Uh-oh — I don’t think the metaphor works

This cartoon does sort of accurately illustrate how Muller’s Ratchet works, but like any metaphor, it drags in a lot of baggage.

Here’s what I don’t like: every step involves intent. In the metaphor, every change has a purpose and brings in a functional element in addition to the random noise and degradation, and the recombination step is portrayed as combining specific, useful parts. Intelligent design creationists will love this cartoon, which is too bad.

I guess my primary objection is that last panel that says “…if you have an image editor that lets you spice together parts of two images, you can make a new version with the best parts of both“. Recombination doesn’t discriminate what parts are best. It’s random.


  1. blf says

    @1, …sadly their families objected — one was Pixels and the other Vectors, and from the different lands of RGB and CMYK — but with no enlightened friar…

  2. naturalistguy says

    If you’re a fundamentalist determinist though, intent is also a random thing. That won’t please the god-botherers who want a super-natural designer behind it all.

  3. says

    You’re quite right to worry that the analogy involves intent. Miller’s Ratchet Is undone by recombination, because error-free copies keep getting generated after they have disappeared from the population. But they are not instantly recognized as better and made into the template for the future. They just on average have more descendants, because natural selection. And recombination also generates copies that contain more deleterious mutants.

  4. PaulBC says

    I’m not a biologist, but after looking up Muller’s Ratchet, I think I understand the principle. Noise accumulates when copying, but noisy copies can be combined to attain one with less noise. This works for bit strings, for instance, and is easier to see in that context than with memes.*

    I think if one’s goal is to understand it, trying to map it into folk wisdom (i.e. using a complex and potentially flawed analogy) is less effective than abstracting away all the biology and talking about bit strings with uniform probability of errors. This has the advantage that you can develop models that are very easy to simulate and observe. You can conclude: I know it works here, so it’s plausible that it works in a more complex system. Taking the analogy route, you have to concede uncertainly about both your metaphor and its supposed application.

    *I was really hung up on whether memes really get altered all that much. In my experience, it’s much more often the same photo with different text. And I thought the point was going in an entirely different direction. If I had stopped in the middle I would have been left thinking Muller’s ratchet was an observation about rhetoric or politics like the Overton window.

  5. PaulBC says

    In the 70s, when I was very little, my father had a series of old xerox copies that started with a blank paper and accumulated first a few spots and propagated those to the next copy along with new ones until the final copy was just a big blotch of black. He died a long time ago, so he’s not here to explain to me what that was for, but I think he tried to at the time. He did some signal processing in the 60s so maybe this was part of his work. It may have also been something he made out of personal curiosity.

    Before computers were ubiquitous, there was lot of analog copying going on, so people would have been familiar with problem of accumulated error in n-generation copies of a music tape. The Philip K Dick story Pay for the Printer (1956) seems to carry this suspicion about copies of copies.

    It was this kind of intuition that left me believing at least until college that it was paradoxical or impossible to make a faithful copy of anything. In fact, most of this is solved in a digital context with error-correcting codes. Assuming you can make multiple copies from an original, any amount of noise less than a 50% chance of inverting a bit (which is a complete loss of information) can be reduced to an arbitrarily small percentage with enough copies. Oddly, the first place I read about triple redundancy was in an old, possibly original, copy of L. Ron Hubbard’s Dianetics article that was also lying around the house! (But I did study Hamming codes later and know at least some introductory information theory.)

    Today, we expect to be able to transmit gigabits from the other side of the world without any loss. The errors are happening constantly, but they’re getting corrected and probably a lot of people just aren’t aware there’s a hard technical problem being solved for them. So I think there’s a cultural shift on what we expect to be able to copy with high fidelity. Maybe an analogy about memes is necessary for those who are not old enough to remember how noisy any copy is unless you apply some kind of redundancy and error correction.

    (I also realize that living things do not “correct” in the sense of trying to match an original, but are more likely to pass on the traits better adapted to the environment, which may or may not be the original ones.)

  6. birgerjohansson says

    BTW, if nature worked like that to escape signal degradation, I assume we would get neither cancer nor ageing, but no benign mutations either. The result would be a perfectly healthy precambrian soup of microbes.

  7. blf says

    @6, Indeed. However, the copied data may itself be lossy, e.g., JPEG images, which can incur loses each time it’s recompressed (rewritten) even if no intentional change was made. Whilst I haven’t tried this myself, my understanding is you can see something akin to that photocopy example by opening a JPEG image in your favourite image editor, rewriting the image (without making any changes), again and again… supposedly, and I find this very plausible, the image will degrade and degrade. This is despite the storage, memory, etc., being “error-free” (e.g., in the case of SSDs, “protected” by a suitable ECC).

    (Some quick searching suggests both the photocopy and JPEG examples are examples of “generational loss”.)

  8. stroppy says

    If cultural evolution is just an expression of advanced chemical evolution, then perhaps intent is just an artifact of that process,

  9. stroppy says

    @ 9
    JPG. Lossy says it all. If my degrading memory serves, it’s less of a problem of TIFF RLE.

  10. PaulBC says

    birgerjohansson@7 I’m sure this is obvious to biologists, but it only really struck me a few years ago that since clearly there is no error correction (or nothing that would be considered very reliable) you need natural selection just to preserve viability (among other things, explaining why some regions are more conserved than others).

    Unlike digital error correction, there is no mechanism that favors similarity to the original sequence several generations back. You have a good chance of preserving the same sequence because it’s better adapted than most random mutations. But a mutation that works just as well would have the same chance.

    I suppose recombination throws in the additional factor that most extant copies of a region will look like the original, so they can often crowd out mutations that are just as well adapted, reducing the amount of drift.

    My take on creationists is they believe the passage of time is always the straying from an ideal determined at creation. This principle has been expressed in terms of Platonism, it shows up in Golden Age myths, and it has been shoehorned to fit into a misunderstanding of entropy. There is also a teleological element to it: e.g. a hinged device that functions as a pair of pliers must be used as pliers. It cannot be used as a hammer, and an assembly that falls by chance into the shape of such a hinged device can never really be a pair of pliers. Its functioning as pliers is proof that it was designed to be.

    It is a worldview that shows up very natural questions like “What is this for?” or “Am I doing this right?” It’s reinforced by beliefs about religion, purpose, and hierarchy. I think the hazard with nearly any analogy about evolution is that most people are going to map it to their notion of teleology unless they are working hard to avoid it. This is why I would prefer simplified, formalizable models over metaphors.

  11. blf says

    @11, RLE is lossless. TIFF supports multiple compression methods, all(?) of which are lossless except JPEG’s (which is supported by TIFF but I believe is rarely used). The space savings from RLE aren’t necessarily great (there are “more compressive” lossless algorithms, some of which are also supported by TIFF (e.g., LZW, which I believe is what is usually recommended for TIFF if storage space is an issue)), but RLE’s big advantage — from my lossy degrading memory — is it’s extremely fast (both compressing and decompressing).

  12. says

    If you’ve got mutations that are mostly deleterious piling up, without there being any natural selection against them, yes, the genome deteriorates. But that’s not the situation described by Miller’s Ratchet. That happens when there is deleterious mutation, natural selection, a small enough population size, and no recombination. As PZ is pointing out, the copy machine analogies people are using are not close enough, especially when natural selection is replaced by humans picking and choosing.

  13. PaulBC says

    Joe Felsenstein@15 This seems like a case in which any analogy is more likely to confuse than illuminate. I think the comic does capture a useful point about using redundancy to recover signal from noise. However, the fact that I read it that way may illustrate that I am missing the main point of Muller’s Ratchet. At least, the more I think about the analogy, the further I probably get from understanding the biology. Maybe it’s best just to study the actual system in question.

  14. unclefrogy says

    the problem is the over simplification leaves out details for explanation reasons and is in turn misleading as a result. It works as a description of the photo meme and it’s popularity but only selectively

  15. says


    It very much depends on the “quality” setting that the JPEG encoder uses.

    I performed an experiment using ImageMagick‘s convert program, where I re-encoded an image ten times with the default quality setting 80%. When comparing the original to the last copy (using ImageMagick’s compare) there were some differences. However, visually I could not spot the difference. There also was very little difference between the file sizes.

    But at at 40% quality setting, you will see visual artefacts after a single conversion.

  16. blf says

    rsmith@21, Thanks ! I was contemplating doing precisely that same trial myself — convert &tc — but couldn’t quite work up the energy to bother (in part because I tend to use PNG and TIFF myself, both of which are nominally lossless).

    I’m not sure, but I think(? vaguely recall?) JPEG’s compression algorithm is specifically for images and uses, e.g., “area” similarity and prediction (which reminds me of MPEG’s video compression), unlike LZW and RLE, which can used for arbitrary data. That is, it “knows” how the data being compressed will be interpreted after decompression, and uses that knowledge (which would presumably include human visual capabilities / limitations (again, like MPEG video)) to guide, e.g., what can be “lost” and what shouldn’t be. As such, I speculate it might be possible to craft an image which, perhaps after an initial few rounds which do degrade, no longer degrades. (It’s perhaps worth pointing out the 80% example did degrade, suggesting that after a “zillion” rounds, it will have degraded to the point where the artefacts are visually apparent — albeit something like the speculated stasis might be reached instead, as possibly suggested by the minimal file-size shrink.)

    (I’m clearly now outside my area of expertise, and so will shut up.)

  17. PaulBC says

    blf@22 If you had sufficient AI to reconstruct what it was supposed to be a picture of, it would be possible to correct for noise in a convergent way and not degrade after a certain point. Maybe there’s something that would work at a lower level too just to recognize artifacts and avoid reproducing them. I have no idea what happens with a real JPEG. You could just add a signature to indicate it was already the product of lossy compression and cut the Gordian knot that way.

  18. stroppy says

    some interesting degraded jpg images.

    You can sometimes improve jpgs by scaling up, adding a small about of blur and then scaling down again, a sort of DIY antialiasing. In any case, I think they’re working toward better upsampling with AI. I would think this kind of tech would, as you suggest, aid reconstruction, maybe along with deep fake.

    That and most programs also have a de-noise feature among other enhancements. If you’re nerdy, they also let you create small convolve filters which might or might not be helpful in that regard (I really couldn’t say).

  19. PaulBC says

    stroppy@24 When I look at the second picture I conclude: It looks like a mountain with the horizon in the back, and those floating rectangles are more likely to be compression artifacts than a flotilla of invading extraterrestrials.

    If I had any artistic skills in addition to that, I could touch it up and produce something closer to the original picture.

    If I can do that, so can AI (as you have suggested). I suspect there are simpler ways to avoid having it get that bad in the first place though. It’s probably not a high priority since most people understand that if it’s been compressed once, there is not much point in repeating the process.

  20. says

    Here is one issue with the analogy: Suppose, in considering M[u/i]ller’s Ratchet, you have a population of haploid nonrecombining genomes, and a few are mutant-free. Then an intelligent picker might be able to make sure to breed from those. But natural selection might not always succeed in doing that — genetic drift might sometimes overwhelm it and result in the nonmutant alleles getting lost. And then the Ratchet advances some. So the Ratchet can work in cases where an intelligent picker would defeat the Ratchet.

  21. says

    I need to look at the book to find the right word, but I think it was “intent” or similar that Antonio Damasio had no problem ascribing to microorganisms, and thus intent was a factor before brains. I’m on board with that so yeah, the intent behind the changes can be looked at in an evolutionary context, but the choice of terminology muddies the responsability the person who acted with intent. This isn’t a useful way to model bad behavior.

  22. blf says

    @25, I know I said I would shut up in @22 — plus, of course, this seems further and further OT — but in response to “[m]ost people understand that if it’s been compressed once, there is not much point in repeating the process.”

    You may be correct in “most people”, but that is certainly not “all people” / “all tools” — Not that long ago I stumbled across a discussion about JPEG degrading (at a site I cannot recall now, Sorry!), where the commentators clearly misunderstood. Despite observing the degrading effect (using blinking (seemingly amazingly sophisticated for that community?)) concluded it cannot ever be serious. One point of confusion (not the only one!) was that “closing” an unmodified image necessarily caused degrading. (It’s fairly obvious the tools being used (none of which I’m familiar with) were being used in a such a manner the JPEGs were being rewritten (i.e., recompressed), despite being (allegedly) unmodified.)

    Despite banging my head on the desk, I refrained from commenting (partly, as I now recall, because I was only a “visitor” with no “commenting rights”? (nothing a priori wrong with that policy!)).

  23. PaulBC says

    Without the part under the panel, and l a lot of explanation, I am not even sure I would grasp that this xkcd comic was about evolution. (Since I had never heard of Muller’s Ratchet.) It’s about memes and about recovering signal from redundant noisy copies. That part I get. When he says “recombination”, my immediate reaction is to think about biotech and maybe he’s making a point about using genetic engineering to recover extinct species from fossil DNA. The final panel does not capture anything I would expect evolution to do undirected–though maybe if “best” is a proxy for having adaptive traits and “image editor” is… well I am not really sure… the recombination happens many times in an undirected way, and this makes it very unlike “you have an image editor.”

    I like xkcd in general, and this one is food for thought as much as anything, but I think I can state categorically that without commentary from PZ and Joe Felsenstein, I would not have had any chance of getting the point, assuming I have gotten it now.

  24. says

    I remember now, it was “value”, not “intent” that microorganisms have. I’d argue that intent and value are connected in that the intentions are driven by value and a microorganism changing direction towards food or away from a toxin display both.

    Otherwise humans still choose to act with their intentions and values.

  25. susans says

    stroppy@8, Why is it that with all the obvious outward physical body changes the Asgard suffered with multiple generations of cloning, their brains and reasoning capacity apparently survived intact? Is it the same hand wavium that deems when something something allows beings to walk through walls, they don’t fall through the floor?

  26. blf says

    @34, The mildly deranged penguin values cheese, and intends to eat it. She’s admittedly larger, on Earth, than an Earth microorganism (measured by essentially any metric), and is adamant she’s not a feathered long pig. She also meanders without a point, rather like this comment…

  27. says

    @blf 36
    Tangential thought, a bane of adhd.

    Speaking of moving towards something.
    “A cell type–specific cortico-subcortical brain circuit for investigatory and novelty-seeking behavior”
    It’s nice to have something positive to tie to the periaqueductal grey after that article on predator and prey behavior. It’s like 10:1 bad feeling things to good feelings things when it comes to opportunities to objectify the mind. I know that medicine focuses on the negative but it would be nice to have more than nightmares to process more often.

  28. stroppy says

    @35 susans
    That’s a good question. My questions about the series start right off with the whole ancient Egyptian alien pyramid thing. But from another point of view Egyptomania is an interesting cultural phenomenon in it’s own right, so that’s my excuse for watching it and I’m sticking to it.

    Actually egyptologist Bob Brier has written a book about it which I recommend as good light reading. Brier himself is a collector of such material from the sublime to the ridiculous.

    Egyptomania: Our Three Thousand Year Obsession with the Land of the Pharaohs

  29. stroppy says


    Perhaps the Asgard were merely coasting on past achievements. They apparently weren’t smart enough to prevent their own extinction.

  30. Rob Grigjanis says

    susans @35:

    Why is it that with all the obvious outward physical body changes the Asgard suffered with multiple generations of cloning, their brains and reasoning capacity apparently survived intact?

    Not so sure they did survive intact. Apparently projectile weapons were beyond them. I think they were coasting on old technology.

    the same hand wavium that deems when something something allows beings to walk through walls, they don’t fall through the floor?

    That question is addressed in the Stargate SG-1 episode “Wormhole X-Treme!”, about the show-within-the-show;

    Yolanda: I’m having a little trouble with scene 27. It says I’m out of phase, which means I can pass my hand through solid matter, or I can walk through walls.
    Director: Yeah, yeah, ’cause you’re out of phase.
    Martin: Um, exactly.
    Yolanda: So… how come I don’t fall through the floor?
    (long, awkward silence as Martin and the director stare at Yolanda, look at each other, and back at her)
    Martin: We’re gonna have to get back to you on that one.
    Yolanda: Right.

  31. Rob Grigjanis says

    stroppy @38: The show had holes you could fly a Goa’uld mothership through. All transplanted Earthlings speak English with North American or English accents, regardless of when or where they were taken from Earth. If there are any universal translators around, I didn’t hear about them.

    But I don’t care. I liked all but the last two seasons of SG-1, all of Atlantis, and even the first season of Universe (a couple of episodes introduced me to Alexi Murdoch’s music).

  32. chrislawson says

    Joe Felsenstein–

    Nice to see you commenting on this thread as Wikipedia credits you with being the first to use the term “Muller’s ratchet” in the scientific literature. I also love that your typo is so on topic.

  33. says

    @chrislawson: Thanks. Coming up with the name wasn’t much of a stretch, because Muller had described the effect by saying that “an asexual population incorporates a kind of ratchet mechanism, such that it can never get to contain, in any of its lines, a load of mutations smaller than that already existing in its at present least-loaded lines.” However I did give a better population-genetic explanation of how it works, and I did do the first computer simulation of the Ratchet.

  34. birgerjohansson says

    Hmmm. Assuming some alien science making nanotech DNA correction possible, you might use umbilical cord cells to get a zero-age reference genome, and make sure the immense number of adult cells do not stray from that template.
    So real extra-terrestrials might have fixed ageing. But at that level of knowledge, they would have strong AI, so why bother with any leaky bags of organs?
    Still, it is fun to ponder “Culture”* -level tech for altering the body.

    *I still grieve for Iain Banks.