|
Author
|
Topic: Demonology and Biology
|
Janitor@MIT
Member
Member # 125
|
posted 17. April 2002 15:45
Adami, Christoph, Charles Ofria, and Travis C. Collier. “Evolution of biological complexity,” Proc. Natl. Acad. Sci. USA, Vol. 97, Issue 9, 4463-4468, April 25, 2000.
Adami, C. & Cerf, N. J. 2000. “Physical complexity of symbolic sequences,” Physica D 137, 62-69.
Another attempt to stir the recalcitrant demon to his task.
Is it possible to measure the information-complexity of the environment independently? The authors don’t apparently think this is important to supporting their thesis. Again we are dealing with a tautology rendered in silico (accompanied by a pointless diversion into Kolmogorov complexity).
It wouldn’t be fair to the authors (or the “Brainstorm” participants) to reproduce here the seven pages (!) of notes I took on these two articles because, as I indicated, the misconception of information theory is pathological to Darwinism, not any particular researchers.
What is distressing is that I can highly recommend Adami’s “Introduction to Artificial Life” (1998). Like Schneider, I know Adami knows better! Its only when we attempt to get Shannon to shake hands with Darwin that we discover Darwin’s hiding a joy buzzer. Darwin’s fooled us all! (And he always looked so grim in those photos. Who’d have guessed he was such a prankster?!)
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 15. January 2003 18:22
The moderator allowed me one more post on Schneider's ev program. Anticipating that this topic would resurface, I saved that opportunity for the future, but with Frances's recent reintroduction of the topic and Janitor@MIT's continuation of his/her misrepresentation of Schneider's work, I think now is the right time.
In the ev paper, the following parameters are described:
y = the number of binding sites in the genome (16 in the simulation)
G = the length (in nucleotides) of the genome (256 in the simulation)
L = the length of a binding site (6 in the simulation)
To find a binding site, one must locate its starting point. The amount of information required for this is
Rfreq = -log2(y / G).
(Although it does not seem to be noted in Schneider's paper, this is an approximation. An exact calculation would involve a binomial coefficient, but the given equation for Rfreq is probably more relevant to the way cellular processes actually locate binding sites.) In order for the binding sites to be located, they must be different from the rest of the genome in some way and it is this difference which provides the information necessary to locate them. In order to calculate, a la Schneider, how much information is actually provided by the binding sites, we need to know the following quantities:
p(b) = the frequency of nucleotide b in a specific genome
f(b,l) = the frequency of nucleotide b at position l in the binding sites
Ignoring small sample effects, we can then define
Hbefore = -L * SUM p(b) log2(p(b)), (summed over b=A,C,G,T)
Hafter = -SUM f(b,l) log2(f(b,l)), (summed over l = 1,...,L and b=A,C,G,T)
With these definitions we have
Rseq = Hbefore - Hafter
and this quantity is a measure of how much the binding sites stand out from the rest of the genome (for this to be exact, it must be assumed that the rest of genome is well-described by p(b)). Now, Schneider's hypothesis is that, approximately, Rseq = Rfreq, i.e. that the binding sites stand out just as much as they need to to be reliably located.
In the simulation, Rfreq is constant (the number of binding sites does not change nor does the genome length). Because the initial genomes are generated randomly, with each nucleotide chosen independently according to a uniform probability distribution, it is clear that Rseq starts out at approximately 0 in each genome. Schneider then noted that Rseq evolves up to Rfreq when genomes enabling more reliable binding site recognition get to reproduce more than others. Ignoring the possibility of a significant programming error, this is simply an indisputable mathematical fact. Rseq does indeed evolve up to Rfreq and nothing can be said to change that. No matter how many insinuations about Schneider ID advocates can think of, no matter how much equivocation over the word "information" is employed, etc. it will still be true that Rseq evolved from zero up to Rfreq. Learn to live with it.
What can be said, however, is the following: "Sure, Rseq evolved up to Rfreq, but I find these quantities uninteresting. I would have preferred that Schneider had instead tracked the quantity... [insert definition of a quantity here]." That a quantity is named "information" does not necessarily mean that it is interesting and significant. Some are quantities named "information" are interesting, others are not, and all of them would benefit from being demystified by being given different and more telling names. Janitor@MIT is of course perfectly free to not be interested in Rseq and/or Rfreq as defined by Schneider. But it is unreasonable to insinuate (especially repeatedly and after having been corrected) that Rseq somehow did not evolve up Rfreq. The only potentially reasonable way to criticize the ev simulation is to define a different quantity and argue that Schneider should have studied that quantity instead. Challenge: If this topic simply must continue to resurface, at least define that quantity and calculate its value at the beginning and end of the simulation.
In particular, arguments of the following kind are unreasonable:
quote:
Now, Dembski talks about “smuggling” information and I imagine that smuggling is some covert operation. But there’s no “smuggling” here at all! This is being done right before my eyes! This information, the target Rfreq, a population, recognizer, etc., etc., altogether must constitute some ponderable amt of information. What happened to the “zero information” we were going to begin with? Is any of this information, other than Rfreq, measured precisely by the experimenter? He said he was going to do that.
Now, why is it that neither the reviewers, the editors, nor Frances (among many others) sees this?
(The quote comes from this thread.)
The answer, of course, is that Rseq measures neither Dembski's nor Janitor@MIT's favourite kind of information. Rseq measures how much the binding sites stand out from the rest of the genome. There is no coherent sense in which Rseq--the only quantity claimed to start out at zero--could take into account factors like how the simulation was programmed, the intentions of the programmer, the construction of the hardware used to run the simulation, etc. To require that it should be included in Rseq is nonsensical and to imply that Schneider claimed, or intended, to include it in the quantity starting out at zero is a direct misrepresentation of his work.
Erik
IP: Logged
|
|
Janitor@MIT
Member
Member # 125
|
posted 17. January 2003 12:12
Golly, I better stop my insinuations and misrepresentations! I’ve already permitted Erik to have the last word on this matter, and will attempt to ignore that my obvious character flaws have become an issue and simply reiterate one point: The interesting measure of the information in this experiment is not the “distance” Rseq/Rfreq, but this distance over the number of trials required to close it. Am I being unreasonable in seeing this as the relevant measure? Can anyone, besides myself apparently, not see the significance of this measure for evolutionary theory? If RM&NS incorporate information into genomes (add information) than whatever information contained in the genome is appropriately measured wrt to this process, which is definitively the set of all possible instances and those actualized by selecting from among them. The proper measure of information in the genome is only these numbers and nothing else. (And, sorry, but I won’t be convinced otherwise. My mind’s closed on this one.) Schneider does indeed provide some estimates of what I would consider to be the relevant measures in the “Discussion” section of the NAR paper. But they tend to highlight, despite the spin put on the numbers by Schneider, just how embarrassingly trivial the amts of information we are treating here, i.e., the amt of information generated—even conceding, for the sake of argument alone, the point that information is “somehow” added. Appears to me that Schneider, in his “argument with the angels” has fairly conceded the argument to their Earthly advocate—Dr. Dembski. LOL
Of course the problem Dr. Dembski has highlighted is not even a problem for biological science if biologists don’t recognize it as a problem. (LOL Science, whatever it is, is what we say it is, nothing less and nothing more.) Dembski has done well to emphasize why natural scientists don’t even recognize that the problem exists—and the failure to recognize it has far more to do with the intrusion into our thinking of non-scientific factors than to the existence of any rigorous solution to be found in biology. There is neither a problem nor a solution there. I believe the problem is real and that there is a solution to the problem within an evolutionary-information theoretic context. And as I have repeatedly emphasized this requires an extension/revision of received theories. If that means persons operating within the narrow confines of the extant paradigm consider me “dishonest,” well, what can I say—My feelings are really hurt.
IP: Logged
|
|
|
Printer-friendly view of this topic