I see that in the unending TSZ and Jerad Thread Joe has written in response to R0bb
Try to compress the works of Shakespear- CSI. Try to compress any encyclopedia- CSI. Even Stephen C. Meyer says CSI is not amendable to compression.
A protein sequence is not compressable- CSI.
So please reference Dembski and I will find Meyer’s quote
To save Robb the effort. Using Specification: The Pattern That Signifies Intelligence by William Dembski which is his most recent publication on specification; turn to page 15 where he discusses the difference between two bit strings (ψR) and (R). (ψR) is the bit stream corresponding to the integers in binary (clearly easily compressible). (R) to quote Dembksi “cannot, so far as we can tell, be described any more simply than by repeating the sequence”. He then goes onto explain that (ψR) is an example of a specified string whereas (R) is not.
This conflict between Dembski’s definition of “specified” which he quite explicitly links to low Kolmogorov complexity (see pp 9-12) and others which have the reverse view appears to be a problem which most of the ID community don’t know about and the rest choose to ignore. I discussed this with Gpuccio a couple of years ago. He at least recognised the conflict and his response was that he didn’t care much what Dembski’s view is – which at least is honest.
Imagine that gpuccio comes across strings generated by Lizzie’s program. He figures out that they are functional (the product of their head-run lengths is very high) and specified (very few strings can do that). He does not know that they can be generated by a simple algorithm. What will he conclude? Will he change his conclusion when he reads the description of the program?
Okay.
That’s right. The landscape is an abstraction of an environment, which is something outside the population of replicators. (Word Mutagenation was written to respond to a very specific claim about words introduced by an ID proponent.) Think of it as a map to be traversed.
You yourself reference landscapes, such as when citing studies of functional complexity in proteins. Kairosfocus also references landscapes when he points to his “isolated islands of function in vast seas of non function”. By the way, Word Mutagenation addresses these isolated islands of function. They are traversed, not laterally, but vertically through inheritance.
Obviously. This relationship is represented by a fitness landscape which returns relative fitness for a given phenotype. You could use a physical environment instead, such as experiments with protein evolution, bacteria in the lab, or birds in the wild.
You seem to be confusing the model with the thing being modeled. Word Mutagenation can’t address biological evolution specifically, but it can address general statements about evolutionary processes, such as “isolated islands of function in vast seas of non function”.
Not confused on that point. But if you didn’t know the evolutionary origin of nylonase, you would conclude design, a false positive. Worse, you would know it with certainty!
But is it a *deterministic* explanation per #4? We’re not quibbling over the use of the word “deterministic. We thought you were using it broadly, and believe that is still your meaning, but you aren’t being clear, and have recently changed your nomenclature.
Which emphasizes that you are excluding known evolutionary transitions per #4 of your definition. Is that correct? Is your “deterministic explanation” dichotomous with design?
There’s information in the relationship between the replicator and the environment. That’s what we mean by selection. So if your notion of complexity means including the natural environment as well as the genome, well, you left a few steps out of your definition.
In any case, you are still confusing the model with the thing being modeled. The fitness landscape is just a table of fitness values for each phenotype. While no complete fitness table is available for biology (though there are many for aspects of natural biology), we can still explore how evolution works with evolutionary algorithms. When you make generalized statements about evolution, that’s when an evolutionary algorithm may be useful.
Since we know that many random sequences code for functional proteins, how do we know how many bits of change is required to optimize a sequence. It is quite possible that no more than a few are required.
Vertical means through inheritance. The connection between disparate groups can be found in common ancestors.
Of course it is. We have heredity, sources of variation, and a relative fit to the environment which determines reproductive success. we can even model very specific biological situations, such as the effect of antibiotics on the evolution of bacteria.
Is so. (Handwaving isn’t an argument.)
That’s not correct. Genetic algorithms can include drift, chance, relaxed or no selection.
That’s not quite correct as the statement only applies to an infinite population. In a finite population, fitness can decrease even if natural selection drives all evolution (which it doesn’t).
OMG! 🙂
It’s not supposed to run!
It’s to show you what is implicitly being done.
The CSI is calculated according to UD terms.
If the “digital functional specific information” reaches the UPB threshold, then CSI is “asserted”, as per Joe and gpuccio.
The “specific functionality” Lizzie was looking for has been attained as indicated by the dFSCI, (ask gpuccio what this means), and therefore CSI is asserted, whether implicitly, explicitly or “wink wink/nudge nudge”, the result is that the program has finished generating a string containing CSI.
As I said, I , Toronto, relabeled it, not to “add” code to someone else’s program, but so it would be clear to you, where this CSI calculation was being done, but you again, have let me down.
As KF says, please try harder.
I actually thought you would thank me. 🙂
Whatever you believe your skill set can handle. 🙂
Then so is gpuccio and kairosfocus.
If I have dFSCI above an agreed-upon UPB, I can safely say that the string containing that dFSCI, exhibits CSI, and that’s according to what I have read from gpuccio, and with different terminology, KF.
Many evolutionary algorithms include drift, chance, relaxed or no selection. Not sure why you think otherwise.
We just recently described a simple evolutionary algorithm that includes no selection whatsoever. It shows how diverging descent with modification leads to a nested hierarchy.
Not just an assertion, but an algorithm that anyone can follow to verify the assertion, even recreating the algorithm independently.
As we said, we are using the word vertical to refer to refer to a common population diverging and climbing separate peaks, rather than a population traversing laterally from one peak to another.
Huh? As we said above, and as your citation supports, the statement that fitness can never decrease only applies to infinite populations (which don’t exist, but provide a useful limit) *and* when natural selection drives all evolution (which it doesn’t). When a population is finite then fitness can decrease even if natural selection drives all evolution (which it doesn’t).
Yes, there is “specific functionality” and that is a product of values in the string that result in a number larger than “1.0e60”.
See this line in the program: #define FITNESS_THRESHOLD 1.0e60
Simple. We asked if evolutionary processes are included in #4 of your definition. We also asked whether your use of the term “deterministic explanation” is dichotomous with the design explanation. You made a long comment, and we don’t see a clear answer.
Yes, that is correct. Do you understand why? Keep in mind that Shannon Information is the theoretical basis of all modern digital communications, including the Internet. Why would a random sequence have more Shannon Information?
Which program? Word Mutagenation uses various types of selection, such as length or Scrabble® score. Valid words have positive fitness, while strings that don’t spell perfect words have zero fitness. Another program, as we mentioned, doesn’t use selection whatsoever, but only drift. Still another rewards poetic phrases (iambs, alliteration, rhyme, etc.). Which model we use depends on what aspect of evolution we are investigating. More complex models we’ve worked on include all of these aspects.
No, that is not correct. Random sequences are generally incompressible, but Shakespeare is quite compressible, one of many simple tests of randomness. The English language is full of patterns, which is why evolutionary search is so effective.
Because the genomes change over time. However, there is no adaptation without selection, of course.
You don’t seem inclined even to define sets, much less a nested hierarchy.
Joe,
It is as functional as a “string” of DNA that is “code” for a functioning human.
If a string of DNA contains “information” then so does Lizzie’s.
I am feeling both honoured and immortalized.
Look in “double fitness(genome_t *genome) {…”
In main though, the test is made for the “dFSCI” filled in by his fitness function: “while (genome_array[0].fitness < FITNESS_THRESHOLD) {..”
The point is that the winning string exhibits the “functionality” required by the “environment”, (i.e. FITNESS_THRESHOLD) before the program can exit.
OMG! 🙂
Whether its set to 1.00e+58, 1.00e+59, 1.00e+59 in Lizzies’s program, or 1.0e60 in the C program, makes no difference to the “algorithm”, which is what we’re testing, not language syntax or program structure enforced by a specific language.
The algorithm is the same, and that is to generate a string whose structure reflects a “….set of values then when multiplied together result in a value exceeding a certain threshold….”, and thus allow you to survive in your environment and have children.
Amazingly, both programs seem to have converged on the same “dFSCI” bit pattern which means the algorithm, regardless of language implementation, is consistent.
When testing, its sometimes more productive to have shorter run-times which means thresholds get adjusted simply for test purposes.
I’m guessing that’s why she set it to (10**58) instead of (10**60).
Yes; the run times were getting long. MatLab is relatively slow.
The maximum possible threshold for this algorithm is 4100 = 1.6 x 1060 (100 groups of THHHH). If it were set higher than that, the program would never halt.
The likely reason the program takes longer and longer to approach the maximum threshold is because of small fluctuations in the populations of offspring. One could make it converge to the maximum more quickly by the use of “latching” or by diminishing the probability of a mutation as the populations approach the maximum. Such a constraint might apply to situations in which kicking particles out of a well becomes less likely the deeper they settle into the well (i.e., they dissipate energy as they fall in). Latching could also correspond to something like radioactive decay in which there is no reactivation of the decay product.
But that feature was not in Elizabeth’s program.
Notice that the computer program never specifies how the heads are grouped. That is an emergent phenomenon that is not part of the program’s algorithm.
Mung,
Here is the important line: “product *= (double)(j – i);”
“product” ends up in each “<genome[x]>.fitness” which then gets sorted so that we end up with the highest “fitness” in “<genome[0]>.fitness” for testing against the “threshold”.
Again, consider this pseudocode since I’m not looking at the actual code as I type this.
The resulting “500 bit pattern” is termed by gpuccio to be “dFSCI” and the threshold test, if successful, asserts CSI for that bit pattern.
You’ll see in the program, the threshold test is actually done on the “functionality” but “dFSCI” is implicit due to the fact that “functionally specified” strings are of a known length.
So we don’t actually “calculate” CSI, we “calculate” “dFSCI” which is compared to a “threshold”.
If you want to “calculate a value” for CSI, ask gpuccio how to do it, since he claims CSI is not a scalar value, but rather a boolean.
keiths has posted a great comment with his “bucket of CSI” analogy.
An IDist has a bucket of things containing CSI that have no known “deterministic mechanism” explaining their existence. As soon as he finds a reason for a thing’s existence, he takes it out of the bucket.
What’s left in the bucket?
All the things he can’t explain! 🙂
What does he do next?
He attributes their existence to an “intelligent designer” that he can’t explain.
So if you can’t explain something, the default position is ID!
Mung,
“dFSCI” is not just the fact that it is in this case a 500 bit string, but the “specific functionality” of the 500 bit pattern, which in this case is the information that results in a “product of terms embedded in the pattern, that exceeds THRESHOLD”.
Mung:
It’s a parameter, Mung. You can change it.
It happens to be set to 2 in the version I posted to Codepad because I was testing STEP_MODE, which displays the genomes every n generations, and a population size of 2 was most convenient for that purpose.
Mung,
Since you’re still confused about ‘latching’ (aka ‘partitioned search’, aka ‘locking’), this is a good place to start reading:
Dembski Weasels Out
The ‘latching’ fiasco is one of the more amusing episodes in ID’s checkered history. Dembski and Marks embarrassed themselves by wrongly claiming that Richard Dawkins’ ‘Weasel’ program employed and depended upon latching. They even immortalized their mistake by publishing it in an IEEE paper. That’s gotta hurt.
To the best of my knowledge, they never retracted their erroneous claims.
Kairosfocus also got burned by claiming that Weasel latched. Instead of just admitting his mistake and moving on, he compounded his misery by insisting for weeks (and maybe still does even now) that he was right and that Weasel latched. It’s just that it used “implicit quasi-latching” instead of “explicit latching”. No kidding. Those are his phrases.
I guess an “implicitly quasi-latching” program is one that doesn’t latch but fools IDers into thinking that it does.
Good times. We still laugh about that over at AtBC.
Since Mung actually asked a serious question, I will try to answer it.
The “latching” or the decreasing of the probability of a change in the string as it approaches the “target” corresponds to situations like particles falling into wells and remaining there. In order to do so, energy is gradually shed so that the particle doesn’t have enough kinetic energy to pop out of the well again. For example, it could be a simulation of system of atoms or molecules condensing into a liquid or a solid. So the algorithm is simulating a phenomenon that actually occurs in these kinds of systems.
”Latching” is roughly analogous to the case of radioactive decay in which the atoms are not reactivated by an environment of radiation. Once they decay, that’s it; they don’t reactivate unless the are immersed in some intense radiation environment.
However, in the case of organisms “condensing” toward a different environment (i.e., being selected for fitness, in the language of random variation in the presence of selection), the phenomenon that is operating is a roughly fixed rate of mutation regardless of how “fit” the current generation is relative to the new environment.
In other words, the mutation rate continues despite the fact that the current generations are close to being the “fittest” relative to the current environment. If the environment (simulated by the target in the program) changes in the course of the evolution of the population, the evolution changes direction and the population starts to converge on the changed environment (new target in the program). You can easily add an outer loop to the program that changes the target in the course of running the program, and you can watch the population track the change.
Mutation rates are roughly constant over the course of history of an evolving population. Some of that is due to background radiation involving gamma rays or UV. Other causes include simply the probabilities in soft-matter systems that bonds will be broken or swapped or whatever happens in such complex systems simply because they are immersed in a thermal bath.
So, for an evolving soft-matter system such as a living organism – a system that adapts by producing offspring that are slight variations of itself rather than simply adjusting itself to the new conditions – it is more appropriate to keep the rate of mutation roughly constant. That is closer to being more realistic in the case of the evolution of living organisms. “Latching” is not appropriate in this case because it misrepresents what is actually going on with real populations. That would be equivalent to freezing the organisms to match the environment. It is supposed to be soft matter adapting by producing surrogates of itself. It has to stay “soft” in order to track changes in the environment.
Genetic algorithms include whatever laws of nature apply to the systems being modeled. If those algorithms are relatively good approximations of the laws that apply, what falls out of the GA program is close to what falls out in nature even though it may be impossible to predict it or mathematically model it.
These kinds of program have been around for a long time. They often went under the name of Monte Carlo simulations in the past. They were use on the earliest electronic computers, such as the ENIAC, to do calculations for designing the atomic bomb.
Mung et al,
A problem has arisen! I tornado has torn up my office. I’ve managed to put everything back together, but typically there is one last thing.
I have two documents left over:
Document 1
Document 2
But I have only one file that is missing a page! The file is entitled “Properties of DNA”. We’re about to spend much money researching the data on this page. But to me they both look very similar, no way to tell between them at all.
I’m not sure where the other document came from. Perhaps from another office, but they do all sorts here so no telling what it is.
Would you be able to help me, Mung, and determine which page is the correct page? Which page should I investigate further and which should I discard, as that’s the choice (limited budget don’t ya know). Which page is more interesting then the other? If you discover that design factors into it, are both designed? Neither? One but not the other? Which?
For bonus points, anything further you can tell me about the contents of either document would be appreciated.
Thanks!
OMTWO
And if you map each genotype or phenotype to its relative fitness, you create a fitness landscape. We can do this with actual phenotypes, such as protein function maps, or abstractly in some evolutionary algorithms.
An intern has just come up with a brilliant idea! Perhaps it’s one page after all that was simply torn in two. Can’t tell by the edges, the tornado messed it all up. So we’ll have to go by the data.
Is that possible Mung? We’re dealing with a single page, not two?
We need to be sure, so please explain your reasoning.
What is worse is, latching hardly matters in Dawkins’s Weasel program. The number of steps it takes to get to the target would be affected in only a minor way by it.
Nevertheless Dembski, Marks, and others pointed to it as the reason that Weasel did so much better than pure random search. They publicized latching as an important property of the Weasel program, one that anti-ID and anti-creationists were trying to cover up.
However, the Weasel program never latched at all. The reason for its success (compared to pure random search) was … selection. The very thing it was advertised to be about.
A mathematician could probably tell them why latching doesn’t affect performance.
Yes, but which segment is the top?
It had some work done at the time.
http://austringer.net/wp/index.php/2009/04/18/the-weasel-saga-with-math-part-1/
Good question!
Quite possibly, but Mung asked “If I take a 504 bit string and “randomize” it, I’ve generated Shannon Information?” The answer to that is yes. And whether that is important to claims about CSI or not, it is certainly relevant to information technology.
Start with a single genome of significant length. Each digit represents a gene for our purposes. Replicate the sequence with reasonable rates of mutation. (Try a single mutation in every other genome for starts.) You can add some limits to population by random culling. What pattern would occur in the descendants?
Most traits would be preserved by simple heredity, while no particular allele would be better preserved than any other. Here’s the result of such a process after four generations. (Previously, we used commas to make it easier to see the differences. This is harder to read, but allows more precision in grouping.)
abcdefghijklmnop
abcdefghijkCmnop
abcdefghijklmNop
MbcdefghijklmNop
abcdefghijklmnIp
abcdBfghijklmnIp
abGdefghijklmnIp
abGdeCghijklmnIp
Turns out that you can reasonably reconstruct the genealogical relationships from the nested pattern.
They can be very far apart. Keep in mind that most interesting landscapes have many dimensions, so the relationships are not always intuitive.
Because there are many different niches.
In order to show fitness can decrease in a finite population doesn’t require a computer simulation. Consider a simple example. The environment has resources to support only two individuals. Their genotypes are AA and AA. Each generation, they mate and produce three offspring, of which only two survive to the next generation. This goes on for many generations. Then one day, a mutation occurs and one of the A-alleles is reduced to an inferior a-allele. Now our population has AA and Aa. By chance alone, the offspring might look like this: Aa, Aa, Aa. The survivors are Aa, Aa. These reproduce again. By chance alone, the offspring are aa, aa, aa. The A-allele has disappeared. (The odds of this happening depend on whether the a-allele is recessive. If it is, then it can persist over several generations, even with strong selection.)
Yes. You might start with the clay tokens preceding cuneiform script. But that’s hardly “modern digital communications, including the Internet”.
From the definition.
The reason why Shannon defined it this way is described in his seminal paper, A Mathematical Theory of Communication, The Bell System Technical Journal 1948.
Shannon’s theory is fundamental to all modern digital communications.
Not necessarily. It depends on the contents of the first string, and the context of the message.
Try it, and let us know. By the way, Shannon (and subsequent studies) showed that English, in context, only transmits about 1 bit per letter to a human reader. Amazing, huh? (See Sajak & White, W h – – l o f F – – – – n -.)
Or due to neutral drift. Nor does it have to go all the way to fixation, but just a significant number.
There are many complex biological structures for which we can trace the history. A common example is the mammalian middle ear, where each step is selectable, while the final result is irreducibly complex.
Of course. It’s well-established that recombination is essential for traversing rugged landscapes.
Yes, apparently natural selection is capable of evolving quite adequate proteins — even with one hand tied behind its back!
In the past, you’ve rejected any experiment showing mutation is random with respect to fitness, such as Lederberg & Lederberg, Replica Plating and Indirect Selection of Bacterial Mutants, Journal of Bacteriology 1952.
Along with making unwarranted extrapolations of the number of “required” steps, gp ignores recent research indicating that protein domains are themselves modular.
I asked Joe recently if he thought there was such a thing as a fair die. Or set of dice even.
He said that there was such a thing.
I wonder how he can possibly know that?
What testable hypothesis did he test to determine that some dice are random I wonder.
And I also wonder why that method, whatever it is, can’t be extended out to other systems.
What say you Joe? How did you determine that your dice are fair and why can’t anybody else do a similar thing according to you?
Are you special Joe? Only you can arbitrate chance/not chance?
Indeed; the latching changes only the rate of convergence and narrows the distribution of the populations that approach maximum fitness.
There is very little difference between a genetic algorithm that maximizes “likeness” or one that minimizes “difference.” That change in perspective lies entirely in the thought processes that go into representing what is being modeled. Whether one maximizes “fitness” or minimizes the differences between the current population and the “template” – genotype, phenotype, or whatever trait is measurable – that stands in as a representation of the new environment, the result is the same.
The “target” could be a map of the potential well into which particles are condensing or it could represent a “template” of an organism that is consistent with a given environment.
The “latching” might be a more accurate representation of particles settling into potential wells by losing the kinetic energy that would kick them out again. But living organisms have a probability of changing even though they are close to being “fit” relative to the new environment. That is what makes them “pliable.” Evolution is closer in analogy to a soft material sagging into the shape of its current container. Move it to another container, and it begins to conform to the new container.
As far as a genetic algorithm is concerned, the main difference is that a pliable material is thought of as being the same object in successive generations whereas replicating organisms replace themselves with approximate surrogates of themselves in successive generations. To the computer program, there is very little difference unless one is also modeling the intermolecular forces in a pliable material.
The major reason for fitness peaks instead of potential wells is because, in biology, fitness is the objective measure of how a population relates to a given environment. That is a measure that increases; hence fitness peaks rather than potential wells. Yet ultimately, they are simply mirror images of each other reflected in the horizontal plane. To the algorithm, there is little difference.
Watching the churnings over a UD – although it is both nauseating and amusing at the same time – does give some insight into why people caught up in ID/creationism have so much trouble understanding things like genetic algorithms. It is because none of them has any hands-on experience with the real world. Instead, they have spent their entire lives in word-gaming without ever reaching out to grasp reality. So they have nothing in common with the experiences of those who have immersed themselves in studying the world around them. Most of the ID/creationist followers seem to hate science despite what they claim.
Joe seyz
And that rules out it’s distribution being the product of an algorithm internal to the dice how exactly? And in place of “check it’s balance” we could have “design an experiment”, right? Just like Lederberg & Lederberg, Replica Plating and Indirect Selection of Bacterial Mutants, Journal of Bacteriology 1952. Fer instance. So how come you can come to a conclusion of fair/not fair yet dispute outright anybody else’s ability to do the same?
Joe,
Perhaps you’d like to apply your “design detection” skills to the question I posed in this comment, in this very thread?
I’m happy to accept input from anybody on this, it’s a real poser.
That’s exactly what the Lederbergs showed wasn’t the case. The mutations were not due to environmental clues. You could also look at the Luria–Delbrück experiment.
That doesn’t change that the mutations were random with respect to the environment. But feel free to explain how intercellular communication explains the Lederbergs experiment. Please be specific.
Please be more specific. What is communicated? How is each bacteria to know what to bring?
What if we isolate each colony? Are you saying we would have a different result?
Are you saying that if we looked at the actual mutations, the rate of the mutation for antibiotic resistance would not be the background mutation rate?
Now that we have established some ground rules, i.e. that Joe agrees that a die can be said to be fair by examination of it’s roll distribution etc, I’d also be interested in Joe’s answer to this.
Mung,
Joe won’t help me out and use his “design detection skills” with my problem. He said:
Apparently that’s his reason. Not an excuse at all. You up for it Mung? Just thought I’d ask…
http://theskepticalzone.com/wp/?p=1352&cpage=2#comment-16703
Joe,
In fact, you could take a quick glance at the two documents I have and the problem I posed. I’ll donate $10 to the charity of your choice should you deign to apply your skills to my trivial problem.
But if not then I guess your principles are more important to you then making me look a fool. Understandable.
Looks like Joe will just keep repeating the hollow refutation he can’t support.
Joe, falling back to that already? A minute ago it was all going so well, then things got specific and you fall back to this? C’mon.
Joe
No, you can just look at the data. It’s all contained in the 2000 characters. The security team won’t let me release any more details. Nothing. Not a thing. It’s just, you know, what with all the talk about “500 bits” and “needle in a haystack the size of the cosmos bit string probability” I thought, you know, you might be able to just examine the data itself.
I understood ID to have many tools at it’s disposal with regard to “messages” and “strings”. I thought we could at least discover something interesting via ID with regard to the puzzle. You know, how you would have to do if you received those strings from a radio telescope. If you were in charge would you need to travel to the origin of the signal to conduct a “proper investigation”?
If that’s the case then ID is not going to be a whole lot of use if that ever happens is it? And you claim that SETI is doing ID right now? Ha! Just disproved that haven’t we…