A “powerful, visual statement” of a fallacy

ETA 2015/12/05:Rubik’s Cube Is a Hand-Sized Illustration of Intelligent Design,” featured in the Discovery Institute’s Evolution News and Views, illustrates clearly the fallacious “conservation of information” reasoning discussed in a post of mine, “The Law of Conservation of Information Is Defunct.”

Checking for the next round in Pooh Bear’s popgun defense of Idlet, I find that Evolution News and Views has seen fit to feature a post in the mathematics category, “Rubik’s Cube Is a Hand-Sized Illustration of Intelligent Design,” but not to identify the author. The work is unmistakably that of Dennie O’Sneery, reprocessing reprocessed technical material that s/he does not understand, and reassuring the dimmest of blessedly assured wits that they are filled with the Holy Sass. Although it’s generally amusing to smash Dennie’s ambiguous little thingy to a bloody pulp, when s/he masochistically points it my way, I don’t have time for sadism now. Not to discourage you from whacking away (biohazard suit and ten-foot pole recommended), I’m treating this as a teachable moment.

1. What is the single most important point in my argument that the Conservation of Information Theorem does not apply to scientific investigation of nature?

2. O’Sneery conflates engineering of a process that generates a desired outcome with emergence of a natural process that generates a remarkable outcome. Do you see how s/he tacitly illustrates misapplication of the theorem?

24 thoughts on “A “powerful, visual statement” of a fallacy

  1. Note: There are two notions of what constitutes a move in solving the puzzle. If each move is a quarter-turn of a face, then the maximum number of moves required for a solution is 26. If a half-turn (two quarter-turns) of a face counts as a single move, then the maximum number of moves required is 20. O’Sneery plays mix-and-match with the two definitions of a move. There are even sillier errors in his or her post, and I hope you will identify them.

  2. I’ll just park this here:

    http://www.genetic-programming.org/hc2010/7-Borschbach/Borschbach-Evo-App-2010-Paper.pdf

    “Abstract. Solutions calculated by Evolutionary Algorithms have come
    to surpass exact methods for solving various problems. The Rubik’s Cube
    multiobjective optimization problem is one such area. In this work we
    present an evolutionary approach to solve the Rubik’s Cube with a low
    number of moves by building upon the classic Thistlethwaite’s approach.
    We provide a group theoretic analysis of the subproblem complexity induced
    by Thistlethwaite’s group transitions and design an Evolutionary
    Algorithm from the ground up including detailed derivation of our custom
    fitness functions. The implementation resulting from these observations
    is thoroughly tested for integrity and random scrambles, revealing
    performance that is competitive with exact methods without the need
    for pre-calculated lookup-tables.”

    Note to Phoodoo – the rest of the paper is below the abstract, available to all, which you could read, but presumably won’t.

  3. Winnie-the-Pooh and the Killer Bug

    Synopsis. The Pooh Bear of Intelligent Design loses sphincter control at the sight of a scary bug, makes dookey in the honey pot, and feeds Idlet a tasty new treat.

    Relevance to this post?

    The GUC Bug does worse than random queries, due to getting stuck in a local optimum rather quickly. I hardly need to rehearse the insufficiency of even large numbers of random queries to solve biological problems. This model will have an even harder time solving problems.

    Pooh is a stupid little bear — so stupid that it would never occur to him, no matter how long he spent in his thinking place, that stupid little bears make very bad liars.

  4. Richardthughes,

    Or read this:

    “The GUC Bug finds a sequence better than 99.98% of all other sequences, which may sound impressive. But consider, as do English and Felsenstein, this algorithm running for fifty generations. That corresponds to 150,000 different sequences. If we simply took 150,000 random genotypes, we’d expect to find one better than about 99.999% of all the other genotypes. The GUC Bug does worse than random queries, due to getting stuck in a local optimum rather quickly. I hardly need to rehearse the insufficiency of even large numbers of random queries to solve biological problems. This model will have an even harder time solving problems.”

    But no, instead of addressing this, Tom decides he can do better by thinking of his hilarious alternatives for the name O’Leary. What a scholar.

    Maybe you can teach him how to write butthurt too huh? That’s always a reliable retort isn’t it Richard?

  5. phoodoo:
    “The GUC Bug finds a sequence better than 99.98% of all other sequences, which may sound impressive. But consider, as do English and Felsenstein, this algorithm running for fifty generations. That corresponds to 150,000 different sequences. If we simply took 150,000 random genotypes, we’d expect to find one better than about 99.999% of all the other genotypes.
    . . . .”

    In addition to Ewert’s latest, anyone interested in this topic should read Joe Felsenstein’s original post on The Panda’s Thumb.

    Ewerts’s claim that you quoted is not correct. JF and TE consider a genome length of 1000 with four possible bases. There are 1.148 times 10 to the power of 602 such strings. Choosing 150,000 strings at random out of that large a pool is highly unlikely to result in “one better than about 99.999% of all the other genotypes.”

    Hopefully the two authors of the original work will correct me if I’m wrong.

  6. Patrick: Ewerts’s claim that you quoted is not correct. JF and TE consider a genome length of 1000 with four possible bases. There are 1.148 times 10 to the power of 602 such strings. Choosing 150,000 strings at random out of that large a pool is highly unlikely to result in “one better than about 99.999% of all the other genotypes.”

    Hopefully the two authors of the original work will correct me if I’m wrong.

    Sorry, but the raw size of the space of genotypes is irrelevant here. Only the percentage matters. I haven’t checked the calculation, because it’s merely an evasion that Ewert has ginned up. I’m working on a relatively polite technical response, which I want to segregate from the nasty response to the Click Whore of Babylon. Though it won’t show in my post, addressing you directly has gotten me a draft of some text. For some weird (or not) psychological reason, I’m much better at writing to persons than peoples. So thanks for the comment.

  7. Tom English,

    Sorry, but the raw size of the space of genotypes is irrelevant here. Only the percentage matters.
    . . .
    Though it won’t show in my post, addressing you directly has gotten me a draft of some text. For some weird (or not) psychological reason, I’m much better at writing to persons than peoples. So thanks for the comment.

    I don’t see my error, but I’ll wait for your full response to Ewert. I’m happy to continue babbling incoherently at you if that will help.

  8. Patrick,

    When you randomly choose 150,000 genomes, the probability of getting at least one genome that is better than 99.999% of the others is equal to one minus the probability of not getting a single genome that is better than 99.999% of the others.

    The latter probability is equal to .99999^150,000, or 0.223.

    1 – 0.223 = 0.777, or a 77.7% probability of getting a genotype better than 99.999% of the others.

  9. keiths,

    When you randomly choose 150,000 genomes, the probability of getting at least one genome that is better than 99.999% of the others is equal to one minus the probability of not getting a single genome that is better than 99.999% of the others.

    The latter probability is equal to .99999^150,000, or 0.223.

    1 – 0.223 = 0.777, or a 77.7% probability of getting a genotype better than 99.999% of the others.

    Thanks. I’m not sure where my head was after reading Ewert’s post. I’ll try it again tomorrow.

  10. Tom English,

    Oh she is the Click Whore of Babylon is she then? Well, why didn’t you say so earlier!

    Now I really am going to deeply regret your technical response. Whoa! Are you go to employ elements of Richards “Butthurt” strategy?

  11. Patrick,

    Thanks. I’m not sure where my head was after reading Ewert’s post. I’ll try it again tomorrow.

    Yeah, the problem isn’t that Ewert’s calculation is wrong — just that it’s irrelevant. He’s comparing apples to oranges.

    In the GUC model, the current genotype is effectively thrown into a pot with its 3,000 immediate neighbors in the fitness landscape. Selection’s role is to pick the fittest genotype out of the 3,001 candidates. That genotype then replaces the current genotype. If the current genotype is fitter than all of its 3,000 neighbors, then it “replaces” itself.

    To do an apples-to-apples comparison between the GUC model and a random search, Ewert should have compared the above scheme to one in which the current genotype is replaced by one chosen randomly from the 3,001 candidates.

    Such a scheme would do far worse than the GUC model, of course.

    Ewert didn’t do the apples-to-apples comparison. Instead, his method effectively works like this: at each iteration, choose 3,000 genotypes at random from the entire genotype space. Throw them into a pot with the current genotype, then use selection to pick the best one, replacing the current genotype with it.

    Ewert employed selection without even realizing it — or without acknowledging it, if he did realize it. He wasn’t comparing selection to random search, he was comparing selection with a constrained search space (neighbors only) to selection with the same number of candidates taken from an unlimited search space. Of course the latter performs better on average with a random fitness landscape, but that’s just one form of selection performing better than another. It says nothing about the merits of selection vs. random search.

    Ewert also could have made an oranges-to-oranges comparison, but didn’t. The oranges-to-oranges comparison would have looked like this:

    In the GUC bug case, choose 3,000 genotypes at random from the entire genotype space, not just the immediate neighbors. Throw them into a pot with the current genotype and pick the one with the highest fitness. Use it to replace the current genotype.

    In the random search case, choose 3,000 genotypes at random from the entire genotype space. Throw them into a pot with the current genotype, then pick one at random and use it to replace the current genotype.

    Again, it’s obvious that the second scheme will do much worse than the first.

    It was a silly mistake, but Ewert makes a lot of those.

  12. phoodoo:
    keiths, What’s the definition of fitter?

    Fitter is always comparative and depends on the niche that is being tested or considered. You can only sensibly talk about relative fitness with regard to individuals or populations in a niche. Sharks are not good at survival in deserts, for example.

  13. Alan Fox,

    Then how do you have an algorithm that selects fitter?

    Furthermore, I disagree with your premise. In evolution, the thing that survived is fitter. There is no definition for fitter other than this, because there is no definition for any environment-environments are fluid and random.

    This has been discussed here many times, and no one has ever been able to define fitter. Particularly since an organism is a combination of so many traits, are long teeth fitter, at the same time as rough fur, at the same time as short legs, at the same time as propensity towards a certain cancer, at the same time as the ability to jump higher, at the same time as a bigger nose? If it breeds it is. Evolution is not choosey.

  14. phoodoo: Evolution is not choosey.

    This is wrong. The niche is choosy. That’s the non-random aspect of selection. Sharks are much fitter in the sea than in the desert. The desert niche environment eliminates sharks.

  15. Alan Fox,

    Yea, that all sounds good and reasonable doesn’t it Alan. Heck, its like we could say healthy is more fit than sick, beautiful is more fit than ugly, thin is more fit than fat, fast is more fit than slow.
    Unfortunately you can’t say that with evolution Alan. All you have is what survived. Fat ones and skinny ones and slow ones and fast ones,…

    There is no way to make such definitions in evolution’s (so called theory). What is fittest is what survived.

    So who does the algorithm decide what is fitter before the fact?

  16. keiths,

    To do an apples-to-apples comparison between the GUC model and a random search, Ewert should have compared the above scheme to one in which the current genotype is replaced by one chosen randomly from the 3,001 candidates.

    Indeed. I re-read his post this morning. The first time I think I was expecting more confusion on Ewert’s part about the power of cumulative selection (most IDCists still don’t understand Dawkin’s Weasel). My bias, my bad.

  17. Patrick: The first time I think I was expecting more confusion on Ewert’s part about the power of cumulative selection (most IDCists still don’t understand Dawkin’s Weasel). My bias, my bad.

    You’re really on a roll today Patrick!

    First the claim that design arguments have been refuted and now this.

    I can’t wait for the third shoe to drop! I feel a new OP coming on. 🙂

  18. AIUI, the fallacy is that to evaluate the performance the GUC bug, we only consider the 50th genotype, not all the genotypes explored (of course this is because the last one could never be less fit than any of the other genotypes explored). But Ewert is asking us to compare the final* position of the GUC bug with every genotype traversed by his random bug at some point. This is absurd. The final position of the random bug would NOT do better than that of the GUC bug.

    (*) final in the sense that the program was stopped after 50 generations. There is nothing special about the fiftieth generation, obviously.

  19. Vanellus vindex:

    AIUI, the fallacy is that to evaluate the performance the GUC bug, we only consider the 50th genotype, not all the genotypes explored (of course this is because the last one could never be less fit than any of the other genotypes explored). But Ewert is asking us to compare the final* position of the GUC bug with every genotype traversed by his random bug at some point. This is absurd. The final position of the random bug would NOT do better than that of the GUC bug.

    Hi Vanellus,

    Welcome to TSZ.

    I think you and I are making the same point from different angles.

    Your point is that to do a fair comparison, we must look only at the 50th generation of both the GUC bug and the random bug. That’s true, and it’s equivalent to my point, which is that the random bug can’t “cheat” by using selection.

    The only way for the random bug to achieve Ewert’s 99.999% performance claim is if it uses selection. But if uses selection, then it isn’t a random bug.

Leave a Reply