EV Ware: Dissection of a Digital Organism

_{Baylor Bear

December 28, 2008

Intelligent Design}

Share: Facebook; Twitter; LinkedIn; Flipboard; Print; Email

Can undirected Darwinian evolution create information?

In a celebrated paper titled “Evolution of Biological Information,” a computer program named ev, says yes. It claims to illustrate the following properties of evolution.

“[Ev shows] how life gains information.” Specifically “that biological information… can rapidly appear in genetic control systems subjected to replication, mutation and selection.”
Ev illustrate punctuated equilibrium: “The transition [i.e. convergence] is rapid, demonstrating that information gain can occur by punctuated equilibrium.”
Ev disprove “Behe’s … definition of ‘irreducible complexity’ … (`a single system composed of several well-matched, interacting parts that contribute to the basic function, wherein the removal of any one of the parts causes the system to effectively cease functioning’. “

In a wonderful friendly GUI (graphic user interface) from the people at EvoInfo.org, it is easy to show that without front loaded programmed information about the search, ev simply will not work. These claims therefore ring hollow.

The goal of ev is to identify a given string of bits (ones and zeros). The reason the ev program works is because of the structure created by the writer of the ev computer program. A Hamming oracle in ev, for example, tells you how close your guess is to the correct answer. Contrast this to undirected random search where you either told: No, your guess is wrong or, Yes, your guess is right. At a trillion trials per second, it would take “about 12 566 000 000 000 000 000 ([over] twelve and a half quintillion) years” to find the ev target using undirected random search. To identify the target string of bits, the Hamming oracle allows reduction of the number of trials to thousands, hundreds, and even tens.

EvoInfo’s EV Ware GUI works on your browser and is easy to use.

A tutorial on the EV Ware GUI is here: http://EvoInfo.org/Resources/EvWare/
And the GUI is here: http://EvoInfo.org/Resources/EvWare/ev_simulation.htm

ALSO: See the GUI autopsy results for Dawkins’s METHINKS*IT*IS*LIKE*A*WEASEL at EvoInfo.org

Description of Weasel Ware: http://EvoInfo.org/WeaselWare.html
Weasel Ware GUI: http://EvoInfo.org/WeaselGUI.html

Comments

Venus Mousetrap: please, see also my points at #19 about NS. I believe there are two fundamental errors in your reasoning. The first, and most important, has been clearly shown by Atom in #24 and 25. The problem of metrics is essential, and corresponds to the situation in my second case of "artificial" NS in GAs: "b) The results are “measured” for some property related to the known solution (so called “fitness function”). That would be the case in the ev program, if I understand correctly." Now, you can have two different kinds of metrics. In one case, you can return a number which measures the real distance from the result. That is a very explicit kind of oracle, and obviously requires perfect knowledge of the solution. It is no different from the weasel situation. In the second case, let's say a generic fitness function, the measurement is not absolute, but relative: it can measure the emergence, or the increase, of a pre-determined function, with great sensitivity and specificity. Here again, even if the exact information for the solution is not needed, you need knowledge of two very important and almost equivalent things: what is the function you want to attain, and how to measure it efficitently, even at very low levels. Two understand the importance of just knowing what the function is, let's take the example of protein engineering through partial targeted random variation, which I quote in my post #19. Just imagine what would happen if the engineers, after applying a round of random mutations, had to measure any possible emergence or increase of any possible function, even at very low level, and then select each case. The method would immediately become empirically useless. Instead, if you can measure for only one function, the one for whichh all the experimental setting (including choice of the original library) was programmed, and promote any tiny sign of that function against all other possibilities, then system works. So, the advance choice of what to look at is fundamental, and is an explicit act of engineering. And you shouldt also recognize the importance of being able to measure the function having a specific measurement procedure, intelligently programmed. That allows you to measure any observable function variation in the limits of the sensibility of your measurement system, and you will be careful to intelligently provide a measurement system which is sensible enough. Even if you choose to use in your GA a binary oracle, based on a threshold, the setting of the threshold is still an indirect form of metrics. You can object that NS is something like that: it works with a threshold, and utilizes a binary oracle. That's correct, but the important point is that the measurement in NS is neither specific nor sensitive, because it is extremely indirect: indeed, it is not in any way connected to the searched function itself (also because there is no searched function: NS is by definition blind), and it selects on the basis of any generic improvement in survival. Please, reflect that improvement in survival implies an extremely high threshold for any function: the function must be present, relevant, and must be, alone, capable of significantly increase the survival of the whole replicator. And the measurement is also extremely slow. To understand how fundamental are these differences, let's take the second example which I quote in my post #19: antibody maturation in the immune system after the primary immune response. Here the system, after having applied targeted partial random variation, selects for increased affinity to the pre-determined antigen (which is known to the system, because it is the antigen which primarily activated the immune response). In a few months, the antibody affinity for that antigen greatly increases. But just think if the immune system had to find what clone has expressed the increased affinity just by measuring the survival increase: in that case, assuming as a thought experiment that the new clones could be genetically transmitted to different individuals, the model would work only if the individuals with a higher affinity for that antigen invariably could survive better. That is not only a slow way of measurement, it is a way which would never work, because there is no reason that a higher affinity in an antibody to that specific antigen must be a factor which improves survival in a generic situation. That's why the immune system builds a generic low sensitivity library of antibodies for each one of us (the primary repertoire), and engineers higher affinity antibodies in each case where a specific antigen has been met (the secondary response). That makes absolute sense, and is based on careful engineering which allows the system to work on two basic informations; what the antigen is (an information given by the primary immune response), and how to measure the affinity of each clone against it (a very intelligent information coded into the immune system itself). So, I conclude pointing at the second error in your reasoning, which should by now be self-evident. You say: "The next question, of course, is how much information does an environment create by natural processes? I believe it will be a lot more than the few hundred bits of ev." No, it's exactly the contrary. The environment passes not a lot of information, it is just as it is, and that becomes information for the replicator only indirectly through the rough measurement implied in survival. But consider the case of antibodies. If the replicator received information from the environment about all possible existing antigens, that would be meaningless to it. It's only the restriction of information (this antigen is more important now) which allows the replicator to react intelligently through a specific measurement system. In other words, useful (functional) information is a restriction of possibilities, not an increase of them. You gain useful information when you can restrict the field of possibilities. That's why CSI (and in particular FSCI) is the most useful information: it gives you the possibilities which work, and you can work without trying all the others. That's why ev is so powerful. Because it has so many bits of CSI in it. Remember, bits of CSI are a measure of how improbable that information is, in other words they are a measure of how much you are restricting the field vs a random search. I quote from the important paper by Durston, Chew, Abel and Trevors: "Theoretical Biology and Medical Modelling": "The measure of Functional Sequence Complexity, denoted as ?, is defined as the change in functional uncertainty from the ground state H(Xg(ti)) to the functional state H(Xf(ti))" and "The resulting unit of measure is defined on the joint data and functionality variable, which we call Fits (or Functional bits). The unit Fit thus defined is related to the intuitive concept of functional information, including genetic instruction and, thus, provides an important distinction between functional information and Shannon information" It's that simple. A program like ev contains many Fits of functional information. The environment doesn't.gpuccio_{December 29, 2008
December
12
Dec
29
29
2008
11:27 PM
11
11
27
PM
PDT}

Another old post relevant to this topic: Dodgen DailyPatrick_{December 29, 2008
December
12
Dec
29
29
2008
09:28 PM
9
09
28
PM
PDT}

DaveScot (18),
“Evolution” (Darwinian or otherwise) produces not a single simplest fastest reproducer (due to variations in environments) but if the winner is declared by metrics of biomass and/or number of individuals then the simplest creatures, bacteria, also thought to be the first creatures, are the winners.
Ah, so word gets around. Yes, DaveScot, there is an evolutionary free lunch. Wolpert and Macready did not take time into account. Going from the simple to the complex is just the way nature works, and it happens to be the best way to proceed in optimization of a "really black" black box. There is no reason to impute design to a natural process that goes from simplicity to complexity.
Predation, weaponization, and defense might conceivably account for a departure from simplicity and reproductive speed.
These are successful departures from simplicity and reproductive speed. I would describe biological complexity as a pyramid, with each level continually generating variations that usually fail and sometimes succeed, but everyone would imagine far too small a base. It is one thing to consider that unicellular organisms account for most of the earth's biomass, and quite another to consider that they account for almost all of its reproductive trials.Sal Gal_{December 29, 2008
December
12
Dec
29
29
2008
05:30 PM
5
05
30
PM
PDT}

Thanks Patrick, but dead bodies (I know you were kidding) does not tell you how far away from a functional subset you are. The points you bring up in your other post are apropos here: not every fitness function will lead you to the functional subset, therefore you need the correct type of fitness landscape to hold for a given duration of time for self-selection (differential reproduction) to have any reasonable hope of ever locating that functional subset.Atom_{December 29, 2008
December
12
Dec
29
29
2008
03:35 PM
3
03
35
PM
PDT}

An environment does not return a value
How about the number of dead bodies? :P More seriously, environmental conditions can funnel a search but the problem is twofold: 1) the funneling is usually not balanced like an intelligent search would be. In engineering evolutionary searches should use a generalized (not explicitly defined) target as to find solutions to problems that we might not have conceived of otherwise. The problem is that nature is usually too generalized and cannot efficiently optimize the search within reasonable time constraints. 2) Environmental conditions necessary for funneling the search may not be available for all functionality. I wrote about this at length fairly recently but it was ignored...possibly because the comment was quite lengthy.Patrick_{December 29, 2008
December
12
Dec
29
29
2008
03:26 PM
3
03
26
PM
PDT}

In the above post: "(uncertainty)" technically should read "(reduce uncertainty)" in the last paragraph...Atom_{December 29, 2008
December
12
Dec
29
29
2008
03:15 PM
3
03
15
PM
PDT}

Continued: Now let's examine fitness functions themselves. You may argue that nature provides information about "closeness" by rewarding those genomes that are "closer" to a functional target with more offspring. But then your fitness function (the fitness slope itself) would encode the information about closeness to target. Indeed, there are many fitness slopes which would not work: ones with sparse islands of functionality, ones where there are no smooth paths to highly functional states, ones that rewarded genomes that were actually far away from functional states (relative to whatever function you are trying to develop), etc. What chose this particular fitness function/landscape out of all the possible ones? Any time you exclude possibilities (reduce uncertainty), you are mathematically providing information (in the Shannon sense.) We have now pushed the problem up one level to selecting the proper fitness function.Atom_{December 29, 2008
December
12
Dec
29
29
2008
03:13 PM
3
03
13
PM
PDT}

Venus Mousetrap, Let me cut to the heart of your criticism. You wrote:
The Hamming oracle is giving out no more than a value. It has to be a value with a range, because organisms must compete. So yes, it’s giving information, but no more than an environment does (an awful lot of fitnesses can be summed up with a single number, after all. Leg length. Adhesive stickiness. Body weight. etc.)
An environment does not return a value when evaluating fitness that is like "ten DNA bases away from a functional organ." (I know you aren't saying it does, but to be similar to Ev, it would have to.) Ev is giving the distance from a small functional subspace (how many binding sites the genome string missed), and this information alone is enough to quickly find the target, in N + 1 steps, using Hamming Search. No evolutionary search is needed and indeed, using one is inefficient. The math and GUI decisively show this. Try it yourself. Remember, the Hamming Oracle is not a free source of information. It needs information about the target to work properly. (If not, then how could it ever provide information about "distance" from the target? It would degenerate to a needle-in-the-haystack oracle.) So if nature acts as a Hamming Oracle, what encoded the information about the target space into nature (the fitness function)? What chose the correct fitness function that encodes this information about the target space? Again, you are only pushing the information problem up one level. AtomAtom_{December 29, 2008
December
12
Dec
29
29
2008
03:08 PM
3
03
08
PM
PDT}

I'm not sure about the Hamming oracle being so bad. People seem to be hung up on details and overlooking that it is just a simulated fitness function. I mean, I've seen some of the challenges people make here over 'active information' and how much of it can be put in. For some people, the mere act of writing a genetic algorithm is enough to load in Active Information, which means that no simulations of nature can ever be run. The Hamming oracle is giving out no more than a value. It has to be a value with a range, because organisms must compete. So yes, it's giving information, but no more than an environment does (an awful lot of fitnesses can be summed up with a single number, after all. Leg length. Adhesive stickiness. Body weight. etc.). Atom: And you are correct, organisms in the real world cannot hope to have this information given to them, unless an intelligent agent somehow provided it or programmed it. This is clearly untrue! What if dna encodes for body size, and mice have to escape down some small holes? If I simulated that on a computer, I would of course be testing the digital bodies against digital holes, and telling them how well they did. That's active information! Are you telling me that this invalidates the real life example, or are you agreeing that actually, natural information is easily passed from environment to lifeforms? Because I'm sure the population of mice will notice that there aren't as many fat ones around. Saying that fitness functions encode this information only pushes the problem and information back one level to the fitness functions. Which in nature, arise as a result of lawlike processes (hills have slopes, some trees are taller, etc.). If you're saying that ev loads in active information, this is directly analogous; you are saying that environments load in information to lifeforms. The next question, of course, is how much information does an environment create by natural processes? I believe it will be a lot more than the few hundred bits of ev.Venus Mousetrap_{December 29, 2008
December
12
Dec
29
29
2008
12:06 PM
12
12
06
PM
PDT}

Addendum: jdaggs wrote:
While it looks like from the ev source code that a comparison between already existing DNA strands does occur. This is a completely untargeted relative comparison. There is no target against which each DNA strand is compared.
The "comparison" is based on a metric, in this case the Hamming Distance. The Hamming Distance is based on the divergence from a fixed target (phenotype, or output, string.) So in short, your statement "This is a completely untargeted relative comparison" is mistaken. To return to the example I gave earlier, let's say we have 10 searchers in our field for the eggs. They each wander to a spot in the field, and then I assign a number to them based on their exact distance from the egg. You then do you "relative" comparison among these searchers using the Distance numbers I assigned earlier, eliminate the 5 farthest (furthest?), replacing them by 5 new searchers placed randomly one foot away from each of the remaining searchers. Then repeat the process. It is clear where the Oracle's information is input: it is at the "evaluation" step, which is assigning a number based on a fixed target. The selection step uses this number in the comparisons. Again, even though the information is hidden one step back in the process it is still there. Therefore, Ev does not demonstrate a free source of information; it only demonstrates the inefficient use of Active Information given by the Hamming Oracle. AtomAtom_{December 29, 2008
December
12
Dec
29
29
2008
08:15 AM
8
08
15
AM
PDT}

jdaggs, Thank you for your thoughts. The "target" in Ev is a string that is bound at only 16, fixed binding sites, and is not bound at any other location. Indeed, the "Error Count" shows the Hamming Distance or the distance from this target. (It counts how many sites it is bound at that it shouldn't be bound at as well as how many it still needs to be bound at.) Once you have this information, you can zero in on a target extremely rapidly as the Hamming Search demonstrates. Hamming Search mode doesn't even use an "evolutionary" reproduction scheme; it just efficiently extracts the information already given to us by the same Hamming Oracle that Ev uses. Armed with that information, we can find any target of a fixed length in N + 1 queries. In regards to your question about where the information input is in the source code, it is given by the oracle that evaluates the search space and a given genome string and returns the Hamming Distance (Error Count) of this genome, so that selection can do its work. I can highlight the code, but it is easy to find (I included a link to the source code at the bottom of the GUI. It is in the "ev_simulation_array_based.js" file.) Does the Oracle itself represent a free source of information? No, because the oracle itself needs to be given the information about the target. There is no free source of information, we just push the problem back one level. Furthermore, as mentioned before, the very data structure used by Ev to represent a digital organism and "binding" produces phenotypes with lots of "zeros" (not bound) and few "ones" (bound). This increases the odds of finding a string with few ones and lots of zeros. Atom PS As far as I know, you will only be banned if your become disrespectful or if you disagree with a certain mod in a political post. Just steer clear of politics on the site and be as courteous as you would face to face with people, and you will be here a while.Atom_{December 29, 2008
December
12
Dec
29
29
2008
08:02 AM
8
08
02
AM
PDT}

1. Biological information is related to biological function. 2. Shannon information does not care about meaning nor function. And I would say: 3.- Biological information is not the sequence of nucleotides. 4. The sequence is important only in carrying out the instructions embedded on the DNA via the pre-determined genetic code- as in which codons represent which amino acids, start and stop positions.Joseph_{December 29, 2008
December
12
Dec
29
29
2008
07:24 AM
7
07
24
AM
PDT}

jdaggs: This is just a request for information, in order to uderstand better. Indeed, I don't know in detail the ev program, so I would like to be sure I understand how it works. I have tried to read the paper, and i am interested to understand how the selection process works, because I think that is the most relevant point. I quote here a couple of phrases from the paper: "...the program arbitrarily chose the (16) site locations, which are fixed for the duration of the run" I I understand well, that would mean that the program decides the 16 site locations and knows them. "The organisms are subjected to rounds of selection and mutation. First, the number of mistakes made by each organism in the population is determined. Then the half of the population making the least mistakes is allowed to replicate." So, it seems, selection happens by determining how much the mutated organism can localize the site locations known to the program. In other words, these simple organisms are mutated (both the localizing "gene" and the sequence where the localization has to be made) in order to match a pre-fixed distribution of localization sites. It seems to me that the pre-fixed distribution is indeed a target. And the measurement of the mistakes is certainly an oracle. And the measurement can be done only because the program already knows where the sites have to be localized. Am I wrong? (this is not a rhetorical question: I just want to understand) And now, some personal comments. Again, I would like to emphasize that selection in these simulation programs seems to be based on one of two methods: a) The results are directly matched against the information (the "solution"). That should be the case in the weasel example. b) The results are "measured" for some property related to the known solution (so called "fitness function"). That would be the case in the ev program, if I understand correctly. Well, I believ that neither a) not b) are models of natural selection. For a) it is rather obvious. The program knows the solution, and uses it directly to get the solution. Nothing could be more trivial. But for b) too, the program has to know the solution, in order to measure against it, although more indirectly. So here too we have a lot of active information in the program. We are here in a context similar to the intelligent selection in protein engineering, ot to the intelligent selection in antibody maturation: a random variation is applied to a well specified target, and a very intelligent selection is made in order to get a pre-fixed result. In the case of protein enigeering, that can be the measurement of the desired function. In the case of antibody maturation, that would be the measurement of the affinity for the (known) antigen, whose information is stored in the immune system (probably in the antigen presenting cells). In both cases, the system knows what it wants to select, measures it, and selects. That's the big difference with NS. In NS, the system is supposed to be completely blind. Both the environment and the replicators have no idea of what they should achieve. The emerging function must emerge without any pre-conceived accommodation, only by virtue of a random acquisition of some unexpected advantage in the existing, blind system formed by the environment and the replicator. I think the difference is really big. It means that the programmer must know nothing of what will be selected. It means that the selected function has to emerge on its own merit. It has to be functional, and not only "measurable". It is easy to "evolve" measurable functions, when we know what the function must be, we restrict our target, use random variation on it, and then are able to meausre any possible increse, even if very small, of our expected function. The immune system does that in antibody maturation. Protein engineers do the same thing. GA programmers do the same thing. But none of that is in any way a model of NS. And none of that is a model of CSI generation witout active information.gpuccio_{December 29, 2008
December
12
Dec
29
29
2008
07:13 AM
7
07
13
AM
PDT}

The "Oracle" in nature is differential reproduction. The information it provides is whether, given either a change in the environment or a change in the organism, the result is more or less reproduction. The child's guessing game of "warmer or colder" is the only answer an oracle imitating natural selection needs to give. Warmer means more successful reproduction and colder means less successful. This seems to logically result in the simplest, fastest reproducer coming out on top every time in an environment with limited resources. Indeed that's what origin of life experiments with self-replicating ribozymes has shown - you get something uber-simple that takes over the whole flask. "Evolution" (Darwinian or otherwise) produces not a single simplest fastest reproducer (due to variations in environments) but if the winner is declared by metrics of biomass and/or number of individuals then the simplest creatures, bacteria, also thought to be the first creatures, are the winners. Predation, weaponization, and defense might conceivably account for a departure from simplicity and reproductive speed. The bottom line for explanations of evolution, as always, remains with how the first reproducer that was able to play the warmer/colder game came about and in that regard probably the most difficult thing to explain is how the abstract genetic code came about. One of the quotes on the EvoInfo page zeroes in on this:
“The information content of amino acid sequences cannot increase until a genetic code with an adapter function has appeared. Nothing which even vaguely resembles a code exists in the physio-chemical world. One must conclude that no valid scientific explanation of the origin of life exists at present.” Hubert Yockey, “Self Organization Origin of Life Scenarios and Information Theory,” Journal of Theoretical Biology 91 (1981): 13.
Some of the other quotes make related points. The selection of quotes on the EvoInfo home page is excellent, by the way, and makes for an enjoyable reading.DaveScot_{December 29, 2008
December
12
Dec
29
29
2008
05:09 AM
5
05
09
AM
PDT}

...furthermore, I cannot see that any computer algorithm can establish (in principle) that digital organism is biologicaly viable at EVERY stage in it's "search" towards the target.inunison_{December 29, 2008
December
12
Dec
29
29
2008
02:34 AM
2
02
34
AM
PDT}

Sev: This might help:
"Now we believe that the DNA is a code. That is, the order of bases (the letters) makes one gene different from another gene (just as one page of print is different from another)"
Source? Oh, a certain to-be Sir Francis Crick, in a letter to his son Michael, March 19, 1953. [Cited, Thaxton, here.] GEM of TKI PS: Defining FSCI and showing its roots in OOL research on the informational macromolecules of life, c. 1970's - 80's. (FSCI is the relevant subset of Orgel's CSI.)kairosfocus_{December 29, 2008
December
12
Dec
29
29
2008
01:06 AM
1
01
06
AM
PDT}

Seversky: You ask: "The question is, what is meant by “information” in this context? Information in DNA, for example, if it can be said to exist at all, does not appear to be the same as the information being conveyed in these posts. There is no ‘meaning’ in the sense of that which is intended by a sender or that which is apprehended by a recipient." Your question shows probably a lack of familiarity with the ID concepts. Of course there is meaning in DNA, and that meaning corresponds to the specified information. As you probably know, information in the ID theory means that some result is fixed out of all possible theoretical results (in a system). So, if we are talking of a binary string of 130 bits, for instance, like in the example Atom makes commenting the GUI, any single random string is information with a complexity of 1 : 2^130. That kind of information is only a probability, and has nothing to do with meaning. Indeed, in Shannon's information theory, meaning is not even an issue. Shannon's theory is a theory about information in this blind sense, and not about meaning. On the contrary, specified information corresponds broadly to our intuitive concept of meaning. Specified informations is a subset of all possible information, usually a very small one. "Specified" means all information which has some properties which allow us (intelligent observers" to distinguish that information from a generic random information. There are many ways that information can be specified (see Dembski). Bit for our purpose, only one is important: functional specification. An information is functionally specified when, in the right context, it can do something which would be impossible without it. Going back to your example (DNA and these posts): both are examples of functionally specified complex information. These posts are information which, in the context of english language, transmit to the reader some specific knowledge or thought. DNA (the protein coding genes) are information which, in the context of the language of the DNA code, transmit to the translation system the correct functional sequence of a protein. In both cases the meaning is abstract, and is encoded in a symbolic language. Both cases are examples of a functional message being conveyed through a symbolic language. Both cases are CSI. Just to show you the similarity. I can use this post to send a message to you, a fellow biologist, saying: Hey friend, this is the protein whose properties you should study. Just synthesize it and study how it folds. Here is the sequence: GTGCTGTGAACTGCTTCATCAGGCCATCTGGcCCCCTTGTTAATAATCTAATTACeCTAGGTCTAAGTAGAGTTTGACGTCCAATGAGCGTTT As you can see, I have used this post exactly to do what DNA does; to convey a specific useful information. I can agree that these posts can convey a grater variety and complexity of meanings, but after all DNA is only a static mass memory, while we are using these posts to communicate in almost real time. But there is CSI, and therefore meaning, in both.gpuccio_{December 29, 2008
December
12
Dec
29
29
2008
12:06 AM
12
12
06
AM
PDT}

Atom, Ok, I'll bite. I have a few thoughts on Ev I would like to share. First, I think The EvInfo.org site misses the point of the ev program somewhat. The ev program was designed with the intent of showing how evolutionary processes can increase the amount of information (Shannon Entropy) in a simulated DNA strand. It does this by selecting, randomly mutating and reproducing many strands over many generations. Your presentation on EvInfo.org doesn't make that clear that this is the intent of the ev program only that it's function is to find a match for patterns or targets. While it looks like from the ev source code that a comparison between already existing DNA strands does occur. This is a completely untargeted relative comparison. There is no target against which each DNA strand is compared. Second, about your general objection, You stated the program functions by means of an oracle which gives the program information about the search space. If I understand your objection correctly this somehow invalidates the ev program as an evolution simulation because the oracle is providing the program with information that a true evolution simulation would not otherwise have. If true, I disagree. The "oracle" or fitness function in ev is not adding any new information directly or indirectly to any of the DNA strands. It is simply selecting which organisms make it to the next round of mutation and reproduction. That is it. There really is no "target" that ev attempts to find. The oracle really is not an oracle it really should be called an "environment" or fitness function instead. The fitness function or "oracle" in ev exists simply to reproduce the environment that any real DNA strand might have to exist in. Just as an avalanche rushing down the mountain is selecting strongly rooted trees from weakly rooted trees for survival to the next round of reproduction, the selection mechanism or fitness function in ev is doing nothing more. No, additional active, free or non free information is being added to the trees that live or die just decisions on which ones lives and which ones die. Just like in nature. I don't care necessarily if you respond to my critiques above However, I would ask that you show me and the others here on this forum where in the source code of ev active free or non-free information is being added to the DNA strand. P.S Will this get me banned? I heard it was a tough crowd over here :) jdaggsjdaggs_{December 28, 2008
December
12
Dec
28
28
2008
11:43 PM
11
11
43
PM
PDT}

Atom: Wonderful work! As it is evident from most of the discussions here at UD, demonstrating that algotithms cannot generate CSI remains the main point of ID. While we rely on the work of our theorists (Dembski and Marks) to get ever better theorical demonstrations of that, your practical implementation showing clearly to us non mathematicians what is really at stake is extremely useful. I have often used your weasel ware GUI to help friends who are not familiar with mathematical concepts what is really happening in the different models. The ev ware is another precious tool. I can't understand why some people find it difficult to understand the fundamental intuition behind these analysis: these softwares already know the target! They just refrain from giving you immediately the correct answer, because otherwise there would be no game, and give it to you in small pieces. But they know the answer! The evolutionary process, as it is conceived, does not know the answer. Indeed, it is not even interested in it. Natural selection can only select function, not information. GAs, instead, select information. Their only meaning, in practice, is to sow that: if I already know information, I can select it. What an achievement! I have always thought that the only true evolutionary simulation should be like that: take a system (a computer) and implement in it simple digital replicators subject to random variation (possibly at an adjustable rate). And then just wait for their "evolution". We have all that is necessary. One could say: but where is NS? Well, NS is in the same place where it is supposed to be in natural history: it is in the rules of the system and in the rules of the replicator. The replicator has all the chances to become more efficient by random variation and profit of the rules of the system to become something better. So, just wait! But the moment the programmer, tired of that infinite wait, starts saying: well, let's help it a bit; after all, we know what we want to achieve. Well, I suppose that's exactly what a patient designer has been doing...gpuccio_{December 28, 2008
December
12
Dec
28
28
2008
11:37 PM
11
11
37
PM
PDT}

Seversky, Welcome to UD. An evolutionary search (or any heuristic search method) attempts to improve the chances of a search over brute force exhaustive search (which is not possible for large search spaces) or random sampling (which also doesn't work fast enough on average for large spaces.) To improve the chances of a search, you must limit the search space by eliminating some sub-space where the target is not likely to be found. Any time you eliminate possibilities, you impart information, which is mathematically the reduction of "uncertainty" or possibilities. With enough information, you can eliminate all sub-spaces but the target. Again, using my analogy, you can find any easter egg in any field given enough information. Depending on the quality of the information (how much sub-space it eliminates) and how well you use that information, you can greatly reduce the time it takes to find the target. Ev uses information given by the Hamming Oracle to retain those genomes "closer" to a target space. It knows it is closer or further away because the Hamming Oracle gives it this information. Without this information, the search doesn't work. (See Random Output mode in the GUI) And you are correct, organisms in the real world cannot hope to have this information given to them, unless an intelligent agent somehow provided it or programmed it. Saying that fitness functions encode this information only pushes the problem and information back one level to the fitness functions. So saying that Ev is a viable model for evolution or somehow represents a "free" source of information is false and misleading. AtomAtom_{December 28, 2008
December
12
Dec
28
28
2008
10:39 PM
10
10
39
PM
PDT}

Platonist @ 6
Can someone pleasedumb this down so that I can understand it? Thanks.
I agree. The question is, what is meant by "information" in this context? Information in DNA, for example, if it can be said to exist at all, does not appear to be the same as the information being conveyed in these posts. There is no 'meaning' in the sense of that which is intended by a sender or that which is apprehended by a recipient. Atom @ 7 offers as an illustration:
For example, pretend you are in a field 20 acres large and you want to find an easter egg. You ask the oracle “Is the egg here, where I’m standing?” and the oracle can reply in two different ways: she can say “No”, and leave it at that or she can say “You are 23.5 feet away.”
As a metaphor for what is happening it is helpful but the difference is that in neither the computer program nor evolution are there intelligent agents asking or answering questions in this way. It is analogous to what is happening in the program but not the same. By the same token could it be that what is happening in the program is analogous to what is supposed to happen in evolution but not the same?Seversky_{December 28, 2008
December
12
Dec
28
28
2008
09:01 PM
9
09
01
PM
PDT}

Plantonist, What is being said is that the Ev program does not represent a "free" source of information. This has implications for other evolutionary algorithms (if they use similar oracles), but the article/GUI is primarily concerned with Schneider's Ev. I disagree; 2008 was a good year for ID, since ID research was being done. Politics and opinions do not change truth; as long as the ID project is moving forward, then ID is having a good year. The only bad year for ID is when ID research is not being done. AtomAtom_{December 28, 2008
December
12
Dec
28
28
2008
02:50 PM
2
02
50
PM
PDT}

Thanks Atom. So what is being said is that the computer program shows nothing. Only that information from an intelligence is required. With that said. Lets all hope that 2009 is good to ID. 2008 perhaps wasn't the greatest year, but lets not give up hope. Darwinism is bankrupt and God exists.Platonist_{December 28, 2008
December
12
Dec
28
28
2008
12:11 PM
12
12
11
PM
PDT}

The Dawkinites should be careful. If evolution can be proven by computer program, they will be proving intelligent design because the programmer is (moderately, in this case) intelligent. I think I will write a program to use evolution to evolve a tracked vehicle, thereby proving that automobiles and similar conveyances were not designed, but evolved.William Wallace_{December 28, 2008
December
12
Dec
28
28
2008
11:29 AM
11
11
29
AM
PDT}

Platonist, The Ev program uses what is called an "oracle" that gives it information about the search space. Oracles can give more or less information, depending on how they answer a question. For example, pretend you are in a field 20 acres large and you want to find an easter egg. You ask the oracle "Is the egg here, where I'm standing?" and the oracle can reply in two different ways: she can say "No", and leave it at that or she can say "You are 23.5 feet away." Obviously the second way of answering provides much more information, and you'll be able to find the egg much faster. The Ev simulation uses the second type of oracle which returns a "Hamming Distance", telling the algorithm how far away from a target the "organism" is. Using this information, one can zoom in on the target rather quickly. This oracle is a HUGE source of information. Just using the Hamming Oracle and disabling the "evolutionary" part of the algorithm, you can zoom in even faster. (See "Hamming Search" in the GUI.) This means that the evolutionary part of the search actually hinders the process, it doesn't add anything to it. Ev turns out to use the available information inefficiently. In addition to this, the structure of the digital organism is pre-disposed to create genomes with few binding sites (zeroes in the GUI.) This helps reduce the search space as well, although the information provided by the oracle dwarfs the information gained by the structure. You can test this yourself: after a few thousand runs using both "Ev Bindings" (few ones, lots of zeros) and "Random Bidnings" (equal parts on average of zeros and ones), Ev will take longer on average to find random target sites than it will just using "ev" binding sites. (I ran the tests, I can post the numbers if you'd like.) You can also randomly converge on a target of all zeroes, which should be impossible unless there is a bias towards genomes with lots of zeroes.Atom_{December 28, 2008
December
12
Dec
28
28
2008
11:24 AM
11
11
24
AM
PDT}

Can someone pleasedumb this down so that I can understand it? Thanks.Platonist_{December 28, 2008
December
12
Dec
28
28
2008
11:06 AM
11
11
06
AM
PDT}

And all this demonstrates once again that the great majority of Darwinists do not understand beans about information systems like DNA and should not be allowed to teach biology! Ha! ;-)Borne_{December 28, 2008
December
12
Dec
28
28
2008
10:23 AM
10
10
23
AM
PDT}

But I guess that's why Baylor said it was "much celebrated"... I just didn't recall that paper. I must have missed the party! :PJGuy_{December 28, 2008
December
12
Dec
28
28
2008
06:51 AM
6
06
51
AM
PDT}

By the way, that paper is 8 years old.JGuy_{December 28, 2008
December
12
Dec
28
28
2008
06:47 AM
6
06
47
AM
PDT}

From the discussion of the above paper "Evolution of biological information" by Thomas D. Schneidera. We find: " The ev model shows explicitly how [...] thereby completely answering the creationists." "The ev model can also be used to succinctly address two other creationist arguments." "This situation fits Behe’s (34) definition of ‘irreducible complexity’ exactly [...]" "So, contrary to probabilistic arguments by Spetner [...]" ----------- ...but I thought ID and creationist arguments were suppose to be untestable. Just saying. :PJGuy_{December 28, 2008
December
12
Dec
28
28
2008
06:28 AM
6
06
28
AM
PDT}

Prev 1 2 3 Next

You must be logged in to post a comment.

Leave a Reply