|
Author
|
Topic: Design Inference Game II
|
Rex Kerr
Member
Member # 632
|
posted 28. January 2004 15:32
Cre8 has no justification for assuming that the "bits" in his picture have equal probability of being ones and zeros, and therefore his probability calculation is broken.
If the probability of one mark is p and the other mark is (1-p), then the total information (entropy) of one mark, in bits, is p*log2(1/p)+(1-p)*log2(1/(1-p)) from basic information theory. This is less than 1 for all values of p except p = 0.5.
IP: Logged
|
|
Cre8ionist
Member
Member # 140
|
posted 28. January 2004 20:13
RBH, you sure spend a lot of time here for not thinking it's worth much ![[Confused]](confused.gif)
FYI, this is a thread about doing what you said you don't think's worth doing......
Rex, I'll have to talk to you later, but let me suggest that something is wrong with your view.
Your view can't discern a message from randomness, and while you might find comfort behind some formula, your formula wouldn't work for SETI, and it won't work here. Meanwhile, I still maintain that when Dembski's filter attributes something to design and it is being used correctly, that it will be right. If and when a conclusive example comes up to refute Dembski I will certainly drop my support (as will most everybody who now supports it), but at this point I don't see it happening. As I said SC comes from SC, despite the computer shell game that's now being played.......I'm working on a post for another thread at the moment and that's all the time I have for this one today...................Cre8
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 28. January 2004 20:30
Cre8 wrote quote:
RBH, you sure spend a lot of time here for not thinking it's worth much [[Confused]]
FYI, this is a thread about doing what you said you don't think's worth doing......
In fact, I spend considerably less time 'here' than I used to. I read and post on what interests me, and unfortunately, less and less interests me here. The time and energy necessary to do the kind of Filter calibration program you described is considerably more than the 10 or 15 minutes I spend in a day composing posts for ISCID. The former is the level of effort I don't see worth expending.
RBH
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 29. January 2004 02:01
Cre8: Meanwhile, I still maintain that when Dembski's filter attributes something to design and it is being used correctly, that it will be right.
In other words, if it fails, it was not used correctly?
Cre8: If and when a conclusive example comes up to refute Dembski I will certainly drop my support (as will most everybody who now supports it), but at this point I don't see it happening. As I said SC comes from SC, despite the computer shell game that's now being played
I see no shell game. What I see is that claims about the filter are being made which 1) remain unsupported 2) are likely erroneous.
In fact SC has been shown to be able to arise from non-SC so it may be time to reconsider? Or one can always move the goal posts of course.
IP: Logged
|
|
Cre8ionist
Member
Member # 140
|
posted 29. January 2004 08:17
RBH, 15 minutes for composing, how much for reading, how much for pondering etc...It could obviously be a significant amount of your valuable life being spent on something you consider worthless, forgive me but I'm still puzzled.
Pim, I'm not much of a goal-post mover, I just want to see things proceed fairly. I pointed out that there's a difference between what MN can do and what ID can do and even if Dembski were completely wrong, and I don't believe he is, in order to more fully understand our universe, we'd need to develop methods of detecting and quantifying this difference.
Rex, let me post a few more thoughts here, as I short changed you a little last night and I apologize for that.
First, it occurs to me that perhaps RBH could help you more than I could here, as you might listen more.
RBH said wrt my analysis of the basketball's text quote:
Further, there's no justification offered for Cre8ionist's selection of the two reference classes he identifies (letters of the alphabet and the English language). It might be the case that the appropriate reference sets are an alphabet of just 8 symbols: {i l n o s t u w} and a language with just two permitted combinations of those symbols: {wilson solutions}.
Italics mine
RBH apparently at least understands that you can't just apply Shannon's definition of information as decisive when you're dealing with meaning. You've got to do other work. Shannon's definition is incomplete wrt the information here, in that it doesn't deal with meaning. That is, two texts of equal length, one loaded with meaning and one which is random nonsense, are treated as exactly equivalent wrt to their information content by Shannon's theory.
This is how you're approaching the Martian code. And I don't think (and RBH may agree) this concept of information is suitable for evaluating the information content of a meaningful text.
My question to you is, do you really think SETI would use your method of evaluating possible codes? I could just see you there with your fellow researchers holding up the Martian code photo saying "I see a bunch of dashes and dots. But how were they produced? I don't know. Must they have all come out the same? I don't know. I could postulate a one-dash or one-dot marking process, but I really don't know that. How do I account for the regular array? Should I read left to right, top to bottom or in some other order? " ![[Razz]](tongue.gif)
I suppose, we're ready for the next pic anytime............Cre8
[ 29. January 2004, 08:25: Message edited by: Cre8ionist ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 29. January 2004 10:23
Cre8 wrote quote:
RBH apparently at least understands that you can't just apply Shannon's definition of information as decisive when you're dealing with meaning. You've got to do other work. Shannon's definition is incomplete wrt the information here, in that it doesn't deal with meaning. That is, two texts of equal length, one loaded with meaning and one which is random nonsense, are treated as exactly equivalent wrt to their information content by Shannon's theory.
This is how you're approaching the Martian code. And I don't think (and RBH may agree) this concept of information is suitable for evaluating the information content of a meaningful text.
But I thought the object of this exercise was to apply Dembski's Explanatory Filter. "Complexity" for Dembski's EF is just improbability, a transform of Shannon information. Meaning doesn't enter into it. And I quite agree that this concept of information is not suitable for evaluating a meaningful text. But Dembski's EF provides scant guidance for how to incorporate 'meaningfulness.' I suppose it must somehow enter via background knowledge involved in finding a specification, but since that's a pretty subjective process I can't see a principled way to resolve it. The same problem arises with the notion of "specification" in specified complexity, particularly when functional specifications are invoked.
RBH
IP: Logged
|
|
Matthew J. Brauer
Member
Member # 819
|
posted 29. January 2004 12:32
Cre8,
I freely admit my ignorance as to how the EF is to be applied. I'm especially unclear as to how the probability of a phenomenon arising by chance is to be calculated.
For this reason, I'll have to pass on taking up your challenges. (Sorry!)
However, I'd like to invite you to look at an old thread of mine, recently bumped, titled 'Solar eclipses and CSI'. I'd be interested in knowing how onw might evaluate such a phenomenon in terms that are useful to application of the filter.
-Matt
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 29. January 2004 12:40
Cre8: Pim, I'm not much of a goal-post mover, I just want to see things proceed fairly.
Same here.
Cre8: I pointed out that there's a difference between what MN can do and what ID can do and even if Dembski were completely wrong, and I don't believe he is, in order to more fully understand our universe, we'd need to develop methods of detecting and quantifying this difference.
Something I fully support but given the evidence it seems obvious to me that so far the proposed methods of detecting, namely the design filter, has yet to show that it can reliably detect these differences. In addition, despite claims to the contrary, SETI and archaeology or criminology do NOT use the design inference filter to infer design, on the contrary, these approaches use methods which include motives, opportunity, methods to infer design. I suggest you read Del Ratzsch book. In appendix A he describes in quite some detail Dembski's filter and the many limitations of the filter. In short, the filter is a logical argument in which design is merely the " set theoretic complement of the disjunction regularity-or-chance". In other words, even if the design inference were to infer 'design' it has no causal link between 'design' and 'intelligent designer' and in fact it has not even a causal link between 'design' and 'designer'.
Del reaches the following conclusion "That Dembski is not employing the robust, standard, agency-derived conception of design that most of his supporters and many of his critics have assumed seems clear."
Dembski's contributions to providing a logical framework to design should be acknowledged but the extension of this filter to 'intelligent design' is quite another step.
Wimsatt, who was featured on the cover of Dembski's first book comments as follows on the claims made in NFL
quote:
Unfortunately "popular" presentations of "Intelligent Design" have tended to give the impression that it rested solely on mathematical demonstrations. Anyone who could have succeeded in showing that natural selection is incapable of generating biological structures according to standards from mathematics or logic would have constructed a mathematical proof that would have dwarfed Godel's famous Undecideability theorem in importance. As one who read Dembski's original manuscript for his first book, found much to like in it, and had appreciative remarks on the dust jacket of the first printing, I can say categorically that Demski surely has shown no such thing, and i call upon him as a mathematician to deny and clarify the implications of this advertising copy.
Link
In other words, while the approach chosen by Dembski is an interesting example of a logical analysis, the extension of its claims that it can 'reliably detect intelligent design' has to be rejected.
As RBH points out, ID proponents have had many years of opportunity to show that the filter can be succesfully applied and so far the extent of such an excercise can be found mostly in this thread. This thread shows that the issue of probability estimates (aka Dembski complexity) are non trivial and fraught with problems and complications.
quote:
So typically, patterns that are likely candidates for design are first identified as such by some unspecified ("mysterious") means, then with the pattern in hand S picks out side information identified (by unspecified means) as releavant to the particular pattern, then sees whether the pattern in question is among the various patterns that could have been constructed from that side information. What this means, of course, is that Dembski's design inference will not be particularly useful either in initial recognition or identification of design
Del Ratzsch Nature design and science p. 159
[ 29. January 2004, 12:43: Message edited by: Pim van Meurs ]
IP: Logged
|
|
Matthew J. Brauer
Member
Member # 819
|
posted 29. January 2004 13:04
My submission to the "inference game".
Is the "smiley face" on Mars specified?
How about the Cydonia "face" (in the Viking imagery, pre-MGS)?
IP: Logged
|
|
|
|
Cre8ionist
Member
Member # 140
|
posted 30. January 2004 09:00
RBH, the difference is between solely using probability, which would lead to Rex like solutions, and the combination of probability and specificity. With the latter incorporating something akin to a pattern(meaning) search. This is why, all the pictures thus far have a definite pattern, they have the potential for being non-ad hoc. Dembski calls them specifications as opposed to fabrications.
Nobody is likely to look into their bowl of alphabet cereal and think "Somebody's been tampering with my cereal," generally. But if you look into your bowl and see the following,
it changes things. You then say to yourself, after pattern recognition kicks in, that it is highly unlikely that this meaningful text appeared by law or chance, or some combination of the two (and the more text you see, the closer you are to absolute confirmation according to the filter).
Though you may not use those exact words. Point being, your average bowl of alphabet cereal presents no cause for special design detection to occur, it's only after some sort of pattern recognition that you would even try the filter. This is why a smiley face or a basketball might be proposed for evaluation by the filter, but not something like a normal pile of beach sand.
Now to the smiley face, I don't personally see a way of determining the degree of complexity in the Martian smiley face, but to me Matt, even if we were to use the enhanced version,
Matthew Barton, 1998
I would have trouble putting it through the filter. People are a little mistaken I think, in their thoughts about what the filter can and can't detect. I don't think it's really designed for this type of structure, perhaps I'm mistaken here, but the EF is designed to detect CSI, and while the smiley face is certainly specified I think that it would probably fail in any accurate assessment of it's complexity, were we able to do an accurate assessment. However, if we can get an accurate assessment of it's complexity, and it's specificity were a bit clearer, then I think it could be run through the filter. Methods must be developed to quantify complexity in these types of cases. As yet I haven't seen an accurate method.
This inability of the filter to evaluate all examples should not dissuade us from seeking examples which it can evaluate however. Where the rubber will actually meet the road (and I think most of us can foresee this), is when the filter is used on biological examples. Not just the flagellum, but on far smaller examples, such as certain proteins or genes. If ID is successful and we detect, quantify and qualify examples from biology as possessing the "difference", a.k.a. the designsignature, we certainly will have a strong argument for ID in biology. The filter seems especially well suited for this kind of design detection, however, like the shell game that's now being played with computer codes (evolutionary algorithms etc...), savvy biologists will play shell games with life's code. Therefore, we will need IDers to continue to put on their "Amazing Randy" caps and do some more debunking..................Cre8
IP: Logged
|
|
brauer
Member
Member # 398
|
posted 30. January 2004 10:02
Hi Cre8,
This is what troubles me about the concept of CSI, and specification generally. It seems that there are whole classes of phenomena for which the procedure "doesn't work".
Most critically the problematic phenomena usually turn out to be those phenomena for which we already have conclusions regarding design.
This makes it very difficult to test and/or calibrate the performance of the filter.
I've had a very difficult time for instance, getting anyone to conclude whether or not solar eclipses are "complex" (sensu the EF) or specified. The problem seems to be that "the EF works best for objects such as biological organisms."
But this is where the contention is! Why should it be that the phenomena whose "designedness" everyone can agree on isn't amenable to the operation of the filter?
Given the extreme "no false positives" claim I find it telling that there is such resistance to applying the filter to non-designed phenomena.
IP: Logged
|
|
|
|
RBH
Member
Member # 380
|
posted 30. January 2004 12:08
Cre8 wrote quote:
People are a little mistaken I think, in their thoughts about what the filter can and can't detect. I don't think it's really designed for this type of structure, perhaps I'm mistaken here, but the EF is designed to detect CSI, and while the smiley face is certainly specified I think that it would probably fail in any accurate assessment of it's complexity, were we able to do an accurate assessment. However, if we can get an accurate assessment of it's complexity, and it's specificity were a bit clearer, then I think it could be run through the filter. Methods must be developed to quantify complexity in these types of cases. As yet I haven't seen an accurate method.
Nor can I see an accurate method. But those kinds of stimuli are precisely the stimuli of interest! All these examples of letters floating in soup and dots and dashes on a stone are contrived, and do not veridically model the phenomena about which we really want to know, the biological phenomena that ID repeatedly claims are designed. Sure, if I saw "Manufactured by Og Nanobiotics, Inc., on Beta Centaurus. Patent applied for October 22, 4004 BCE" encoded in a cryptogram embedded in a string of mitochondrial DNA, then I'd say "Hey! There it is! Wowie Zowie!!" But one hasn't yet seen that, and it is the only kind of example of reliable design detection we are repeatedly offered.
RBH
[ 30. January 2004, 12:10: Message edited by: RBH ]
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 30. January 2004 12:50
The cereal bowl example however does not seem to use the Design Filter to reach a conclusion of 'design'. So I am not sure why we are moving back and forth between a variety of examples, none of them too relevant to the issue at hand
Cre8: This inability of the filter to evaluate all examples should not dissuade us from seeking examples which it can evaluate however.
But if these examples are all 'trivial' then we have to ask, what are the limitation of the filter. Del Ratzsch has given us a very likely answer.
Cre8: Where the rubber will actually meet the road (and I think most of us can foresee this), is when the filter is used on biological examples.
So far some attempts have been made to apply the filter to the flagellum but these cases use a model which, while amendable to probability estimates, seems unrealistic biologically.
Cre8: Not just the flagellum, but on far smaller examples, such as certain proteins or genes. If ID is successful and we detect, quantify and qualify examples from biology as possessing the "difference", a.k.a. the designsignature, we certainly will have a strong argument for ID in biology.
But given the problems of applying the design filter, one may argue, that given the likelihood of false positives, even a design inference in biology may not help us much. In fact, in Dembski's filter, a design inference itself cannot exclude a natural designer for instance. After all the design inference is a logical approach not a causal and the link between design and designer needs to be bridged. In other words in nature the question is not design in nature but the nature of the designer.
Cre8: The filter seems especially well suited for this kind of design detection,
How have you reached this conclusion if the filter cannot live up to its claims in these trivial examples?
Cre8: however, like the shell game that's now being played with computer codes (evolutionary algorithms etc...), savvy biologists will play shell games with life's code. Therefore, we will need IDers to continue to put on their "Amazing Randy" caps and do some more debunking.
Rather than do 'debunking' ID could benefit from a more positive approach and do a thorough analysis of the limitations of the filter to non-trivial cases. This has nothing to do with 'shell games' but all with good science or the lack thereof.
[ 31. January 2004, 16:47: Message edited by: Pim van Meurs ]
IP: Logged
|
|
|
Printer-friendly view of this topic