|
Author
|
Topic: Evolving Inventions
|
Frances
Member
Member # 169
|
posted 03. February 2003 00:06
Dear John,
I would rather that you spent your time focusing on dealing with my arguments and claims and refrain from suggesting incorrectly I may add that I have not read your paper or that I am not familiar with the discussion on ARN.
Nevertheless I will ignore your ad hominem approach and continue focusing on the real issue namely your suggestion that evolutionary algorithms are stuck to a particular hypervolume. In fact as I and others have shown, this is incorrect although some of the simpler GA algorithms may indeed be so constrained but evolution/evolutionary algorithms are not constrained and indeed show plenty of evidence of exploring different hypervolumes through such processes as gene duplication, cross overs etc. So while evolutionary algorithms in their most simplistic forms can be great ways to solve 'routine' problems, they are in no way restrained in such manner.
That you may be disappointed by my insistance that your claims are erroneous despite your efforts to address them may point to the weakness of your efforts and not a weakness in my understanding of your claims. On the contrary, I believe that I understand your claims but I am not impressed by them. But that by itself is hardly sufficient to invalidate them hence me pointing out that evolutionary algorithms can explore higher hypervolumes, in spite of your insistance otherwise, seems highly relevant.
So lets not move the goal posts until you are willing to revise your claims in the papers about evolutionary algorithms and hypervolumes. After we have reached some consensus of this we may explore if evolutionary algorithms can be considered inventive according to the criteria of TRIZ.
As far as gene duplication and other mechanisms of increasing the hypervolume are concerned, I have yet to see any relevant rebuttals of this simple observation.
I hope that you will find the time to discuss these issues in a relevant manner and will refrain in further communications from a response which in many ways is both ad hominem and insulting.
So lets see if we can get the discussion moving forward again in a scientific manner. Others have shown as well how your claim about evolutionary algorithms and hypervolume seem to be incorrect. Lets explore this first and then determine the logical next steps.
From your paper
quote:
This observation suggests that we may consider any genetic algorithm to be operating within a certain n-dimensional hypervolume, and certain fixed parameters completely determine that hypervolume ahead of time. Furthermore, any particular n-dimensional hypervolume is completely isolated and separate from any other m-dimensional hypervolume (m ≠ n).
Would you agree with me (and several others) that while this may apply to a subset of GA's it is not correct for the larger set of GA's, especially those relevant to biological evolution?
Let me repeat quoting from a paper which I referenced before
quote:
In designing a state space of possible designs is implied by the representation used and the
computational processes that operate on that representation. GAs are a means of effectively
searching that state space which is defined by the length of the genotype’s bit string. Of
particular interest in design computing are processes that enlarge that state space to change
the set of possible designs. This paper presents one such process based on the generalization
of the genetic crossover operation.
Adaptive Enlargement of State Spaces in Evolutionary Designing by JOHN S. GERO AND VLADIMIR KAZAKOV
[ 03. February 2003, 00:09: Message edited by: Frances ]
IP: Logged
|
|
John Bracht
Member
Member # 5
|
posted 03. February 2003 01:22
Frances,
I did not say that you haven't read my paper--just that you seem not to have understood it. Big difference. Since I've addressed your arguments, and you have not addressed the points I made, I don't see why I should re-state my original arguments. If you look at the references I provided (especially my earlier discussions of my paper) you'll see why I argue that gene duplications don't buy you the ability to add dimensions to the hypervolume. Here's a hint: think about angel wings.
As for inventiveness and computer programs: I addressed this too. Basically, the definition of inventiveness in TRIZ theory is highly specialized and applies only to situations in which a technical contradiction is overcome. You keep claiming that evolutionary algorithms can "invent" by the TRIZ definition, but you have nowhere shown how a technical contradiction was overcome, or how the hypervolume of possibilities was re-engineered by any one of your examples. I am only asking you to demonstrate what you claim--not, as you say, "shifting the goalposts". Sorry, but if you want to sustain your argument, this is what you have to show.
I have more to say--much more--but I'm way, way too busy now and for the next week or so to do more than dream of posting. My P.I. is on my back to get some important results in the lab, and I'm helping with the recruitment comittee this coming week. I'll try to address some of the issues that have been raised when I get a chance. Frances, please do go back and do me the honor of trying to understand what I'm saying, and of responding to my latest response to your argument. Re-stating your initial argument, no matter how forcefully, simply doesn't cut it--you've got to respond to my response. If it's not clear to you exactly what my counter-argument for a given point is, please say so and I'll try to clarify. Honestly, I'm confused about why you seem to think I haven't addressed your points and I want to get to the point where we make progress.
When I get time.
John
P.S. I'd be careful, Frances, about making the claim that I'm arguing "ad-hominem". I made no statements trying to discredit you as a person, I only made statements about how I see the state of our exchanges--that you seem intent on repeating the same issues without addressing my responses to those very issues. Whether or not you agree with my assessment, this is a statement about the argument, not about you as a person (as would be the case in a true ad-hominem argument). Obviously, you can disagree with my assessment (and you clearly do), but please do be careful not to throw around logical fallacies that do not apply.
[ 03. February 2003, 02:44: Message edited by: John Bracht ]
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 03. February 2003 15:49
Charlie,
The PCP degradation pathway cannot be compared to a vast network of systems, like the flagellum for example. The fact that the information to make a wing can be protected by secondary functions does not attest to it's origin, as even the reports admitted. We're interested in how the functions got there in the first place. That some pathways can buffer mutation due to secondary functions can speak to both sides of the fence in this area.
What would convince me is if critics stopped showing us examples of quirky 3 component pathways started showing us sophisticated mutli-component systems arising through adoption of different function quite easily. And showing that it is a likely, and common event.
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 03. February 2003 16:27
Nelson_Alonso wrote:
quote:
I have not read the paper on the cold sensitive pod mutants, but I think that sensitivity to low tempertures is not as impressive as reducing the function of a gene and still having a viable organism. After all, we do have lethality with the cold-sensitive pod. I find it very difficult to believe that reduction of function of a polarity gene would still yield a viable organism.
But cold-sensitivity arises from reducing the function of the gene. The organism can "get by" with only a fraction of the full normal activity (which is wholly consistent with the broad robustness observed in biological systems).
It doesn't do so well if it is missing the gene.
John said:
quote:
Basically, the definition of inventiveness in TRIZ theory is highly specialized and applies only to situations in which a technical contradiction is overcome. You [Frances] keep claiming that evolutionary algorithms can "invent" by the TRIZ definition, but you have nowhere shown how a technical contradiction was overcome, or how the hypervolume of possibilities was re-engineered by any one of your examples.
Maybe Frances has not, but others of us have at least cast doubt on your claim that developmental processes are immutable. We have stick-insect wings, variable numbers of hox genes, polarity mutants, and so on.
You also have not really responded to my earlier point that
quote:
It is critical to realize that the topology of performance metrics on conceptual space is not like the topology of fitness metrics on genome space! Thus if (cognitive) "trial and error" typically fails to generate "inventive" solutions to human tasks, it does not follow that random search will fail to generate "inventive" solutions to fitness problems.
except by reiterating that inventive solutions are needed. Isn't biology replete with examples of solutions that you would expect would need an "inventive" solution, yet are accomplished with only a couple of genes?
Again: what evidence do you have that the topology of conceptual space is like the topology of fitness space? Or, alternatively, what evidence do you have that anything in biology was ever "inventive" in the way that you mean "inventive"?
For example, in your icebreaker example in the TRIZ paper, the solution would suddenly not be inventive if there were simple instructions for "put a hole through the ship here". Yet in biological systems, instructions very similar to that ("put wings here") are carried out in a remarkably complete way.
Thus, what might seem to be "inventive" to a human designer may be "routine" in terms of genetic changes required. It is the mapping of conceptual space to result, vs. genome space to result, that is at issue here.
quote:
If you look at the references I provided (especially my earlier discussions of my paper) you'll see why I argue that gene duplications don't buy you the ability to add dimensions to the hypervolume.
Unless I am mistaken you are referring in part to your ARN post which says, in part,
quote:
I consider the hypervolume to be determined by the functions of the genes operating upon embryological development. The duplication of genes is of no consequence unless those new genes are somehow modified and integrated into the existing structure and process of development such that novelty is produced.
Doesn't Hox gene homology (with several examples of apparent duplication followed by functional divergence), together with the robustness of development to modest mutations, at least strongly suggest that these genes can in fact be modified and integrated?
I think I understand your point, but it seems to me to boil down to an unwillingness to consider that the hypervolume might change, despite evidence to the contrary.
For example, the paper Ronshaugen et al., "Hox protein mutation and macroevolution of the insect body plan." Nature. 2002 Feb 21;415(6874):914-7 says:
quote:
In sum, full-length Artemia Ubx provides an 'abdominalizing' function in the Drosophila embryonic epidermis, but has little repressive effect on thoracic limb development in Drosophila embryos. Further, the limb-suppressing difference between Drosophila and Artemia Ubx is at least partly mediated by their different abilities to transcriptionally repress the Dll gene.
Basically, a shrimp Hox protein has different leg-suppressing capability than a Drosophila protein. Now suppose you duplicate the Drosophila protein--you then have extended your hypervolume to include both the shrimp and Drosophila phenotypes, which increases your available body plans if you differentially express each.
If you want a larger theoretical change in hypervolume than that--in one gene duplication!--maybe you need to explain in more detail how radical of a change you're looking for.
If you mean: once you have a multicellular metazoan with body plan development dictated by Hox genes, and high levels of specialization for each segment of the body plan, you cannot throw out the Hox system and replace it with a novel system, then you may be exactly right.
But then again, as far as we know, all bilaterians have Hox genes. (Possibly all eumetazoa as well--I'm not sure.) So if you say it can't happen, well, it hasn't.
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 04. February 2003 00:04
Dear John,
I notice once again that you have not addressed my simple observation that contrary to your claim that evolutionary algorithms cannot increase the hypervolume, gene duplication, crossovers and other approaches have shown this to be incorrect.
So lets first establish that your claim about genetic algorithms was incorrect.
Now we can address your moving goalpost claim that gene duplication is irrelevant in biological evolution as far as increasing the hypervolume is concerned but as I and others have shown that seems to be incorrect as well.
Btw if you take the time to read your original response then you would realize that indeed you suggested that I may not have read the paper
quote:
Your reply shows again that you haven't really taken the time to read or understand my paper or the ARN discussion of that paper.
This claim is really ad hominem in nature but I understand that you are working under the pressures of time and will thus let it pass.
Lets at least agree that your sweeping claim about evolutionary algorithms
quote:
This observation suggests that we may consider any genetic algorithm to be operating within a certain n-dimensional hypervolume, and certain fixed parameters completely determine that hypervolume ahead of time. Furthermore, any particular n-dimensional hypervolume is completely isolated and separate from any other m-dimensional hypervolume (m ≠ n).
is incorrect?
Perhaps it may be time to address the links I provided you with in which it is shown that genetic algorithms can and do increase the dimensions of the hypervolume?
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 04. February 2003 02:12
Interesting snip from John Bracht’s TRIZ paper:
quote:
Notice how the sifting of variants (trial and error) could not move to a new hypervolume because it was restricted to varying the characteristics of the existing system. Adding a new piece to the system re-engineered the hypervolume in such a way that the solution was accessible. This re-engineering of the hypervolume is the essence of an inventive solution or invention. And it is precisely this re-engineering of the hypervolume that is out of reach of the Darwinian process because it requires the ability to step outside the hypervolume and make external changes. Endlessly sifting variations of some basic structure will never create a new structure because it moves from point to point within the hypervolume itself—it can never break out of the hypervolume to find something new. This is another way to re-state the problem of morphological inertia: trial and error is locked into variations of existing structures. Trial and error is shortsighted in the sense that it can only “see” to the edges of the hypervolume—and it can go no farther.
I find it interesting that in John’s response, he feels that he has answered the objection that Francis raises in some location. I cannot find it in the TRIZ paper, or in the ARN discussion.
What I do find is that I have personally asked a number of questions of John that relate to precisely this sort of expansion of the search hypervolume (and associated issues). And nowhere, in either the ARN or this discussion thread, have any of the questions been answered.
For example I point out that the mind itself could easily be using a kind of “evolutionary” process in finding new solutions. I think that the world of inventions is characterized by the addition of a new concept drawn from one place into concepts that are drawn from other locations.
In John’s TRIZ paper, he opens with an example of the design of an ice breaker ship. But the description is in terms of a “hollow tube” and of an action that is very suggestive of how a knife cuts through objects.
Consider these as mental bits of information floating around. Is there some reason to believe that the mind would not ever focus on the knife action, when trying to design something to cut through ice? I think that it is a possibility. Then the hollow tube concept, that is one that is very familiar to anyone who has ever built a ship or a lamp or just about any mechanical system.
There are two tricks -- thinking of them at the same time, and the “flash” of realizing that they fit together into a solution. Now the mind, when it has such a flash, is recognizing that the concepts have fit a form of consistency -- and so the mind is capable of doing a sort of simulation with the concepts and by this simulation finds that the combination is consistent with the goal it is working on.
In evolution (real evolution, not simulations or GA algorithms for some other purpose) there is no particular goal except for the greater rate of survival will produce a greater degree of reproduction. (I’m not now talking about any ultimate goal of creator of the universe or the like now -- the issue of regress of causality and displacement issues to some ultimate cause is not at issue.) But that greater survival rate is precisely equivalent to the effect of the consistency check done by the mind’s simulation of the problem -- the only difference is that in physical evolution the trial “concept” is actually constructed and is not simply a mental model.
My main point is that the combination of aspects is similar to what one might get from a mutation or activation of a duplicated piece of genetic material. A change in a structure occurs because of the new dimensionality that came from the added genetic material is then varied. In essence the duplicated genetic material gives a new dimension for variation to proceed with, and appears to reasonably represent an increase in the so-called “search” hypervolume.
I have asked questions of John in this thread on ISCID that bear on precisely on these issues. And on the original ARN thread that John gave us a link, I also asked related questions. Neither here, nor in the ARN thread, did I ever receive a reply on these issues. At least for Frances there was a reply, though not answering the issue specifically. I agree with Frances that nowhere has John actually addressed these issues of possible ways for a genetic algorithm or some similar process could increase the hypervolume.
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 07. February 2003 16:36
Nelson:
After all, we do have lethality with the cold-sensitive pod. I find it very difficult to believe that reduction of function of a polarity gene would still yield a viable organism.
Rex:
But cold-sensitivity arises from reducing the function of the gene. The organism can "get by" with only a fraction of the full normal activity (which is wholly consistent with the broad robustness observed in biological systems).
It doesn't do so well if it is missing the gene.
Nelson:
This is not what the tempereture shifts show. What the temperture shifts indicate is at what time in development the gene functions. Pod-1 and Pod-2 are key genes and are essential for function. Which is why the very authors of the paper stated:
quote:
"If the asymmetrical division is screwed up, the embryo is never going to make it," says Akiko Tagawa, a graduate student and the lead author of the paper, which also includes Chad A. Rappleye, another graduate student working in Aroian's laboratory. "It quickly dies."
In addition, the cells containing these mutant genes are vulnerable to bursting and shriveling. This is due to the inability of the defective embryo's cell membranes to maintain the cell's osmotic pressure, which can produce an undesirable flood of fluids into or out of the cells.
http://www.cs.columbia.edu/~bgrundy/mcpod.htm
I really do think that you are remembering the pod-1 example wrong, and I think that you misinterpreted the temperature shift experiments involving pod-2.
Rex:
Maybe Frances has not, but others of us have at least cast doubt on your claim that developmental processes are immutable. We have stick-insect wings,
Nelson:
Stick-insect wings have nothing to do with the inventiveness of evolution. They don't even have anything to do with their origin, as the reports themselves admitted.
quote:
The authors note that the new study does not speak to the biggest question about such complex features - their origin. That is, no cycles of losing and regaining of wings can explain how insects first evolved their wings.
Rex:
variable numbers of hox genes
Nelson:
If you are talking about growth and development of sponges, very little is known about it. I wouldn't stake your hox bucks on sponges.
, polarity mutants
Nelson:
As I described above, the data shows that polarity mutants result in a messed up development. I don't think this particular example casts doubt on Bracht's statement.
Rex:
It is critical to realize that the topology of performance metrics on conceptual space is not like the topology of fitness metrics on genome space! Thus if (cognitive) "trial and error" typically fails to generate "inventive" solutions to human tasks, it does not follow that random search will fail to generate "inventive" solutions to fitness problems.
Nelson:
I don't speak for John but I think that this makes it worse for evolution not better. The topology of the genome is not as "free living" as the topology of conceptual space. I am only half-way through the paper (not even half-way) but I think that John is claiming that most solutions that we see evolutionary mechanisms capable of are technical contradictions, things that would improve the system but not necessarily generate something new and unique. For example, with the ice breaker ship, an immediate solution would be to add on high powered engines, but that is not very cost efficient.
As a side note, I think that if TRIZ can be put into the language of mathematics and somehow plugged in to the Pdco calculation we would be getting closer to how truly improbable bacterial flagella are.
[ 07. February 2003, 16:51: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 07. February 2003 17:43
Nelson:
quote:
As a side note, I think that if TRIZ can be put into the language of mathematics and somehow plugged in to the Pdco calculation we would be getting closer to how truly improbable bacterial flagella are.
Perhaps a bit early to jump to this conclusion since there is so far no probability calculations that would support such a rush of judgement.
In fact more and more we find out about the flagella, the less and less likely a design inference seems to become.
Niicholas has provides us with some links and resources
Also see Musgrave's Evolution of the Bacterial Flagella
A good resource "Faguy, DM and KF Jarrell. 1999. A twisted tale:the origin and evolution
of motility and chemotaxis in prokaryotes. Microbiology 145: 279-281"
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 07. February 2003 17:47
Ged:
My main point is that the combination of aspects is similar to what one might get from a mutation or activation of a duplicated piece of genetic material. A change in a structure occurs because of the new dimensionality that came from the added genetic material is then varied. In essence the duplicated genetic material gives a new dimension for variation to proceed with, and appears to reasonably represent an increase in the so-called “search” hypervolume.
Nelson:
I'm curious as to how this relates to your question:
quote:
"We are not aware , however, of any convincing evidence that the majority of duplicate copies have acquired new functions that did not already exist in the ancestral genes."
Genetics 1999 Apr;151(4):1531-45 Preservation of duplicate genes by complementary, degenerative mutations
If what is occuring in gene duplications points to a more complex ancestor where the function was already present, does duplication really point to Darwinian inventiveness?
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 07. February 2003 17:52
I was planning on writing up a new topic but I guess there's always time for that tomorrow. I'll use up my third post to reply francis just to set the record straight about Flagella.
I don't want to take this thread off topic so this is my last post on this topic in this thread unless Francis wants to start up a flagella thread discussing this series of articles:
http://www.idthink.net/biot/flag1/index.html
Francis cite:
A good resource "Faguy, DM and KF Jarrell. 1999. A twisted tale:the origin and evolution
of motility and chemotaxis in prokaryotes. Microbiology 145: 279-281"
It's been a while since I read it but I think this is talking about archael flagella which is irrelevant.
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 07. February 2003 20:54
Alonso Nelson said:
quote:
If what is occuring [SIC] in gene duplications points to a more complex ancestor where the function was already present, does duplication really point to Darwinian inventiveness?
Alonso, first your quote from Genetics is about a “majority” of duplicated genes. It says nothing about the capabilities that can occur in lower probability. Also that the presence of the new function “already exist in the ancestral genes” is not a pointer to a “simpler ancestor”[EDIT -- I meant to say “more complex ancestor” or simpler descendent], rather it is a pointer to an ancestor of the same complexity. So what about the minority? What happens there?
My experience is not in biology, so I will comment only a bit from the genetic standpoint. (Otherwise I shall leave this to the experts.)
I am drawing a comparison to an earlier post in this thread where I asked John Bracht questions (that he has not responded to).
In my previous post, I raised the issue in another domain, of building logic for a field-programmable logic device. In that case, the algorithm could duplicate a section of logic. Then that duplicated section would of course perform the same function as the original copy. But the duplicated, as well as the original portions of the logic are both subject to mutation or other change operations.
Of course a majority of mutations of the logic connectivity would not provide a better functionality in the FPGA. (And they could be selected out if they perform in a less adapted manner.) But my point is that the duplicated section of logic is a new dimension from which variation can occur. It provides another dimension of variability. And if each section of logic varies independently, each “dimension” can vary with different types of functional responses in the FPGA case. Are there analogs in the genetic case?
Johh Bracht’s TRIZ paper suggests that there cannot be any new dimensionality in solution types. That only a majority of cases are not new is not a statement of capability or limitation.
Also remember that genome is for the most part “digital”. If the new genetic code could develop by a sequence, it pretty much would produce the new organism with the new functionality. So the issue is whether the new genetic code could develop by the evolutionary processes of mutation and other forms like duplication, it not a question of whether a modification of the genome could produce the new functionality.
The TRIZ paper suggests that there is no mechanism for new dimensionality. I am suggesting that duplication is one such mechanism for new dimensionality -- but must be followed by additional modification within that dimensionality. (Addition of a new dimension does not in itself confer movement in that dimension.)
Bracht’s TRIZ paper describes the development of the ice breaker by combination of shaping of various parts of the hull. A part alone could become narrow and cutting, but it would not carry the load of the ship. It could alone be shaped to carry the load, but would not become cutting. But now consider a duplication into two sections of hull. One could vary toward thin and cutting, the other could vary toward shape to carry the load below the ice. This is a new capacity for dimensional change that did not occur before the hull was divided into two sections. This is somewhat analogous to the duplication case in the FPGA, and possibly in the genetic case as well.
It shows how a new dimensionality can occur, without the new dimension initially conferring any new capability.
The TRIZ paper already admits that once the new dimensionality is present, that the processes can “sift” through them to find a more fit organism.
[ 08. February 2003, 19:29: Message edited by: gedanken ]
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 08. February 2003 06:21
Nelson: regarding pod-1, I am both right and wrong. There is a temperature sensitive allele that is only partly lethal at 20C, but the polarity defects don't appear until higher temperatures.
So it wasn't as good of an example as I'd hoped. (pod-1 has all kinds of non-polarity defects too, making it a particularly unfortuante choice. Pretty obvious that I don't work in development!)
However, the pod-2 data is unambiguous. First, even at the cold-shifted temperature, there are still a few survivors--about 2-4% (see Table 2 in the reference I provided). Second, the cold-shifting was done for a different reason--hence the different emphasis given in the quote you gave--but there are clearly intermediate lethal phenotypes when shifting out of cold -6, -4, and -2 hours relative to the first cell division. If you interpret this as less time to make functional protein, then the protein levels are less, showing robustness. If you interpret this as properly-folded protein having less time to act, it shows robustness. If you interpret this as a variably-timed critical period, it also shows insensitivity to major changes in timing (albeit during oogenesis, not later development, in this case).
With respect to Hox genes, it ranges from two in sponges to 20+ in mice, with many intermediates, so there is plenty of evidence that the number is variable.
However, to be fair, it's not yet clear how perturbable the system is. In Galis & Metz, J Exp Zool 2001 Aug 15;291(2):195-204, injections of teratogens (powerful morphogen) caused peak lethality in mice, rats, and hamsters during the phylotypic stage (approximately the stage drawn in Haekel's (in)famous embryos), suggesting significantly greater sensitivity during that stage than other stages. This is exactly the stage during which the Hox genes first activate. (It is also the stage during which tissues are sensitive to teratogens, so in a way, the chemical they used specifically selected for the effect that they wanted.) They got some fairly messed up animals to live, but there were also certain fairly minor changes (e.g. a cleft lip) that were surprisingly strongly correlated with death.
Anyway, I'm certainly not going to try to argue that development isn't a rather sensitive process. However, just how sensitive--both now and historically--is of great importance. Given how well fruit flies survive with spurious shrimp Hox genes in them, you have to speculate that the sensitivity is not so great as to prevent gene duplication followed by modification of function.
Where that will get you is another issue, but it pushes the question back another level: you have your new hypergeometry, and a new set of bodyplans that potentially come available, if any of them have non-lethal paths to them. Ruling out new hypergeometry or any early developmental changes is too simplistic.
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 08. February 2003 17:40
Hi Rex and Gedankin.
Going to forego replying to Rex today as he brings up some good points I want to get deeper into, but won't have time to today. Thanks for your reply regardless.
Ged:
Alonso, first your quote from Genetics is about a “majority” of duplicated genes. It says nothing about the capabilities that can occur in lower probability.
Nelson:
I think Bracht has addressed this at the ARN thread which parallels the point I want to make:
quote:
The duplication of genes is of no consequence unless those new genes are somehow modified and integrated into the existing structure and process of development such that novelty is produced. This is what Richard Dawkins notes when he comments that angel wings will never sprout from a human's back.
What I consider the minority pointed to by the paper is most likely the antifreeze protein example. But this indeed was addressed by John.
quote:
The antifreeze example is even weaker. The lowering of the freezing point of a solute only requires that there be some solvent dissolved in it, so the only requirement for an antifreeze protein is that it be water-soluble. There is not even a need to alter the protein itself, it just has to get accidentally mis-directed into the bloodstream. This would definitely not qualify as an invention in TRIZ; I'm not even sure it would get a patent in the USA. Certainly, no contradiction was overcome and no invention was made.
Ged:
Also that the presence of the new function “already exist in the ancestral genes” is not a pointer to a “simpler ancestor”, rather it is a pointer to an ancestor of the same complexity.
Nelson:
I never claimed that this points to a "simpler ancestor", I quite clearly stated:
quote:
If what is occurring in gene duplications points to a more complex ancestor where the function was already present
I think that what Rex and the rest of the critics on this thread are proposing is what this paper calls the classical model. Where gene duplications result in the novel inventions mentioned in the TRIZ paper, by mutations and aquisition of new functions. However, the DDC model, in contrast, states:
quote:
However, empirical evidence from genome duplication events suggests that gene duplicates are preserved in genomes far more commonly and for periods far in excess of the expectations under [the classical model], and whereas some gene duplicates clearly evolve new functions,
there is little evidence that this is the most common mechanism of duplicate-gene preservation. An alternative hypothesis is that gene duplicates are frequently preserved by subfunctionalization, whereby both members of a pair experience degenerative mutations that reduce their joint levels and patterns of activity to that of the single ancestral gene.
That is, if we intepret gene duplication as the partitioning of function that was already present in the more complex ancestor, how does this apply to novel inventions in TRIZ? In order for gene duplication to be at all relevant to TRIZ, we would need to resort back to the classical model, which as we have seen, do not give examples of jumping over technical contradictions.
Ged:
I am drawing a comparison to an earlier post in this thread where I asked John Bracht questions (that he has not responded to).
Nelson:
As a side note, John has already clearly stated that he is quite busy, there is no need for you to keep repeating the fact that he hasn't gotten to you yet.
Ged:
In my previous post, I raised the issue in another domain, of building logic for a field-programmable logic device. In that case, the algorithm could duplicate a section of logic. Then that duplicated section would of course perform the same function as the original copy. But the duplicated, as well as the original portions of the logic are both subject to mutation or other change operations.
Nelson:
No, it would not retain the original function, just part of it. If we stick to the DDC model, that section of logic that would be duplicated contains instructions that are, say binary. 1 for true, 0 for false. Since a duplication event partitions these functions, we have one piece which encodes for true(1), the other encoding false(0). Since this is most likely the common mechanism of gene duplication, what does this say about inventiveness according to TRIZ? This further supports the claim that trial-and-error processes are restrained within the hypervolume but never jump to a new "higher-order" hypervolume. This is not of course, to say that a rare event in some time in the past has never happened that comes close, as rare unlikely events do happen.
And this also applies to your main point:
quote:
Ged:
So the issue is whether the new genetic code could develop by the evolutionary processes of mutation and other forms like duplication, it not [SIC] a question of whether a modification of the genome could produce the new functionality.
leading to my main point, and the point of the paper (as I see it at the moment) that trial and error processes do not lead to higher order dimensionality, or the type of inventiveness discussed by the TRIZ paper.
P.S. Sorry for the "right back at ya" way I included a badly formed sentence in your paragraph. But hopefully this will teach you that on an internet board we are going to make typos and mistakes with grammar, there is no need to obnoxiously point it out, unless you are having trouble understanding the text.
[ 08. February 2003, 18:36: Message edited by: Nelson_Alonso ]
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 08. February 2003 19:53
Let me state my case once again:
Gene duplication amongst others is a way for genetic algorithms to increase their hypervolume. Bracht's paper suggested, incorrectly imho, that GA's were limited in the hypervolume they could explore.
Now the goal posts seem to have shifted to: Gene duplication is not enough in _evolution_ but that was not the point of objection.
Perhaps it would be helpful to see if we can agree that GA's can indeed increase their hypervolume.
We can then explore the importance of this observation to the creativity of GA's. But lets first see if we all accept the fact that GA's can increase their hypervolume.
Nelson: You suggest that the paper's focus is that trial and error processes do not increase hypervolume, as I have shown this seems to be incorrect in general. Actually even point mutations themselves can increase dimensionality.
[ 08. February 2003, 20:25: Message edited by: Frances ]
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 08. February 2003 20:33
Sorry, Alonso, I didn’t read what I wrote carefully. (See edit in last post, but left the mistake visible).
I meant to say:
Your quote from Genetics is about a “majority” of duplicated genes. It says nothing about the capabilities that can occur in lower probability. Also that the presence of the new function “already exist in the ancestral genes” is not a pointer to a “more complex ancestor” (or simpler descendent), rather it is a pointer to an ancestor of the same complexity. So what about the minority? What happens there?
You haven’t answered that question, except embedded in your quote from “DDC model” (I assume the Genetics article again), with my emphasis added:
quote:
However, empirical evidence from genome duplication events suggests that gene duplicates are preserved in genomes far more commonly and for periods far in excess of the expectations under [the classical model], and whereas some gene duplicates clearly evolve new functions, there is little evidence that this is the most common mechanism of duplicate-gene preservation. An alternative hypothesis is that gene duplicates are frequently preserved by subfunctionalization, whereby both members of a pair experience degenerative mutations that reduce their joint levels and patterns of activity to that of the single ancestral gene.
Alonso, you continue to focus on the aspect that provides no new function while ignoring the statements in your own quotes that do provide new functions. Once again, that a majority or the higher frequency or probability of duplication events not producing a new dimension of variability says nothing about the possibility, the capacity, or the capability of those duplicated genes. Once gain, if the low to mid probability event of a new function occurring (“whereas some gene duplicates clearly evolve new functions”) still occurs regularly, then a new dimensionality of change has been enabled by the gene duplication. In other words, the hypervolume has increased. (And if it did not, then the hypervolume was semi-infinite in the first place.)
quote:
This is not of course, to say that a rare event in some time in the past has never happened that comes close, as rare unlikely events do happen.
What is the basis of suggesting that this virtually never happens except for extremely rarely? In the GA generating an FPGA code, this could be a small fraction of the times (thus far from “majority”), and yet be a very important fractional case of regularity of creating novel new behavior. No evidence has been presented that the rarity is any lower than the rarity of new functions appearing in evolution, as in the fossil record -- that is the issue.
Sorry about the “[SIC]”, I was feeling obnoxious. As to John Bracht not responding, my first related posts were in December 2001 on the ARN thread he links to, and he didn’t answer those either.
So as Frances says “Now the goal posts seem to have shifted to: Gene duplication is not enough in evolution, but that was not the point of objection.” I think this would be my point, with regard to the GA and FPGA duplication example as well. I leave the genetics questions to those expert in the field. I don’t think that direct evidence is as rare as Alonso suggests.
[ 08. February 2003, 20:58: Message edited by: gedanken ]
IP: Logged
|
|
|
Printer-friendly view of this topic
