|
Author
|
Topic: Retention of Complexity vs. Origination
|
Atom
Member
Member # 1840
|
posted 18. December 2005 11:45
Hey everyone, I'm a brand new member, a few weeks new to this board (reading), and this is my first post. I am a computer scientist (software engineering) and generally supportive of the ID research program.
I was thinking today about the problem of generating CSI by stochastic mechanisms, which led to a train of thought that included thinking of the standard Darwinian response that "natural selection" can produce CSI. This is sometimes taken for granted by the thinkers I encounter.
Dembski has done much of the groundwork in showing why CSI cannot be derived from unguided forces, so I won't rehash his points here. My train of thought led me over the claim that the idea of Natural Selection boils down to a tautaulogy: Those organisms which reproduce the most (are the fittest) will reproduce the most. I will henceforth refer to this as the NS tautaulogy. I noticed something about this framing (which in my eyes, makes it a possible strawman in the making). It does not include an important corollary, which deals with the net effect of NS. Basically, NS says one more important thing:
"Those organisms which reproduce the most will reproduce the most, and over time, the population will consists of mainly these individuals."
So, not only will the most reproductively successful reproduce the most, but their hereditary material will over time dominate the population gene pool.
Now, ID advocates are right in correcting the popular use of NS to include creative processes. NS isn't a creative force; it is merely a retentive force. So, stochastic process must do any unguided creative work in the Neo-Darwinian framework.
To see why this distinction is important in discussions concerning the advent of CSI rich artifacts, I offer the following illustration:
Let's go to a island hundreds of years ago, where a large tribe lives. These tribal Germanic ancestors are constantly at war with eachother, and as many die in this tribe from battles amongst themselves as from disease. Let's also assume this population is isolated from other islands and tribes.
Now, one day we posit that a man from the tribe is born with a ray gun for an arm. This gun has two features: kill mode, which vaporizes anything he shoots it at, and food mode, which produces food from the end, both vegtables and meat. Now let's assume this ray gun arm is a dominant trait, so when he reproduces, his children inherit this arm feature.
Now it can be safely assumed that this trait will confer on him and his offspring tremendous reproductive success. He can kill any threat, and has a constant supply of nutrients. So over time, applying NS with it's corrollary, we see that his genes will begin to dominate the tribal gene pool (he can outreproduce any competitor by killing him), and soon most of the population will have a ray gun arm.
End of illustration.
So what did we learn from this regarding the origination of CSI structures? We saw that NS (in non-tautaulogical form) does actually provide a good explanation as to why a given trait is seen in an entire population. But what does it fail to explain? The origination of the ray gun arm in the first place! We began by assuming the appearance of a complex trait, but never explained its appearence. In the same way, NS cannot be said to explain the origination of CSI in nature, but may be useful in explaining its spreading.
I hope this post was within guidelines. I'd like to hear any thoughts on this, or corrections if I missed something.
IP: Logged
|
|
Christopher D. Beling
Member
Member # 723
|
posted 18. December 2005 19:59
Hi Atom,
I very much agree with your analysis and follow Dr. Demski's reasoning that stochastic processes in themselves cannot lead to CSI. NS can only conserve CSI (and even then with some degradation) but cannot produce it. However, although I am convinced - I find the formalization in terms of using NFL (No Free Lunch) theorems as a means of ruling out GA (Genetic Algorithms which employ NS) only partially acceptable as a response to the proposed neo-Darwinian mechanism. I suppose the reason is that most neo-Darwinists would accepts some sort of "fitness landscape" as a given fixed commodity - which in principle permits the use of a GA in producing CSI. I would thus be very interested in helping discuss here ways of substantiating the complete inadequacy of GAs as providing a means of real life complexity increase in bio-evolution. To do this one needs to try and calculate the probabilities of hitting an item of new CSI given existing replicating CSI within a given fitness landscape - perhaps putting in real numbers. I have some ideas on how this might be done but I would be interested to here from you and others. - Chris
[ 18. December 2005, 20:04: Message edited by: Christopher D. Beling ]
IP: Logged
|
|
Atom
Member
Member # 1840
|
posted 18. December 2005 20:54
Hey Chris,
I think the problem boils down to whether or not a continuous fitness landscape exists. In the case of functional protein groups, it seems that they are actually highly isolated (at least according to Behe's 1996 paper, and Meyer's new "Higher Taxonomic" paper), which would make the fitness landscape essentially discontinuous, with non-functional chasms. That being said, mutation is still doing the actual work of producing the novel "increases" in CSI, while NS still only acts as a retentive force, in agreement to my original point.
But, I'd like to hear your ideas concerning fitness landscapes and GAs. Do you know if No Free Lunch addresses this issue? (I haven't read the book, only his paper on the NFL regress.) I would like to hear what you had in mind, and maybe we can work through some math, or code some simulations. (I write code for a living.) So I'm open to ideas. I'll continue to think on this point, though, as it may indeed represent a weakness in the NFL case.
IP: Logged
|
|
David L. Hagen
Member
Member # 323
|
posted 18. December 2005 23:18
Atom
Your note that the origin of the CSI is not explained by mutation is critically important.
You observation
quote:
Meyer's new "Higher Taxonomic" paper), which would make the fitness landscape essentially discontinuous, with non-functional chasms.
is also vitally important.
I expect that biological "function" will show mesa or table top type plateaus with rapidly dropping sides to the "chasm" between tabletops. Sometimes a single mutation in DNA space results in no protein being formed, or forming a protein that cannot fold properly resulting in NO functionality. i.e., the drop in function being a step jump in folded protein space. Then there are corresponding tabletop configurations in DNA space that generate those configurations. (I am thinking towards how to formalize these definitions and methods.)
I think that if anything, this will strengthen the NFL case, not weaken it.
Mathematically, Genetic Algorithm's are just one one method of randomized populating multidimensional space. With conventional nominally somewhat smooth functions, they are robust in eventually finding global minimums, but are very slow.
I wonder as to how practical GAs will be with a highly step function mesa tabletop type objective function in DNA space mapping onto another mesa table top protein sequence space and then to another strong mesa type function in folded protein space. This may end up being a very highly randomized search effort with very little "gradient" of change in the objective function in folded protein space to help with the optimization.
Encourage careful use of terminology re mutations. They are often referred to as a source of "information" (sic) but they do not generate CSI. Rather from a Design viewpoint, they degrade functionality.
On you interest incoding/calculating probabilities, I look forward to exploring that with you, possibly in a new forum focusing particularly on the the numerous issues involved.
(I have been thinking towards the underlying formalism needed.)
Encourage you to read Michael Behe's Darwin's Black Box (if you have not already) where he addresses some of the improbabilities involved.
1) One very major factor that is conveniently overlooked is the long reproduction cycle for beings with complex body parts. e.g., typically 20 years for humans.
2) Redundancy with the two DNA strands and 3) Degeneracy in the DNA Codons (64 combinations for 20 proteins) substantially reduce the probability of mutations being passed on.
4) Then there is the major challenge of finding folding protein sequences to that map to folded protein space.
There is a major challenge in modeling these such as finding self folding protein strings vs those needing chaperone molecules. I think modeling protein folding is currently very much in the grand challenge category capable of utilizing all computing resources etc.
1)-3) may be reasonably modeled to arrive at upper bounds. I don't know that 4)is currently quantifiable within any reasonable computational time frame.
Then factor in Behe's Irreducible Complexity where many different functioning proteins must be brought together in a precise sequence before any functionality is obtained that can be "selected" for.
See Bauer's paper on specificity where he explores some of the issues involved.
http://www.iscid.org/boards/ubb-get_topic-f-6-t-000561
So in sum, I support you thoughts, suggesting focusing on the key differences between mutations .html"increasing information" (sic) in evolution, vs degrading functionality in ID, coupled with the very long time steps involved with humans and other species with complex body systems.
These are key to some of the issues that need to be modeled and demonstrated. While it may be possible to model upper bounds, serious modeling and calculating of actual probabilities of increasing (sic)/decreasing functionality by mutations I think is a very long way off.
Testing and finding some practical search methods that can practically explore some of these mesa type objective functions might be a useful exercise in "coding" and exploring. That is probably worth a major thesis and grant application in itself.
[ 18. December 2005, 23:27: Message edited by: David L. Hagen ]
IP: Logged
|
|
Atom
Member
Member # 1840
|
posted 19. December 2005 01:04
Hey David,
Thank you for your input. I have read DBB, but a few years ago. I'm sorry for my use of information in my post (you kept putting the sic on it), but I think we can agree that artificial selection acting on random mutation can produce CSI sequences. (The "Methinks it a weasal" type of forward selection, where we see if a string is "closer" to a functional target we want, then move in that direction.) This is in contrast with NS, which does not forward select. It only retains actual current fitness, not future fitness.
These questions all play out in fitness landscapes, which we can model with continous functions. As you also pointed out David, GAs are slow for finding optima, and usually a type of heuristic is used in searches (once again, intelligent artificial selection). Dembski's work on the NFL regress problem was brilliant, in my humble opinion, for capturing this distinction.
Anyway, yeah, I am right there with you. I see the work of calculating the actual probabilities as daunting, and probably a little ways off. But I had some insight as I drove home tonight. This may provide a way of simulating similar scenarios to the ones we're discussing, so let me run my brainstorm by you guys:
I has the insight that fitness and function may not always be synonymous. Function refers to a state machine which takes a given input, and produces a given output. An enzyme is a state machine which takes given input (chemical compound(s)), performs state transitions according to its state table (physical and chemical laws), and produces an output (catalyzed compound(s)). Now, this is a function performed. On the other hand, let's say you have a patch of blue skin on your forehead that women find irresistable. It attracts them to you, and so increases your reproductive fitness. Yet it was static. It didn't change state, it just was there. So it increased your fitness, but didn't necessarily "function" in the usual sense. So fitness may or may not presuppose function. Or maybe it does, and the function can be said to be getting you some action. I haven't formally developed this enough to be sure.
Anyway, moving on, I found a mechanism (I think) where we can simultaneously test fitness and function independently. In such a simulation, we would need to go through functional stages to reach a targetted fitness level, but this targetted fitness level is not merely a "sum" of these functional intermediates. The target is itself functional, but does not have to have the same function as the beginning stages.
My thought is to begin with a few lines of computer code (functional), and mutate it. Our fitness target will be defined as "A function which takes in a variable, and returns it to the 11 power". Now, we can see that the question of global function (grammatical code) has been divorced from final function (to take a number and return n^11). It has also been divorced from fitness (we have a number of ways which we can get the results such as n^11, (n*n*n)*n^8, (n/n)*1*(n^(2+9)), and there exists levels of fitness there.) There are some formulas which are non-functional (non-grammatical), and some that do not perform the function we want. Some maybe do it "closer" than others (n^10 vs. n^3), so we can define fitness in relation to proximity to the final product (n^11). We can map that across a given number of input values, and those whose output forms a curve mapping closest to our n^11 curve will be considered more fit.
This would also imply a mesa-island type of functional space, since most mutated sequences will result in non-grammatical computer code. We can incorporate point mutations, frame shifts, duplication, and any other proposed mechanisms.
Ok, I'm done spilling my thoughts. I'm sure this has some kinks, so I'd appreciate you guys helping me work them out. In the mean time, I'm off to sleep.
Atom
IP: Logged
|
|
Atom
Member
Member # 1840
|
posted 19. December 2005 10:57
Ok, I thought about this some more last night. It seems that the blue patch does indeed have a function: it takes as input white light, absorbs certain frequencies, and outputs blue light.
Thinking more generally, I think I now see the pattern of the relationship between fitness and function. Fitness presupposes function, but function does not imply fitness. In other words, function is a necessary but not sufficient condition for fitness. (If this seems painfully obvious, I apologize, but I'm trying to work through these issues and be precise.)
The proof of this relationship can be seen as follows:
Imagine we have a blackbox, which represents a biological structure. This BB is assumed to increase fitness of an organism over its competitors. Now we also claim that this BB has no function. By my definition, the input of this BB will match the output. So, in layman's terms, it does nothing, except somewhow increases fitness. We can find another BB, called NBB, which matches this input-output mapping exactly. What is NBB? Nothing. It is a non-existent black box, a patch of air if you will. It will merely pass along the input unmodified. So if BB can be replaced by NBB, and the input-output mapping remains unchanged, then the inner workings of NBB can replace the inner workings of BB. So whether or not BB is in fact present (it may be nothing, replaced by air, the NBB), the presumed increase of the fitness is still there. Therefore, BB cannot be said to increase the fitness, since its presence or non-presence will not change the level of fitness present. Its non-presence will not reduce the level of fitness present, and you can start without it, attain level of fitness F1, then add it, and you still have level of fitness F1. There is no gain by adding BB, so it does not increase fitness, which contradicts our original premise.
Now there is a further consideration. Since the relationship between fitness and function is asymmetrical, function does not imply fitness. We can have a functional unit that does not do anything "useful" for an organism's survival or reproduction. For example, let's say you have an enzyme in your body which catalyzes a reaction that turns the fat in your abdomen a reddish tint. This reaction doesn't change the properties of fat, other than color, and it retains the same basic functions. So this enzyme can be said to produce a useless effect, changing the color of something you will not see, nor will anyone else probably. Since it leaves fat as fat, it does not harm the organism, or benefit its survival chances. It has function, but does not change fitness.
Ok, these thoughts should help tighten up the distinction and relationship between function and fitness. Point me in the right direction if I'm just rehashing topics that have been thoroughly dealt with before...
Atom
IP: Logged
|
|
David L. Hagen
Member
Member # 323
|
posted 19. December 2005 13:47
"Fitness" is difficult to predict or pin down. Among humans, I expect that it will be dominated by religious and social parameters rather than be productive capability.
For "function", I prefer to use the term "Design Function" where we can define an explicit function for the component separate from any evolutionary assumptions. e.g., the design function for each component of a flagella.
Also distinguish the "fitness function" or objective function used in optimization or evolution. See: Fitness Function
http://www.iscid.org/encyclopedia/Fitness_Function
Any statement that mutations increase "information" begs the quesion of whether mutation with natural selection is capable of such. e.g., If you examine the origin of reproducing cells with all their inbuilt "information" in the DNA and cell structure, there was nothing to "select" before it began reproducing. Rather this appears to be an example of Design based CSI.
IP: Logged
|
|
Atom
Member
Member # 1840
|
posted 19. December 2005 14:52
Valued points David, and I'll attempt to incorporate the appropriate vocab and concepts into my further development of these ideas. Thanks for the link...I didn't know about the ISCID encyclopedia (I may have seen it, but didn't think to pay it attention), but I will definitely check that out.
Atom
IP: Logged
|
|
David L. Hagen
Member
Member # 323
|
posted 19. December 2005 17:49
See:
Hubert P. Yockey, "Information Theory, Evolution, and the Origin of Life", 2005 Cambridge as advertised by ISCID. (I just obtained a copy). Yockey has alot of good material on information theory.
However, from a cursory view, the logic underlying his statements on evolution, ID and irreducible complexity need to be very carefully examined.
See also ISCID discussion at:
Wade: Growth of Complexity at
http://www.iscid.org/boards/ubb-get_topic-f-6-t-000573#000002
IP: Logged
|
|
Atom
Member
Member # 1840
|
posted 19. December 2005 18:21
Thanks. I was looking at that Yockey book, and it looked very interesting. I may just purchase it.
Thanks for all the info.
Atom
IP: Logged
|
|
|
Printer-friendly view of this topic