|
Author
|
Topic: Evaluation of neo-Darwinian Theory with Avida Simulations
|
Salvador T. Cordova
Member
Member # 959
|
posted 13. August 2004 17:52
quote:
At mutation rates of 0.5, contrary to what you originally believed you had found, AVIDA organisms die.
Nope, they lived and reproduced.
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 13. August 2004 19:07
quote:
cd:
At mutation rates of 0.5, contrary to what you originally believed you had found, AVIDA organisms die.
STC:
Nope, they lived and reproduced.
The following is the explanation by Evan Dorn, who works with Dr. Adami (one of the creators of AVIDA) of Salvador's findings, from this ARN thread (emphasis mine): quote:
Experimental Result #1 (Pmut = 0.5)
We were given a copy of Mr. Cordova's actual experimental results for examination. In this run, he set POINT_MUTATION_RATE to a probability 0.5 of random mutations per genome site per update, and turned off all other sources of mutation.
Examining his output files:
Item 1) The directory "genebank" contains only one new file, named 020-aaaaa, which contains a copy of the ancestral genotype. New genotypes are saved in this directory as soon as three copies appear in the population. Since none were saved, it implies that no new self-replicating genotypes arose during the experiment. In a normal run, this directory will fill with hundreds or thousands of files.
Item 2) The last line of the output file "totals.dat" contains these values:
50000 -560808571 98513 98513 1 1
(Update, total # of instructions executed, total # of creatures, total # of genotypes, total # of threshhold genotypes, total # of species)
# of instructions is negative because the number grew so large that the long integer which stores that number rolled over. See Experimental Result #2 for an explanation of binary integer rollover.
# of creatures and # of genotypes are equal, # of threshhold genotypes = 1. This mean every execution of the "divide" command created a new and unique genotype, presumably by splitting a random string in two at a random location.
Item 3) The first appearance of a new "cell", as evidenced by "totals.dat", did not occur until update 440. The second happened at update 1100. In normal, living runs, even at high (i.e. near lethal) mutation rates, reproduction occurs within only a few updates.
Explanation: Put simply, the population was dead. What Mr. Cordoba observed was slow, random activity of non- replicating random strings.
Whether or not the strings of instructions existing in grid cells genotypes are valid, replicating organisms, Avida will dutifully execute their instructions. In this case, the ancestor organism was thoroughly randomized in the first update, as would be expected at Pmut=0.5. However, the execution of random computer instructions will occasionally include the execution the "allocate" instruction, creating new room at the end of the "genome" if other instructions happen to have put a non-zero number in the appropriate register. Also, the occasional "divide" will execute when other instructions have loaded integer values into the appropriate register ... meaning some piece of the randomized genome will get split into another grid cell. Without the actual code for organized replication, this will occur very slowly and all of these divides will contain random-length groups of random instructions. While yes, this process will slowly fill the grid cells, it is in no way analogous to life: nothing is replicated with fidelity, and no information is conserved.
Conclusion: Mr. Cordova has experimentally demonstrated that there is in no mechanism protecting genomes from randomization by killing the ancestor organism with a high mutation rate and observing the "brownian motion" activity of dead, random, sequences of computer instructions.
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 14. August 2004 13:57
Dead organisms can reproduce offspring?
If one accepts that characterization, Avida has now supposedly proven dead organism can reproduce offspring through "brownian motion" of random computer instructions.
Now, in respect to Mr. Dorn who is one of Dr. Adami's graduate students, I'm trying to refrain from engaging Mr. Dorn directly and limit my direct commentary on his posted statements. If Dr. Adami or Schneider or Lenski or Penncock visit, I will address their responses more pointedly.
ISCID would very much welcome their participation.
IP: Logged
|
|
charlie d.
Member
Member # 159
|
posted 14. August 2004 15:06
Salvador, you are grasping at straws here. Evan explains pretty clearly why what is observed at mutation loads 0.5 is not replication. I am sure he would not mind listening to your counter answers, if you had any.
IP: Logged
|
|
Scott
Member
Member # 1222
|
posted 14. August 2004 17:26
charlie d,
I can understand what you are saying, but I can also understand why Salvador is questioning the results. So perhaps some further clarification would help resolve the dispute.
The results show new "cells" being formed, but they are not really new cells, they are an artefact of the program.
The results show 98,513 "organisms," but they aren't really organisms in any conventional sense because none were capable of self-replication.
The results show 98,513 "genotypes," but these were all "failed" genotypes because they did not code for any sort of successful organism (in the sense of being able to self-replicate). They were pretty much random attempts at getting a start on life, all of which failed.
Do I have that about right?
Now, how do we know that none of these 98,513 genotypes were capable of self-replication?
There were no new genotypes, because none were saved, and yet there were 98,513 new genotypes. Where did they come from?
Why were none of the 98,513 saved? It's not because there weren't any new genotypes, it's because, apparently, there were not three which were the same. Considering the high mutation rate, this is understandable, but does the conclusion then follow? I don't see how it does.
What is the difference between the "divide" instruction and self-replication?
It's too bad that the mutation rate must remain constant. Surely it has changed throughout the history of self-replicators.
Salvador, an interesting experiment might be to attempt to identify the maximum and minimum permissible mutation rate which will lead to new successfully replicating organisms (at least three copies), if that has not already been done. Feel free to address my questions to Mr. Dorn over at ARN.
regards
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 15. August 2004 04:31
While I'm not actively participating in this thread, I'll briefly address one question. Scott asked quote:
Now, how do we know that none of these 98,513 genotypes were capable of self-replication?
There were no new genotypes, because none were saved, and yet there were 98,513 new genotypes. Where did they come from?
Why were none of the 98,513 saved? It's not because there weren't any new genotypes, it's because, apparently, there were not three which were the same. Considering the high mutation rate, this is understandable, but does the conclusion then follow? I don't see how it does.
I don't know about all 98,513, but in a replication of Salvador's 0.5 point mutation rate run I paused the run after 3,000 cells had been filled by the 'brownian motion' and checked the code in a sample of 100 of the cells. Not one of those 100 cells contained code that could reproduce. While that does not establish that none of the 98,513 creatures in Salvador's run was incapable of replication, it provides strong evidence that none were.
The apparent discrepancy (and I emphasize "apparent") is that new viable genotypes are saved, but cells can be filled with junk by the random brownian motion. If a genotype is non-viable, in the sense that it cannot reproduce, it isn't saved out. In Scott's summary quote:
There were no new genotypes, because none were saved, and yet there were 98,513 new genotypes. Where did they come from?
As I read that, the first use of "genotypes" refers to viable genotypes while the second use refers to cells with (non-viable) code in them. Dorn's use of "cell" in two places, including one where he uses the phrase "grid cell genotypes" supports my interpretation. The avida shell executes the instructions in every cell that has an instruction string associated with it. Occasionally those randomized instructions fill a new cell, but the code in the newly filled cell is not a viable critter. In this respect avida does not resemble biology. Dead biological organisms decay; dead avida organisms display random "brownian motion" associated with the execution of randomized instructions. That does not affect the utility of avida as a research platform when various parameters are within those found in biological systems.
RBH
[ 15. August 2004, 04:37: Message edited by: RBH ]
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 11. September 2004 06:47
[charlie d.] Posted July 20:
quote:
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.
[Royal]: I do not claim Avida alleges to simulate biological evolution in detailed accuracy. But several papers in biological journals are claiming that Avida experiments have indeed demonstrated neo-Darwinian processes have and can create real, complex biological novelty.
Consider Richard E. Lenski, Charles Ofria, Robert T. Pennock & Christoph Adami, “The evolutionary origin of complex features”, NATURE , 423( 8), MAY 2003: 139-144
“A long-standing challenge to evolutionary theory has been whether it can explain the origin of complex organismal features.”
And what do the authors then claim?
“These findings show how complex functions can originate by random mutation and natural selection.”
What are the authors trying to accomplish with their Avida platform?
“To examine the evolutionary origin of a complex feature in much greater detail than has previously been possible, we have performed experiments with digital organisms”
Are the principles and claims irrelevant to simulation of biological evolution as charlie d. states above? This does not seem to square with what various authors are claiming:
“By using this tractable system, we aim to shed light on principles relevant to any evolving system.”
I argued in my essays that such conclusions are not justified on the basis of most reported Avida runs. If we modify some parameter settings slightly and monitor outcomes we get a different picture.
Some neglected realities include:
- the effect of rapid genome truncation for miniscule, rapidly reproducing organisms
- the necessity of vastly greater amounts of genetic material which is subject to mutational damage (one cannot damage the Avida shell program via mutations!)
- the degradation of complex functions when (a) selectivities are lowered to realistic values and (b) when the proportion of sequences coding for slightly inferior functions is increased.
The net effect of mutations in the presence of realistic degradation pathways is to less effective functions and eventual loss of many of them. Avida assumes all-or-nothing: a mutation destroys 0 or 100% of the (logic) function, and in the later case most researchers assign a dramatic penalty to ensure the slightly inferior lineage goes extinct.
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 11. September 2004 07:30
[charlie d.] Posted Aug. 6
quote:
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 “irreducible complex” structures still stands.
And earlier:
[charlie d.] Posted July 20
quote:
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.
[Royal]: I must respectfully but firmly disagree. I have already explained why such details are critically important. For example, whether the net outcome of a financial feasibility analysis is profit or loss demands an accurate accounting of income and expense factors. A flawed model and/or parameter settings could "prove" either.
[Salvador] pointed out several times that tautological arguments are being used. Let me elaborate.
One can easily design computer programs to iterate and converge to a desired goal. (A large number of numerical analysis algorithms are designed to do just that). This can be set up so that the intended goal is essentially guaranteed. This is all Avida does: a series of letters are modified in computer memory according to a computer program to produce a planned outcome in spite of considerable stochastic effects.
This may be the point Salvador made earlier:
[Salvador] (July 19):
quote:
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.
One cannot simply imply the computer results are suitably reflective of a process of interest, such as neo-Darwinian theory, and claim this proves the feasibility of the later theory.
Let us be ridiculous to ensure clarity. I will “prove” everyone in the world can become rich by spending an hour a day at a casino gambling. I create a program which presents 3 buttons on the screen and the user has one mouse click per iteration. If the correct one is guessed, $1 million is added to a counter. Incorrect guess are penalized with $1. It costs $2 every time one repeats the game. The outcome of many runs leads to my intended results.
Of course I agree that “this is not true in all details”, “it only illustrates the principle”, “once in a blue moon someone could go bankrupt playing the game”, “it is not a real simulation”, “it is only designed to permit studies of gambling phenomena”, and use such arguments Avida defenders use. Suppose someone publishes such work in professional journals and claims to have demonstrated a solution to the world’s poverty. I would then point out once again what I have stated about many Avida claims: the computer platform has not been mapped to real-world realities (“how are casino running costs financed; who pays the salaries; how realistic is the reward vs. penalty payback proportion).
The program on average leads to the intended outcome but the relevance to what is supposed to be proved is deficient.
Humor aside, mutation/selection evolutionary processes have NOT been shown to be capable of generation new complex functions in real, cellular genetic systems.
As I have pointed out repeatedly. Coding genes must code for the real proteins (with constraints which make a random origin extraordinarily unlikely) which then associate into molecular machines necessary for transcription, translation, DNA duplication, metabolic biochemical chains, cell replication, repair processes, etc.
As I showed by real Avida runs, the proportion among random instructions strings which can code for new computer logic functions is huge. The proportion of random polypeptide chains able to perform a new, valuable function is extremely minute. Most truly new functions actually require multiple new proteins concurrently before any selective advantage can accrue.
The information in real cells is physically coded genetically. All these processes are provided in Avida by an external C++ based shell which is 100% isolated from the instructions in memory which are mutating. Mutations cannot destroy these processes in Avida, but can in the real world.
My discussion partner alludes to this when he wrote:
[charlie d.] posted Aug. 13
quote:
Whether or not the strings of instructions existing in grid cells genotypes are valid, replicating organisms, Avida will dutifully execute their instructions.
Dr. charlie d. still believes:
[charlie d.] Posted July 20
quote:
Until then, AVIDA successfully invalidates the claims that non-teleological evolutionary mechanisms cannot lead to complexity (including IC) and information gains.
I can only repeat, building a computer “funnel” which converges to desired outcomes in the presence of considerable randomizing effects is inadequate for such a claim. I concurr entirely with the insight someone else had:
[Nelson-Alonso] Posted July 21
quote:
“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?”
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 11. September 2004 07:55
[charlie d.]:
quote:
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. 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.
[Royal] : I believe the “No Free Lunch” theorem is being alluded to here.
Let me pretend to “prove” how “natural processes” can generate columns of vertical coins with highly improbable alternating runs of heads (H) and tails (T).
I build a computer program in which conceptually a funnel guides a coin down to a restricted surface. A random number generated produces H or T and is dropped down the funnel. This is repeated for a total of 3200 different column “organisms” distributed over a non-overlapping surface. The program returns to organism # 1. If the next coin is different from the one beneath (TH or HT), the software replicates that column and kills off one of the competing columns if one is found with TT or HH. This provides Lebensraum for better lineages. Otherwise the first organism dies. The selection of better columns in all cases is not deterministic but with a high probability < 1.
This is repeated for all 3200 lineages with additional coins. As alternating HTHTHTH… or THTHTHT… get longer, I reward such columns getting an addition “correct” new coin with ever increasing number of progeny.
In this manner I deterministically ensure via the funnel that coins drop in acceptable locations by external resources [Avida’s non-genetic resources smuggled in do conceptual replication, metabolism, etc. deterministically with no possibility of misshap]. My program merely provide a selection algorithm which will often lead to large series of HTHT… or THTH… sequences. Avida also does nothing more than select desirable sequences in computer memory according to an algorithm.
Suppose I now claim “natural processes” could form such columns. Surely everyone would argue I have hardly proven my point with the computer program described.
Complexity and information have indisputable been smuggled in by the design of the Avida platform and careful selection of parameter settings. One could modify such programs to show that 3200 organisms with large genomes would be killed off in a few generations. Or alternatively that logic functions would degrade and disappear over time. This is not a reflection of biological realities but of computer software design choices.
Others have identified other examples of how the Avida software is biologically unrealistic:
[Salvador]: Posted Aug. 14
quote:
Dead organisms can reproduce offspring?”
Scott was apparently surprised at what others are claiming:
[/B] [Scott]: Posted Aug. 14[/B]
quote:
The results show new “cells” being formed, but they are not really new cells, they are an artefact of the program.
The results show 98,513 “organisms,” but they aren’t really organisms in any conventional sense because none were capable of self-replication.
RBH recognizes such difficulties but has not carried his reasoning far enough to conclude the Avida outcomes are inherent and as Salvador points out, tautologically inevitable, in the run designs:
[RBH]: Posted Aug. 15
quote:
The avida shell executes the instructions in every cell that has an instruction string associated with it.” “In this respect avida does not resemble biology. Dead biological organisms decay; dead avida organisms display random “brownian motion” associated with the execution of randomized instructions.
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 11. September 2004 10:12
Welcome back Royal,
I found a few more typos in the second paper, I'll try to send the corrections to you.
-----------------------
quote:
Charlie D wrote:
Salvador, you are grasping at straws here. Evan explains pretty clearly why what is observed at mutation loads 0.5 is not replication.
A minimal Avida genome has instructions. The instructions are executed by the physical computer and operating system and infrastructure. There is an infrastructure which makes the genome of Avida viable and executes the instruction in the Avida genome.
In a real biological organism the physical infrastructure such as the proteins and replication machinery also have corresponding genes from which they are coded from. In a real organism there is biological hardware and software to do the transcription and duplication and carry out the instructions in the DNA. All of the hardware and software to do this has a corresponding section in the DNA as well. All of these genes are subject to mutation in the real world.
Avida does not even model these genes, much less subject them to mutation! That was the point of the experiment I ran. That's why I could effectively bombard the Avida genomes and dead creatures could still make offspring. It totally confirms this statement in Royal's paper:
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 fact that the Avida group left an embarrasing and misleading artifact in their own code was just icing on the cake for the Avida critics. The important part of the issue was clouded by the volume of posts on their misleading comment in the genesis file which they have said they will fix.
The real problem was not in the documentation or that there is no bounds check on the mutation rate (both indicative by the way of sloppy work, imho), but a fundamentally incorrect way of modeling the very problem they are trying to examine.
Avida was created to help understand the problem of entropy propagation, and off the bat they fail to model a critical aspect of entropy. Not only do they not model the genome of the replicating hardware, they don't even expose it to mutational variation. They avoid modelling the very thing they set out to examine.
The proper way they should have modeled the simulation is to expose the genomes which create the replicating hardware to mutation. Had they done so, there results would be more realistic.
I would venture to say, had they done that, given combinatorial considerations alone, modelling the 99.9% of the genome to mutation which they failed to do would at best result in evolution requiring time beyond the life of the universe and at worst no evolution at all.
I have seen experiment with Avida where we start off with gigantic genomes genomes and watch them contract to the optimum size. I believe the optimum size shrinks as mutation rate increases. This is essentially the Eigen hypothesis which I believe Avida has confirmed. However, now if we factor in the 99.9% of the genome Avida failed to model, what will we expect as an outcome?
Avida explored the Eigen limits of genome length. In the spirit of Brainstorms, I postulate that the majority of genomes are actually operating beyond the Eigen limit already, and thus genomes I expect will deteriorate over time and not increase in functional complexity, contrary to the results of Avida. This is in principle a scientifically and empirically falsifiable statement. Whether such an investigation is economically feasible is another story.
Avida succeeds to the degree it avoids modeling real world facts. At ARN I suggested an empirical approach would be to observe the contraction of biodiversity in the world today and track genomic deterioration in species. I have made a falsifiable prediction in priniciple that genomes will deteriorate and biodiveristy will contract over time. We have reason to believe this is in fact the case.
The superior model to Avida is the real world itself, and I believe the real world is empirically telling us information is being lost not gained.
Salvador
[ 11. September 2004, 10:13: Message edited by: Salvador T. Cordova ]
IP: Logged
|
|
Scott
Member
Member # 1222
|
posted 11. September 2004 14:18
quote:
The proper way they should have modeled the simulation is to expose the genomes which create the replicating hardware to mutation. Had they done so, there results would be more realistic.
It has already been admitted that the simulation is not entirely realistic, and indeed quite probably cannot be so. So I don't see how claims that this or that modification to the program could make the program "more realistic" are a valid criticism.
The question then is whether this "flaw" invalidates any results derived from the simulation. I fail to see how it would. The only way it could would be if claims are made that AVIDA in some way demonstrates how replicating "organisms" can arise.
Why is there a problem with assuming a certain level of functionality, as we do when we study living organisms, and then simulating whether given that certain level of functionality, additional functionality might be evolved?
I also am not convinced that exposing "the genomes which create the replicating hardware" to mutation would help. We could assume that such mutations would almost inevitably be harmful and that such genomes would be selected against, and that the typical non-mutated genome would then dominate, leaving us right back where we started. So as with any simulation, we make a decision that this is not what we are interested in studying and it would merely clutter up the simulation with no real benefit. Again, one would need to show why choosing not to model this aspect of evolution invalidates the results taht we do achieve.
If we are to successfully criticise AVIDA, we need to stick with what it does, I think, and not what it does not do. What it does not do is only important if something it does not do in some way invalidates what it does. I don't think I have seen any demonstration yet which shows this.
But if someone wishes to argue that there is something the program does not do, which it should, and the absence of which invalidates all program results, or some result in particular, perhaps we could discuss those on a point-by-point basis.
AVIDA and Information
It is claimed that AVIDA can increase information. Can someone address how this information is defined and measured?
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 13. September 2004 18:49
Hi Scott,
I appreciate your question, especially knowing that on balance your posts at ARN seem ID friendly.
quote:
It has already been admitted that the simulation is not entirely realistic, and indeed quite probably cannot be so. So I don't see how claims that this or that modification to the program could make the program "more realistic" are a valid criticism.
It is not a matter of "not entirely realistic" it is a matter of grossly unrealistic. Avida is good at modeling aspects of population genetics and phylogenic reconstruction. However, to claim it solves the information problem is stretch. It creates information because it is designed to do so.
quote:
AVIDA and Information
It is claimed that AVIDA can increase information. Can someone address how this information is defined and measured?
The Avida authors have not solved this problem either! They have admitted they haven't been able to arrive at useful metrics for complexity, which is basically a metric for information. The problem is epistemological. I don't think most have realized nor appreciated how subjective information theory is. Subjective mentalist constructs are deeply a part of it, unlike almost any other measurement of physical quantities. The subjectivity does not mean information theory is unscientific, but it puts an unusual perspective on the issues.
We could give a metric in terms of functional bits. We could write a simpler program to get the same or betters results if one just wanted to generate functional code. Avida goes through enormous stochastic gyrations to implement an otherwise simpler program.
Given your Java skills, I'm sure with some effort your could write a program that writes software that ends up creating more bits than the originating program. For that reason, Avida is a completely unremarkable exercise from a computer science standpoint, imho.
One can claim information increase simply by the way they define the information boundaries. There is a lot of fluidity in information theory, thus I advise the IDists NOT to say information increase is impossible. Even though I personally believe in Law of Conservation of Information (LCI), applying it in practice is too unwieldy. So even if LCI is true, applying in practice is too difficult.
quote:
But if someone wishes to argue that there is something the program does not do, which it should, and the absence of which invalidates all program results, or some result in particular, perhaps we could discuss those on a point-by-point basis.
I can't speak for the authors in totality, but Avida has been used to argue the viability of Darwinian evolution. In prinicple it's possible, but again, there is a lot of tautology. "If the circumstance are right for Darwinian evolution to create massive information increase, then massive information increase will happen."
Now, one interesting thing in all this is the Eigen limit. The more functionally integrated information an organism has, the lower it's mutation rate must be for it to have reached that state. But the lower the mutation rate, the slower the upward evolution. Thus there are two competing constraints against the substantial evolution of more complexity. More complexity requires more mutations, but that is constrained Eigen limit. Lower mutation rate requires more time, and some people therefore think there is not enough time.
quote:
I also am not convinced that exposing "the genomes which create the replicating hardware" to mutation would help. We could assume that such mutations would almost inevitably be harmful and that such genomes would be selected against, and that the typical non-mutated genome would then dominate, leaving us right back where we started. So as with any simulation, we make a decision that this is not what we are interested in studying and it would merely clutter up the simulation with no real benefit. Again, one would need to show why choosing not to model this aspect of evolution invalidates the results taht we do achieve.
The simulation achieves phenomenal evolution. In a few thousand generations it can increase the genome by a factor of 10 times. That's like going from bacterium to man in a few tens of thosands of years. The issue is one of how fast evolution can really work.
We actually don't know what would have happened if the mutation rate is allowed to propagate (through to the parts I outlined above), how at risk would be the entire population toward extinction? If it had there is always the chance of population would encounter extinction. We do know from fruitfly radiation experiments this is a possible outcome.
I have not said that Darwinian evolution is utterly impossible to create LARGE increases of information, probably virtually impossible is the right word. I just don't think from an empirical standpoint it could have happened in the real world. It is not a matter of "argument from ignorance" but rather "proof by contradiction."
I believe a good key to creating a simulation of the past is observing evolution today. That is the best simulation possible, albeit imperfect, but better than any that our computers will ever be able to do.
[ 13. September 2004, 20:56: Message edited by: Salvador T. Cordova ]
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 16. September 2004 13:28
Royal Truman and I are not the only ones to question the validity of Avida or it's cousins. Avida is research into Artificial life or Alife. Quoted from Discovery Institute
quote:
34. Rodney Brooks, “The relationship between matter and life,” Nature 409 (2001): 409-411.
Rodney Brooks of the Artificial Intelligence Laboratory at MIT has long been a pathbreaking investigator in the construction of “AI” (artificial intelligence) and “Alife” (artificial life) systems. In this skeptical article, however, Brooks steps back from the bench to look critically at what AI and Alife research has actually demonstrated. He writes:
...both fields have been labelled as failures for not having lived up to grandiose promises. At the heart of this disappointment lies the fact that neither AI nor Alife has produced artefacts that could be confused with a living organism for more than an instant. AI just does not seem as present or aware as even a simple animal and Alife cannot match the complexities of the simplest forms of life. (p. 409)
The failures of these fields, Brooks argues, requires a diagnosis:
We build models to understand the biological systems better, but the models never work as well as biology. We have become very good at modelling fluids, materials, planetary dynamics, nuclear explosions and all manner of physical systems. Put some parameters into a program, let it crank, and out come accurate predictions of the physical character of the modelled system. But we are not good at modelling living systems, at small or large scales. Something is wrong. (p. 410)
After considering several modest “fixes” for AI and Alife (e.g., incorrect parameters, lack of computing power, lack of complexity in models), Brooks turns to a more challenging diagnosis: “we might be missing something fundamental and currently unimagined in our models of biology” (p. 410). He argues:
We would then need to find new ways of thinking about living systems to make any progress, and this will be much more disruptive to all biology. ... So what might be the nature of this unimagined feature of life? One possibility is that some aspect of living systems is invisible to us right now. The current scientific view of living things is that they are machines whose components are biomolecules. It is not completely impossible that we might discover some new properties of biomolecules or some new ingredient. One might imagine something on a par with the discovery of X-rays a century ago, which eventually led to our still-evolving understanding of quantum mechanics. Relativity was the other such discovery of the twentieth century, and had a similarly disruptive impact on the basic understanding of physics. Some similar discovery might rock our understanding of the basis of living systems. (p. 410)
[ 16. September 2004, 13:37: Message edited by: Salvador T. Cordova ]
IP: Logged
|
|
Royal
Member
Member # 1060
|
posted 28. September 2004 11:03
[Scott] Sept. 11, 2004
quote:
I also am not convinced that exposing "the genomes which create the replicating hardware" to mutation would help. We could assume that such mutations would almost inevitably be harmful and that such genomes would be selected against, and that the typical non-mutated genome would then dominate, leaving us right back where we started.
[Royal]: Suppose a mutation in a stretch of DNA became the first step in developing a totally new gene. Use a selectivity factor of 0.01 and notice from my essay, Part 1, that a very large number of generations would be needed before even 10 descendants would build up.
A minimal size, free-living (non parasite) bacteria may have on the order of 500 genes: http://www.icr.org/pubs/imp/imp-340.htm This includes the replicating hardware, etc.
The odds of a random mutation being deadly or deleterious among these 500 genes is much greater than that a new mutation will improve upon this gene set. See:
Truman, R., “Protein mutational context dependence: a challenge to neo-Darwinian theory: part 1, TJ 17(1): 117-127, 2003
http://answersingenesis.org/home/area/magazines/tj/docs/v17n1_proteins.pdf
This means that the new lineage of very few members for many, many generations, would on average become extinct or degrade on average thousands of times, for each new useful mutation occurring which improves the gene being developed.
I pointed out in Part 2 that Avida does not provide a graceful degration path: loss of a logic function is an all-or-nothing proposition, dramatically penalized in published reports (typically 2 to 32 times slower reproduction each generation!). In reality, mutations can often lead to suboptimal proteins which still work. Different suboptimal genes would accumulate throughout the population, decreasing the average competitiveness. Further degradation would sense ever decreasing natural selection.
Each of the 500 genes can slowly become suboptimal if natural selection is truly concentrated on creating a new gene.
I showed in Part 2 that use of more realistic reward levels in Avida parameter settings leads to repeated extinctions of better lineages (genetic drift). My arguments above now indicate we must additionally include the hundreds of other genes in which mutations would be damaged in the real world. An occasional useful mutation just is not going to lead to truly novel complex features, which I have repeatedly pointed out require multiple new genes concurrently anyway.
IP: Logged
|
|
|
Printer-friendly view of this topic