|
Author
|
Topic: ID detecting computer program
|
Cre8ionist
Member
Member # 140
|
posted 27. February 2002 13:56
I've been thinking for sometime now, about how a computerized implementation of William Dembski's work, namely his "Design Filter", might look. I began to think about it on a deeper level when I heard about the work of someone at the University of Virginia.Physics Professor Lou Bloomfield designed a computer program to catch student plagiarists.
It seems to me that Bloomfield's program is similar to the Design Filter in its function (though much simpler to implement).
To learn more about the success of Bloomfield and his program visit both
http://www.cnn.com/2001/fyi/teachers.ednews/11/26/uva.ap/ and
http://www.csmonitor.com/durable/2001/05/17/fp11s1-csm.shtml
Or you can download this software free at http://plagiarism.phys.virginia.edu/
This is not a commercial site or program. Anyway, his program detects plagiarism instead of intelligent design, but both his idea and the type of algorithm he used could be helpful in designing a computer program for Dembski's filter. Bloomfield's program scans documents for sets of words (at least six in a row), in its search for plagiarism. When it finds documents with the same wording it flags them for further evaluation. I envision a program, which, when fed the specificity/information of a given object or string, returns either positive or unsure with respect to intelligent design. This could be helpful to those interested in Seti, cryptography,ID and anyone else who wants to detect design in a quick and reasonably efficient manner. Obviously, as alluded to earlier, it's a much bigger task to determine the CSI of some things than it is for others, but
if the criteria is trustworthy, there's no reason why it couldn't be programmed. In its simplest form it could be used to analyze signals received by Seti, or to analyze certain amino acid sequences of a given protein, and in its most complicated form it could be used as an aid in solving the fine-tuning and irreducible complexity debates.
In any case, I think that an ID detecting program is inevitable and hope some hot shot programmer(s) with time on their hands will take a good solid look at the issue.
IP: Logged
|
|
|
|
Paul Smith
Member
Member # 133
|
posted 27. February 2002 14:15
(Preface: The original post here does not seem entirely consistent with the format the moderator is looking for, but the question is interesting enough that I'm venturing a reply anyway. If this is a no-no, I will humbly accept any reprimands that may come my way...  )I am a programmer -- perhaps not a hot-shot one --but I'll take a crack at this anyway. The major hurdle for any such software is the heuristic needed to measure the CSI in the input. The difficulty with this lies in describing the *meaning* of the input (i.e. the "S" in "CSI".) For an approximation of the information content of some input, one could use a compression algorithm to try to express that same information in its simplest terms. The "CSI" measurement would then be the number of bits it took to describe the input in compressed form. However, there's a fundamental flaw with this approach: When writing, say, a cryptographic algorithm, I can test it out by creating long strings of very compression-resistant ASCII gibberish. Sure, it's aperiodic and complex, but it isn't specified. So the challenge is to propose a mechanism that measures the *meaning* (or specificity) of the input; then some function of that meaning relative to the bytes consumed to describe it might constitute a complexity-measure. The only way you could say that you had a detachable specification in the input would be to compare the input to some list of all independent specifications that the input could conceivably be describing. At this point, you've probably created far more new work to do than would be saved by using a computer program instead of a human mind to make the CSI determination. At a miniumum, user intervention would be required at the stage where "S" was decided; perhaps with that caveat, the problem becomes more tractable.
IP: Logged
|
|
|
|
Paul A. Nelson
Member
Member # 26
|
posted 27. February 2002 14:27
Bill Dembski and I have recently discussed proposing a "Detecting Design" experiment to CSICOP or Michael Shermer's Skeptics Society. We would ask them to compile a database of bit strings (the more strings, the better) from various sources: radioactive decay, SETI radio telescope data, random number generators, tree rings, Wordsworth sonnets, telephone books, genome sequences -- you name it. These strings would all be converted into binary form, and placed in the database with a keycode. CSICOP would hold the key under the strictest security for a year or so. ID theorists or anyone with an interest would then be able to access the strings via the web, to see if they could discover which were the "intelligently caused" strings (e.g., the Wordsworth sonnet, the telephone book entries) versus the naturally caused strings (e.g., radioactive decay). The challenge would be for the ID theorists to avoid false positives.CSICOP could put whatever they wanted into the database: all natural strings, all intelligently caused strings, a mixture of both. The critical methodological aspect of this experiment would be for CSICOP or the Skeptics Society to keep the keycode secure until the specified "string sifting" period (one year, probably) had ended. This experiment would be easy to set up and tremendous fun to do. It would also be powerfully instructive. Note: forgot to say that "CSICOP" is the Committee for the Scientific Investigation of the Claims of the Paranormal. Bill Dembski and I be partipating in an "Intelligent Design versus Darwinian Evolution" panel at their annual meeting, June 20-23, 2002, in Los Angeles. Wesley Elsberry and Ken Miller will speak as the evolution team; Massimo Pigliucci of the Univ. of Tennessee will moderate. [ 27 February 2002: Message edited by: Paul A. Nelson ]
IP: Logged
|
|
nobody
Member
Member # 145
|
posted 27. February 2002 14:48
quote:
Originally posted by Paul A. Nelson:
Bill Dembski and I be partipating in an "Intelligent Design versus Darwinian Evolution" panel at their annual meeting, June 20-23, 2002, in Los Angeles.
Fascinating. Thanks for that information. Will this event be open to the public? Has the exact day been decided for your debate? How many hours have been set aside for you? It shouldn't take you very long. 
IP: Logged
|
|
Cre8ionist
Member
Member # 140
|
posted 27. February 2002 16:17
quote:
Micah, Seeing that the individual programs are already in
the business of design detection, I don't see a reason to develop a general program.
It's difficult for me to imagine that in 50 years we'll still be using a calculator with a pen and paper (or any other primitive means) to detect design in difficult cases. This is precisely the reason a SETI design detector was implemented. Computers have the uncanny knack of simplifying tedious tasks. Of course, as with any other decent program there would have to be a simple starting point, with periodic upgrades to the technology. quote:
Micah,Generalizing algorithms is one of the most difficult tasks for programmers. This is the case with genetic algorithms, general expert systems, etc.
True, but this doesn't stop them from trying. quote:
Paul Smith,The major hurdle for any such software is the heuristic needed to measure the CSI in
the input. The difficulty with this lies in describing the *meaning* of the input (i.e. the
"S" in "CSI".)
I think you're on target about the major hurdle, but it is also dependent upon the type of data you expect to receive from your program. The question of specificity could be answered by the operator in some hypothetical beta version. Example: If I were to enter the phrase "The quick brown fox jumps over the lazy dog" into the program what would the result be? I would obviously already know that the entire phrase was (S)pecified, and could click a button to that effect, yet I still don't know if the (C)omplexity of the string is such that its improbability surpasses the level which would guarantee ID. The program could simply make the necessary calculation and I would know almost instantaneously. The same would be true for proteins etc... In many cases the question of specificity is already answered. I think overcoming the big hurdle, of which you spoke, would be a distant goal, but wouldn't be necessary to overcome in the short term. It seems to me that we'd need only answer two questions , i.e., is it Complex, and is it Specified? It's obvious to me that a program could be written now to determine answer to the first question, the other question will have to await more research. One possibility is that AI may someday be able to fill that void. I came back to correct my own post which had to be completed in haste...................................Cre8  [ 27 February 2002: Message edited by: Cre8ionist ]
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 06. March 2002 01:40
quote:
Originally posted by Paul A. Nelson:
Bill Dembski and I have recently discussed proposing a "Detecting Design" experiment to CSICOP or Michael Shermer's Skeptics Society. We would ask them to compile a database of bit strings (the more strings, the better) from various sources: radioactive decay, SETI radio telescope data, random number generators, tree rings, Wordsworth sonnets, telephone books, genome sequences -- you name it. These strings would all be converted into binary form, and placed in the database with a keycode. CSICOP would hold the key under the strictest security for a year or so. ID theorists or anyone with an interest would then be able to access the strings via the web, to see if they could discover which were the "intelligently caused" strings (e.g., the Wordsworth sonnet, the telephone book entries) versus the naturally caused strings (e.g., radioactive decay). The challenge would be for the ID theorists to avoid false positives.
A very interesting experiment. Another interesting experiment would be the algorithm room as proposed by Wesley Elsberry.
The claim is that there is CSI and apparant CSI. As I understand Dembski's claim, algorithms cannot produce CSI. But what if the algorithm room is a black box. Inside the box is a computer and an intelligent designer. The intelligent designer is given the time to work out the solution by hand and the computer is allowed to resolve the problem through appropriate algorithms. The solution is passed to the outside and it's up to the IDist to determine if the CSI is apparant or actual.
Another interesting test was proposed by Marty Fouts
[ 06 March 2002, 01:41: Message edited by: Frances ]
IP: Logged
|
|
Columbo
Member
Member # 113
|
posted 06. March 2002 02:19
Hello Frances,
Regarding your comment ....
quote:
Originally posted by Frances:
But what if the algorithm room is a black box. Inside the box is a computer and an intelligent designer. The intelligent designer is given the time to work out the solution by hand and the computer is allowed to resolve the problem through appropriate algorithms. The solution is passed to the outside and it's up to the IDist to determine if the CSI is apparant or actual.
Wouldn't this actually be a contest between two designers, the one in the room, and the one that designed the computer?
Columbo
[ 06 March 2002, 02:20: Message edited by: Columbo ]
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 06. March 2002 09:52
quote:
Originally posted by Columbo:
Hello Frances,
Regarding your comment ....
quote:
Originally posted by Frances:
But what if the algorithm room is a black box. Inside the box is a computer and an intelligent designer. The intelligent designer is given the time to work out the solution by hand and the computer is allowed to resolve the problem through appropriate algorithms. The solution is passed to the outside and it's up to the IDist to determine if the CSI is apparant or actual.
Wouldn't this actually be a contest between two designers, the one in the room, and the one that designed the computer?
Columbo
An interesting question. If the statement is that algorithms are unable to generate CSI then how is one going to test this? For instance with the Traveling Salesman problem, how does one show that the algorithm does not generate CSI? Does the fact that someone designed the computer have any relevance to this issue?
IP: Logged
|
|
Columbo
Member
Member # 113
|
posted 06. March 2002 12:17
quote:
Originally posted by Frances:
An interesting question. If the statement is that algorithms are unable to generate CSI then how is one going to test this? For instance with the Traveling Salesman problem, how does one show that the algorithm does not generate CSI? Does the fact that someone designed the computer have any relevance to this issue?
Well, if there is anything to notion that there are laws for information analogous to thermodynamic laws, then one could apply equations corresponding to energy equations, to detect true CSI. If I recollect properly, it was in this way that perpetual motion machines were finally judged to be not just improbable, but categorically impossible. Just as one cannot get energy from nowhere, so one cannot (given ID theory) get CSI from nowhere - "no free lunch." So, it seems to me that the computer in your hypothetical room could not produce CSI at a lower entropy level than that of the computer itself.
Am I missing something here?
Columbo.
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 07. March 2002 00:35
quote:
Originally posted by Columbo:
[QUOTE]Originally posted by Frances:
[qb]
An interesting question. If the statement is that algorithms are unable to generate CSI then how is one going to test this? For instance with the Traveling Salesman problem, how does one show that the algorithm does not generate CSI? Does the fact that someone designed the computer have any relevance to this issue?
quote:
Well, if there is anything to notion that there are laws for information analogous to thermodynamic laws, then one could apply equations corresponding to energy equations, to detect true CSI. If I recollect properly, it was in this way that perpetual motion machines were finally judged to be not just improbable, but categorically impossible. Just as one cannot get energy from nowhere, so one cannot (given ID theory) get CSI from nowhere - "no free lunch." So, it seems to me that the computer in your hypothetical room could not produce CSI at a lower entropy level than that of the computer itself.
Interestingly enough it has been shown that algorithms similar to RM&NS can generate information (Shannon entropy). In fact Dembski argued hard agains this example provided by Tom Schneider but without much success as I see it.
[url http://www.lecb.ncifcrf.gov/~toms/paper/ev/]Evolution of biological information[/url]
[QB][QUOTE]
Am I missing something here?
Columbo.
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 07. March 2002 00:55
I have been thinking further about the issue of algorithms and the generation of CSI. My question is, how would one distinguish between actual CSI versus apparant CSI?
Dembski has placed his bets on the NFL theorem but it seems that the NFL theorem may very well NOT apply to the instances of for instance biological evolution. The NFL theorem applies only to instances were the fitness functions are closed under permutation.
I quote:
quote:
Igel C., Toussaint M. (2001)
xxx.lanl.gov/abs/cs.NE/0108011|
On Classes of Functions for which No Free Lunch Results Hold
Abstract: In a recent paper it was shown that No Free Lunch results hold for any subset F of the set of all possible functions from a finite set X to a finite set Y iff F is closed under permutation of X. In this article, we prove that the number of those subsets can be neglected compared to the overall number of possible subsets. Further, we present some arguments why problem classes relevant in practice are not likely to be closed under permutation.
The paper proves two conditions for problem classes *with Free Lunch*:
a kind of continuity condition (that the value of the function to be maximized does not vary too much between one point and the next) and a
restriction on the number of local maxima. IMHO, both conditions will be fulfilled for realistic fitness functions.
Now the demise of the NFL theorem does not necessarily mean the end of ID but it surely would put a damper on it.
The other comment I have is wrt detecting natural examples of CSI. Gert Korthof proposed the Fibonacci series as found in certain plants. This would surely show CSI in nature. The question now is, did a designer place this here. But this objection raised another issue, what if the designer is not present during the growth, the designer may have placed the information in the genome, but what's different from a mutation placing this in the genome and a designer? Furthermore what if we can show all the way until the first ancestor that there is no need for a designer to have placed this info here, does this disprove ID? Unless one proposes that a designer is directly involved and not operates at a distance. One could always propose that the designer is one step further away than one has looked. All the way back to the Big Bang and there the Quantum Uncertainty will likely not allow us to determine one way or the other. So can ID be disproven? Surely showing that the Fibonacci series originates through a simple algorithm is not enough because the designer may be one step further away.
I believe that unless Dembski takes effort to define CSI in a measurable manner so that we can see if algorithms can increase CSI we will not see much progress be made by the idea of ID as a scientific alternative. But even in these instance, could one not claim that the implementer of the algorithm smuggled in CSI? In fact that is what Dembski seems to be doing. In case of the evolution of biological information paper by Tom Schneider, Dembski suggested how CSI may have been smuggled into the system. Tests by Schneider placed quite some doubt on this but again, if the argument is that algorithms cannot generate CSI but when one tries to implement an algorithm, the excuse of 'you smuggled CSI' into the system is raised, how could one expect science to test this proposition.
if it is Dembski's argument that algorithms cannot generate CSI then should it not be up to him to propose not only a scientific reason why this is the case and provide for ways to test AND disprove the idea?
IP: Logged
|
|
|
|
Frances
Member
Member # 169
|
posted 07. March 2002 12:14
quote:
Originally posted by Cre8ionist:
Just a note about the Schneider link. It's got a tail.....
http://www.lecb.ncifcrf.gov/~toms/paper/ev/
Use the one above but be sure to read Truman's response if you read any:
http://www.trueorigin.org/schneider.asp
![[Smile]](smile.gif)
The same link is found at Tom's webpage. He addressed some of the errors in Truman's assertions..
But the interesting part to me is the exchange between Dembski and Schneider.
Dembski made his assertions on the MetaNexus list
Schneider decided to check to see if Dembski's arguments made sense
It seemed that they did not hold. I understand that Dembski refers to Schneider's work and his MetaNexus article in NFL but does not seem to address the rebuttal.
What I think is interesting is that Schneider claims to have shown what Dembski says is impossible namely that algorithms generate CSI. The problem with Dembski's argument is that in any instance in which an algorithm is tested by science, in fact in any instance science performs a test, one can make a claim that ID was involved thus any outcome of such experiment can be rejected as 'CSI was smuggled in by the experimentor'.
So how does one disprove the claim that algorithms cannot generate CSI if any attempt to do so will require a test by an IDer and thus the claim that "CSI was smuggled in" cannot be eliminated?
[ 07 March 2002, 12:40: Message edited by: Frances ]
IP: Logged
|
|
|
Printer-friendly view of this topic