Betting on the Weasel

… with Mung.   In a recent comment Mung asserted that

If Darwinists had to put up their hard earned money they would soon go broke and Darwinism would be long dead. I have a standing $10,000 challenge here at TSZ that no one has ever taken me up on.

Now, I don’t have \$10,000 to bet on anything, but it is worth exploring what bet Mung was making. Perhaps a bet of a lower amount could be negotiated, so it is worth trying to figure out what the issue was.

Mung’s original challenge will be found here.  It was in a thread in which I had proposed a bet of \$100 that a Weasel program would do much better than random sampling.  When people there started talking about whether enough money could be found to take Mung up on the bet, they assumed that it was a simple raising of the stake for my bet.  But Mung said here:

You want to wager over something that was never in dispute?

Why not offer a meaningful wager?

So apparently Mung was offering a bet on something else.

I think I have a little insight on what was the “meaningful wager”, or at least on what issue.  It would lead us to a rather extraordinary bet.  Let me explain below the fold …

Mung accepted that Weasel programs reach their goal far faster than random sampling.  However Mung also said (here) that

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.

and elsewhere characterized Weasels as succeeding because of “intelligence” as opposed to ignorance.

So let’s imagine what might happen if we took Mung up on the \$10,000 bet.  We would bet that the Weasel would succeed, because of cumulative selection.  Mung would bet that (because of “intelligence” or being “guided”) the Weasel would succeed.  The stake would be held by a house of some sort, which would not take a commission.

The Weasel would be run.  It would succeed.  So the house would declare that we had all won.  The stake would be given to the bettors, in proportion to their bets.  But alas, no one actually bet against the Weasel.  So the winnings would be zero.  Everyone, Mung and the rest of us, would get their stake back, and that’s all.

To bet against Mung, we have to come up with some event that distinguishes cumulative selection from “intelligence” (or being “guided”).  That seems to be the issue on which Mung was offering a \$10,000 bet, and declaring (here) us all to be “pretender[s]” because we would not put up or shut up.

So there it is.  We’re all betting on the same side, and no one will win or lose a penny.  Unless Mung can come up with some test that distinguished “intelligence” or being “guided” from cumulative selection.

Now I am possibly misunderstanding what the bet actually would be.  I hope that Mung will straighten us out on that, so that we can understand what test is proposed, and place our bets.

664 thoughts on “Betting on the Weasel

  1. Joe Felsenstein: The change may have no target or conscious goal but it is nonrandomly in the direction of higher fitness.

    And what counts as higher fitness is based upon random factors that have nothing to do with natural selection. So when evolutionists claim that evolution is non-random they do so in a deeply disingenuous manner.

  2. The mutations that selection can work on are random with respect to fitness. So are environmental changes. Unless you want to argue that the random environmental changes would always be in the direction of making the organisms in the environment less fit. Natural selection is surrounded by and buffeted by randomness on all sides. The overall effect is randomness. For all intents and purposes natural selection may as well be random.

    That organisms are always and forever getting more fit is a myth.

  3. Mung: So when evolutionists claim that evolution is non-random they do so in a deeply disingenuous manner.

    Oh, come on, mung. Trying to be charitable, I’ll allow you’re mixing how with why. Science won’t tell you why, only how.

  4. Mung: The mutations that selection can work on are random with respect to fitness. So are environmental changes.

    Yes. And natural selection is the process by which the phenotypical effect of those mutations are sampled in that environment.

    Unless you want to argue that the random environmental changes would always be in the direction of making the organisms in the environment less fit. Natural selection is surrounded by and buffeted by randomness on all sides.

    What does that even mean, “surrounded by and buffeted by”? Your whole post hinges on this very imprecise and strange statement.

    The overall effect is randomness.

    That doesn’t follow from anything you’ve said. The closest you came to supporting this claim is when you said natural selection is “surrounded by and buffeted by randomness on all sides”.

    But that sentence is nonsensical, as natural selection doesn’t have a location that can be surrounded, and it doesn’t have sides. So what the hell you meant to say there is anyones guess, but the conclusion you derive from it doesn’t actually logically follow.

    For all intents and purposes natural selection may as well be random.

    That doesn’t follow. The fact that some of the inputs in the equation are stochastic variables doesn’t mean you can conclude the output “may as well be random”.

    That organisms are always and forever getting more fit is a myth.

    Yes it is, a myth you just made up. I’ve read quite a bit of evolution literature and I’ve never come across the claim that organisms are always and forever getting more fit. Where do you get this shit?

  5. Female cowbirds whose eggs are ejected from host nests often come back and destroy those nests, resulting in a decrease in the contribution of ejectors (a genetically determined trait) in the next generation. Is that natural selection or artificial selection? Is it guided or unguided?

  6. Joe Felsenstein: Erik leaves out an important aspect of artificial selection (which I discussed a couple of times in this thread, one in the comment immediately preceding Erik’s). The breeder selects on the trait (say butterfat percentage in dairy cow milk) but does not have any knowledge of exactly how this will be achieved, which genes will change.

    The emphasized bit is either false (because anyone, such as yourself, can look up the details these days, and even before the modern times breeders knew well what they wanted and they knew when they had got it; they also knew what’s obtainable and what’s not) or irrelevant (because it doesn’t explain away the difference between natural and artificial selection). As it is, it’s both false and irrelevant.

    Joe Felsenstein: And that is why Darwin used artificial selection as a close analogy to natural selection.

    But it’s not a good analogy because artificial selection is not the survival of the fittest. It’s the survival of those who are preferred by the whim of the breeder.

    Joe Felsenstein: Yup, all those slime molds and sequoia trees and algae and sponges are blissfully unaware of what is happening. If that disturbs you, my condolences.

    But breeders know. Hence the difference.

    Joe Felsenstein: So Erik thinks that natural selection isn’t really there?

    It is there, but it’s not what you think it is.

    Joe Felsenstein: Dawkins’s argument was countering the misleading arguments of creationist debaters, who endlessly drum into their audiences’ heads that evolutionary biologists are explaining adaptation by “random” change.

    He’s not doing a good job. The weasel thingie presents a target. Evolution supposedly does not have one. Fail.

  7. Rumraket: I’ve read quite a bit of evolution literature and I’ve never come across the claim that organisms are always and forever getting more fit. Where do you get this shit?

    I get it right here at TSZ of course. From the best and brightest.

    Joe Felsenstein: The change may have no target or conscious goal but it is nonrandomly in the direction of higher fitness.

    Are you, Rumraket, saying it’s in the direction of higher fitness except when it isn’t?

  8. And it’s yet another myth that all organisms in a population are always tracking the same target. Every single member could well be on it’s own evolutionary trajectory. What’s fit for one must be fit for all is nonsense.

  9. Mung: And it’s yet another myth that all organisms in a population are always tracking the same target.

    You’re not making sense here. Are you attacking a strawman consisting of “all organisms in a population are always tracking the same target”?

    Every single member could well be on it’s own evolutionary trajectory. What’s fit for one must be fit for all is nonsense.

    More strawmen! Individual organisms are just taking the opportunity to live – that is presented in the niche in which they find themselves.

  10. Mung: Rumraket: I’ve read quite a bit of evolution literature and I’ve never come across the claim that organisms are always and forever getting more fit. Where do you get this shit?

    I get it right here at TSZ of course. From the best and brightest.

    No, you don’t.

    Mung quoting Joe Felsenstein: The change may have no target or conscious goal but it is nonrandomly in the direction of higher fitness.

    Which is true, selection is a bias that always biases in the direction of higher fitness. But Mung, your mistake here is that nobody is claiming it always succeeds.

    The overall fitness of a population can be in decline, even as selection is biasing the change towards higher fitness. In such a case, the rate of decline is slowed down, not overtaken, by selection.

    Imagine you’re on your bike rolling downhill. That’s the fitness decline of the population over time. But then you take out a small propeller from a toy remote controlled helicopter and turn it on so it produces thrust and you point it so it pushes you in the direction back up the hill. But you’re too heavy for the propeller to totally overtake your downhill roll, but it does apply a force in the direction back up the hill. So it just slows down the rate of decline.

    That way, selection (the direction of thrust of the propeller) is uphill (higher fitness), but the overall fitness (the direction you are traveling on your bike) is still downhill.

    Are you, Rumraket, saying it’s in the direction of higher fitness except when it isn’t?

    No, selection is always in the direction of higher fitness. It’s just that it might not succeed at overtaking (for example) drift, or the changes that happen in the environment are too fast. That doesn’t mean selection is somehow not in the direction of higher fitness, it still is and always will be, it’s just that it can be overwhelmed by another, stronger effect.

  11. Mung: And it’s yet another myth that all organisms in a population are always tracking the same target. Every single member could well be on it’s own evolutionary trajectory. What’s fit for one must be fit for all is nonsense.

    Selection is a population-level phenomenon. There can’t be selection operating on an individual to change it. Individuals might die to selection, but then it is the population as a whole that changes (the frequency of the allele the individual carried, changed). It is not the individual that changed, the individual reproduced (or not) and then died at some point.

    If there are multiple different selective pressures in operation, or just a single one but they don’t uniformly apply to the entire population, you might actually get speciation as some aspects of the population change in a different direction (or just more) from others (and if there’s no geneflow between them of course).

    Congratulations, you’re re-discovering a basic aspect of evolutionary theory: How diversity emerges.

  12. A couple of thoughts:

    1) It would be possible to generate a Weasel program in which the ‘target’ was not chosen in advance, but as the first step.

    — a) Generate a ‘random’ 26-character string.
    — b) Run the Weasel, measuring fitness by closeness of match to this string, preferentially retaining the closer. Methinks it is like %^SA@ds.

    So, ALL implementations of Weasel are guided (or not) to the same extent. The cumulative ‘guidance’ is provided by the fact that there is a smooth slope from any random string to any other, not by the fact that someone has knowingly picked that target, or has any idea what it is.

    2) Any concerns about ‘fine-tuning’ mutation rate could be assuaged by adding mutation rate as a variable – a gene – attached to the strings (as, indeed, in life: variation in mutation rate happens, and is reasonably expected to be under selection, at least in part). Then you allow mutation rate to mutate (itself dependent on mutation rate). My prediction is that the mutation rate would approach an optimum without anyone needing to fine tune it. Strings with the optimum attached rate would increase in the population. Note that this optimum cannot sensibly be zero.

    3). The word ‘random’. Opportunities for talking at cross-purposes abound, and there are plenty of examples in this thread.

    4) Natural and artificial selection are the same thing, not analogous processes. They both bias populations by affecting the retained traits in the breeding pool, the only difference being that, in one, there is definitely an intent to do so, in a particular way. Weasel isn’t artificial selection – especially with the modifications proposed above.

    5) I am somewhat slack in my usage of the word ‘couple’.

  13. Allan Miller: — a) Generate a ‘random’ 26-character string.
    — b) Run the Weasel, measuring fitness by closeness of match to this string, preferentially retaining the closer. Methinks it is like %^SA@ds.

    Huh? This is not defining a target in advance, because first you let a computer select a target in advance??

  14. phoodoo,

    Huh? This is not defining a target in advance, because first you let a computer select a target in advance??

    It is not a designer-chosen target, is actually the point. Nor is it guided.

    Clearly, it is impossible to use Hamming Distance as a fitness criterion without there being a point against which the distance of any given string is measured. Is that the problem with Weasel?

  15. So here’s the heart of the supposed problem: there is no string in the space-of-all-possible strings that cannot be located by ‘Weasel’ faster than random search, using the criterion of Hamming distance as the distinction between candidate strings. But there is no possible implementation of such a distance criterion which is not ‘guidance’.

    Correct?

  16. Allan Miller,

    Oh what poppycock, of course it is a designer chosen target. Its not different than saying, spill a bunch of letters on the floor, then whatever letters fall, those are the ones you have chosen. You still have chosen THAT random sequence that the computer generated.

    Why do you expect thinking people to so easily be fooled by such a silly attempt at disguising a choice of target? Are most of the people you encounter in your daily life so gullible?

  17. phoodoo:
    Allan Miller,

    Oh what poppycock, of course it is a designer chosen target.Its not different than saying, spill a bunch of letters on the floor, then whatever letters fall, those are the ones you have chosen.You still have chosen THAT random sequence that the computer generated.

    Why do you expect thinking people to so easily be fooled by such a silly attempt at disguising a choice of target?Are most of the people you encounter in your daily life so gullible?

    OK, so you think that if there’s a target, that counts as evidence for intelligent design. So a weasel algo that uses a random target, random mutation and cumulative selection produces “intelligent-design”. right?

  18. phoodoo: Oh what poppycock, of course it is a designer chosen target. Its not different than saying, spill a bunch of letters on the floor, then whatever letters fall, those are the ones you have chosen. You still have chosen THAT random sequence that the computer generated.

    If design requires foresight then using a random sequence is not designing the specific target. Is foresight required for design or not?

  19. phoodoo,

    Why do you expect thinking people to so easily be fooled by such a silly attempt at disguising a choice of target?Are most of the people you encounter in your daily life so gullible?

    I’m not attempting to pull the wool over your eyes here, phoodoo. I know you’re far too smart for that. I’m just asking if I’ve framed your problem correctly. You are saying that there is no string in the entire space-of-all-possible-strings that cannot be said to have been ‘chosen’. Even if that ‘choice’ involved generating such a string entirely at random. Is that it?

    What if one used a fitness criterion that did not involve comparison to a specific string? Suppose, for example, that one added up the hex values of the characters in ASCII, and regarded those closer to a sum of 322 as more likely to survive than those further away? Is that chosen? What, then, if that number itself were a random number picked at runtime? Is there no way of implementing differential fitness in a program without ‘choice’? None at all?

  20. Allan Miller,

    You are saying that there is no string in the entire space-of-all-possible-strings that cannot be said to have been ‘chosen’. Even if that ‘choice’ involved generating such a string entirely at random.

    Indeed, every member of the starting population, and every mutated intermediate, has also been phoodoo-chosen. I think this is a somewhat creative use of the term ‘choice’.

  21. newton: If design requires foresight then using a random sequence is not designing the specific target.

    That does not follow at all newton. If you are going to generate the target by means of an algorithm you have to choose and design and implement the algorithm to choose the target, and that requires foresight.

  22. dazz: OK, so you think that if there’s a target, that counts as evidence for intelligent design.

    It means there’s some other cause at work other than cumulative selection. Natural cumulative selection is not supposed to know about long range targets, randomly generated or not. Artificial cumulative selection on the other hand, is allowed to be guided towards a long term goal.

  23. Allan Miller,

    Allan, why can’t you see, as it is so blazingly obvious, that whatever you decide fitness means, is tied to whatever string you chose.

    What don’t you try this, chose a random string, but declare fitness to be whatever is close to entirely different random string. Now see if you get to the first string, by setting fitness as tied to something entirely different.

    So chose a random string of letters say 30 characters long. Then chose fitness as whatever is closest to a second mutated version of your first random string. Then see how long it takes to arrive at : “Methinks it is like a weasel.”

  24. phoodoo: Why do you expect thinking people to so easily be fooled by such a silly attempt at disguising a choice of target?

    iirc, Elizabeth liked to argue that natural selection met Dembski’s definition of intelligence (or something along those lines) because it involved making a choice.

    Now they want to deny that selecting a target, by whatever means, involves making a choice.

  25. Alan Fox: ETA and if you run it on a computer, that was designed too!

    You are of course permitted to use a computer to model or simulate. But the map is not the territory, and people here seem to forget all about that when it comes to Weasel.

  26. phoodoo,

    So, you are saying that if you don’t implement a consistent genotype-related fitness differential, you don’t outperform random search. But we kind of know that. It’s Weasels that do outperform blind search that we are betting on. And you are defining the case, or so it seems, such that it is impossible to cause any non-intentional consistent differential in fitness, by any means whatsoever, in a computer program. That is a minority position.

  27. Rumraket: Let’s get an independent judge we can both accept, and then you put 1000% dollars on the line. If I come up with a definition and a way to measure cumulative selection, to the satisfaction of the independent judge, you pay me 1000% dollars! Deal?

    Do you want me to donate to charity or do you want the money? My offer was a donation to charity.

  28. Mung,

    iirc, Elizabeth […]

    Now they […]

    You do realise that there is not a Church of Darwinists, singing from a hymn sheet? There is, in any case, no inconsistency between regarding NS as tantamount to ‘intelligence’, and ‘choices’ made between genotypes during a run of selection being completely blind and unintentional.

    Killing a disproportionate number of hairless individuals in a cold snap leads to an increase in the proportion of the population that can withstand the cold. It seems an intelligent thing to do. Is it?

  29. Please read:

    Mung: My question is, where did I ever claim that Weasel programs do no better than random search or offer to wager that Weasel programs do no better than random search?

  30. phoodoo (and Mung):

    I previously asked Mung to write out the “search problem” that is solved by the Glass model, which is an extension of Dawkins’s monkey/Shakespeare model (see “Evolution Is Not Search”). Note that

    Problem: Output “METHINKS IT IS LIKE A WEASEL”

    is not a problem. Let’s see you write a problem to which “METHINKS IT IS LIKE A WEASEL” is the solution, without referring to the solution itself.

    In the Glass model, the “target string” is merely a parameter of the fitness function. Dawkins says nothing about fitness. Nonetheless, the “target phrase” to which the monkey compares the “mutant ‘progeny’ phrases” is a parameter of the process. The target phrase is not extrinsic to the process. There would be no process at all without some definition of what is selected and what is not selected in the process of cumulative selection. You’re confusing definition of the process with design of the process to generate a desired outcome.

    Write out the problem that Dawkins designed the program to solve.

  31. Mung: It means there’s some other cause at work other than cumulative selection. Natural cumulative selection is not supposed to know about long range targets, randomly generated or not. Artificial cumulative selection on the other hand, is allowed to be guided towards a long term goal.

    The weasel algo with a random target works just fine, doesn’t it? The programmer could even identify the randomly generated target once it’s finished.
    So cumulative selection plus some randomness (random target generation and random variation) work much better than random sampling, are you saying that this is because some other “cause” not in the code, is making it possible?

  32. So what is the problem? Dawkins knows what the problem is:

    “The problem is that of complex design.”

  33. Allan, it’s directed towards anyone reading this thread who thinks that I’ve been arguing that Weasel performs no better than random search or was willing to make any wager along those lines.

  34. Mung,

    I’m assuming that’s a rhetorical question.

    Of course. But the conjunction ‘first Elizabeth … now they …’ suggests a failure to separate out that which you should be perfectly capable of separating out without assistance.

  35. dazz: The weasel algo with a random target works just fine, doesn’t it?

    It depends.

    The programmer could even identify the randomly generated target once it’s finished.

    If it ever finishes.

  36. Allan Miller: …suggests a failure to separate out that which you should be perfectly capable of separating out without assistance.

    No, I thought you were being sarcastic. Of course pretty much everyone here who opposes ID reads from the Darwinian hymn sheet. It’s the only alternative to ID that they have.

  37. Mung: It depends.

    If it ever finishes.

    OK, so let’s say that it finishes. Is that, according to you, an example of ID at work?

  38. Mung,

    It need never finish. But the ‘random Weasel’ is no more or less likely to finish than the Dawkins Version. They are identical in every respect, except the specific string deemed the ‘target’. One could of course decide on some measure of success – a threshold mean Hamming Distance for the population. But when-to-finish is hardly a problem for evolution.

  39. Mung,

    No, I thought you were being sarcastic. Of course pretty much everyone here who opposes ID reads from the Darwinian hymn sheet. It’s the only alternative to ID that they have.

    And the Darwinian hymn sheet has us concur on whether NS is usefully described as intelligent or not? (That too is a rhetorical question, of course).

  40. dazz: So cumulative selection plus some randomness (random target generation and random variation) work much better than random sampling, are you saying that this is because some other “cause” not in the code, is making it possible?

    I listed a number of other causes in a prior post. Meanwhile, I don’t know what you mean by “cumulative selection” since you haven’t defined the term.

  41. Allan Miller: But when-to-finish is hardly a problem for evolution.

    This is true. And yet another reason why the Weasel program is not like real life.

  42. Allan Miller: But the ‘random Weasel’ is no more or less likely to finish than the Dawkins Version.

    If I were to design a ‘random Weasel’ program it would be more than just the target phrase that was generated at random. 🙂

    Weasel works because it is decidedly not random.

  43. Mung: I listed a number of other causes in a prior post. Meanwhile, I don’t know what you mean by “cumulative selection” since you haven’t defined the term.

    You know full well what is cumulative selection in the algo.
    Do you agree that this random target weasel works by means of randomness plus cumulative selection?

  44. Mung,

    This is true. And yet another reason why the Weasel program is not like real life.

    Was it supposed to be?

Leave a Reply