Uncommon Descent Serving The Intelligent Design Community

Michael Yarus and the Thing that Couldn’t Die

Share
Facebook
Twitter
LinkedIn
Flipboard
Print
Email

MSTMichael Yarus, an emeritus professor at UColorado,  is one of the leading experts on the RNA World hypothesis, which takes the origin of life as flowing from RNA chemistry. His recent book with Harvard UP, Life from an RNA World, contains lots of material responding to ID, though without basic understanding, to say nothing of nuance.

The reason I bring the book up here, however, is to note his extensive use of Dawkins’ famous METHINKS IT IS LIKE A WEASEL evolutionary computing simulation. Yarus changes the target phrase to NOTHING IN BIOLOGY MAKES SENSE EXCEPT IN THE LIGHT OF EVOLUTION, but the essence of Dawkins’ simulation is nonetheless there in all its glory — indeed, Yarus develops this “instance of evolution” more extensively than Dawkins did. Moreover, Yarus sees this simulation as underwriting the power of evolutionary processes.

Dawkins’ simulation has come under considerable criticism both here at UD and at the Evolutionary Informatics Lab, where we have implemented “WEASEL WARE” (go here). Some internet critics have urged that we are beating a dead horse, that this example was never meant to be taken too seriously, and that if we were “serious scientists,” we would be directing our energies elsewhere. Let me suggest that these critics take up their concerns with Yarus.

The reason we keep bringing up Dawkins’ example is because evolutionists themselves won’t let it die. You can find Yarus’ discussion of it beginning on p. 64 of his book. It is available at Google Books here. Or you can view it below:

Comments
Isn't the idea of the "allele" outdated? Isn't it true that such a simplistic conception of genes and their related traits does not reflect reality? Isn't the relationship between genotype and phenotype barely understood to this day? Aren't "alleles" only really useful as a way to begin to teach undergraduates basic genetics?Phaedros
June 9, 2010
June
06
Jun
9
09
2010
02:59 PM
2
02
59
PM
PDT
Petrushka:
But ID has the logically impossible task of proving there are no possible viable organisms with genes or alleles bridging that gap.
Nonsense. ID presents that as an observation and an inevitable result of chemistry and statistics. All you need to do to prove ID wrong is to fill in the blanks in a single case that has been presented as "irreduceably complex". All swans are white. Find me the black one. It's been done with swans. It hasn't been done with biology yet.SCheesman
June 9, 2010
June
06
Jun
9
09
2010
02:38 PM
2
02
38
PM
PDT
Moreover, there are lots of very likely d)s: as I have repeatedly said, a few thousands of independent protein domains which cannot have originated by RV...
You mean they appear not to have a common protein ancestor. I have no way of evaluating that, so I'll pass on commenting. What you call integration I call spaghetti code. If I, as a programmer, write functions that reads differently depending on the entry point or frame shift, I get fired. Or my company eventually goes broke. Eventually programs require maintenance, and density of meaning becomes a problem rather than a sign of cleverness. So if living things do this, they limit their ability to track environmental change. Raise or lower the average temperature five degrees and you get widespread extinctions.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
02:22 PM
2
02
22
PM
PDT
Petrushka: The problem with asserting that “d” exists is that neither evolutionists nor ID proponents have an actual history of such an event. The same problem exists with asserting c). Moreover, there are lots of very likely d)s: as I have repeatedly said, a few thousands of independent protein domains which cannot have originated by RV, and for which there is not even a pale theory for a step by step transition, and all possible evidence against it. And remember, science is about best explanations, not about infinite faith that some day perhaps we could find something which can save us from all our present inconsistencies. And I don't agree about the ID "gaps" becoming narrower with research: research is increasing the level of complexity, of integration and of design present in living beings. In the last few decades, known functional complexity in biology has increased of many orders of magnitude. While materialistic explanations for it are still where they were in the beginning: a useless theory which has no internal consistency and no explicatory power. Micorevolution is still the only thing neo-darwinism can explain.gpuccio
June 9, 2010
June
06
Jun
9
09
2010
02:06 PM
2
02
06
PM
PDT
Petrushka: Asserting that there are living examples of “d” seems to be at the very heart of the ID movement. The irreducibly complex structure. Something that jumped the unbridgeable gap. Here you understand ID correctly.gpuccio
June 9, 2010
June
06
Jun
9
09
2010
01:57 PM
1
01
57
PM
PDT
The problem with asserting that "d" exists is that neither evolutionists nor ID proponents have an actual history of such an event. What you have with structures like the flagellum is a molecular equivalent of a gap in the fossil record. But ID has the logically impossible task of proving there are no possible viable organisms with genes or alleles bridging that gap. The problem gets more difficult with research, because there are dozens of kinds of cilia and flagella, and it turns out there are only a couple of proteins that are both necessary and common to all these organisms. The problem with the gaps argument is that the gaps get narrower with research.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
01:53 PM
1
01
53
PM
PDT
Arthur Hunt (#35): I have read your essay and, quickly for the moment, Yarus' paper. I give you some first impressions: 1) I appreciate your essay, although obviously I don't agree on many things. Some of them are evidently due to our different positions, and it would probably be useless to discuss them as such. But a first point I would like to make is that IMO your emphasis about the central role of the origin of the genetic code in Meyer's book is excessive. I have read the book, and if I had to summarize it in a few words, I would say that the central theme is OOL in all its many aspects. The origin of the genetic code is certainly one of them, but not certainly the only one, and not necessarily the most important, neither in reality nor in Meyer's book. 2) About the Yarus paper, while it can be of some interest, I must say that I am not impressed. Its methodology and its analysis and its conclusions are not convincing, deeply artificial, and obviously motivated by a pre-existing ideology: defending the hypothesis of the RNA world. I have already expressed my "perplexities" about this approach, and I confirm them here. 3) Anyway, some more specific comments are due. I think that Yarus has not proved his thesis in any way (and, I believe, you agree with that in some measure). However, you say that Meyer's statement that there is no biochemical law which binds the genetic codons to the corresponding aminoacids has no supporting evidence, but that's simply the truth of what we observe today, of how the genetic code, transcription and translation work in the real world. In the real world, the codon (or anticodon) do not interact with the aminoacid. It's only the complex of tRNA and aminoacil-trna-synthetases which holds the key to the decoding of the genetic information. These are facts. Yarus' hygpotheses are just hypotheses of how the present situation could have had origin through some transitory state where there was some biochemical connection between a minority of the present codons and some aminoacids. A vague and very incomplete scenario, based as usual on very artificial findings. 4) Even taking for good some of Yarus' findings (and I do that with many reservations), many questions remain regarding his interpretation of those results. a)First of all, only a minority of the codons have a special "enrichment" in the population of artificial RNAs and aminoacid binding sites generated by Yarus. He promptly interprets this fact imagining that those were the original codons, and that the rest was added after. That looks like a very ad hoc explanation. The fact remains that the vast majority of the codons of the genetic code for the 8 aminoacids tested had no special relationship with the corresponding aminoacids. b) Only 8 aminoacids were tested, out of 20. Only six had some relation with some of their codons. Yarus promptly infers that 75% of aminoacids "entered the code in this stereochemical era". A very risky inference, for an era which exists only in his imagination, and for numbers which are so small that do not allow any valid inference. If and when he will test all of the 20 AAs, we will know how many of them have the property he describes (not certainly how many of them entered the code in a stereochemical era of which we have no evidence). c) The results he describes can be interpreted in many different ways. There can be a reason why those specific codons are concentrated in those binding sites. There could be a biochemical reason, but that reason could have nothing to do with the code. It is not clear to me if the analysis has been performed for all 64 codons for each aminoacid. The results of such an analysis, if it was done, are not provided in the paper. The table only shows the statistic probability for the codons specific for each aminoacid, and not for all the rest. And, again, only a minor part of these codons have a significant association, And only in six AAs out of eight. There is more. As you certainly know, there are many different attempts, none of them definitive, to find second codes which can influence the overall behaviour of the DNA molecule beyond the simple coding of the primary structure of proteins. And proteins and nucleic acids certainly interact at many other levels other than translation. there may be biochemical reasons for the specific interactions found by Yarus, which have nothing to do with the genetic code. Or which are necessary for these "second codes", or for regulatory mechanisms. It would be interesting to know if the interactions described by Yarus can find some application and occurrence in what really happens today in existing cells, and not only in artificial RNA molecules created in the lab. d) One final note about the statistical analysis. I cannot comment on it in detail, because I have made no serious attempt to analyze it, and probably I will not (I really can't see any reason to give any priority to that). Moreover, papers rarely give you sufficient data to understand if the statistical analysis is credible, and believe me, many times it isn't. Considering how we arrived to this discussion, anyway, I am obliged to mention that, according to the understanding of probability shown by Yarus in the book which originated this thread, I definitely feel not very inclined to give him credibility in this field. But of course, I hope that at least for the paper he was helped by some competent statistician. :)gpuccio
June 9, 2010
June
06
Jun
9
09
2010
01:53 PM
1
01
53
PM
PDT
c) Any complex result which can be deconstructed into successive molecular steps, each of which is in the range of RV, and each of which is selectable, can be achieved. d) On the contrary, any complex result which cannot be deconstructed in that way cannot be achieved
All you are saying here is stuff that any biologist would agree with. 1.The standard model of evolution requires an unbroken chain of descent. 2. Since about 1940 the standard model of evolution requires that variations be small. No hopeful monsters. A lot has been learned since 1940 about molecular biology, but it is still true that individuals must find a mate among others of its species, and that limits the "size" of a mutation that can contribute to the gene pool of a population. Asserting that there are living examples of "d" seems to be at the very heart of the ID movement. The irreducibly complex structure. Something that jumped the unbridgeable gap.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
01:44 PM
1
01
44
PM
PDT
Thanks gpuccio, that was clear and to the point, and I will cite it because of its clarity.bornagain77
June 9, 2010
June
06
Jun
9
09
2010
12:56 PM
12
12
56
PM
PDT
Petrushka and others: About NS and GA. GAs can compute and find solutions. Nobody denies that. But NS is a very specific GA. Petrushka says: Evolution is embodied in imperfect replication plus fecundity. That’s it. There’s no code specifying goals. It’s an inevitable outcome of any system that replicates imperfectly and in which offspring have differential reproductive success. I perfectly agree with that statement, provided that for "evolution" we mean darwinian evolution. Pwetrushka's definition is bvery correct. There are many equivocations about NS, as though something external is selecting, as though some fitness function is evaluating a result. That may be true for human made GAs, but it is not true of the model of darwinian evolution. In that model, NS means only: a) we have replicators b) we have random variation in the replicating process c) if and when RV creates a replication advantage, that specific result of RV expands in the population d) when RV creates a sufficient replication disadvantage, that specific result is eliminated. This is the algorithm. Of this we must discuss, to understand what it can do and what it cannot do. Human made GAs are nothing like that. They create fitness functions, evaluate results, define specific searches. Nothing of that is in the original model. In the original model, the replicator selects itself by some real function which exploits the environment. In other words, it is the new function in the replicator which selects itself, in a specific environment, and not the environment which selects anything. That has an important consequence: only new functions which spontaneously give a replication advantage can be selected. Now, if we ask what the model can do or not do, the answer is: a) It can select anything which arises through RV, provided that the emerged result gives some replicatory advantage. b) Selected results can certainly cumulate, provided that each partial result can emerge through RV, and be selected. c) Any complex result which can be deconstructed into successive molecular steps, each of which is in the range of RV, and each of which is selectable, can be achieved. d) On the contrary, any complex result which cannot be deconstructed in that way cannot be achieved, The ID central point is: most complex results we observe in the genomes and proteomes are of the d) kind. Indeed, no single protein domain has ever been shown to be of the c) kind. IOW, GAs come in different flavours. Some, like the weasel, are only ill inspired propaganda. Others are useful and serious computational tools. But no human made GA says anything about the "spontaneous" GA which is modeled in neo-darwinism. So, if darwinists want to show what their model can really do, they should really analyze the RV + NS algorithm, and not others which are completely different.gpuccio
June 9, 2010
June
06
Jun
9
09
2010
12:15 PM
12
12
15
PM
PDT
Petrushka (#50): Please, don't put words in my mouth which I have never said. Your reading of Durston is completely wrong. Random proteins can be functional. Durston: "Abel and Trevors have delineated three qualitative aspects of linear digital sequence complexity [2,3], Random Sequence Complexity (RSC), Ordered Sequence Complexity (OSC) and Functional Sequence Complexity (FSC). ... As Abel and Trevors have pointed out, neither RSC nor OSC, or any combination of the two, is sufficient to describe the functional complexity observed in living organisms, for neither includes the additional dimension of functionality, which is essential for life [5]. FSC includes the dimension of functionality." Protein functionality can increase gradually due to mutation and selection. Here is the phrase in Durston's paper about which you equivocate: "In principle, some proteins may change from a non-functional state to a functional state gradually as their sequences change. Furthermore, iso-enzymes in some cases may catalyze the same reaction, but have different sequences. Also, certain enzymes may demonstrate variations in both their sequence and their function. Finally, a single mutation in a functional sequence can sometimes render the sequence non-functional relative to the original function." Here Durston is simply listing some common concepts which are the background for his work. Please note that he saya "in principle", that is "it is theoretically possible that". Indeed, the first statement, that "some proteins may change from a non-functional state to a functional state gradually as their sequences change", is simply the currebt assumption. Durston states it dutifully as a theorical possibility, exactly because his work has the purpose of trying to verify quantitatively at least some aspects of that assumption. Your reading of these words if incorrect and instrumentaL. Moreover, I must say that I find your use of the quoted material veri unfair. You know very well that I have presented the Durston paper as an essential example of how functional complexity can be really measured, and of what is the appropriate unit. Those were your questions. Durston gives the answers, and I have repeatedly re-formulated those answers for you and others. You have never commented on the sibstance of those answers, just avoiding them saying that you are not impressed by statistical arguments, or something like that. Instead, you continuosly misquote Durston attributing to him statements that he has never made, and which are not part of the pertinent work he has done. Papers, whoever the authors may be or may think, are important for the facts and the objective work they present. If you want to comment on a paper, take the time to understand it and to comment on the substance of what is presented there. Finally, in my post #43 I have given an explicit comment about your previous post. As usual, no comment has come from you on the substance of what I have said. I don't pretend that you agree, I don't pretend that you spend your time in a long discussion, but at least a brief acknowledgement of what others say in response to you, and some brief thought, would be appreciated.gpuccio
June 9, 2010
June
06
Jun
9
09
2010
11:49 AM
11
11
49
AM
PDT
One further question, why oh why would clumps of matter need to or want to reproduce in the first place? What is the point? Isn't one carbon atom, or a chain of them, just as meaningless and inert as any other chance arrangement of atoms?Phaedros
June 9, 2010
June
06
Jun
9
09
2010
11:23 AM
11
11
23
AM
PDT
Hmm Petrushka I guess you haven't considered all of the systems needed to sustain just one life and how extremely well they must work. Again an example of the extent to which people have to simplify life in order to accept an inadequate and dated theory.Phaedros
June 9, 2010
June
06
Jun
9
09
2010
11:18 AM
11
11
18
AM
PDT
"The idea of performance near the physical limits crosses many levels of biological organization, from single molecules to cells to perception and learning in the brain,,,," William Bialekbornagain77
June 9, 2010
June
06
Jun
9
09
2010
11:17 AM
11
11
17
AM
PDT
William Bialek - Professor Of Physics - Princeton University: Excerpt: "A central theme in my research is an appreciation for how well things “work” in biological systems. It is, after all, some notion of functional behavior that distinguishes life from inanimate matter, and it is a challenge to quantify this functionality in a language that parallels our characterization of other physical systems. Strikingly, when we do this (and there are not so many cases where it has been done!), the performance of biological systems often approaches some limits set by basic physical principles. While it is popular to view biological mechanisms as an historical record of evolutionary and developmental compromises, these observations on functional performance point toward a very different view of life as having selected a set of near optimal mechanisms for its most crucial tasks." http://www.princeton.edu/~wbialek/wbialek.html Physicists Finding Perfection… in Biology — June 1st, 2009 by Biologic Staff Excerpt: "biological processes tend to be optimal in cases where this can be tested." http://biologicinstitute.org/2009/06/01/physicists-finding-perfection-in-biology/bornagain77
June 9, 2010
June
06
Jun
9
09
2010
11:12 AM
11
11
12
AM
PDT
Petrushka,
The oracle simply grades each individual by total distance traveled.
Please, please, PLEASE do not contribute to the confusion by adopting the "oracle" terminology of Dembski and Marks. A "black box" function (finite domain and codomain) does not explain itself. Observed input-output pairs provide no information whatsoever as to the outputs associated with inputs yet to be supplied to the box. This is precisely why there is "no free lunch" for optimizers. The notion of Dembski and Marks that the black box supplies "warmer-colder" information, as in the child's game, is simply wrong.Sooner Emeritus
June 9, 2010
June
06
Jun
9
09
2010
10:38 AM
10
10
38
AM
PDT
Perfection is not an attribute of living things. Reproduction is just about the only attribute common to all living things. I'm not going to debate the origin of life, because neither side has any detailed record of the history of first life. Once there is a replicator, however, the issue of perfection is moot. So long as some descendants are able to survive it does not matter that some variants are non-viable. There are a bunch of possible red herrings. One is the assumption (unnecessary) that evolution is striving or searching for something. The simple fact is that a lineage either survives or it doesn't. If some variants are more successful than others, the population shifts in their favor. But shifting can also occur because the evolutionary algorithm dithers about a level of fitness without any obvious increase or decrease.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
08:37 AM
8
08
37
AM
PDT
Information is completely unlike a crytstal, however, and and no genetic algorithm can arrange letters into meaningful sentences because there is no benefit until the sentence is nearly “perfect”.
You'll need to take that up with gpuccio. He presented evidence on a previous thread that: Random proteins can be functional. Protein functionality can increase gradually due to mutation and selection. Durston was one of the sources.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
08:29 AM
8
08
29
AM
PDT
Petrushka:
The traveling salesman algorithm would work even if the cities moved or shifted position between trials.
A genetic algorithm finding the solution to the travelling salesman problem is like the formation of a crystal from a fluid. Things are shifted around in a manner which minimizes the energy, and the crystal is the natural and inevitable result. The shortest (or very nearly shortest) route is the inevitable consequence of the genetic algorithms' actions. Information is completely unlike a crytstal, however, and and no genetic algorithm can arrange letters into meaningful sentences because there is no benefit until the sentence is nearly "perfect". There are no short-range advantages to be gained in permuting random strings to other random strings, and the search space is so large that you cannot hope to locate an island of meaning in the sea of meaninglessness. The problem of composing the information in the DNA that undergirds life, indeed the process required to modify the code to achieve real evolutionary change (as opposed to shuffling around existing information, or expressing it differently via epigenetics) is in no way equivalent to the travelling salesman problem, and by extension genetic algorithms are fundamentally unsuited for such a task. Genetic algortithms are great for growing crystals. Only intelligence can create digital code.SCheesman
June 9, 2010
June
06
Jun
9
09
2010
07:27 AM
7
07
27
AM
PDT
There's one other possible outcome of a GA that hasn't seen much discussion. In case where there is no reachable target, only gradations of fitness, the algorithm can reach a certain level and stop improving. It simply dithers up and down in fitness. My itatsi program does this. The fitness scores improve dramatically for a while, then level off, bobbing up and down about a relatively high mean. The initial rapid rise in fitness followed by dithering is suggestive. Some of you have argued that in living things, we seldom see beneficial mutations. This is an inevitable outcome when there is goal or target, just levels of fitness.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
05:51 AM
5
05
51
AM
PDT
That’s certainly true. No one wants to deny that algorithms can be very efficient in finding solutions to problem according to laws of necessity, and GAs are just a subset of algorithm, which incorporate some random search as part of their working. But still algorithms they are
Algorithms, whether written by programmers or not, are embodied. They have machinery that executes the steps. Evolution is embodied in imperfect replication plus fecundity. That's it. There's no code specifying goals. It's an inevitable outcome of any system that replicates imperfectly and in which offspring have differential reproductive success. It can be observed in simple chemical replicators, such as Spiegelman Monster. There's no program specifying improvement or targets.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
05:35 AM
5
05
35
AM
PDT
How can it have no target and no way of knowing which direction it is going, yet also know that some variations are better than others? “Better” according to what standard — what goal?
A GA requires just two components, both of which are blind to goals. It needs a replicator, a component that produces a generation of offspring. And it needs a component to introduce variations. There is a third component, the oracle, which grades individuals in each generation. But the oracle could be written by another programmer. The criteria for scoring fitness could be anything. The replicator is blind to the oracle's criteria.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
05:25 AM
5
05
25
AM
PDT
In fact, you contradicted yourself within two sentences. You first state: “It has no target and no way of knowing how close it is to the best possible solution.”
There's no contradiction. There's no way to know the best solution to the TSP. Unlike the Weasel demo, no one knows the target. The oracle simply grades each individual by total distance traveled. The shortest routes are reproduced with random modifications. There is no end point, just gradual improvement. If you take the trouble to write a Weasel program, you'll discover that this is true even of the simplest form. The algorithm has no knowledge of the target. It simply replicates, with modification, the best scoring individual. The Weasel algorithm doesn't need to know that there is a case where further improvement is impossible.Petrushka
June 9, 2010
June
06
Jun
9
09
2010
05:13 AM
5
05
13
AM
PDT
Arthr Hunt: strange, I have debated with many darwinists here in perfect serenity. Anyway, thank you for the invitation. Give me some time: I will read your essay and answer you, either here or elsewhere.gpuccio
June 9, 2010
June
06
Jun
9
09
2010
04:22 AM
4
04
22
AM
PDT
Petrushka: Thank you for answering my challenge. I can see the meaning of your post. You mean (correct me if I am wrong): even if the Weasel knows the solution, there are other genetic algorithms which can solve very complex problems without knowing the solution in advance. That's certainly true. No one wants to deny that algorithms can be very efficient in finding solutions to problem according to laws of necessity, and GAs are just a subset of algorithm, which incorporate some random search as part of their working. But still algorithms they are. Again, the point here is not simply complexity. It is fucntional complexity (dFSCI). In other words, algorithms, genetic or not, can certainly generate complexity, but they cannot generate dFSCI. Let's take your example. The only dFSCI which I can see in it is the alògorithm itself, the program. It has a function (solving the traveling salesman problem with reasonable approximation), and it is 27.8 KB long (zipped). So, assuming that the functional target is small enough (in computer programs, single bits are probably more strictly specified than in proteins), it is functionally specified (it describes a function), it is complex, it is not significantly compressible (not in the zipped form, at least), it is digital: it is dFSCI, and it was designed. Not so its output. Its output is not dFSCI, it is just data which give the solution to a specific problem in input. It is information, but not functionally specified information. The Weasel output is dFSCI (if we put the threshold not too strict). It is functional (has meaning in English), it is complex, it is digital. That's why no algorithm genetic or not, will be able to output it without having prior information about the solution.gpuccio
June 9, 2010
June
06
Jun
9
09
2010
04:20 AM
4
04
20
AM
PDT
Petrushka: "What evolution doesn’t require is a specific target. Just differences in reproductive success." You have yet to support the assertion that evolution requires no target. Petrushka: "The traveling salesman problem has one significant similarity. It has no target and no way of knowing how close it is to the best possible solution. All it knows is that some variations are better than others." ... according to rules which are defined by a goal -- a target. Or is it just a fluke that the whole class of GAs which are brought together in the aforementioned competition just so happen to come close (some more so than others based on the intelligence of the programmer) to solving that specific problem rather than some other arbitrary problem? In fact, you contradicted yourself within two sentences. You first state: "It has no target and no way of knowing how close it is to the best possible solution." and then you state: "All it knows is that some variations are better than others." How can it have no target and no way of knowing which direction it is going, yet also know that some variations are better than others? "Better" according to what standard -- what goal? As I stated above, tongue in cheek of course, "… and I’m sure the Genetic Algorithm that wins also brags to all his Genetic Algorithm friends about how smart he is …" Do you see the significance?CJYman
June 8, 2010
June
06
Jun
8
08
2010
08:51 PM
8
08
51
PM
PDT
The traveling salesman algorithm would work even if the cities moved or shifted position between trials. Obviously there is some rate of shifting that would result in zero progress, but that happens to living things also.Petrushka
June 8, 2010
June
06
Jun
8
08
2010
08:45 PM
8
08
45
PM
PDT
Don’t believe it? Then find me a genetic algorithm that can produce “Methinks…”, indeed any sentence without a target built-in.
The selecting agent must rank individuals by some definition of fitness. For living things it could be the biochemistry necessary to sustain metabolism, or it could be predation or competition for resources. In all cases, including biological evolution, some sort of differential reproduction takes place. The algorithm needs to know absolutely nothing about how a generation will be ranked. The variation generator needs to know absolutely nothing about what will be useful, necessary or successful. It needs to know nothing about targets or whether the target is fixed or changing.Petrushka
June 8, 2010
June
06
Jun
8
08
2010
08:42 PM
8
08
42
PM
PDT
Petrushka: Genetic Algorithms. A valiant effort, but genetic algortithms rely on two important prerequisites: 1) the ability to calculate small improvements in a penalty function, the mere existance of which implies an over-arching knowledge of the solution space, i.e. distance from the target, and 2) a relatively dense solution space, where each valid solution is but a short step away by random search.
OK, but that's exactly what's observed in Nature -- tiny, almost imperceptible variations among individuals. That's the way plant and animal breeding works. What evolution doesn't require is a specific target. Just differences in reproductive success. The traveling salesman problem has one significant similarity. It has no target and no way of knowing how close it is to the best possible solution. All it knows is that some variations are better than others.Petrushka
June 8, 2010
June
06
Jun
8
08
2010
08:23 PM
8
08
23
PM
PDT
Petrushka: Genetic Algorithms. A valiant effort, but genetic algortithms rely on two important prerequisites: 1) the ability to calculate small improvements in a penalty function, the mere existance of which implies an over-arching knowledge of the solution space, i.e. distance from the target, and 2) a relatively dense solution space, where each valid solution is but a short step away by random search. Neither of these are present for a truly blind evolutionary process to evolve a bunch of letters into an (or any) intelligible sentence, in any language, of any reasonable length, in less time than the universe has been around. Don't believe it? Then find me a genetic algorithm that can produce "Methinks...", indeed any sentence without a target built-in.SCheesman
June 8, 2010
June
06
Jun
8
08
2010
07:33 PM
7
07
33
PM
PDT
1 2 3 4

Leave a Reply