Algorithmic Specified Complexity in the Game of Life, revisited

In 2015, Winston Ewert, William Dembski and Robert Marks published a paper entitled Algorithmic Specified Complexity in the Game of Life.

The paper was a wreck. We examined it here at TSZ and found well over 20 substantive errors in it.

ID supporter Eric Holloway describes it as a “neat paper”. I describe it as an “abysmal mess”.

Eric has been touting the virtues of ASC here at TSZ, so now is a good time to reopen the discussion of this paper.

22 thoughts on “Algorithmic Specified Complexity in the Game of Life, revisited

  1. I’ll be quoting a number of comments from the old thread.

    One of them involves the very first sentence of the abstract, in which Ewert, Dembski and Marks claim that ASC is a measure of meaning:

    6. The very first sentence of the abstract reads:

    Algorithmic specified complexity (ASC) measures the degree to which an object is meaningful.

    Interesting claim. How to test it?

    Well, EDM certainly can’t prove it, so the test has to be empirical.

    Okay, let’s collect a bunch of ASC values (or OASC values, which are lower bounds for the true ASC values), then collect the meaningfulness values for the same objects, and test the strength of the correlation.

    Oops, we can’t do that, because no one knows what the “meaningfulness” values are. That’s what ASC is supposed to provide us: the first viable tool for measuring the degree of meaningfulness of an object.

    Okay, then let’s look at a bunch of objects that Life aficionados are interested in. Surely those objects are meaningful. And we may not be able to assign meaningfulness values to them in order to see how well they correlate with the ASC values, but at least we can see if the ASC values exceed some threshold for all of these “meaningful” objects.

    Aw, crap. Two of the most important objects in the Life universe, blocks and blinkers, have -25 and -29 bits of OASC, respectively. They’re meaningful, but they fail the ASC test.

    So much for this:

    Algorithmic specified complexity (ASC) measures the degree to which an object is meaningful.

  2. Tom,

    The introduction of the published article is different from that of the paper we reviewed.

    Interesting. What’s the nature of the changes?

  3. Another complaint, related to the first:

    2. The name doesn’t correspond to the metric. ASC can be negative. How does it make sense to speak of negative complexity?

  4. Eric,

    Before you get involved in this new discussion, it would be good of you to acknowledge that Joe was right and you were wrong about the conservation of CSI as claimed by Dembski in No Free Lunch.

  5. I’ve perused the old ASC thread, and it seems there are a number of interesting observations. For instance, Dr. English thinks there is a problem in that ASC is a log ratio of a probability and semimeasure. But, if he’s read ‘improbability of ASC’ or even the 3rd party ‘algorithmic significance method’ that he found, he’d know this is not an issue.

    Then keiths thinks the very definition of ASC is problematic, even though it is a generalization of the well known randomness deficiency, which Dr. English has also acknowledged in one of my recent OPs.

    Then there is Dr. Felsenstein’s question about how this infers design. An interesting question, but not one addressed by the paper, so hard to see how this is a flaw in the paper.

    Thus, at least in my brief perusal, it seems the problems identified by keiths and Dr. English go way beyond anything in the paper. If keiths and Dr. English are right, then luminaries such as Kolmogorov and Levin are dealing in fundamentally flawed theory. I believe keiths and Dr. English can at least win a Fields medal by publishing their findings, so I highly recommend they publish their results in some well respected mathematics journal. It’s not everyday someone finds such crucial flaws in widely accepted theorems.

  6. EricMH: Then there is Dr. Felsenstein’s question about how this infers design. An interesting question, but not one addressed by the paper, so hard to see how this is a flaw in the paper.

    I can’t imagine how I ever thought that people discussing ASC were intending it to be used to infer design. How could I have made that mistake?

    Oh, wait:

    EricMH:

    Joe Felsenstein: The issue, for me, is not whether CSI is meaningful, but whether there is some valid argument that CSI can only be produced by Design, rather than by natural selection.

    This is the improbability of ASC proof again. Natural selection is a stochastic process, and thus it forms a random variable, to which the improbability of ASC applies.

    Aha.

  7. In my last comment, I reminded Eric:

    Before you get involved in this new discussion, it would be good of you to acknowledge that Joe was right and you were wrong about the conservation of CSI as claimed by Dembski in No Free Lunch.

    Joe’s argument stands unrefuted. And in failing to refute it, not only did Eric mistakenly attribute CSI to a random permutation function, he also confused the permutation with an XOR mask!

    Mistakes don’t get much more obvious than that. The honest thing would be to acknowledge them.

  8. Joe writes, sarcastically:

    I can’t imagine how I ever thought that people discussing ASC were intending it to be used to infer design. How could I have made that mistake?

    …and then quotes Eric shooting himself in the foot.

    Well, it isn’t just Eric shooting Eric’s foot — the paper does it too. From the conclusion:

    We conclude that many of the machines built in the Game of Life do exhibit significant ASC. ASC was able to largely distinguish constructed patterns from those which were produced by random configurations.

    [emphasis added]

    Note to Eric: If you want to defend a paper, it’s a good idea to read and understand it first.

  9. Eric:

    If keiths and Dr. English are right, then luminaries such as Kolmogorov and Levin are dealing in fundamentally flawed theory. I believe keiths and Dr. English can at least win a Fields medal by publishing their findings, so I highly recommend they publish their results in some well respected mathematics journal.

    This is the second time Eric has tried the goofy “Fields Medal” gambit on me, pretending that I am saying something that contradicts well-established mathematical knowledge. I’m not, of course.

    If I were, Eric would be able to quote me doing so.

  10. Eric,

    When Joe and I tried to get you to talk about CSI, you wanted to talk about ASC instead. Now that I’m trying to get you to talk about ASC, you’re avoiding the issues I’m raising. What’s up with that? You should be eager to defend ASC.

    See these two comments: Link, Link.

  11. Does Eric have any issue with my meagre contribution:

    “1. In Conway’s Life: http://en.wikipedia.org/wiki/Conway%27s_Game_of_Life
    2. There is the Glider-Producing Switch Engine http://conwaylife.com/wiki/Glider-producing_switch_engine
    3. It is coded by 123 “On Cells” but requires a space of 67×60 in a specific configuration.
    4. That’s 4,020 bits, > UPB.
    5. It contains well matched parts : 4bli,3blo,2bee,1boat,1loaf,1ship,1glider http://wwwhomes.uni-bielefeld.de/achim/moving.html
    6. It occurs naturally out of randomly configured dust : http://wwwhomes.uni-bielefeld.de/achim/moving.html
    7. It can evolve from a much smaller entity (“time bomb” – 17 active cells): http://conwaylife.appspot.com/pattern/timebomb

    What are the implications of this?

  12. Regarding the glider-producing switch engine, Richardthughes asks:

    What are the implications of this?

    The implications are dire for the paper.

    In the original thread, I looked at a couple of “naturally occurring objects” that Rich tracked down: the CP-pulsar and the glider-producing switch engine. The following two comments deal with those objects.

  13. Regarding the CP-pulsar:

    16a. EDM’s neglect of the “physics” of the Life universe causes massive errors in their OASC values that lead to false design inferences.

    EDM’s crazy assumption has already been mentioned a few times in this thread:

    In the game, determining the probability of a pattern arising from a random configuration of cells is difficult. The complex interactions of patterns arising from such a random configuration makes it difficult to predict what types of patterns will eventually arise. It would be straightforward to calculate the probability of a pattern arising directly from some sort of random pattern generator. However, once the Game of Life rules are applied, determining what patterns would arise from the initial random pattern is nontrivial. In order to approximate the probabilities, we will assume that the probability of a pattern arising is about the same whether or not the rules of the Game of Life are applied, i.e., the rules of the Game of Life do not make interesting patterns much more probable than they would otherwise be.

    [Emphasis added]

    To demonstrate just how crazy that assumption is, let’s look at a couple of objects. Richardthughes tracked down simple precursors for two large objects, the CP-pulsar, which I’ll discuss here, and the glider-producing switch engine, which I’ll discuss in my next comment.

    The CP-pulsar is item #19 on this page, and its precursor is shown in this video at time 5:57.

    The Game of Life is deterministic, which means that a specific pattern X0 always leads to X1, which always leads to X2, and so on. In other words, once a pattern occurs, all of its descendants are inevitable.

    To determine the true value of P(X), one must therefore account not only for the probability of X arising randomly during initialization, but also the probability of all of its ancestors. EDM don’t do this because they can’t. Even in a simplified world such as the Life universe, the true probabilities are horrendously difficult to calculate, because determining all of the ancestors of an arbitrary given pattern is hopeless. Hence EDM’s crazy assumption.

    How much damage does this faulty assumption do? Consider that the simple precursor of the CP-pulsar fits into a 5×5 box. That means that in EDM’s standard encoding, the representation only requires 25 bits (after the unnecessary width and height information is stripped out — see flaw #12 above). Thus, the true P(X) value can be no less than 1/2^25.

    However, that simple precursor evolves into a 15×15 object, for which the (corrected) standard encoding requires 225 bits. Thus, EDM will attribute a probability of 1/2^225 to it.

    How big is the error? EDM’s methodology has underestimated the probability by a factor of 2^200.

    In other words, the true probability is over
    1,606,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000
    times as large as EDM’s methodology would indicate.

    That is far, far worse than mistaking the width of a human hair for the diameter of the observable universe.

    Does this translate into a bogus design inference? You bet.

    For the 15×15 object, the (corrected) I(X) is 225. Thus, K(X|C) would need to be at least 140 bits in order to prevent a design inference, given the ASC threshold that EDM are using (see flaw #14). K(X|C) is much smaller than that, so EDM will infer design.

    Yet the CP-pulsar is a “naturally occurring” object, meaning that it arises out of randomly-initialized “dust” (not surprising, because the precursor’s probability is better than 1 in 34 million).

    Therefore, EDM’s methodology leads to a bogus design inference.

  14. Regarding the glider-producing switch engine:

    If you thought flaw #16a was bad:

    16b. Using EDM’s methods, we find that “natural” causes can generate infinite amounts of ASC in the Game of Life. In other words, there is no ASC threshold that will avoid false positive design inferences.

    Like #16a, this happens because EDM ignore the effect of “physics” in the Game of Life.

    The ‘glider-producing switch engine’ is described here. It is a complicated entity that moves in one direction, spewing gliders in the direction of travel and leaving ash objects in its wake.

    The amazing thing is that this monstrosity can be produced by the tiny 15×6 “time bomb” precursor shown on the web page, whose small ASC is well within the reach of EDM’s random initialization process and does not trigger a design inference. It has also actually been seen in “the wild”.

    The pattern grows ever bigger over time, so I(X) (according to the standard encoding) is always increasing. K(X|C) is also increasing, but at a much slower rate, because the functional encoding is barely affected. Each iteration merely requires an increment of the ⊕ operator’s parameter. Since I(X) increases indefinitely, and always faster than K(X|C), this means that ASC(X) increases without bound.

    EDM not only have another false positive; they have one that they can’t get rid of, no matter how high they set their ASC threshold, unless they abandon their bogus “no physics” assumption and find a defensible replacement. Good luck to them.

    “Natural” causes can produce infinite ASC in the Game of Life, if we calculate ASC using EDM’s methods.

  15. Eric, July 1st:

    Anyways, I’m happy to discuss the ID math with anyone.

    I’m starting to wonder about that.

  16. keiths:

    Just to give you someone to talk to, since Eric seems to have ridden off into the sunset, here is a question for you on Montanez paper:
    What is the meaning of the word “conservation” used in Thm 2? How does it relates to Dembski’s meaning(s?) of the word in claiming conservation of CSI?

    And how do any of these meaning relate to what “conservation” means in the conservation laws of classical physics or in the conservation of quantum information in the formalism of QM (ignoring measurement)?

  17. Bruce,

    Let’s save the discussion of the Montañez paper for Tom’s forthcoming thread.

  18. Looks like Eric bailed out of yet another discussion when things got too difficult for him.

    Bruce coined the term “pulling an Eric” to describe this sort of behavior.

    It fits.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.