|
Author
|
Topic: Evaluation of neo-Darwinian Theory with Avida Simulations
|
Salvador T. Cordova
Member
Member # 959
|
posted 18. July 2004 16:35
quote:
Royal wrote:
Here is what I am finding: in the absence of any logic function rewards, contra expectations, the genome size starts to increase initially, plateaus, then decreases to a value smaller than the expected lowest number of instructions needed to define replication.
I independently confirm this in Avida 1.3 (Avida 2.0 Beta 6 by RBH's admission is still buggy).
(
It is of interest that in regards to cosmic Ray Point Mutation, RBH is sending those concerns to the Avida Authors as he wishes clarification on the issue that high point mution values of 50000 enabled replication, whereas 0.5 had less replication. It happens on my platform, not on his, but he does confirm similarly odd behavior.
All this to point out that Avida results should be accepted with caution. The software, like all software, has vulnerabilities.
)
In regards to the initial increase which you mentioned, I used a freshly downloaded configuration from the Avida website for Avida 1.3. I commented out all of the lines in the Task File. I did see an initial popups in genome size. SIZE_MERIT_METHOD was at the default setting of 4.
I will run it again to double confirm and post the results more formally.
I will also appraise you Dr. Truman of an interesting application of Avida. The idea is in it's infancy, but Avida, over many generations, I predict, will NOT generate nested hierarchies between the lineages of genomes.
The sequence data, I predict, will not look like that generated by nature as witnessed by The Problem of Equidistance in cytochrome-c hiearchies.
I will start another thread on this, but I figured I would advise you of these developments as they have bearing on Avida and the underlying paradigms of Avida, namely neo-Darwinian evolution.
When I begin that thread in earnest, I will post a link here as I intend to use Avida simulations (Avida bugs and all) to highlight that even under Avidian assumptions, a supposed key piece of evidence for Darwinism, namely hierarchical relationships in molecular sequences, is actually a key piece of evidence for Intelligent Design or at the very least, alorithmically driven descent with modification.
Salvador
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 18. July 2004 22:48
Truman wrote quote:
Here is what I am finding: in the absence of any logic function rewards, contra expectations, the genome size starts to increase initially, plateaus, then decreases to a value smaller than the expected lowest number of instructions needed to define replication. (Emphasis original)
If one considers what happens in the initial stages of an avida run, neither observation is contrary to informed expectations.
Consider starting an avida run with an Ancestor that is CREATURE.BASE, which is 20 instructions long, all 20 being required for replication (tested by knockouts). Assume that all task rewards have been commented out in TASK_SET. Assume also that all mutation rates are GENESIS defaults.
When CREATURE.BASE replicates, the replication process can be hit by one or more of several different kinds of mutations -- a copy error, a divide insertion, or a divide deletion. With all instructions necessary for replication in the initial stages of the run, only mutations that add instructions in places in the genome where they don't interfere with replication will produce offspring that can themselves replicate. That is, only DIVIDE_INSERTION mutations have a chance to be propagated in a lineage early in an avida run. All other mutations are reproductively lethal to the offspring and hence no lineage propagates from them: there are no "grandchildren". As a consequence, only longer genomes can found lineages in the early stages of a run and hence the average genome length of the nascent population increases.
In addition, inter-critter competition for living space in the population is at a minimum at the start of a run with just the Ancestor present, and increases from that initial zero value to a maximum when the array is filled. The selective pressure for shorter genomes starts at zero and increases as the array fills. (To a decent approximation, competition increases as the square of the radius of the occupied area on the avida array.)
So, during the initial stages of a run, only mutations that add genetic material can result in successful replicators, and competition is at (relatively) low levels. That enables longer genomes to appear and reproduce. That increase in average length is a consequence of the truncation of the distribution of genome lengths at the minimum defined by the Ancestor's replication code. But as population size -- and therefore competition -- increases, selective pressure for faster replicators pushes length back down. Eventually the critters evolve more compact replication code than the Ancestor started with.
It is therefore not unexpected that in the early stages of an avida run there is an increase in the average length of genomes. Only later, as evolution proceeds and variants are thrown up by mutations, can the intrinsic pressure for shorter genomes operate to select for variants that code for more efficient replication. And it is surely not the case that genome length quote:
... decreases to a value smaller than the expected lowest number of instructions needed to define replication. (Italics added)
as Truman claims. What reason can Truman offer to support his expectation that avida critters cannot evolve more compact replication code than that written by human programmers?
Both expectations -- an initial increase in average genome length and the later evolution of more compact code -- follow directly from purely evolutionary considerations applied to the specific conditions of an avida run. There's no special pleading or ad hoc assumption being made.
RBH
Late edit for punctuation.
[ 19. July 2004, 13:10: Message edited by: RBH ]
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 19. July 2004 15:01
[RBH]: posted 11. July 2004 16:00
quote:
How about a reference. I don't want to fight my way through the pdf again hunting for what Truman means here.
[Royal]: From Part 2, ref. <2> Ofria, C.; Adami, C.; and Collier, T. C., “Selective Pressures on Genomes in Molecular Evolution”, J. theor. Biol. 222, (2003) 477-483.
http://arxiv.org/abs/quant-ph/0301075.pdf
See Figure 1. Happy reading.
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 19. July 2004 15:21
[charlie d.]: posted 16. July 2004 19:08
quote:
I am not sure I understand what you mean here. Are you saying that natural selection strongly disfavors organisms with large amounts of duplicated, redundant DNA? Or just organims producing a vast eccess of mRNA and proteins from a specific gene?
[Royal]: Asexual, rapidly reproducing organisms with small genomes are most likely to offer neo-Darwinian Theory examples (it is unlikely point mutations will permit evolution of new complex function in elephants).
For an average generation period on the order of 20 minutes, even a couple of seconds longer DNA replication time leads to a measurable fitness disadvantage for that lineage, and for every every generation. (See ref. <9> of Part 2: Truman, R.; and Heisig, M., “Protein families: chance or design?”, TJ, 15(3), (2001), 115.)
This effect only applies if DNA replication represents a significant fraction of the life cycle and if the extra DNA were not to code for advantageous functions.
[charlie d.]: quote:
Also, I am not sure that metabolic costs of mRNA and protein synthesis really apply to AVIDA, in which all the metabolic costs are associated with genome replication.
[Royal]: That is correct. The AVIDA software processes instructions using external C++ programs. These incur no metabolic costs and the processing machinery cannot be damaged by mutations.
Now, I am arguing that creating a computer program which demonstrate that ‘in principle’ new complex functions could arise is not very useful. Simulations which conceptually demonstrate that something might be true can always be programmed with suitable parameter settings.
We cannot, unfortunately, realistically simulate bacteria-length genomes, populations on the order of 10E10 members and point mutations on the order of 10E-11 to 10E-9 per codon. But we should perform some sensitivity analysis to permit a feel for what would happen as we extrapolate towards biologically realistic settings.
See if you agree with my reasoning:
- In biological organisms a novel gene or domain can only offer a selective advantage if transcribed (and in the case of proteins also translated).
- Mutations on DNA which is not expressed cannot be selected for just because it could code for something useful
- Mutations in the Avida system could produce instructions which define a logic function in a particular environment
- In biological terms, mutant mRNA and polypeptides must be generated continually so that whenever something useful in terms of gene sequences is created natural selection is able to recognize it
- The proportion of non-functional mRNA and polypeptides is vastly greater than the biologically useful. Note that both sequence and amount pose real-world constraints: a cell flooded with a useful protein sequence will die; a single copy is unlikely to help
- Therefor extrapolations based on Avida runs must capture the following realities:
* Genomes of minimal size to survive need to posses on the order of 1000 superfluous base pairs (coding + regulator sequences) to permit a novel gene to evolve. That ‘junk DNA’ sequence produces mRNA and polypeptides non-stop as the DNA is mutating over many generations. This represents a considerable metabolic penalty compared to the original and competing lineages which did not receive the extra DNA. This effect is much more interesting than the metabolic cost to maintain more DNA and the slightly longer generations times mentioned above.
* Mutations could deactivate expression of ‘junk genes’, but then natural selection would not be able to reward the future lucky mutations.
* A novel biochemical processes actually requires the origin of multiple genes. And realistic mutational rates for free-living organisms and realistic proportions of coding material among all random alternatives needs to be kept in mind. RBH likes to use point mutations rates of 0.005 per instructions in his Avida runs. Logic functions typically require but a few dozen generations to be produced by mutations, using 3200 organisms. This is computationally convenient, but needs to be extrapolated dramatically for real organisms. In such cases the metabolic disadvantage would have to be carried for zillions of generations for a particular lineage before generating a novel mutant. Deletions anywhere in the junk DNA during this time would be advantageous.
* For such organisms, rapid genome trunction is the natural outcome.
* Let us not forget that real genes are composed of continuous stretches of codons. Avida logic functions can consist of instructions dispersed all over the genome, since no real decoding (transcription and translation) is occurring.
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 19. July 2004 15:25
[Salvador T Cordova]: posted 18. July 2004 16:35
quote:
Royal wrote:
Here is what I am finding: in the absence of any logic function rewards, contra expectations, the genome size starts to increase initially, plateaus, then decreases to a value smaller than the expected lowest number of instructions needed to define replication.
I independently confirm this in Avida 1.3 (Avida 2.0 Beta 6 by RBH's admission is still buggy).
[Royal]: I only work with Avida 2.0 Beta 6. Dr. Richard Lenski put me in touch with one of the Avida developers. I’ve encountered no real errors in this version.
What I almost always find:
(1) Avida runs in the absence of reward lead to increasing genome size initially (more inserts that deletes survive). A plateau is reached, then the genome starts to truncate slowly.
(2) When rewarding logic functions: once all or a large number of logic function have fixed some genome trunction will occur as unneeded instructions are weeded out. But initially, before any logic functions are found, the genome size has already started to increase slowly.
(3) Under default Avida settings the proportion of random instruction sequences able to code for one of the 9 logic functions is very large (compared to DNA based organisms). One needs very little extra junk instructions in these Avida experiments to permit a mutation to stumble on one of functions. By using very strong selection (high rewards in the environment.cfg file) one can generally fix the lucky lineages quickly.
(4) Even under genome trunction settings (SIZE_MERIT_METHOD 0 in the genesis file) one can develop novel logic functions with a little cleverness. For example, always reward any logic function dramatically; provide lots of extra nop-C instructions to the ancestral organism. As if it fails, use a different RANDOM_SEED and try again!
(5) Adding a little biological realism paints a totally different picture. For example, since most random DNA is worthless, don’t reward the simplest 2 or 4 Avida logic functions; use relative fitnesses on the order of 1.01 to 1.1 for the others; start of with relatively little junk instructions. I got tired of trying, but at the end of huge runs no logic functions were present at all.
Antibiotic resistance studies clearly show that natural selection can occur, no one really has a problem with that. Whether this can actually lead to a better genome in the face of multiple challenges is an entirely different matter. But these (artificial) extreme conditions are not representative of the processes which could develop new protein families.
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 19. July 2004 15:38
[RBH]: posted 18. July 2004 22:48
quote:
Royal: Here is what I am finding: in the absence of any logic function rewards, contra expectations, the genome size starts to increase initially, plateaus, then decreases to a value smaller than the expected lowest number of instructions needed to define replication.
[RBH: If one considers what happens in the initial stages of an avida run, neither observation is contrary to informed expectations.
[Royal]: Oops, RBH managed to twist a poor choice of wording and try to make it look like I claimed the opposite of I have said. From all my writings he knows I never found nor expect genome truncation under typical Avida settings (SIZE_MERIT_METHOD 4 in the genesis file) during the first updates. I have have always argued that truncation, especially initially, must be explicitly demanded from Avida runs this the opposite usually occurs even without any logic functions.
I probably share the blame. My “contra expectations” comment was directed towards what it seemed RBH had himself claimed elsewhere (and could no possible represent my view):
RBH posted on ARN 07-01-2004 02:39 http://www.arn.org/ubb/000026 the following:
quote:
In actual fact, in the data below it is obvious that there is selective pressure for shorter genotypes even with the 'neutral with respect to length merit award scheme', contrary to both Salvador's and Truman's claims.
quote:
For the purposes of this illustration I plotted the average size of the genotypes in the population of 3,600 critters in each run. Those curves are shown on this graph.
[Royal]: Please take a careful look at his graphs. For both uniform and non-uniform external fitness RBH’s plot of Updates vs. # Instructions shows an immediate drop in genome size between update 0 and 400. Since this showed “selective pressure for shorter genotypes” in the initial updates I was attempting to argue what I found empirically (and had long since reasoned out anyway): an initial increase in genome size is generally to be expected. These graphs are very misleading since they clearly contradicts what RBH is now (correctly) saying.
But we need not quibble. We all now surely agree: during the critical initial updates, before logic functions are generated, no penalty results for larger genomes. As I pointed out in my post to charlie d., natural selection can only identify new useful genes if these are producing mRNA and polypeptide. Until a useful protein has evolved, the organism must be heavily penalized for the heavy metabolic cost if Avida based simulations are to be applied to real biological systems.
IP: Logged
|
|
Scott
Member
Member # 1222
|
posted 19. July 2004 20:50
Royal: In biological organisms a novel gene or domain can only offer a selective advantage if transcribed (and in the case of proteins also translated).
quote:
Traditionally, a gene was defined as a segment of DNA that codes for a polypeptide chain or specifies a functional RNA molecule. Recent molecular studies, however, have radically altered our perception of genes, and we shall adopt a somewhat vaguer definition. Accordingly, a gene is a sequence of genomic DNA or RNA that is essential for a specific function. Performing the function may not require the gene to be translated or even transcribed.
At present, three types of genes are recognized: (1) protein-coding genes, which are transcribed into RNA and subsequently translated into proteins, (2) RNA-specifying genes, which are only transcribed, and (3) regulatory genes.
- Fundamentals of Molecular Evolution
Royal: Let us not forget that real genes are composed of continuous stretches of codons.
I would like to recommend that you not use "codons" in this context. Perhaps nucleotides?
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 19. July 2004 21:46
Greetings Scott,
Welcome to ISCID!
Royal,
Scott is a good comrade, I feel he can very much add to this discussion.
I am in the middle of running Avida with the following settings. I do not know if it is of interest to you, but I offer it for the sake of completeness.
1. I shut off all the rewards for logic functions by commenting out all the TASK_SET
lines.
2. I set
BIRTH_METHOD 4
DEATH_METHOD 1
AGE_LIMIT 200
this ensured Natural Selection was turned off
3. I set
SIZE_MERIT_METHOD 3
Avida evolved complexity upward from 20 to 28 in the absence of Natural Selection. It seemed to stop at 28 (update 40,000).
Salvador
[ 19. July 2004, 22:03: Message edited by: Salvador T. Cordova ]
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 19. July 2004 23:31
Royal,
For the sake of completeness I report the following with a basic configuration of Avida 1.3.
I set:
1. all lines in TASK_SET to NULL
2. SIZE_MERIT_METHOD 3
Avida by update 9000 was evolved genomes from 20 all the way to 70, and still climbing....
All that is needed for Avida to evolve complexity is to reward complexity.
Even reward of logic functions presumes selective pressures can exist to reward logic functions.
All that says is that if a way to reward complexity exists, then complexity will be favored. Such statements may be physically impossible in many instances, but they are givens in Avida. It is essentially tautologous : "if a selective pressure exists to evolve complexity, then complexity will be evolved."
But there is no justification for these premises because such selective pressures may be more improbable than the genomes they supposedly evolve. Potentially impossible situations are easy givens in Avida!
Setting SIZE_MERIT_METHOD to 3 evolves complexity with less hassle than playing with the TASK_SET. It is not very different than arbitrarily assigning EQU the largest merit multiplier, except one gets lots of junk along with functional complexity.
There are other weakness in the Avidian-Darwinian model.
I concur with this from your paper:
quote:
5) Dramatic rewards are provided for new logic functions • New, more complex functions, like EQU, once formed are essentially guaranteed to perpetuate, i.e., fix in the population due to very high relative fitness rewards.
The model approximates the summation of rewards linearly. In the case of backup systems, total rewards may well be inversely proportional to complexity, the exact opposite of the rewards for EQU.
Avida-Darwinian models reward complexity by making an approximate linear superposition (an addition) of rewards associated with the appearance of each complex system. That is if
System A commands 32 reward points
System B commands 32 reward points
then presumably an organism with A and B have
32 + 32 = 64 reward points
Biologically speaking this is indefensible in many cases. Why? Consider the numerous backup systems in biology such as the developmental pathways for the vulva in Nematodes. There are two independent complex systems to express the same organ, yet there is no selective advantage for evolving backup complexity, yet it's there.
In such cases:
System A commands 32 reward points
System B commands 32 reward points
System A and B combined yield at best 32 reward points, and in actuality less with respect to immediate fitness, all other things equal, because of metabolic demands, etc.
Again, modelling fitness as the linear superposition of rewards is inappropriate given aspects of backup systems.
Salvador
PS
I cite for completeness:
Michael Denton molecular genetist in Nature's Destiny writes page 337-338:
quote:
Another very intriguing aspect of development in higher organisms which has become increasingly apparent over the last ten years, and which is bound to impose additional constraints against any sort of bit-by-bit undirected change, is the use of partially or totally redundant components to buffer organisms against random mutational error and ensure relieability, particularly during development. As one authority points out: "The idea that redundancy may be quite common in cell and develoopmental biology has its origin in Spemann's (1938) idea of double assurance, a term taken from engineering."
The strategy of using several different means to achieve a particular goal where each of the individual means is sufficient by itself to achieve the goal is used in all manner of situations to guarantee that the goal will always be achieved, even if one or more of the means fail. Missiles, for example, are often guided to their targets using a number of different automatic guidance systems, including ground-based radar, map matching, inertial guidance systems, following a graded signal (heat-seeking). Even if one fails, the missile will still home in unerringly on its target. Reliability of information storage on computer discs is increased by encoding the information in two or more different ways. The functional reliability of complex machines such as aircraft and particularly space vehicles invariably involves the use of redundant components. The space shuttle's on-board inertial guidance system, which it uses during boosting into orbit and during reentry, consists, according to the McGraw-Hill Encyclopedia of Science and Technology of "five redundant computers and three inertial measurement units. Dual star trackers are used for periodic realignment in space....A radar backup system is provided for safety during launch and landing." (My emphasis.) Another instance where redundancy is exploited to increase reliability is in human and animal navigation, where most often a number of different and individually redundant clues are followed to minimize the risk of navigational error, which might accrue from following only one type or set of clues.
It now appears that a considerable number of genes, perhaps even the majority in higher organisms, are completely or at least partially redundant. One of the major pieces of evidence that this it the case has come from so-called gene knockout experiments, where a gene is effectively disabled in some way using genetic-engineering techniques so that it cannot play its normal role in the organisms's biology. A classic example of this came when a gene coding for a large complex protein known as Tenascin-C, which occurs in the extracellular matrix of all vertebrates, was knocked out in mice, without any obvious effect. As the author of a paper commenting on this surprising result cautions: "It would be premature to conclude that [the protein] has no importan function ...[as] it is conserved in every vertebrate species, which argues strongly for a fundamental role." The protein product of the Zeste gene in the fruit fly drosophila, which is a component of certain multiprotein complexes involved in transcribing regions of the DNA, can also be knocked out without any obvious effect on the very processes in which it is known to function.
The phenomenon of redundant genes is so widespread that it is already acknowledge to pose something of an evolutionary conundrum. Although in the words of the author of one recent article, "true genetic redundancy ought to be, in an evolutionary sense, impossible or at least unlikely," partially redundant genes are common. As another authority comments in recent review article: "Arguments over whether there can be true redundancy are moot for the experimentalist. The question is how he functions for partially redundant genes can be discovered given that partial redundancy is the rule." (My emphasis.)
And it seems uncreasingly that it is not only individual genes that are redundant, but rather that the phenomenon may be all-pervasive in the development of higher organisms, existing at every level from individual genes to the most complex developmental processes. For example, individual nerve axons, like guided missiles or migrating birds, are guided to their targets by a number of different and individually redundant mechanisms and clues. The development of the female sexual organ, the vulva, in the namatode provides perhaps the most dramatic example to date of redundancy exploited as a fail-safe device at the very highest level. A detailed description of the mechanism of formation of the nematode vulva is beyond the scope of this chapter, suffice it to say that the organ is generated by means of two quite different developmental mechanism, either of which is sufficient by itself to generate a perfect vulva.
It seems increasingly likely that redundancy will prove to be universally exploited in many key aspects of the development of higher organisms, for precisely the same reason it is utilized in many other areas--as a fail safe mechanism to ensure that developmental goals are achieved with what amounts to a virtually zero error rate. A very high degree of redundancy in the specification of the development of higher organism is almost certainly not in the least bit gratuitous, but rather of necessity. Probably no system remotely as complex as a higher organism could possibly function without a large measure of redundancy in many or even every aspect of its design.
Now, this phenomenon poses an additional challenge to the idea that organisms can be radically transformed as a result of a succession of small independent changes, as Darwinian theory supposes. For it means that if an advantageous change is to occur, in an organ system such as the namatode vulva, which is specified in two completely different ways, then this will of necessity require simultaneous changes in both blueprints. In other words, the greater the degree of redundancy, the greater the need for simultaneous mutation to effect evoutionary change and the more difficult it is to believe that evoutionary change could have been engineered without intelligent direction. Redundancy also increases the difficulty of genetic engineering, as it means that the compensatory changes that must inevitably accompany any desired change must be necessarily increased.
[ 19. July 2004, 23:44: Message edited by: Salvador T. Cordova ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 20. July 2004 18:57
I am at a meeting so i don't have time to respond in detail. I'll just answer this basic statement: quote:
RT:
Now, I am arguing that creating a computer program which demonstrate that ‘in principle’ new complex functions could arise is not very useful. Simulations which conceptually demonstrate that something might be true can always be programmed with suitable parameter settings.
We cannot, unfortunately, realistically simulate bacteria-length genomes, populations on the order of 10E10 members and point mutations on the order of 10E-11 to 10E-9 per codon. But we should perform some sensitivity analysis to permit a feel for what would happen as we extrapolate towards biologically realistic settings.
This is true if one wants to simulate biological evolution, but AVIDA does not. AVIDA only models evolutionary processes, and shows that they can generate complexity and information.
For a biologist, this is an interesting model because it allows to study how complex features are generated via evolutionary principles, so that the same processes can be later studied in (not blindly extrapolated to) real organisms. It is irrelevant how big the AVIDA genomes are, what they are made of, what the replication and metabolic costs are (if any), what the mutation rate is. Those all become issues when the process is studied in nature, not in AVIDA. Certainly, only a very incompetent biologist would expect to apply information from AVIDA directly, without modification, to the real world. As far as I know, no one has done this.
The more important issue here (where we deal with creationist and ID claims) is that, according to many creationists and prominent ID advocates, blind evolutionary processes (no matter in what medium they are applied) cannot generate complexity and increase information in principle. This is so much of an absolute statement that Dembski even coined his very own new "Law" about it.
If, as you say, AVIDA accomplishes its information feats because of "suitable parameter settings" (i.e. by "smuggling" in information), then one has to show how the information is smuggled. Until then, AVIDA successfully invalidates the claims that non-teleological evolutionary mechanisms cannot lead to complexity (including IC) and information gains. Saying that AVIDA does not faithfully model specific aspects of biological evolution does not challenge this conclusion, it just changes the subject in a direction that is irrelevant to the specific point we are trying to make here.
IP: Logged
|
|
Nel
Member
Member # 614
|
posted 21. July 2004 02:37
charlie d:
quote:
If, as you say, AVIDA accomplishes its information feats because of "suitable parameter settings" (i.e. by "smuggling" in information), then one has to show how the information is smuggled. Until then, AVIDA successfully invalidates the claims that non-teleological evolutionary mechanisms cannot lead to complexity (including IC) and information gains.
But Avida doesn't lead to the kind of complexity that most IDers are concerned with(some of the "parts" of the system it lead to were easily achievable by chance alone). And when it comes to IC, what claim are you referring to?
[ 21. July 2004, 03:21: Message edited by: Nelson-Alonso ]
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 04. August 2004 00:05
For the sake of completeness, because the point mutation issue is important to me, I finish out an issue raised earlier.
quote:
RBH wrote:
Salvador blew it. See here.
RBH
A resolution to the issue was provide by the Avida group thanks to the RBH. It turned out with respect to the paramamer settings, what I postulated was correct. I didn't blow it after all.
quote:
Footnote: documentation
There is an erroneous note in the default "genesis" file distributed with Avida 1.x. The line specifying point mutation rate has a comment (x10^-6), implying that the user value is multiplied by this quantity before application. This is in error; the specified point mutation is used as-is as a probability. This erroneous comment has been removed in Avida 2.x releases.
We will shortly be releasing a bug-fix of Avida 1.6 (the stable, current version of the Avida 1.x codebase) which corrects this error in the documentation.
I'm pleased to say, that now 3 of the participants (Royal, RBH, myself) on this thread have had direct contact with the Avida group. We have made a contribution (albeit miniscule) to their efforts because of these discussions. My comments led to an improvement in the documentation and clarity of Avida 1.3.
Evan Dorn of the Avida group wrote to me:
quote:
Evan wrote:
quote:
Salvador wrote:
Evan, again thank you for your detailed reply. I appreciate that you've taken the time to respond to me, especially knowing that I am a critic of your project.
You are most welcome. No good scientist fears critical analysis of his or her work, because the truth is what we all seek.
As far as the documentation error is concerned, the (x10^-6) comment in the genesis file is unquestionably in error upon examination of the code. I do not believe that it affects the interpretation of any currently published results.
quote:
Salvador wrote:
You actually addressed far more than what I was personally was pursuing.
The documentation error is unfortunate and must be corrected. I personally am glad to have it cleared up.
Entire Exchange with Avida Group.
Salvador
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 06. August 2004 10:51
I was away for a while, but from what I read the situation is currently the following:
1. There was a mistake in the documentation of AVIDA 1.6 (the 10^-6 correction factor), which was corrected in later versions.
2. The only "bug" that has been identified in AVIDA is that when an impossible value is inserted for mutation rates (probability >1), instead of returning an error message that says the value is impossible, the program automatically resets the value to 0.
3. Within the realm of possible mutation rate values, the program does in fact perform correctly, and the initial puzzling results obtained by Salvador were due to misinterpretation of the output data.
4. No published data obtained through AVIDA have been invalidated, and in particular the conclusion of the Lenski Nature paper that undirected mutation/selection evolutionary processes in AVIDA can lead to increased information and the generation of "irreducibly complex" structures still stands.
Am I correct, or am I missing something?
[ 06. August 2004, 10:52: Message edited by: charlie d. ]
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 13. August 2004 12:31
quote:
Charlie asked:
I was away for a while, but from what I read the situation is currently the following:
1. There was a mistake in the documentation of AVIDA 1.6 (the 10^-6 correction factor), which was corrected in later versions.
Later versions were created to add features. In the software industry version 1.0, 1.1, 1.2 are improvements to a basic design. When you see numbers like 2.0, 2.1, there is a fundamental change somewhere. Evan Dorn assured me that the current version 1.3 will be upgraded with my recommended change to 1.6. The reason 1.3 is used by RBH and I, is on RBH's recomendation because in his words, Avida 2.6 Beta is "buggy".
Being a information technology person, I don't intend to make a fuss of such foibles. Happens all the time. I commend the Avida group for their response to me....
quote:
2. The only "bug" that has been identified in AVIDA is that when an impossible value is inserted for mutation rates (probability >1), instead of returning an error message that says the value is impossible, the program automatically resets the value to 0.
That is materially correct. RBH believed I was in error, and I claimed it was the problem of the "sketchy documentation" in the defualt parameter files programmed by the authors.
I won the ice cream bet: I was right, RBH was wrong. I commend RBH though for settling the issue by going to the Avida authors. Had he not stepped forward (him being respected by them) the issue would not have been solved. Kudos to him!
quote:
3. Within the realm of possible mutation rate values, the program does in fact perform correctly, and the initial puzzling results obtained by Salvador were due to misinterpretation of the output data.
I fully expected Avida to behave the way it did, that's why I cranked up the mutation rates to prove Royal's point here:
quote:
(3) Critically important functions for “life” are difficult to destroy by mutations • All functions necessary for an organism to replicate and survive are carried out initially by only 15 instructions. The physical machinery to transcribe, translate and perform metabolic processes are not coded genetically and thus not subject to mutational damage.
The Avida group would dispute that characterization by Royal (and myself) unsurprisngly, but the fact I could crank up the mutation rate and these organisms kept replicating shows that Royal is right. I fully expected that behavior from Avida given what I know of information technology. That's exactly why I conducted the experiment.
Here is a funny situation in that both I and the Avida authors will claim Avida behaved as expected, yet reach different conclusions as to it's significance.
quote:
4. No published data obtained through AVIDA have been invalidated, and in particular the conclusion of the Lenski Nature paper that undirected mutation/selection evolutionary processes in AVIDA can lead to increased information and the generation of "irreducibly complex" structures still stands.
There is a saying in the software industry, "it's not a bug, it's a feature". The findings of Avida are reasonably valid deductions from their premises. That is to say : "if complexity is rewarded over simplicity, complexity will evolve and persist". But that view is tautologous.
The existence of such selective environments is possibly more remote than genomes evolving through random chance. This fact has not gone unnoticed by the advocates of neutralist poplulation genetics.
It's actually fairly easy to write and design a complexity generator in software ( essentailly a random noise generator being fed through some sort of filter which extracts and preserves interesting results). That's all Avida ultimately is, I'm afraid.
quote:
Am I correct, or am I missing something?
You're mostly correct. I think you're mostly correct, except for the things in my comments that might be at variance with what you said. I'd say 80% correct. The issue, ultimately, is: do selective pressures in the real world really exist to create LARGE increases in complexity.
My position is increases in function complexity are the exception rather than the rule. Natural Selection tends to erode away if not totally extinct complexity.
[ 13. August 2004, 12:35: Message edited by: Salvador T. Cordova ]
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 13. August 2004 17:32
quote:
Later versions were created to add features. In the software industry version 1.0, 1.1, 1.2 are improvements to a basic design. When you see numbers like 2.0, 2.1, there is a fundamental change somewhere. Evan Dorn assured me that the current version 1.3 will be upgraded with my recommended change to 1.6. The reason 1.3 is used by RBH and I, is on RBH's recomendation because in his words, Avida 2.6 Beta is "buggy".
Being a information technology person, I don't intend to make a fuss of such foibles. Happens all the time. I commend the Avida group for their response to me....
Of course, beta versions can be buggy because that's what they are for - to be tested in real world applications to find bugs. It's like complaining that a prototype doesn't work properly. Duh. quote:
I fully expected Avida to behave the way it did, that's why I cranked up the mutation rates to prove Royal's point here:
quote:
--------------------------------------------------------------------------------
(3) Critically important functions for “life” are difficult to destroy by mutations • All functions necessary for an organism to replicate and survive are carried out initially by only 15 instructions. The physical machinery to transcribe, translate and perform metabolic processes are not coded genetically and thus not subject to mutational damage.
--------------------------------------------------------------------------------
The Avida group would dispute that characterization by Royal (and myself) unsurprisngly, but the fact I could crank up the mutation rate and these organisms kept replicating shows that Royal is right. I fully expected that behavior from Avida given what I know of information technology. That's exactly why I conducted the experiment.
Here is a funny situation in that both I and the Avida authors will claim Avida behaved as expected, yet reach different conclusions as to it's significance.
As I understand it, the AVIDA group would dispute that characterization because it's false. There is no inherent protection against mutation damage in AVIDA critters. For all mathematically possible mutation probability values (between 0 and 1) AVIDA performs correctly. At mutation rates of 0.5, contrary to what you originally believed you had found, AVIDA organisms die. Neither you nor Royal could "crank up the mutation rate and these organisms kept replicating". What you found is that for certain mathematically impossible mutation probability values (namely, 50,000), instead of crashing the program resets itself to mutation probability 0. This does not mean however that AVIDA organisms are protected from high mutation loads. AVIDA most definitely does NOT "circumvent critical considerations with respect to the propagation of destructive entropy". quote:
It's actually fairly easy to write and design a complexity generator in software ( essentailly a random noise generator being fed through some sort of filter which extracts and preserves interesting results). That's all Avida ultimately is, I'm afraid.
Yes, and that's how evolution works, I'm afraid. That the filter is provided by the interaction of organisms with their everchanging environment is the beauty of it. No one is claiming otherwise. quote:
The issue, ultimately, is: do selective pressures in the real world really exist to create LARGE increases in complexity.
Sure. That's why evolution occurs gradually. Do you think there is a theoretical limit, outside computing power, in the complexity that an AVIDA-like program can generate by gradual increases? Do you think one could in principle design an AVIDA variant that generates organisms twice as complex as the current ones? Ten times as complex? Ten thousand times as complex? quote:
My position is increases in function complexity are the exception rather than the rule. Natural Selection tends to erode away if not totally extinct complexity.
That's just a statement - neither you nor anyone else has shown that NS necessarily behaves that way. AVIDA empirically shows it doesn't.
Within AVIDA, non-directed mutation and selection mechanisms act to increase complexity and information content. This fundamental conclusion stands.
IP: Logged
|
|
|
Printer-friendly view of this topic