simpler logical components. Rather, it seems instead to hijack portions of other logic functions. (Or, put another way, a random assembly of logic-function-bits has a much higher chance of generating EQU than a random assembly of Avida-instructions.)
This is exactly the indirect, circuitous route Behe talked about. It's enough to knock a probability of 10^-620 up to about 1/2 after a few tens of thousands of generations of a population of a few thousand. (We would expect the probability to be on the order of 10^-613 if the system had to be created directly.)
So here's the critical question: if circuity works so well for Avida, why doesn't it work for biological organisms?
This is a question about the structure of the underlying space that's being searched. We often come across problems where the easiest solution is also one of the best: for example, if you are trying to boil water, you get a container, and some flame, and put the container over the flame. If you do a bunch of other stuff, almost none of it will yield boiling water for you. This is the type of problem where the "circuitous route" can be neglected.
However, there are other problems that have a very different structure. For example, in the Traveling Salesman Problem, you have a bunch of cities that you want to visit, and the goal is to choose the shortest route to visit them all. Typically, there are extremely few "best" solutions. However, there are ridiculously many solutions--you could visit cities by alphabetical order, or even at random, and you would end up visiting them all. It just would be inefficient. Here, a (literally) circuitous route gets you the answer almost all the time, simply because there are so many workable circuitous routes.
Interestingly, evolutionary algorithms work very well on the Traveling Salesman problem--to find a solution, they generate a circuitous route and then optimize it over succeeding generations until an efficient route remains.
Which type of search-space do complex macromolecular structures inhabit? Behe seems to assume that they're in the boiling-water class. But this seems like a dangerously unsafe assumption, for a number of reasons.
Avida is one, showing a prevalence of cooption in a system where one would otherwise be tempted to predict a direct route (since we know the "right" answer--when I sat down to come up with an EQU algorithm on my own, I picked the direct route as the "most obvious").
Another is the prevalence of components in a complex that aren't essential for function. One of the hallmarks of a circuitous route--clearly seen in Avida--is that you end up with a bunch of non-essential components that are largely historical artifacts in addition to the irreducible core. A good fraction of flagellum components are in this category.
And this finally brings us back to Joshua's paper, and the earlier claim that in addition to all the good suggestions he has about investigating irreducible complexity, it is absolutely necessary that biological systems be evaluated for hallmarks of circuitous development if IC is to be applied as evidence against evolution.
[ 10. June 2003, 22:14: Message edited by: Rex Kerr ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 11. June 2003 00:42
Let me respond to Mike Gene's remarks with the same organization he used.
FUNCTION
"EQU" is a logic operator that evaluates whether two inputs are equal, with a truth value of "1" if they are and a truth value of "0" if they are not. A program - the program of one Avida critter - that performs EQU accepts two inputs and if they are equal (in the case of the Avida simulation, are both 1 or are both 0), the program writes the value "1" to the output of the virtual machine of the critter whose program is currently running. If one of the inputs is 1 and one is 0, the program writes a "0" to the output of the virtual machine of the critter whose program is currently running.
I made a misstatement when I used the phrase "logic function" in the context of this discussion. That allowed Mike to confuse two senses of the word.
Thus the programs evolved to perform the function of evaluating a pair of inputs to ascertain whether they satisfy the conditions defined by the truth table for the logic operator "EQU".
PARTS
Mike uses the word "parts" in two different senses, as did I, and that confusion needs to be clarified.
The 26 primitive instructions comprise a set of raw materials in much the same sense that a bunch of bins of Lego components or tinkertoy components (for the elderly amongst us) comprise a set of raw materials that can be used in the construction of many different structures. Think of 26 bins, each bin containing one kind of instruction, with a bunch of identical components in each bin.
Suppose we have a bunch of bins holding physical components, one bin containing a bunch of identical wood bases, one bin a bunch of springs, and so on. Within the bins all the components are identical. One builds a mousetrap by selecting an appropriate number of components from appropriate bins and assembling them in a particular configuration. The unordered bunch of components is redefined by the assembly operation as "parts" of a mousetrap. (A mousetrap is actually an atypical machine because it requires just one of each kind of component. Most machines contain several instances of some components.)
Programs are assembled by drawing components from the various bins and assembling them in an appropriate sequence. The final product contains as many parts as there are instructions in sequence, each instruction in the sequence being a "part."
Similarly, the two identical screws that hold the handle of my wife's hair dryer to the body are interchangeable parts of the hair dryer, drawn from a larger supply of such identical and interchangeable components. Like a machine, the 60 parts of the Case Study program can be dissasembled (perhaps by listing them alphabetically) and still exist, all 60 of them, though in that dissasembled state they could not perform the function of the program just as Mike's hair dryer won't dry hair when dissasembled. My remarks about "instructions in sequence" means that when one re-assembles them, particular parts have to go in particular places in the sequence, just as in re-assembling his hair dryer Mike has to put the various parts in the right places in relation to one another. I see no confusion or difficulty with that correspondence.
When the raw materials are assembled into a sequence such that when run on the virtual machine of an Avida critter the machine correctly performs the function of evaluating two inputs to ascertain whether they satisfy the logic operator EQU, they are "parts". The Case Study program has 60 "parts," the program being an ordered aggregation of various numbers of each of the 26 components drawn from the bins.
So the Case Study program has 60 parts, of which 35 (IIRC) are indispensable to the performance of the function of evaluating pairs of inputs to ascertain which of the relationships the logic operator EQU's truth table defines the inputs satisfy.
Human-written programs typically have several hierarchical levels - primitive instructions, functions and procedures (no, Mike, not "functions" in the sense of Behe's term - "functions" in the sense of mathematical functions), and higher levels of organization. (In human programming nowadays the assembly language and machine language levels of analysis are generally hidden from the programmer. Forty years ago one had to be adept at mentally translating binary to octal quickly in order to remember long strings of 1s and 0s. The Polaris A-series autopilot computer was not programmed in C++!)
The Avida programs quite possibly have some hierarchical organization, but it is very difficult to tease it out since there are no constraints on the evolution of the programs similar to those imposed on human programmers by the syntactic requirements of the high-level languages humans work in. Argon's 'functional domains' are a (potential) level of analysis one step above the individual instruction level in the programs, and it's a reasonable speculation though difficult to evaluate in practice.
CORRESPONDENCES
I am not entranced with any particular strict point-for-point correspondence between the instructions of an Avida program and biological entities nor with the parts of a mousetrap or any other machine that one might be inclined to identify. In fact, as I noted above, the mousetrap atypical in that it uses just one instance of each non-identical part, and because of that it is responsible for some of the confusion here. In that respect it is a poor example.
The useful 'analogy' between biology and the Avida simulations is not at one or another specific biological component level, but at the evolutionary processes level. The simulation represents the processes held to account for evolution, not one specific biological system. What the various suggestions for biological entities analogous to Avida instructions - genes, amino acids, and so on - illustrate is that the instructions are roughly at those levels of analysis and not, for example, at the level of carbon, nitrogen, hydrogen, and oxygen atoms.
A FEW MISCELLANEOUS NOTES
My example of a brick pillar was not an analogy for an IC structure, and if Mike had read reasonably carefully he'd know that. It was meant to illustrate that two parts of identical form that are in different places in a mechanical structure can play different roles with different consequences for the functioning of the structure as a whole. That's all. It was most definitely not intended as an example of irreducible complexity. Mike is merely obfuscating in his remarks on that topic.
Mike asked why other functions with EQU-like complexity didn't evolve. Selection, Mike. That venerable evolutionary operator. Read the paper.
He asked how many parts of the programs were associated with performing the simpler functions. It varied from function to function and from program to program. In many of the programs, perhaps all of them, some of the parts - individual instructions in a particular place in the program's sequence of instructions - participated in performing several of the logic operations the programs performed.
Many of Mike's "relevant questions" would be answered if he would deign to actually read the paper and its associated Supplementary Information. But since he is unfamiliar with computer programs, especially at this level, and is unwilling to read the original work with its Supplemental Information, I'm not surprised that he's been confused by my attempts to summarize it for his benefit.
Moreover, whose "consensus" is necessary? I don't much care if Mike agrees with me or not, given his self-professed ignorance of the nature of the Avida simulation. I care about his posts only to the extent that they raise interesting questions, identify issues that I need to clarify in my own thinking, or misrepresent an issue or interpretation. I'm not much interested in whether he agrees with my interpretation of the study. I'm satisfied that the programs that evolved in the Lenski, et al., study meet all the relevant conditions of Behe's original definition of irreducible complexity, and that they evolved in a simulation context where the relevant constraints on what could occur are those identified by evolutionary theory as likely to produce such structures. As such the results of the study constitute an existence proof for the proposition that irreducibly complex structures not only can evolve, they do so quite readily, much more readily than the various probability estimates and qualitative remarks about "exponentially unlikely" that are thrown around would imply. I have seen no reason to change that judgment given what I have read in this thread.
RBH
P.S. Since Rex posted his remarks while I was writing offline (with my machinery off for a while during a short thunderstorm) I'll add this: What Rex said! ![[Smile]](smile.gif)
[ 11. June 2003, 00:48: Message edited by: RBH ]
IP: Logged
|
|
Jerry D. Bauer
Member
Member # 756
|
posted 11. June 2003 16:35
A few notes on function:
I would think that very few IC systems are actually reduced to the IC core and we need turn to the lab in order to deduce these core systems.
In mammals, the cardiovascular system is IC as it cannot function without blood to pump, a heart to pump it, veins and arteries to carry blood around, lungs to oxygenate the blood, kidneys to clean it out and a brain to make it all work together. Yet, the cardiovascular system is much more complicated than this; it just cannot be reduced past this core.
Organisms are IC systems. They can be reduced but not past a certain point.
I can lose an arm and still function, but my degree of function has gone down. I could lose all four limbs and still function although I wouldn’t function very well and energy would have to be inputted into the system in the form of work (A nurse to feed and cloth me) to keep me functioning.
This is why I’ve proposed that function be shown mathematically. In RBH’s post, if I understood him (and even if I did not we can still consider the scenario) something in computer programming consists of 60 parts, but only 26 of these are the IC core.
If I use this formula: Ft = (FM – Fm) + .001, where Ft is total function, FM is maximum function defined as a system with all it’s component parts intact, and Fm is minimum function defined as the number of parts of a system that are irreducibly complex.
At full function that system would work out in total function to 60 – 26 + .001 = 34.001.
If FM drops to 35 parts so will total function drop: 9.001.
When the system is down to its core components, the formula will still show it as a functioning system, but just barely because Ft = .001.
But if total component parts drop below our core element needed for the system to function then Ft drops to < 0 showing no function at all.
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 12. June 2003 00:03
My reply to both RBH and Rex were similar so I'll leave it as one post.
I really don't see the relevance of the Behe quote that Rex cited. The quote does indeed require that you specifically identify the parts that are needed for the function of the system. The definition of a part does not have to be precise, however, it does have to be useful in determining what can be removed before the system loses function. And this is the crux of Mike's concern, identifing an instruction as a part is not even analogous to anything that we see in the molecular machines of biology, or even of human made machines. The instructions are interspersed whereas, in machines such as bacterial flagella, we don't see, say, rod proteins in the basal body, motor, etc. Each part is discrete.
That is the problem with calling the EQU system IC, we don't see the utter plasticity (which points to the simplicity of attaining the function) in any of the IC systems that Behe or Mike Gene discuss. For example, Rex writes:
quote:
First, let's consider the "goal" of EQU. Is this even complex? The authors claim that their best attempt to write an EQU function uses 19 instructions out of 26 possibilities. I don't see a shorter way to do it, personally, although there are a few different ways to do it at 19 instructions. The probability of getting one of these is 26^-19, or about 10^-27. This is definitely not below the universal probability bound, but getting 23 different versions has a probability of about (10^-27)^23 or on the order of 10^-620. So you should *never* see multiple instances of EQU coming up.
And yet, if you read the paper, the organisms were able to evolve a 17 "part" EQU program.
quote:
The number of instructions required for EQU ranged from 17 to 43, with a median of 28 instructions.
I've mentioned this before, we see so much plasticity, from a 35 instruction EQU, to a 19 instruction EQU, to a 17 instruction EQU, and different kinds of EQU (not from radically different "components" but using the the same set of instructions. What is the equivalent of this with the bacterial flagellum? We either see all 20 parts, or none of it's parts. We don't see different kinds of bacterial flagella, some less simple, some more complex, each of the essential components are all conserved. It seems for EQU, as might be expected from the difference between a set of instructions and a set of machine parts, is that there seems to be no end in sight for how plastic the EQU function can be. There is not really a "good fraction" of flagellar components that are non-essential, some are just luxury items, some are simply not needed because some bacteria do not have an outer membrane. However, most of the components simply cannot be removed and yet you still have function. You don't see the same plasticity here.
RBH writes:
quote:
Human-written programs typically have several hierarchical levels - primitive instructions, functions and procedures (no, Mike, not "functions" in the sense of Behe's term - "functions" in the sense of mathematical functions),
But what is the difference? In Behe's term and in the programming term, a function is equivalent. If I have a program that writes my name, a simple one for the sake of argument, I could write it in two procedures,
int add(int a, int b)
{
return a+b;
}
int subtract(int a, int b)
{
return a-b;
}
In the "main" we see the "assembled" program where two numbers are subtracted and added. It's function is to do both, however, it has two subfunctions, one adds, the other subtracts. One can posit, in like manner to EQU, how one procedure was co-opted from the other through subfunctions, which finally led to the final printing out of my full name in correct order.
Of course this is simply illustrative, however, my point is clear. You can see how hard it is to tell what is a part from this and what is not. Each instruction isn't necessarily discrete, the logic is essentially interspersed throughout the program itself, and can be combined and used for many different purposes given so many lines of code and opportunities to choose from it. To do EQU, the organism simply outputs a 1 if two input strings are both 0 or both 1, and a 0 if they are not. I for one don't really see this as the circuitous route for a complex system whose probability did not fall extremely low despite the fact that it is IC.
[ 12. June 2003, 00:07: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 12. June 2003 00:43
If you read the paper more carefully, you'll notice that the 17 alone are not enough to produce EQU; they only came up with 17 essential components by single knockout because of redundancy.
Furthermore, not all EQU have the same evolutionary origin. Related EQUs are similar, and unrelated ones are less similar. Flagella and cilia perform the same function in many bacteria, yet they too are not that similar and not thought to be closely related. "Flagellum" is not a function. "Motility" is; there are multiple solutions to motility, and multiple solutions to EQU.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 12. June 2003 01:11
Two minor corrections to Nelson's post first. He wrote quote:
I've mentioned this before, we see so much plasticity, from a 35 instruction EQU, to a 19 instruction EQU, to a 17 instruction EQU, and different kinds of EQU (not from radically different "components" but using the (sic) the same set of instructions. (Emphasis original)
Nelson is here again conflating "EQU" and "program that performs EQU." There are a number of programs that perform EQU; there is only one "kind" of EQU. And the numbers he gives are for the irreducible core as determined by Behe's knockout criterion. We don't know that the 17 instructions of the shortest irreducible core are all that are involved in that program's performance of EQU. Also recall the authors' mention of epistasis: one can define the irreducible core with the knockout procedure, but that often misses interactions and the redundancies Rex mentions. So it provides a minimum estimate of the number of components of the irreducible core, not necessarily the exact number.
And he wrote quote:
We don't see different kinds of bacterial flagella, some less simple, some more complex, each of the essential components are all conserved.
Well, now we actually do see all sorts of flagella attached to motile single-celled entities, ranging from the 'standard' 9+2 to the 3+0 of Diplauxis hatti. See p. 142 of Finding Darwin's God for examples of a number of variants.
His main criticisms of the interpretation of the study as indicating that IC can readily evolve seem to be three. First, he rehearses the 'it's hard to tell what the parts are' argument.' He gives a short program as an example, saying quote:
Of course this is simply illustrative, however, my point is clear. You can see how hard it is to tell what is a part from this and what is not. Each instruction isn't necessarily discrete, the logic is essentially interspersed throughout the program itself, and can be combined and used for many different purposes given so many lines of code and opportunities to choose from it.
His illustration, however, is of a high-level language, not the assembly language of Avida. Hence its force is questionable. Whether the parts of the example he gave seem difficult to discern is irrelevant to the language of the simulation. Change the level to that appropriate to the simulation and the analogy loses its force. This is argument by shifting grounds.
Second, he re-visits the 'it isn't really that improbable' line of argument that both John Bracht and Micah tried out: quote:
To do EQU, the organism simply outputs a 1 if two input strings are both 0 or both 1, and a 0 if they are not. I for one don't really see this as the circuitous route for a complex system whose probability did not fall extremely low despite the fact that it is IC.
This is not a little incoherent. Is the antecedent of "whose" the system or the route? I can only suggest that he reread Rex Kerr's posting. Nelson seems to be conflating the probability of a "route," by which I assume he means the evolutionary pathway to one or another of the programs that evolved to perform EQU, and the probability of formation by chance of a program that performs EQU. Regardless of which he means, I encourage him to read about the control condition which tested exactly the hypothesis that a program that performs EQU could emerge 'directly' as a relatively high probability event absent intermediates. It didn't. So it isn't a real high probability event. I do appreciate his "...despite the fact that it is IC," though. That means we might agree on one question, though I'm not super-sure of the antecedent of "it." Something surely is IC here, though. And what evolved, what performed the function, and what passed the knockout test, is the assembly language programs.
Finally, he apparently objects to the fact that instructions are re-used; that the several parts of the programs are not all different, and are not unique just to programs that perform EQU. Re-quoting a longer segment partly quoted above, he wrote that quote:
I've mentioned this before, we see so much plasticity, from a 35 instruction EQU, to a 19 instruction EQU, to a 17 instruction EQU, and different kinds of EQU (not from radically different "components" but using the (sic) the same set of instructions. What is the equivalent of this with the bacterial flagellum? (Emphasis original)
One could equally accurately say "we see so much plasticity, from a 9+2 flagellum to a 6+0 flagellum to a 3+0 flagellum, all using the same parts (microtubules)." Biology really does re-use stuff in various combinations and permutations. This argument implicitly abandons the generality of the notion of IC, appearing to confine it to a description of structures composed of parts that are unique, different from the parts of all other structures. That restriction is going to have real problems in biology.
RBH
(Edited for waaay too many typos: It's late here.)
[ 12. June 2003, 01:29: Message edited by: RBH ]
IP: Logged
|
|
John Bracht
Member
Member # 5
|
posted 12. June 2003 02:40
Rex,
Be careful. You're getting a little sloppy with your statistics, I think:
quote:
First, let's consider the "goal" of EQU. Is this even complex? The authors claim that their best attempt to write an EQU function uses 19 instructions out of 26 possibilities. I don't see a shorter way to do it, personally, although there are a few different ways to do it at 19 instructions. The probability of getting one of these is 26^-19, or about 10^-27. This is definitely not below the universal probability bound, but getting 23 different versions has a probability of about (10^-27)^23 or on the order of 10^-620. So you should *never* see multiple instances of EQU coming up.
This problem with this logic is that you're confusing a fabrication with a specification. The figure of 1 in 10^620 only works if we're trying to evolve precisely the set of 23 different EQU algorithms we got. But in reality we're just selecting for generic EQU, and any set of 23 EQU algorithms will do. And the fact that the field contains many potential solutions makes the probability of evolving any one solution greater, not less.
What needs to be done is to estimate the number of ways to get EQU (within some complexity limit) and compare that target area to the total number of possible arrangements of instructions (within the same complexity limit), thus giving a correct probability of evolving EQU once. Then, take that number and raise it to the 23rd power to get the probability of getting EQU 23 times.
John
[ 12. June 2003, 02:51: Message edited by: John Bracht ]
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 12. June 2003 12:05
Well, that's annoying. UBB ate my post when it complained that I had a tag that wasn't permitted, and I forgot to save a copy.
It'll have to be much shorter this time, and only present the most generous probability calculation (generous to those who wish the probability to be high).
The only way to get EQU is through nand instructions. Each nand requires data to be saved and/or retrieved, for about 4 operations total. (push-pop-nop, swap-swap-nop, push-swap-pop, etc.) The first one is an exception, but requires IO, so we predict about 4 operations per nand. In a chunk of length K, bounded by IO operations, with N nands, we need about N nands, 2(N-1) pushes pops or swaps, and N nops. This computation is likely to be derailed by spurious pushes, pops, swaps, or nands, in addition to inc and dec and add and sub, so in the K-(4N) other slots, we'll say we have a flat 6/26 chance of messing something up. Also, nops have to be pretty carefully positioned after the instruction they modify, and we get about two choices of nop and two choices of instruction per nop. Our probability is then something like C(K,N)*C(K-N,2(N-1))*4^N*(1/26)^N*(1/26)^n*(1/26)^(2(N-1))*(24/26)^(K-4N+2) to get the components needed for a working logic operation in a stretch of size K. For K in the range of sizes of Avida, this works out to be about 10^-12 for N=5 (the minimum), 10^-14 for N=6, and so on. Many of the remaining instructions will be nonsensical (e.g. push push pop push pop pop pop) for any logic calculation, but we'll ignore this.
Not all potential logic calculations of suitable length actually perform EQU. Of those potential calculations that actually load data properly into each nand operation, about 0.3-0.5% actually implement EQU (as determined by explicitly computing the result of every logic operation with 2-10 nands). Thus the probability of getting EQU is about 10^-14 under the most wildly generous estimates (instead of 10^-27). Each Avida run involves about 10^8 different organisms, giving a probability of about 10^-6 per run of coming up with an EQU by chance (assuming random sampling, which doesn't happen; sampling is historically biased away from lots of nands and push/pop/swaps). The chance that 23 of 50 will come up EQU is C(50,23)*(10^-6)^23 = 10^-124 or so.
Still wildly improbable, even using unjustifiedly generous assumptions. Note that including all experimental conditions that could lead through evolution to EQU would raise the numbers from 23 to over a hundred, and push the probability well under the universal probability bounds.
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 12. June 2003 12:31
John Bracht states: The figure of 1 in 10^620 only works if we're trying to evolve precisely the set of 23 different EQU algorithms we got. But in reality we're just selecting for generic EQU, and any set of 23 EQU algorithms will do.
Does this mean that John also agrees that the perturbation calculations performed by Dembski to calculate probabilities are equally 'sloppy'? After all should Dembski not be selecting for generic flagellum and any set that matches such flagellum would do? Certainly Dembski's NFL suggests that CSI is to be calculcated with respect to a uniform probability function. His 'Shannon-like' complexity measure is a simplified form of the general formula limited to uniform probabilities for instance. And despite inquiries by Wein and others as to the exact nature of the relevant probability function, Dembski seems to be wavering between the uniform probability function and the actual probability function. Of course when using the actual probability function it would be trivial to show that ID cannot generate CSI since the probability would be close to 1.
Wein argues that
quote:
Complex specified information (CSI) is a concept of Dembski's own invention which is quite different from any form of information used by information theorists. Indeed, Dembski himself has berated his critics in the past for confusing CSI with other forms of information. This critique shows that CSI is equivocally defined and fails to characterize complex structures in the way that Dembski claims it does. On the basis of this flawed concept, he boldly proposes a new Law of Conservation of Information, which is shown here to be utterly baseless.
Thus perhaps it may be helpful to clarify first how CSI is to be calculated and compare this with how Dembski calculates said measure.
Dembski states in one of his metaview articles that
quote:
What this means is that even though with respect to the uniform probability on the phase space the target has exceedingly small probability, the probability for the evolutionary algorithm E to get into the target in m steps is no longer small. And since complexity and improbability are for the purposes of specified complexity parallel notions, this means that even though the target is complex and specified with respect to the uniform probability on the phase space, it remains specified but is no longer complex with respect to the probability induced by evolutionary algorithm E.
If that were the case then extending this logic to ID, one has to conclude that it remains specified but is no longer complex with respect to the probability induced by intelligent design.
And does the example not show that natural selection can be a probability amplifier in that it reduces, that is weeds out, pathways?
[ 12. June 2003, 13:04: Message edited by: Pim van Meurs ]
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 12. June 2003 21:40
Rex,
Actually they don't say specifically whether the 17 instructions were required or if they underestimated this, they simply point out that careful examination of the genome yielded a lot of redunancy. It looks like for this paper, they wanted to make sure to err on the side of claiming too few instructions rather than too many. Nonetheless, 17 instructions alone is what they found. There is another paper that goes a lot more into this type of analysis here:
Wilke CO, Wang J, Ofria C, Adami C, and Lenski RE, Evolution of
Digital Organisms at High Mutation Rate Leads To Survival of the Flattest Nature 412, 331-333 (2001)
where less is more so to speak.
As far as eukaryotic cilia go, you are comparing apples and oranges, whereas with these different types of EQU programs, you are comparing oranges and oranges. Eukaryotic cilia are not rearrangements of flagellar components (and vice versa), they are completely different and unrelated line of descent. All of these organisms arose from the same ancestor and use the same instructions in different combinations.
When you contrast the function bidirectional motor propeller with the general function of "motility", and in the case of eukaryotic flagella, a linker-motor-MTs (flagellin and MTs are completely different structures), then you will see how irrelevant simply bringing up "different types of motility" is when it is these machine configurations we need to explain. I don't think that co-option does the trick when it comes to complex machines.
Also, I think you completely misunderstood Bracht's reply, he's basically saying that your calculation treats the event as if one organism was able to evolve all 26 possiblities of EQU, that would certainly be improbable.
It's the mathematical equivalent of saying something like: The color green can make a bit of a noise, but a combination of blue, green, and red is unlikely to eat lunch before breakfast. It is so without reason that I don't even know where to start explaining that it doesn't make sense.
However, with 26 possibilities, many individual organisms going through replication and mutation will likely hit upon all of them, seperate organisms of course, that is likely and quite probable, especially when there are 26 possibilities and the function isn't all that complex.
With the flagellum the situation is different, there is only 1 way to get bi-directional motor propeller etc, which is why it is IC. It would be more convincing if we see in Biology, different ways that natural selection accomplished this among bacteria, in fact, that is exactly what co-option predicts. For example, A with parts EF, not BC, another with parts BCD not A. THat is what we see with EQU, we don't see that with flagella. Instead, we see completely new innovations that are universal among the respective kingdom.
[ 12. June 2003, 22:00: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
|
Printer-friendly view of this topic