|
Author
|
Topic: Simulating Self Assembly
|
warren_bergerson
Member
Member # 262
|
posted 05. February 2003 17:20
RBH,
The ‘one to many’ terminology was incorrect. I intended to note that some interpretations of neo-Darwin assume both a functional(mathematical sense) genotype to phenotype mapping and a functional (mathematical sense) phenotype to genotype mapping. The combination of the two would suggest a one to one mapping. I am not sure what concept of mapping Rex is using nor do I know what concept of mapping you suggest should be used. You are welcome to enlighten us.
As I discussed in the Modified GA thread, the analysis here is based on a TA system which 1)does not limit selection to natural selection, 2)does not limit variation to random variation, 3)recognizes logically equivalent operations, and 4)operates on causal relationships not code strings. If you can assure me that Avida incorporates all four modifications then I will take another look at it. Otherwise, the issue is not relevant to this thread.
I have to this point been defining an experimental design rather than specific experiments. The experimental design is one, that as far as anyone has indicated, is not currently being used in biology. It also an experimental paradigm using techniques which are apparently not widely understood by biologists. I therefore seems appropriate to discuss the experimental paradigm and the techniques used before considering actual experiments.
You, as I understand it, are arguing that a new experimental design should not be considered unless the experimenters first demonstrate they comply with all the rules, standards, and methodologies of biology. I have made it clear that in my opinion experimental designs should be evaluated on the merits. If you have questions or comments on the substance of what being discussed here, your questions and comments will be addressed.
gedanken- Thanks, you comments do clarify my position regarding Avida.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 05. February 2003 17:53
warren wrote quote:
You, as I understand it, are arguing that a new experimental design should not be considered unless the experimenters first demonstrate they comply with all the rules, standards, and methodologies of biology. I have made it clear that in my opinion experimental designs should be evaluated on the merits. If you have questions or comments on the substance of what being discussed here, your questions and comments will be addressed.
Um. Sorry, you don't understand it. I said nothing whatsoever about "rules, standards, and methodologies of biology." I'm not sure how I could have been clearer. You say you have an experimentally verifiable model; I say provide the verification. You say analyzing and interpreting existing knowledge is first; I accept that and look forward to seeing it. You say you have a new formal simulation methodology (that so far, given Rex Kerr's analysis, doesn't yet seem new); I say provide it. What's to misunderstand? It's real simple: I'm requesting that you support your own claims. A real working model "verified" with real experimental data. That's all. Simple.
Speaking as a former professor, it is much more effective to introduce new material with a concrete example rather than in abstractions. I suggest you consider that approach.
RBH
[ 05. February 2003, 17:58: Message edited by: RBH ]
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 05. February 2003 18:37
quote:
The ‘one to many’ terminology was incorrect.
In fact, it was correct, in that it correctly represented what "neo-Darwinians" think. There is a one-to-many relationship in each direction.
quote:
I intended to note that some interpretations of neo-Darwin assume both a functional(mathematical sense) genotype to phenotype mapping and a functional (mathematical sense) phenotype to genotype mapping. The combination of the two would suggest a one to one mapping. I am not sure what concept of mapping Rex is using nor do I know what concept of mapping you suggest should be used.
Mapping is used ambiguously to denote both single-valued functions and relations.
Since the use of the word "mapping" seems to be causing confusion, I will define terms and use them consistently from here on:
A relation R on elements between sets A and B is a subset of A x B. A single-valued function f is such a relation that obeys the property: [(a,b1) in f and (a,b2) in f] implies b1=b2; we write this (unique) element b as f(a).
(It is typical to define a mapping M as a relation that for each a in A there exists some b in B such that (a,b) is in M--i.e. for each input there is at least one output. However, it is also typical to use "mapping" as a synonym for "function", and for both to be assumed to be single-valued, so I shall avoid the terminology.)
A relation is 1-1 iff [(a,c) in R and (b,c) in R] implies a=b.
However, as I have mentioned previously, everyone recognizes that the genotype-to-phenotype relation is not 1-1, and that therefore the inverse relation is not a single-valued function.
So I'm really not sure how you conclude that anyone assumes a "one-to-one mapping". Even highly simplified model programs typically have multiple genotypes that can produce a single phenotype, negating the possibility of a single-valued inverse function.
Furthermore--although this level of complexity is not usually included in model programs--biologists are acutely aware of the impact of the environment and/or randomness on the final phenotype. That is, the genotype-to-phenotype relation is not a single-valued function. (Hence the whole "nature vs. nurture" debate, and twin-studies.) Depending on the question you are trying to address, this may or may not be relevant. If the question is, "What is a basic genetic algorithm capable of producing?", then it is not really relevant. If the question is, "What is the best description of the evolutionary processes in play in reality?", then it is relevant. A variety of approaches might be used, ranging from picking according to some probability distribution from a range of phenotypes, through to modeling developmental processes. These are all completely compatible with the traditional ("neo-Darwinian" as you like to call it) model. I think development has not been added to models yet largely because we don't understand real development in enough detail to model it--and coming up with fake/wrong models of development probably wouldn't be too instructive.
Just out of curiosity, am I allowed to use the word "mapping" for a potentially multi-valued function? Or will that cause confusion again? For example, inverse sine is a mapping:
sin^-1(0) = 0 , +-pi, +-2pi, ...
but is not a function.
Edited to fix mixing of function/relation notation.
[ 05. February 2003, 18:57: Message edited by: Rex Kerr ]
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 06. February 2003 11:14
Rex,
I think you provided a very good discussion of the issue of mapping. I agree that the issue here is not mapping in GA’s but Quote: "If the question is, "What is the best description of the evolutionary processes in play in reality?", then it is relevant."
I realize that biologists are aware that a one to one genotype to phenotype map is unrealistic. The issues to be addressed are 1)what are the alternatives and 2)how does the alternative mapping reconcile with the a model or theory of evolutionary change.
I think it can be agreed that it becomes more and more difficult to model/explain evolutionary change as the ratio of phenotypes to genotypes increases. The problem of evolutionary change becomes even more difficult if there are a large number of ‘survival essential phenotypes’ and ‘each survival essential phenotype is very complex or improbable’. The problem of modeling or explaining evolutionary change is made even more difficult if even ‘simple’ evolutionary changes, the types that can be produced with a few generations of selective breeding, can be shown to involve ‘large numbers of changes in essential phenotypes’.
As I stated earlier, the paradigm being defined is a step by step process for moving toward an understanding of both developmental processes and evolutionary change processes. We can start with a simple one to one mapping assumption and a simple RM&NS type change. As we gradually make the genotype to phenotype map more complex- or as we make the egg to operating organisms more realistic, we look at what type of change process is required to fit the more realistic developmental model.
I think we are in agreement that an egg cell has to transform into a very large number of ‘survival essential phenotypes’(cell types as a simple example). What apparently I have not yet convinced you of are 1)the incredible complexity or improbability of the transformation from a set of genes to an adaptive phenotype, and 2)the incredible complexity of the change process required to change or evolve a new phenotype.
One of the fundamental issues here is the difference between 1)viewing a complex transformation as a single causal event, and 2)viewing a complex transformation as a set or series of ‘change in state’ causal events. The second fundamental issue is ‘how do assembly instructions change to produce different transformations. For the time being, let us limit the discussion to the first issue.
I think it will be recognized that statements regarding complex transformations such as ‘an egg transforms to a mature organism’ are simplifications. A complex transformation, ‘A transforms to B’, I suggest, can be translated into "There exists a set or series of causal relationship C containing members c1,c2….. where cx in C denotes ‘sy causes rz’ denoted as cx=(sy, rz)" I believe the suggestion that complex transformations "A transforms to B" are reducible, at least in theory, to sets of simpler causal relationship is non-controversial. Agree?
The question is how far can you usefully and practically break down or reduce a transformation. I am proposing here that we break transformations down to the finite set of ‘assembly instructions’ needed to simulate a completely automated transformation in the laboratory. I originally used the term self assembly. I then used the term automated assembly or manufacture. Maybe an even better term would be ‘computer controlled automated assembly’.
Computerized or automated assembly, as I am using the term, involves a finite set of instructions of the general form ‘ if input S then generate output R’. This type of instruction can be viewed as operating continuously during the assembly or transformation process. The system continuously monitors for the presence of S and if S is detected the response R is generated. If S occurs on multiple occasions, then R is generated on multiple occasions. As far as I know, any automated or computerized assembly process can be modeled using this ‘finite set of instructions’ approach. I was under the impression that this ‘change of state’ approach to computerized automation was, if not widely used, at least is not terribly obscure. Would you agree?
One of the major advantages of the use of this ‘automated assembly’ approach is that it provides an objective verifiable measure of complexity. The form "A transforms to B’ does not provide a basis for measuring how much information is required for the transformation. With the assembly instruction approach, we have a verifiable technique for measuring ‘how many instructions’ are required and ‘how complex or improbable each instruction’. We do not have to pursue this subject very far before we realize that biological systems are far more complex than suggested by the traditional ‘transformation’ type analysis.
This I think is enough for one day.
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 07. February 2003 02:17
quote:
Computerized or automated assembly, as I am using the term, involves a finite set of instructions of the general form ‘ if input S then generate output R’. This type of instruction can be viewed as operating continuously during the assembly or transformation process. The system continuously monitors for the presence of S and if S is detected the response R is generated.
This makes sense--it is the notion of causality. But I fail to see how this is different from any process in biology, including the relationship between various genotypes and various phenotypes. Your model of the relationship improves as you add a more complete picture of environment and development into R.
Maybe you didn't mean to imply that self-assembly is different from the neo-Darwinian position. I can't think of a way in which it is.
For example, if you take codons as assembly instructions, mRNA translation to make protein fits perfectly into the category of "assembly instructions".
quote:
One of the fundamental issues here is the difference between 1)viewing a complex transformation as a single causal event, and 2)viewing a complex transformation as a set or series of ‘change in state’ causal events.
I'm fully willing to think about complex transformations as multiple causal events. And, in fact, even most GA systems built for evolutionary research(such as the programs that run under Avida) have replication split into multiple causal events.
And if we take translation as our assembly instructions, the transformation from genotype to phenotype becomes exactly a series of change in state causal events.
So it seems as though at least in this area, conventional methods fit your criteria quite well.
Edited because I hit "post" before I meant to.
[ 07. February 2003, 02:21: Message edited by: Rex Kerr ]
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 07. February 2003 08:51
Rex,
Quote: And if we take translation as our assembly instructions, the transformation from genotype to phenotype becomes exactly a series of change in state causal events.
I would agree that what I am proposing and neo-Darwinism are not incompatible with respect to setting up or defining the problem. A genotype to phenotype map should logically be reducible to a set of causal relationships or assembly instructions of the type discussed. The automated simulation criteria simply provides an objective verifiable technique for quantifying the complexity or improbability of a map and/or set of assembly instructions. It is worth noting that techniques to quantify the complexity or improbability of assembly are of importance to individuals like Dembski. It should also be noted that, at least IMO, the techniques defined here are apt to produce measures of complexity far greater than the those developed by Dembski.
If we accept that the ‘genotype A to phenotype B’ map is one representation or model of a transformation, and we accept that ‘C’ the set of assembly instructions needed for fully automated assembly is a second model of the transformation, then we should also be will to accept a set ‘P’, a partial or incomplete set of assembly instructions. P would be a set of instructions which, it is reasonable to assume, would be involved in assembly but which are not in themselves sufficient for completely automated assembly. A set P, which is much easier to put together given our current limited knowledge, would provide an estimate of the level of complexity. Based on P, it could be concluded that the identified map or assembly is at least as complex or improbable as the measurement based on P.
The difference between neo-Darwinism and what I am proposing is not, and should not be, in how the problem of evolution is defined, but in the type of explanation, model or theory developed to explain evolutionary change. Which brings us to a couple of the major differences between GA and TA models.
As I believe has already been discussed, TA systems operate on causal relationships which might be referred to as input-output relationships, stimulus-response relationships, assembly instructions, and/or operating instructions. The TA approach defines and addresses change and information processing in biological systems in terms of behavior or actions rather than in terms of codes and physical properties. As we have discussed, this is simply a more detailed but logically equivalent representation of the same phenomena.
GA systems are generally used to analyze changes in one map at a time. GA systems are generally designed to find one optima under one set of environmental conditions. As has been pointed out, GA’s are not necessarily limited to finding one optima or operating under static environmental conditions.
TA systems, by contrast are designed to analyze systems involving very large numbers of causal relationships in a constantly changing environment. GA’s, if used to analyze evolutionary change, show that variation-selection processes can explain change in one or a limited number of maps. TA systems are designed to analyze what types of processes could account for change in systems involving ‘billions of complex assembly instructions’ where ‘10’s of millions’ of these complex instructions can ‘evolve’ in a single generation. But I am getting ahead of myself again.
As I stated TA systems are designed to address large sets of S-R relationships. One of the key operations performed by a TA system is to execute or operate S-R relationships. As discussed above, the first step in analyzing biological self assembly is to construct a set C of the assembly instructions needed for automated self assembly. In most instances, it will be more practical to start with a set P of partial assembly instructions.
Let us assume for the moment that for some biological assembly process we have identified a set of automated assembly instructions C (or a partial set P) containing instructions c1, c2,…,ck. The first hypothesis being tested is that the entire set C of assembly instructions was stored in the egg. If the assembly instructions are all of the form ‘sx causes ry’ then the execute portion of a TA system would be represented as follows[ Note: In many applications, it is more useful to represent an input output relationship as a mathematical function F where F(s)=r for some domain S and range R. ]
EXECUTION SECTION OF TA MODEL GIVEN C
1. LN1: Let j = 0
2. READ CURRENT ENVIRONMENTAL STIMULUS AND ASSIGN VALUE TO SC [SC is one of the s1,s2,..sn stimuli recognized is distinguished by the system]
3. LN2: let j=j+1
4. If j>k go to LN1 [k is the number of assembly instructions in C]
5. Let CC=cj [read input-output relationship from stored file]
6. Let ST = sx from CC
7. Let RT = ry from CC
8. IF SC=ST then generate response RT
9. GO TO LN2
It will be noted that at this point in our analysis we have not yet begun to evaluate models and theories of evolutionary change. We are still looking at and modeling the biological assembly processes. At this stage we can begin looking at questions such as "Does DNA have the capacity to stored the volume of information required for assembly?[we can evaluate this issue because we have techniques for quantifying both the information in assembly instructions and the information that can be stored in DNA.]" and even more interesting questions such as ‘What are the physical mechanisms associated with storing, accessing, and executing assembly instructions?".
As should be apparent, the experimental paradigm proposed here leads to pretty much the same types of laboratory research that are currently being performed. What is being added or changed are the techniques used to interpret the results. Biologists are doing a good job of studying developmental processes. The techniques being proposed here make it possible to quantify the information and information processing associated with development. The ability to quantify information, I suggest and/or predict, will show that we need to reinterpret or reevaluate how we view developmental processes.
So far we have viewed development or assembly as a static process. The next step is to look at possible explanations of how these assembly processes change or evolve. But enough for today.
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 08. February 2003 09:54
Having agreed, I believe, that a genotype to phenotype transformation can be represented by a set C of ‘automated assembly instructions’, it is time to move on to the analyzing the processes that can 1)create a new ‘biological design’ C or 2)transform an existing design or transformation Ct0 into a new design Ct1. Before moving on to change processes, however, it might be useful to review the benefits of expressing and analyzing complex phenomena in terms of sets of causal relationships.
QUANTIFYING COMPLEXITY
The assembly instruction approach makes it possible to calculate (or at least estimate) a numerical value for the complexity of any phenomena. The discussion here has been in terms of modeling physical objects in terms of assembly instructions. The concepts and techniques can be applied equally as well to modeling complex actions in terms of ‘operating instructions’. One such interesting application involves the analysis of ‘operating instructions’ associated with neurons. The technique of reducing a complex phenomena or process to automated assembly or operating instructions makes it possible to apply uniform measures of complexity to any type of ‘design’ observed in nature.
Defining, analyzing and modeling complex phenomena in terms of automated assembly or automated operating instructions may not be a technique widely known or practiced in biology, but it is an established if somewhat complex process known by one name or another in robotics and computer simulation.
As should be obvious, many of the Darwin/ID issues are resolvable given objective verifiable techniques for quantifying complexity.
COMPARING BIOLOGICAL AND NON-BIOLOGICAL DESIGN
Using the proposed measurement techniques, it is possible to quantify the complexity of both biological designs such as a protein and non-biological designs such as snowflakes, diamonds, and even hurricanes. Performing such analysis, I predict, will reveal the basic nature of the differences between biological and non-biological systems. I predict that if reduced far enough, essentially all the assembly instructions used by biological systems occur in non-biological systems. Biological systems are defined or characterized, I predict, not by the types of assembly instructions, but by the speed at which these instructions can be modified to create new and different assemblies.
ANALYSIS OF IDENTIFIABLE ‘MATERIALISTIC PROCESSES’
One of the key issues in biological design is the issue of materialistic processes and/or compatibility of ‘intelligent design’ with the laws of physics and chemistry. The automated assembly process approach insures that at least at the level of ‘observed designs’ the processes operating are all materialistic and compatible with the laws of physics. The question remaining is whether ‘changes in assembly instructions’ can be explained in terms of known physical processes. Clearly the automated assembly approach will also be helpful in this analysis because we can identify specifically what changes need to be explained.
SUMMARY
The automated assembly instruction and automated operating instruction approach provides a technique for reducing complex biological processes and operations (complex biological designs) down to ‘simple’, known physical processes. This I suggest is an essential first step in analyzing and understanding the processes that create and modify biological designs.
As should be obvious, automated assembly is a complex approach involving modeling concepts and techniques that are not widely used, and thus probably not widely understood or appreciated in biology and probably note widely understood or appreciated in most academic fields of mathematics.
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 08. February 2003 13:38
Warren
quote:
As should be obvious, automated assembly is a complex approach involving modeling concepts and techniques that are not widely used, and thus probably not widely understood or appreciated in biology and probably note widely understood or appreciated in most academic fields of mathematics
But it is not obvious despite Warren's 'claims'. But I would encourage Warren to continue on his quest to apply his theoretical ideas in a more experimental manner allowing for actual confirmation of some of his ideas rather than rely on pure speculation. Since Warren stated that it is now possible to quantify complexity using his proposed measurement techniques, I am waiting his results with much excitement. Could Warren enlighten us when he believes such results will become available?
Added
Warren in his comments below seems to ignore the relevant questions namely
1. Warren states that these ideas are not theoretic since they are used in other fields. What Warren has failed to show is that his theoretic ideas have relevance to biology.
2. Additionally when it can be shown that his theoretic ideas are applicable to biology Warren is invited to show that "it is possible to quantify complexity using his proposed measurement techniques" by actually applying his theoretical ideas.
3. This discussion IS about speculation as is evidenced by Warren's own postings in which he makes 'predictions' based on his theoretical ideas.
4. Warren suggests that I do not understand his techniques or claims. While I DO understand his ad hominem approach, I would suggest that he rather spend his time and effort on applying his theoretical ideas as proposed. Perhaps Warren may want to explain why this logical step seems to generate so much avoidance?
The proof is in the pudding as they say and I am encouraging Warren to follow the path proposed by him and apply his theoretical ideas to biological examples to determine if his predictions have any relevance.
So many claims that would benefit significantly from a follow up as proposed in Warren's own postings.
Can these speculative ideas be not only formulated in a mathematical framework but can they also be applied to resolve some of the many issues surrounding Warren's theoretical ideas and speculations.
[ 08. February 2003, 14:58: Message edited by: Frances ]
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 08. February 2003 14:30
Frances,
Quote: But it is not obvious despite Warren's 'claims'. But I would encourage Warren to continue on his quest to apply his theoretical ideas in a more experimental manner allowing for actual confirmation of some of his ideas rather than rely on pure speculation
The discussion here is not about some ‘theoretic ideas’. The discussion is about a form of analysis which I have labeled ‘automated assembly. Variations of this technique are used in some fields of robotics and with some types computer simulation.
As with any proposal to apply known concepts and techniques from one field to another, there are always issues involving terminology and lack of awareness and understanding from the field where the new technique is to be applied.
The discussion here does rely on speculation but rather on the discussion of concepts and mathematics.
The purpose of this thread is to discuss applicability of a known technique or approach to the analysis developmental transformations. If you either don’t understand the technique being proposed or if based on your understanding of the technique it will not work then you should present either your questions or informed comments. Confirmation or rejection of the ideas presented here will be based on evaluation by people who make the effort to understand the concepts and mathematics involved.
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 08. February 2003 15:11
quote:
Confirmation or rejection of the ideas presented here will be based on evaluation by people who make the effort to understand the concepts and mathematics involved.
Where is the mathematics? How can "the mathematics" be confirmed before it is presented. All I have seen has been handily explored by Rex Kerr already.
I suggest that what is being presented is speculation that such a proposition can be made, and not a proposition itself. As such it should not be labeled as more than speculation that such a presentation could be made.
This is one of the difficulties in making ISCID ever have any appearance of approaching scientific validity in its presentations. These presentations need to be backed up by detail, and the detail needs to be checked with observable physical reality to be relevant to science. Is this going to happen here, or is this just going to continue as speculation that such a presentation could be made?
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 08. February 2003 18:08
Warren,
I think you might be confusing typical computer science uses for genetic algorithms with the general form of a genetic algorithm.
quote:
The TA approach defines and addresses change and information processing in biological systems in terms of behavior or actions rather than in terms of codes and physical properties. As we have discussed, this is simply a more detailed but logically equivalent representation of the same phenomena.
It's not even more detailed. It's just a logically equivalent representation. Any behavior or action you want can be represented by a code and a physical property. I will grant that in some cases you might want to not worry about how an action is physically implemented, in which case you would probably want your codes to specify actions. But this is just one possible way (albeit a potentially interesting way) to use a genetic algorithm.
quote:
GA systems are generally used to analyze changes in one map at a time. GA systems are generally designed to find one optima under one set of environmental conditions. As has been pointed out, GA's are not necessarily limited to finding one optima or operating under static environmental conditions.
TA systems, by contrast are designed to analyze systems involving very large numbers of causal relationships in a constantly changing environment.
You have presented no evidence that TA systems actually do what they are designed to do. (You haven't presented the design of any TA system, either.) It's a nice idea, but why can't genetic algorithms do just as well?
Also, as I've asked before, how do you come up with S-R relationships? How do you modify them? How do you decide fitness? It is the answers to these questions that will distinguish TAs from GAs. Right now, there seems to be no difference on a formal level, and only modest difference at the level of envisioned implementation.
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 09. February 2003 07:39
Gedanken,
Quote: This is one of the difficulties in making ISCID ever have any appearance of approaching scientific validity in its presentations. These presentations need to be backed up by detail, and the detail needs to be checked with observable physical reality to be relevant to science.
You use lofty phrases such as ‘backed up by detail’ and ‘observable physical reality’ but it will be noted that you offer not a single argument to counter the pages of material presented here explaining the automated assembly technique for quantifying complexity. You conveniently ignore the vast body of knowledge that exists on 1)the complexity of developmental process, 2)the complexity of automating even simple assembly processes, 3)simple and well known techniques for expressing assembly instructions, and 4)simple and obvious techniques for quantifying the complexity of formally expressed assembly instructions.
There is an extensive body of applied science knowledge supporting the assertion that a developmental process ‘A transforms to B’ can be expressed or denoted by a set of automated assembly instructions C or a set of partial instructions P. There is also an extensive body of evidence showing that for even relatively simple ‘automated assembly’, the levels of information/complexity are extremely high. Again, you have offered nothing but your subjective opinion as a counter to this verifiable approach for defining and measuring information and complexity. You offer neither counter arguments nor alternative measures.
It is useful to note at that the ‘generally accepted’ or ‘academically accepted’ concepts of measuring and quantifying information and complexity in biological systems do not lead to techniques for modeling, simulating and explaining evolutionary change. The generally accepts concepts and techniques support the questionable conclusion that ‘evolutionary change is too complex to permit the development of predictive model’. In rejecting the proposed techniques and concepts, you are in effect not attempting to support not some alternative method of modeling and explaining, but the claim that no explanation is possible.
Rex,
Quote: It's not even more detailed. It's just a logically equivalent representation. Any behavior or action you want can be represented by a code and a physical property
I have no problem considering TA’s as logical extensions of GA’s as long as you recognize and accept a GA 1)operating on input-output relationships, 2)operating on very large sets of input-output relationships, 3)utilizing types of selection other than natural selection etc.. It should also be recognized that TA’s model individuals not populations. It is useful to recognize that TA systems and GA systems are in some senses logical variations of the same general type of mathematical machine. It may, however, avoid confusion to distinguish the two concepts since some of the constraints and concepts imposed on GA’s are not applicable to TA’s.
Quote: Also, as I've asked before, how do you come up with S-R relationships? How do you modify them?
Good questions. Note that in the experimental design proposed here the first step is to use the automated assembly technique to find a set C of assembly instructions or S-R relationships capable of simulating the assembly or transformation being analyzed. Once we know C, or more likely P, then we can begin to evaluate the creation and modification of the S-R relationships making up P or C.
Consider as a simple example some reproducible change readily generated by selective breeding in a few generations. Let P1 represents assembly instructions at the beginning and P2 represents assembly instructions after selective breeding. By comparing P1 and P2 we can identify which assembly instructions changed or evolved and what types of changes occurred. Using variation-selection techniques we can then measure how much information processing was required to produce the change from P1 to P2. We can then evaluate whether the observed change could be generated by certain types of processes such as basic RM&NS. If simple RM&NS could not work, then we look at what other types of processing will fit the observed changes. Having developed a model and/or hypothesis of an alternative change process we would look for evidence support or refuting the hypothesized mechanisms. But we are getting ahead of ourselves. So far we have defined or modeled an assembly process as a set C or P of S-R relationships. This leads to sets P1 and P2 of S-R relationships before and after a change has occurred. This leads to S-R relationship px1 in P1 changing to px2 in P2. As the next step in the discussion, it will be useful to consider mathematical models of the px1 to px2 change.
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 09. February 2003 10:26
Warren says:
quote:
It is useful to note at that the ‘generally accepted’ or ‘academically accepted’ concepts of measuring and quantifying information and complexity in biological systems do not lead to techniques for modeling, simulating and explaining evolutionary change. The generally accepts concepts and techniques support the questionable conclusion that ‘evolutionary change is too complex to permit the development of predictive model’. In rejecting the proposed techniques and concepts, you are in effect not attempting to support not some alternative method of modeling and explaining, but the claim that no explanation is possible.
And in Warren’s opening post, he said:
quote:
The purpose of this thread is to discuss the experimentally verifiable hypothesis that "biological development is a biological design process which can not occur without the presence of information generating ‘intelligent design processes’". The hypothesized existence of scientifically measurable and verifiable intelligent design processes as part of biological development processes would appear to contradict most of the widely held beliefs in both biology and ID.
Warren, in physics there are statistical predictions of aggregate ensembles of molecules. Such notions as “pressure” are the result of such ensembles and those notions. Now it is possible to predict, within limits, the individual pathways of individual molecules, if sufficient information were known of an initial state. Such many-body problems can be simulated (but not easily analytically) with a fair degree of accuracy for short periods. After some time progress, the simulation will become less accurate (just as do weather simulations -- a very similar enterprise). Reasons include supersensitivity and quantum indeterminacy. (In other words the long-term results are supersensitive to initial conditions, and when taken at a sufficiently detailed level even quantum indeterminacy could provide deviations that affect supersensitive or chaotic dynamics sufficiently as to make the predictions fail to hold up in that longer term.)
One could have a greater knowledge of the progress of a dynamic pressure-based event if one were able to simulate these events in great detail. And the weather and climate simulation problems are prime examples. It is useful to attempt to do so, but one must recognize the inherent failure that will result if one expects to “predict” in great detail any long term sequence (for the reasons just given). One can also model the events of the individual lives of organisms in much greater detail. Such modeling would, like the modeling of pressure or weather, possibly give greater detail in understanding, but would similarly be expected to fail to provide long term accurate predictions of behavior due to supersensitivity. (And indeed even quantum indeterminacy again would make these predictions inconclusive in principle, not simply in practical terms.)
It is not my place to accept or reject an un-presented concept that you may have. You reply in angry terms when I challenge that you have not presented details such as to be able to evaluate any such concept that you may have. You said “Confirmation or rejection of the ideas presented here will be based on evaluation by people who make the effort to understand the concepts and mathematics involved.” Yet you have not presented any such mathematics in sufficient detail for me or anyone to “reject” them.
Asking for the detail does not constitute a rejection. But we may fairly reject that you have any such concept -- and certainly that we should accept any particular conclusions that you may wish to speak of, if you do not actually present details that we can observe. So far I have seen no detail that would suggest anything beyond an equivalent of modeling the progress of individual particles in weather forecasting.
Warren, please realize that the readers will not be satisfied with conclusions that are claimed when the details are not presented sufficiently to support those conclusions. It is not my place to provide those details, and I do not “reject” the conclusion by requesting those details.
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 09. February 2003 23:55
Well, we seem to be getting somewhere, albeit excruciatingly slowly.
quote:
I have no problem considering TA?s as logical extensions of GA?s as long as you recognize and accept a GA 1)operating on input-output relationships, 2)operating on very large sets of input-output relationships, 3)utilizing types of selection other than natural selection etc.. It should also be recognized that TA?s model individuals not populations.
(1) and (2) are perfectly compatible with genetic algorithms. (3) is not--but this is irrelevant unless you present some alternative to the method used in GAs. So far you have not. I also do not understand how TAs can operate on individuals in a way different than GAs.
quote:
We can then evaluate whether the observed change could be generated by certain types of processes such as basic RM&NS. If simple RM&NS could not work, then we look at what other types of processing will fit the observed changes.
The specific example you give is an example of evolution, but you don't need any mutations to occur during the selective breeding. Selective breeding works (or is thought to work) by shifting allele frequencies in a population.
quote:
As the next step in the discussion, it will be useful to consider mathematical models of the px1 to px2 change.
That's what I have been waiting to see.
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 10. February 2003 10:13
Rex,
You raised several issues, but I am limiting my response to the concept of selection.
Quote WB: 3)utilizing types of selection other than natural selection etc..
Quote : (3) is not--but this is irrelevant unless you present some alternative to the method used in GAs
One of my many complaints about GA’s is the ambiguous use of the concepts of selection and natural selection. At times selection appears to be interpreted in the narrow ‘natural selection’ operation involving the non-survival and non-reproduction of an individual. On other occasions it seems to mean any process which accepts one option and rejects another. I leave the GA definition of selection to others. I will, however, attempt to clarify how the concept of selection is defined for TA’s.
To begin, in TA’s the expressions ‘variation-selection’ or ‘variation-selection-storage’ are labels referring to complex causal processes or sets of complex processes. Variation-selection, in TA’s is not just the interaction of two processes but a label referring to the interaction of many different logical operations. A variation-selection operation is expressed or modeled by a logic machine program which controls a TA system. Each line of code in such a program represents a complex set of operations (as should be apparent, there are many different ‘programs’ or sets of operations which can be used to express the same program). As the expression is interpreted here, variation-selection refers not to two elementary operations but to a large set of complex logical operations.
In general terms, ‘selection operations’ in TA’s refer to ‘sets of operations which accept some options and reject other operations based on input received from or extracted from the external environment’. This operation can be separated into sub-operations- 1)recording environmental values and 2)performing the logical accept-reject processing. A selection process involves ‘selecting from two or more options’. In terms of pseudo code this can be represented by
CODE FOR A SELECTION PROCESS
1. record environmental value V1 for options 1
2. record environmental value V2 for option 2
3. select option 1 or option 2 (and reject the other option) based on a comparison of values V1 and V2
Using this definition of selection, you can have many different selection processes operating simultaneously in a single system based on recording different types of environmental values. The assumption in modeling biological systems is that environmental values V1, etc. reflect ‘survival value’. Survival values need not, however, be a variable or set of values. We can define, identify, large hierarchies of indirect survival variables. A hierarchy of ‘variation-selection’ processes operating on a hierarchy of survival variables can generate a very complex or improbable ‘adaptive solution’. It should be obvious that a hierarchy of rather simple ‘variation-selection’ processes have the potential to generate a very complex or improbable adaptive solution.
It will also be noted that the above definition of selection is compatible with the concept of rapidly changing environmental conditions. A TA model assumes that the values the recorded environmental values reflect the values at the time the values are recorded. The value V1 for option 1 can be very different at time t=1 than at t=0. Using TA models, adaptive solutions can change rapidly.
A third important feature of the TA definition of selection is that ‘variation-selection processes can operate on variation-selection processes’. Consider the pseudo code defined above. A variation- selection operation could operate to change the method of recording environmental values or it could operate to change the logic used in selection.
Finally, it will be noted that an S-R relationship of the form ‘if sx then ry’ is itself a selection operation(if an environmental value sx then select ry else select not-ry’
OBSERVABLE BIOLOGICAL REALITY
Examples of the types of variation-selection processes described above can be 1)created artificially, and 2)can be identified in biological systems. It should not be difficult to think of illustrations of these properties of variation-selection processes. I will be glad to provide illustrations if you can’t think of them.
SUMMARY
TA systems involve multiple variation-selection processes occurring simultaneously. This is possible because TA systems involve multiple selection variables being measured or recorded at the same time. Is this or is this not a departure from your concept of GA models?
IP: Logged
|
|
|
Topic Closed
Printer-friendly view of this topic
