189 thoughts on “Richard Dawkins’ Software

  1. Richardthughes: How do you think they keep finding the correct answers, Mung?

    Why not try to convince me by presenting actual objective empirical evidence to support “their” claims demonstrating that “their” claims are in fact correct?

    You can’t. So you won’t. It really is that simple.

  2. keiths:
    Mung: Your tests are subjective. They are not objective empirical tests.

    keiths: Incorrect.

    Publish your tests. The fact is that you have no objective empirical tests. If you had written any such tests you could post them. They don’t exist.

  3. keiths: Mung meltdown.

    You have no tests, and your claims about testing are false. Sure, it’s due to a Mung meltdown. Whatever it takes for you to convince yourself that you are right.

    Your WEASEL program assigns fitness. Deal with it. You have no objective empirical tests. Deal with it.

  4. My demands that keiths support his claims can be dismissed as a “Mung meltdown.”

    This from someone who has asserted that he always support his claims. You always support your claims, right, keiths? That is your promise to all of us that for some reason or other, whether justified or not, doubt your integrity?

  5. Richard Dawkins has made his programs from The Blind Watchmaker and Climbing Mount Improbable available via a web browser interface.

    Sadly for our friends at Uncommon Descent, the Weasel is not included.

    I don’t see anything here to be sad about, unless we ought to be sad about Patrick’s conspicuous absence from a thread where he was the author of the OP.

    The programs that were made available were the ones judged to be relevant to biological evolution. The WEASEL program failed. No surprise there.

  6. Poor Mung and his inability to grasp the simplest concepts. So in the PI algo the target is PI, even though the solution only allows for digits and arithmetic signs! It’s almost like Mung’s god can find impossible targets, praise the lawd!

    And of course the average fitness for a given selection coefficient obtained by the algo matching the one predicted by the model doesn’t count as a test! wait, it’s even worse, apparently he’s never heard about such a test!

    How could that possibly happen?

  7. The “test” Mung is asking for is to just look at the code and then simply run the weasel program.

    That way you are first determining, empirically, whether it has been coded to do cumulative selection (it does) and you are empirically testing whether the program works when you run it.

    That’s it. Done.

  8. Mung: The programs that were made available were the ones judged to be relevant to biological evolution. The WEASEL program failed. No surprise there.

    But it didn’t fail to be relevant to biological evolution. It is not entirely like biological evolution, it is very much a reduced version of it. That doesn’t mean it is wholly irrelevant, it just means it does not capture all aspects of biological evolution. It is not supposed to capture all aspects of biological evolution, it is only supposed to show the strength of cumulative selection to produce ever more fit phenotypes, as opposed to no selection.

    More biologically realistic versions of the WEASEL program, without target sequences but instead with some sort of physical simulation or “environment” the “organism” adapts to, have been made. The Steiner-tree version of the WEASEL is one such version.

  9. dazz: Poor Mung and his inability to grasp the simplest concepts.

    True and False. Your claims are False. Your claims are testable or they are not testable. True or False. Your claims are false, it really is that simple.

  10. Mung: You have no tests, and your claims about testing are false. Sure, it’s due to a Mung meltdown. Whatever it takes for you to convince yourself that you are right.

    Your WEASEL program assigns fitness. Deal with it. You have no objective empirical tests. Deal with it.

    LOL. Of course it assigns fitness, and it does it according to the model, do you want to test that functionality too? or will you keep pretending no one provided any after I post some test cases?

  11. Why is it more correct to say it is “assigning” fitness to a string, rather than testing the strings fitness against the target?

    If it was merely assigning fitness to a string, it could just make up an arbitrary score regardless of how fit it was, or randomly pick a number from a list of “fitness values”. That’s what comes to mind when you say fitness is being “assigned”.

    But that’s not what is happening, there is a test being made. The test constitutes in the comparison of the string with the target and the fitness is calculated therefrom.

    Maybe this is semantics, but I don’t see why saying this process constitutes an “assignment” of fitness value to the string is more correct han to say it consitutes a “test” of the strings against the target, and the fitness being calculated from that test? Actually I’d say to call it merely an “assignment” of fitness is wrong.

  12. OK, here’s the proposed “test” of whether Weasel programs do better than random search.

    1. Mung and I agree on a version of a Weasel program. Say one with 1 adult, 10 offspring, and mutation rate per site 0.04. (That means an average of 1.12 mutations per offspring). Its random mutation is controlled by a pseudorandom number generator.
    2. Since we are in the same city, we also agree on an intermediary. As we will see the intermediary need do mostly nothing.
    3. We agree on a random number seed of 4 digits (say I supply the first and third digits and Mung supplies the 2nd and 4th).
    4, We both run the program until the target sequence is found, or until 10,000 generations have passed.
    5. If the target sequence is found before 10,000 generations, Mung owes me $5, otherwise I owe Mung $5.
    6. Repeat steps 3 through 5, to a total of 20 times.
    7. Mung drops of his payment with the intermediary, in cash, or I drop off my payment. The recipient (me, of course, I predict) is told by the intermediary, and comes and picks it up).

    This has the advantage of maintaining Mung’s anonymity. Someone like Winston Ewert would be a good intermediary, since Mung says Mung already met him. And there is no need for me to protect my identity from Ewert. It also has the advantage of giving me $100 to buy chocolate. The results will be publicized here.

  13. Joe Felsenstein: Someone like Winston Ewert would be a good intermendiary

    I think not.

    1. Dembski (rightly) complains that you ignored the most recent of the references he gave in his seminar talk, a paper by Dembski, Ewert, and Marks (DEM).
    2. Dembski tells me that he will stick with the definitions of DEM.
    3. We adopt DEM’s definitions, and clobber ID with the GUC Bug.
    4. Ewert assembles quotes from different parts of our post, and fabricates a convenient claim about the GUC Bug.
    5. Ewert reverts to the defective definitions in the very first publication by Dembski and Marks (which are quite unlike those of DEM, and which you had not ignored), and dismisses the fabricated claim about the GUC Bug.

    Swell guy. I won’t speculate on why Mung wanted give him a hug.

  14. Mung:
    I am mocking people like Patrick who don’t believe anything without “objective empirical evidence.”

    Of course it’s much better to have faith in the absence of or counter to the evidence. Humans made so much more progress when the church enforced that on everyone.

  15. Patrick: Of course it’s much better to have faith in the absence of or counter to the evidence.

    I am skeptical. But then, you, as the King of Skeptical here at TSZ, have actual evidence for your claims. Except when you don’t.

    Please support your claim with objective empirical evidence or retract it.

    I predict that you won’t.

  16. Joe Felsenstein: OK, here’s the proposed “test” of whether Weasel programs do better than random search.

    Weasel programs perform better than blind search because they are guided. I didn’t think the performance was in dispute, nor why the performance was better.

    Can we objectively measure the performance difference, and if so, how?

  17. Rumraket: Why is it more correct to say it is “assigning” fitness to a string, rather than testing the strings fitness against the target?

    Because the “fitness” is assigned, and it’s a matter of fact.

    genome->fitness = fitness(genome);

    Even keiths knows that his WEASEL program assigns the fitness.

  18. dazz: Of course it assigns fitness…

    Yup. That’s what I said. Except for the “Of course” part. keiths claimed that GAs measure fitness. His own GA assigns fitness.

  19. “Always nice to watch Mung flitting about like a little monkey, chattering his mocking one-liners and flinging his tiny turds everywhere.” – Astute commenter at AtBC

  20. Mung:

    Joe Felsenstein: OK, here’s the proposed “test” of whether Weasel programs do better than random search.

    Weasel programs perform better than blind search because they are guided. I didn’t think the performance was in dispute, nor why the performance was better.

    Can we objectively measure the performance difference, and if so, how?

    That’s funny. I seem to recall a call for “objective empirical tests” by one of us. The proposed bet could only be won by me if the Weasel would have a first passage time to the target that is more than 10^{35} times smaller than a purely random search.

    The fact that I am willing to take a bet that I can only win if the Weasel gets to the target that much faster than random search, and that Mung avoids the bet, is objective evidence of a huge performance distance between the Weasel and purely random search.

    (Minor technical note — by making the criterion be getting all the way to the target phrase I am actually making the bet less favorable for myself. Weasels end up on average a little short of the target because of back-mutation. But they do hit the target once and a while and that is enough to win me money.)

    As for whether the real question being whether the word “guided” is appropriate, that of course is not solvable by running some test suite. Mung was calling for empirical tests, but suddenly that has changed into mere word games.

    On the issue of whether Weasels get to their targets far faster than purely random search, Mung has conceded that point by not taking the bet. And if Mung has not conceded that point, the very one that Dawkins was making, well then … the bet is always open.

  21. Mung: Because the “fitness” is assigned, and it’s a matter of fact.

    genome->fitness = fitness(genome);

    Even keiths knows that his WEASEL program assigns the fitness.

    The implementation of the model is not the model itself. Dawkins makes no reference to fitness in his description of the monkey/Shakespeare model of cumulative selection. Dembski acknowledges as much in No Free Lunch, and introduces a fitness function in order to treat the model as an evolutionary algorithm.

    By the same token, keiths is entirely wrong when he tells you to read his code to learn about the model.

  22. Joe Felsenstein: On the issue of whether Weasels get to their targets far faster than purely random search

    Well worded, except in the very end. An implementation may run indefinitely. It need not have the capacity to detect occurrences of the target. What matters is that we can tell, running the program, when the target occurs.

  23. You’re right. I should have called it “blind search”. That is usually defined as a search that has no information aside from its current position and whether or not the goal has been achieved.

  24. Tom:

    The implementation of the model is not the model itself. Dawkins makes no reference to fitness in his description of the monkey/Shakespeare model of cumulative selection.

    The fact that he doesn’t use the word ‘fitness’ doesn’t mean that the concept is absent from the model. Here’s Dawkins:

    The computer examines the mutant nonsense phrases, the ‘progeny’ of the original phrase, and chooses the one which, however slightly, most resembles the target phrase, METHINKS IT IS LIKE A WEASEL.

    In other words, it selects the fittest among the progeny, where fitness is defined in terms of closeness to the target phrase.

    By the same token, keiths is entirely wrong when he tells you to read his code to learn about the model.

    That’s silly. Of course you can learn about the model by reading the code.

  25. keiths,

    Of course Mung would have learned about a special case of the Wright-Fisher model, if only he had read your code, and grasped how it works — even before Joe made the connection. The concepts must be in there somewhere. And how would you have put them in there, unless you’d already grasped them?

    Yeah. That’s the ticket. You understood a special case of Wright-Fisher all along. You just didn’t know you did.

    The Weasel program is a disposable pedagogical tool. Joe used it to introduce something much more substantive. You, on the other hand, continue to parade your Weasel as an emblem of Brite supremacy. None of your pretenses of instruction are convincing.

    The notion that, because you learned something from the exercise of writing the program, others will learn something by reading what came of your exercise, is beyond silly.

  26. Mung,

    It’s time for you actually to commit to something.

    The best way to understand the bet that Joe’s offered is with a statistical model of runs of the Weasel program. It’s relatively easy, though not ideal, to do this in a spreadsheet. There’s a row for each generation. The columns are numbered 0, 1, … 28. The entry in column j of row i is the probability that the survivor of generation i matches the target in j positions. The probabilities in each row, except the first, depend only on the probabilities in the preceding row, the mutation rate, and the number of offspring. What concerns you particularly is the probability of 28 (perfect match of the target) in generation 10,000.

    I think I’ve still got a spreadsheet that I did a number of years ago. I am NOT saying that you should read the cell formulas to learn the statistical model. I AM saying that you can see how the probabilities change from generation to generation, and also that you can do whatever plots suit you. If you genuinely wanted to see the model, which depends only on elementary probability theory, I would lay it out for you.

    I don’t want you just to shut up about the Weasel. But if you’re not going to invest in understanding it a small fraction of the effort that you’ve wasted arguing about it, then you really should shut up.

    Would you bother studying the spreadsheet if I bothered to supply it?

    ETA: Would an animation of the sequence of probability distributions, depicted as bar graphs, be of any value to you? Would you eat them on a plane? Would you eat them on a train?

  27. Tom:

    Of course Mung would have learned about a special case of the Wright-Fisher model, if only he had read your code, and grasped how it works — even before Joe made the connection. The concepts must be in there somewhere. And how would you have put them in there, unless you’d already grasped them?

    The WEASEL model is simple, Tom. You’ve read Dawkins’ description. Didn’t you notice?

    Whether Mung is smart enough to grasp that simple model is an open question, but if he is, then understanding the code would certainly help him understand the model.

    As I said:

    Of course you can learn about the model by reading the code.

    Tom:

    You, on the other hand, continue to parade your Weasel as an emblem of Brite supremacy.

    LOL.

    The notion that, because you learned something from the exercise of writing the program, others will learn something by reading what came of your exercise, is beyond silly.

    That isn’t my reasoning at all. Where did you get that odd idea?

  28. Trolling troll won’t take a bet it’s guaranteed to win if trolling troll actually believes what it is saying?

    What a shocker.

  29. Mung,

    Wouldn’t it be easier for all if you just told Tom, which appears to be what he’s asking? Don’t worry, we can get WJM to change it with his mind later if it doesn’t work out.

  30. Mung: Even keiths knows that his WEASEL program assigns the fitness.

    keiths seems loathe to admit that this is in fact the case.

    Your WEASEL implementation assigns the fitness, isn’t that in fact the case, keiths?

  31. Mung:

    Your WEASEL implementation assigns the fitness, isn’t that in fact the case, keiths?

    Good grief, Mung. The program evaluates the fitness of each mutated string, assigning the result to a variable.

    Now do your patented impotence dance and try to make something out of that fact.

  32. There have been threads here where ID supporters and creationists argued that it’s not a real model of evolution unless you also simulate the interactions between organisms and their environments that bring about the fitnesses. However this thread was not one of them. Mung was raising the issue of whether there was “empirical evidence” that the Weasel programs reached their targets faster than blind search (or got significantly closer to them). Mung kept saying that there were “no objective tests”.

    Until I offered a bet, with $100 at stake. Then Mung said that

    I didn’t think the performance was in dispute, nor why the performance was better.

    Can we objectively measure the performance difference, and if so, how?

    Sure we can. We can see how much money Mung loses.

  33. Joe Felsenstein:
    There have been threads here where ID supporters and creationists argued that it’s not a real model of evolution unless you also simulate the interactions between organisms and their environments that bring about the fitnesses.However this thread was not one of them.Mung was raising the issue of whether there was “empirical evidence”that the Weasel programs reached their targets faster than blind search (or got significantly closer to them).Mung kept saying that there were “no objective tests”.

    Until I offered a bet, with $100 at stake.Then Mung said that

    Sure we can.We can see how much money Mung loses.

    Mung – put up or shut up?

  34. Mung,
    As a programmer you will know the answer to this one.

    When you calculate a value but don’t assign it, what happens to that value?

  35. keiths: Good grief, Mung. The program evaluates the fitness of each mutated string, assigning the result to a variable.

    So I was correct. Nice of you to confess that I was right. Now make of that what you’re able.

    Now how does your program “test” the fitness of genotypes?

    keiths: a GA tests the fitnesses of genotypes

    Your program assigns a “fitness” to each genotype. What on earth then does it mean to say that it “tests” the fitness of genotypes?

    Do you not see any difference between assigning a fitness and testing the fitness?

    What is the fitness of a genotype in your program before you assign it a fitness?

  36. Gotta love that Impotence Dance of yours, Mung. You’re runnning back and forth, trying to avoid both Joe’s wager and my challenge.

    It’s great entertainment for the audience, but what do you get out of the repeated humiliation?

    Why do you suppose that your omnipotent God — if he exists — is content to watch you fail, again and again?

  37. Mung:
    What wager?

    The one I quote four comments above yours. Objective morality in action / inaction.

Leave a Reply