Subjects: Evolutionary computation. Information technology–Mathematics.
The introduction to this series ended with a promise of insights into evolutionary informatics that the forthcoming book by Marks, Dembski, and Ewert is unlikely to afford. There will be little doubt at the end of the fourth installment that I have delivered the goods. First I want to assure you that, although I subscribe to the philosophy “Into Each Life, Some Math Must Fall,” the downpour of abstract notions, Greek letters, and squiggly marks will be intermittent, not unrelenting.
Sunset in the Garden of Id
Marks et al. address algorithmic specified complexity (ASC) in the chapter titled “Meaning of Meaning.” Applying the methods of their paper “Measuring Meaningful Information in Images: Algorithmic Specified Complexity” (2015) to Sunset in the Garden of Id, I have obtained a result of 22 million bits. When the full-resolution image is scaled to the size of the image displayed above, the ASC drops to 5 million bits. I think it’s safe to say already that whatever is being measured is not the meaning of what appears in the image.
For Marks et al., “Meaning Is in the Eye of the Beholder.” That’s the title of a section of their forthcoming book. In their paper “Measuring Meaningful Information in Images,” the beholder has a library of digital images, and uses it to compress a given digital image as much as possible (without losing the ability to recover the image perfectly). If the beholder compresses the image, using the library, much more than a standard compressor does, without using the library, then Marks et al. regard the image as high in meaningful information for the beholder. In my measurement of the algorithmic specified complexity of Sunset in the Garden of Id, the beholder has a library of images relevant to the intelligent-design movement. There is a short computer program that converts the library into Sunset. The program is, for the beholder, Sunset in compressed form. That is, the beholder can run the program, supplying it with the library as input, and obtain a perfect copy of the the image. The standard image compressor reduces Sunset to 22.1 million bits. The difference of 22.1 million bits and the length of the (binary) computer program is the approximate ASC of the image.
There is, of course, a punchline that I’m saving for later. And there is, of course, a story about the double-plus-goodness of algorithmic specified complexity that I have not told. In reality, the authors have only one theorem for ASC (stated coherently and proved by me), and it says much less than they say. What they do, lacking a formal characterization of the measure, is to provide example applications that work out the way they like. But they ought to expect, lacking a formal characterization of the measure, that there are example applications that do not work out the way they like. And they ought to look as hard for negative examples as they do for positive examples. This ought is not the ought of a lawyer seeking to win a case in an adversarial system of justice. It is the ought of a researcher seeking to report the truth. Call me idealistic, but I cherish the notion that researchers attempt to produce evidence against their own claims, not just in support of them.
The Series
Evo-Info review: Do not buy the book until…
Evo-Info 1: Engineering analysis construed as metaphysics
Evo-Info 2: Teaser for algorithmic specified complexity
Evo-Info sidebar: Conservation of performance in search
Evo-Info 3: Evolution is not search
Evo-Info 4: Non-conservation of algorithmic specified complexity
Evo-Info 4 addendum
Great example. I was going to do a whole post on the issue of ASC, but you beat me to it, fortunately.
The other definitions of Specified Complexity have to do with things like fitness or degree of adaptation. And indeed, the whole point of evolutionary theory is to provide an explanation, in terms of natural processes, for how birds can fly well and fish can swim well, because, as we know, fish gotta swim ands birds gotta fly.
But the ASC has always puzzled me. Why should there be some connection between how simply something can be described and how fit it is? What role does ASC play in explaining evolution of adaptation? The answer you suggest is that it is needed because Marks and Dembski are putting forth some theology of information, and they need it for that.
Maybe we should take a photo of Marks and a photo of Dembski, and see which one can be compressed more easily, so as to find out which of them has more information.
Joe Felsenstein,
There’s an “About the Author(s)” tab in the ad for the book. I’ve cropped the bottom edge of Winnie to make it the same size as Willie.
Willie converted to PNG: 8
17,454 bytes = 139,632 bits.
26,935 bytes = 215,480 bits.
Winnie converted to PNG: 8
In the context of the ad, the ASC of each image is not much less than its size in PNG format.
(Wille NOR Winnie) contains more meaningful information than Winnie, if you believe in ASC. (The size of the PNG image below is 30,105 bytes.)
And you said you were saving the punchline for later. A gagagaga..
Tom is a comedy writer!
phoodoo,
Can you give us your summary of Tom’s argument, Phoodoo?
Tom English,
That’s a Scream!
Moved a post to guano.
Neil Rickert,
But Neil, I prefaced it with ” I feel like…” so what’s the problem?
You know sort of like:
Allan Miller,
Just messing with their heads.
Since you are the one creating OP after OP about the same subject, it seems to be them who have gotten into your head.
You’re slightly confused about primacy. My work in the field that Marks and Dembski dubbed evolutionary informatics was between 1996 and 2004. I published a half-dozen relevant papers, and gave tutorials on the topic at a couple of major conferences. Dembski and Marks published their first paper in 2009.
Tom English,
So they forced your hand to start one OP here after another after another then did they?
Diabolical!
I enjoy Tom’s posts. As they write books not papers, it’s important to keep them honest.
Richardthughes,
I think he includes colorful photos in them so you won’t feel left out.
Fixed that for you.
lol
Mung,
I’ve got a Ruby script that makes it easy to mess with their heads in sundry ways. But it will run for you only if you’ve got ImageMagick installed.
Let me expand slightly the comment I made (#1 in this thread). For other cases of specified complexity, some sort of scale of goodness. Here’s William Dembski on page 148 of the 2007 edition of his book No Free Lunch: Why Specified Complexity Cannot Be Purchased without Intelligence:
Specified Information is then computed using (the negative logarithm to the base 2 of) the probability that some “chance” process would result in a value on the specification scale equal to, or greater than, the value observed in the organisms.
The whole point of this is to see whether we can make some argument that values this high cannot be achieved by natural evolutionary processes.
In light of this Algorithmic Specified Complexity is very strange. It measures to what extent something ostensibly complex can actually be described simply. Compared to the other scales, this one is quite odd. Why should there be a premium on simple description? Does it relate to viability, or to biochemical function, or to the reproduction of genes? Fairly self-evidently not. So why should we pay attention to ASC when there are much more meaningful types of Specified Complexity around? It seems obvious to me that ASC is a non-starter, devoid of interest to analyses of evolutionary processes.
I wonder whether someone like, say, Jonathan Bartlett, or VJTorley, or even Winston Ewert himself could enlighten me on this issue.
‘Fraid I can’t help you Joe. That’s one reason I’m looking forward to the new book. 🙂
Your confusion stems from the fact that ASC is meant to support ID, it has nothing to do with Evilution…
…
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
tumbleweed
There are no biological applications in the four published papers on ASC. I can match all of the sections of the “Meaning of Meaning” chapter to content of the papers. There is some ambiguity in the titles of the last couple of sections. So I won’t bet you 100 thousand dollars that there will be no biological applications of ASC in the book. But I will bet you 100 bucks. 🙂
If an event is more likely to occur in an evolutionary process than they think it “naturally” ought to be, then they say that the process generating the event is high in active information. If that doesn’t work out for them, are they going to turn around, and convert the low probability of the event into specified complexity? I think they know that they won’t get away with playing it both ways. That is, they cannot say both that
1. design elevates the probability of an implicit target, and that
2. design is all the more evident, the more improbable a specified target is.
I believe that Dembski contributed little or nothing to the development of ASC. Considering the debacle of Ewert’s master’s thesis, it’s likely that Baylor administrators extracted a promise from Marks and Ewert that Dembski would not be involved in Ewert’s dissertation work. Furthermore, the papers on ASC are rife with silly errors that Dembski would have seen immediately. I doubt that he reviewed them before they were submitted for publication. As you know, I am quite willing to say good things about Marks, Dembski, and Montanez. But, as a 30-year member of the Institute of Electrical and Electronics Engineers (senior member since 2004), I have never seen a Ph.D. electrical (and computer) engineer as awful at mathematics as Ewert is. He seems to be an avid programmer, and I would not be surprised if he were good at that. I feel no need to say that he’s devoid of talent. But his responses to us on mathematical matters are out-and-out surreal.
I think ASC is an attempt to find some use or another for Dembski’s prior work. Ewert may be uncomprehending enough of the situation to try to apply both ASC and active information to biological evolution. But I think Marks is savvier than that. He’s the one who’s been talking about the “meaning of meaning” in the context of ASC.
If you are in Oklahoma we could probably hook up one of these days. My mother is in Oklahoma and I come down to visit some times.
Seriously, Mung, I’d be glad to meet you.
We could make that happen, but you might find out how really dumb I am. 🙂
I usually fly in to Tulsa but if OKC is nearer I could always go through there. One of these days maybe I’ll take time to meet Johnny B.
Mung,
Building bridges is good. I’ll stand any IDer a beer.
One of these days I might set up a meeting with Joe F. We could probably enjoy a good scotch together. He and Tom could compare notes and discover that I am dumber than either one thought. 😉
Richardthughes,
I haven’t forgotten your invitation, Rich. It appeared for a while that I wouldn’t be able to travel anymore (and I haven’t wanted to talk about it). The latest word is that the problem will pass. I keep thinking of the Chicago River dyed green. But alcohol doesn’t have much of a place in my chemical regimen these days. “St. Paddy’s Day” and “moderation” somehow don’t go together in my mind.
As meaningful as the complicated neural network of varying sized pebbles washed up on a shore? THAT meaningful?
Amtrak gets me to Tulsa. But, more interestingly, it gets me to Chicago, and then to Seattle, and then to Vancouver, and then to LA (I have family there). The 30- and 45-day passes are good deals. That let’s me break the travel into smaller segments, and take things at my own pace. Perfect! I’m thinking May and June.
phoodoo,
This is not about the meaning of meaning. The issue is whether the ASC of an image, as measured by Marks et al., is sensibly regarded as the amount of meaningful information in the image. All I am doing is to apply their methods to a case that they have not considered. You evidently did not understand my example in the second comment of the thread. I’m not saying that to insult you. I skipped some details that I’ll include, someday, in my explanation of how I generated Sunset in the Garden of Id. I’ll say more about the calculation here, if you actually want to know about it.
Tom, I’d be more than happy to drive you around to some of Washington’s many places of interest. I’m not into humping no whales though.
I hope you like metal music. Or blues. Or classic rock.
This looks a bit like The FMM game.
Although I never figured out the point of his game.
Professor Felsenstein asks: “Why should there be a premium on simple description?”
That’s a very good question. If you look at Dembski’s 2005 paper, “Specification: The Pattern that Signifies Intelligence,” you’ll find that he appeals to descriptive simplicity as a handy, intuitive way of explaining way a random sequence of coin tosses (heads =1, tails =0),
(R) 11000011010110001101111111010001100011011001110111
00011001000010111101110110011111010010100101011110
need not be designed, whereas the pseudo-random sequence,
(ψR) 01000110110000010100111001011101110000000100100011
01000101011001111000100110101011110011011110111100,
is almost certainly designed. The difference is that the latter sequence actually conforms to the Champernowne sequence, which can be constructed “simply by writing binary numbers in ascending lexicographic order, starting with the one-digit binary numbers (i.e., 0 and 1), proceeding to the two-digit binary numbers (i.e., 00, 01, 10, and 11), and continuing until 100 digits were recorded.” Dembski apparently thinks that the reason why the latter sequence can be confidently declared to be designed is that it is not only massively improbable, but capable of being described succinctly. He writes: “It’s this combination of pattern-simplicity (i.e., easy description of pattern) and event-complexity (i.e., difficulty of reproducing
the corresponding event by chance) that makes the pattern exhibited by (ψR) — but not (R) — a specification.”
For my part, I have two problems with the descriptive complexity requirement.
First, low descriptive complexity isn’t necessary to warrant a design inference. Dembski himself admits in his paper that “prespecifications need not be descriptively simple.” He then proceeds to exclude them from the scope of his definition of specification – which is fair enough. But surely, if we were to come across what looked like a staggeringly complex (non-replicating) machine on another planet, we’d infer that it was designed, even if we didn’t know its function and hence had no way of describing it concisely. Dembski’s low descriptive complexity requirement would also entail (strangely) that we would never be justified in imputing design to an irreducibly complex system, no matter how many parts it possessed, until we found a way of succinctly describing its function.
Second, low descriptive complexity isn’t sufficient to justify a design inference. Dembski acknowledges in his paper that many patterns have low descriptive complexity merely because “our natural language enables us to describe them simply.” But what this overlooks is that:
(i) natural language fails to adequately describe objects at the bottom level – for instance, calling something a snake tells me nothing about the complexity of its internal anatomy or the structure of its cells. Likewise, the description of a bacterial flagellum as a “bidirectional rotary motor-driven propeller” fails to specify its constituent proteins;
(ii) more importantly, the argument that low descriptive complexity indicates design assumes that our language is evolution-independent. It is not. An evolutionist could argue that the reason why we have a single, handy word in our language for describing legless, carnivorous reptiles of the suborder Serpentes is that they posed a danger to our ancestors, which meant that evolution tended to favor individuals with a ready-made snake-recognition system in their brains. It should be obvious that our hardwired visual categories would be reflected in our language. In short: our way of slicing and dicing reality into simple concepts is (at least partially) a reflection of our evolutionary past. Hence the fact that objects which have evolved on our planet can be described concisely in our language cannot be adduced as an argument that they were designed. (Objects on other planets, maybe.)
All of the above. Seeing humpbacks isn’t my thing. But I will go island hopping. I’ll contact you and Rich and Joe when it’s sure that I can make the trip.
There’s a passage mentioning Tononi in the paper I’ve addressed.
They’re both pretty sharp, but don’t put yourself down. I think you’re the smartest theist posting here and at UD.
VJTorley:
Thanks for responding to my inquiry. I interpret your response as being that you too do not see a high level of ASC as a particularly good indicator of Design.
Fundamentally the Specified Complexity argument is that we see a level of adaptation so high that we can show that this could not have been achieved by natural evolutionary forces. (I think that the details of their reasons why are not persuasive, but anyway that’s what they are attempting to prove).
Your response, I think, agrees with me that there is no persuasive way to have ASC connect somehow with the level of adaptation.
Joe Felsenstein,
VJ also believes in evolution.
I think he does not think that evolution is explained by normal evolutionary mechanisms, however much he may believe that evolution has occurred and that there is common descent of organisms. So it is worth asking whether he thinks that the ASC variant of the SC argument is relevant.
Just as asking you the same thing would be relevant.
Tom English,
Cool. I’ve got a guest room in the heart of downtown 😉
I’ll let you introduce Tom to Big Bertha then.
Looks like the book is already sold out on Amazon.