|
Author
|
Topic: Does intelligence imply “motive”?
|
gedanken
Member
Member # 594
|
posted 08. September 2003 02:02
NEWS FLASH!
(In our earlier fictional example of a slot machine that produces a string of “D”s and “R”s – look back on page 1.)
The laboratory discovered something wrong in the slot machine! Something that had not ever been considered in the analysis. They ran the machine for hours with a logic analyzer attached (device that records detailed microprocessor activity). And then it failed in the lab – it produced a sequence of 500 “D”s.
And now the cause was revealed. There was an instruction in the program that changed slightly as a single bit in the memory was unreliable. And that bit failure was not one that ever appeared on the various tests, nor was it one that anyone had been able to figure out as a possibility. It only occurs at just the right temperature and other conditions. It did not, as is often the case, send the machine of into never-never land of noticeable failure, except that it always produced a “D” symbol. This is why the failure was never previously detected, nor did the analysis turn up the possibility. The lab and software people, all going through all the possibilities they could come up with did not consider the actual case here.
Now let’s apply to the EF. Of course – the EF is tautologically true – the event did (in our simulated example) have a low but not exceedingly low probability by a mechanistic (non-intelligent) process. Any described event with actually low probability will occur with low probability!
But it was one that was not recognized. Everyone kept claiming that such a process could not exist. They produced prodigious analysis to show that the case could not exist. Yet what was wrong was they missed an important consideration that had not been thought of before!
And how do we know to keep looking? Well the issue of “motive” was an important clue in our investigation. There did not seem to be any motive for any known designers to make the machine behave that way. It did not benefit the club, because the club lost players when it failed, thus loosing money. It did not benefit the manufacturer, who got a bad reputation. It did not benefit the players, who lost money when they played. In fact we knew, using “motive” as a side-consideration, that intelligent design was not a very reasonable conclusion – though it could have been a possibility.
So in our normal process of evaluation of the case, we used “motive” as a clue. The process was not strictly eliminative in our normal evaluation of the case. ID enthusiasts’ claims that the EF represents the way we actually think about such problems is false. And we have a clear (hypothetical) example of the kind of failure of the EF in real-world application.
What the Easter Island example provided was a discussion of difficulty with lack of clarity in “specification”. Here we have a problem with the knowledge of the probabilities.
The explanatory filter has two different types of hidden knowledge.
One is that the “ID” cause to the event has to actually happen. Presumably it has an effect in the physical world. So when the analysis is done on the natural-non-intelligent causes, the eliminative approach ignores two possible cases:
1) The ID case is ignored, but claimed as result when natural-non-intelligent causes described meet probability criterion.
2) Natural process causes can occur that were not analyzed or were analyzed incorrectly.
The missing information of #1 is pretty much obvious – just it is explicitly ignored in terms of any discussion in the eliminative explanatory filter.
The missing information of #2 is just that – missing from the discussion. Why is #1 different from #2? They are both cases not considered in the analysis of the probabilities of the natural-non-intelligent events that were considered.
[ 08. September 2003, 02:10: Message edited by: gedanken ]
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 08. September 2003 02:46
Gedankin, you bring up an interesting point (one that I think has been answered before in other discussions (perhaps not satisfactorally), although it will be interesting to see if people also recognize it as the same point and offer the same answer).
However, I have two questions. First, I didn't quite understand what you were getting at with point (1). Could you clarify/re-explain? Second, your example is not low probability relative to the universal probability bound. Does that impact your conclusions?
IP: Logged
|
|
Micah Sparacio
Member
Member # 6
|
posted 08. September 2003 08:02
quote:
This new criterion for specification ('non-specified if there is(are) some angle(s) of view in which the object doesn't match the independent pattern') apparently makes it virtually impossible for anything to be said to be specified, since one can almost always find some view or level of abstraction at which the independent pattern and object under test do not match.
This is precisely why I brought up the issue of warrant in epistemology.
There might always be some angle or view that makes our belief unjustified.
Say I'm driving down a road and I see what I think are a bunch of black sheep. I stop the car and walk over to the edge of the fence to get a closer look. They still look like sheep. Now, at this point, should I be satisfied that I know that they are sheep or should I get a closer look? Perhaps it is really just a wolf dressed up as a sheep.
When can I say that I *know* what it is? How many angles do I have to investigate? There are better analogies, and more sophisticated problems, but the issue is this: what RBH sees as a problem for the design inference is just as much a problem for epistemology in general.
Also, as Mike Gene points out: when an experimental program encourages us to investigate multiple angles, we are all the better for it. In fact, this is the very nature of Academia and intellectual curiosity. We "make do" with our current epistemological situation while striving to make better models, investigate more angles, etc.
The design inference isn't to be(or at least shouldn't be) viewed as the end of the road, but as an explanatory element along the road.
Let's take the design inference into high speed: say my first view of the Old Man in the Mountain was from precisely the right angle to see the face. Well, my preliminary inference might be that the rocks were designed to look that way, though I wouldn't put as much weight on this inference as one that I had the chance to investigate more thoroughly. Now, as I move around the mountain (or move closer), I become increasingly doubtful about my original inference. This shouldn't be a problem: I'm merely modifying my knowledge as I discover more information about the structure. My inference about the structure becomes more reliable as I investigate further.
Isn't this the nature of human (and scientific) knowledge, afterall? I fail to see the implication that RBH has alluded to. Rather than be a problem, the opportunity to investigate objects/events in more detail lends itself well to an ongoing empirical science.
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 08. September 2003 09:57
Rex asked:
quote:
However, I have two questions. First, I didn't quite understand what you were getting at with point (1). Could you clarify/re-explain? …
Let’s examine all possible causes for event E. By the way, everybody noticed I said “ALL”? I’m not meaning to be impolite, I am just emphasizing that at this point I mean causes that we might have not talked about, causes that we might not think of for a hundred years, causes that have intelligent origin as well as non-intelligent origin, etc. …
You see that to know about all causes is to be omniscient. If one is omniscient, the very notion of “probability” of an event changes. One knows the details of how a coin is flipping, perhaps years before the mechanism performing the flip actually does it (by that I mean to consider even people who could carefully flip a coin to make it come out a specific token, or any other process that seems to produce a random result because we don’t know all the details.)
But we are not assuming such knowledge in the normal case of the explanatory filter – we must by the very nature of our lack of knowledge leave something out. We must by necessity reduce our standards somewhat from that notion of knowing all causes. In fact “chance” is precisely a formalized lack of knowledge, formalized in a specific way so that various events can be categorized by the “probability” relation, but we do not know details of the specific event that will control the selection within those possibilities except in a generic manner. We have to know the generic formulation of those event possibilities in some manner, or we cannot estimate the probabilities, however.
So I want to categorize a couple of kinds of knowledge that are left out of the explicit formal procedure of the strictly eliminative explanatory filter. We could list the event’s causes we want to categorize as C1, C2, C3… The probabilities of these causes acting to produce the event E could be designated P(C1), P(C2), P(C3) …
Now in the explanatory filter, we have categorized the events C1, C2, C3… explicitly as “natural-non-intelligent”. We have, for example, not tried to categorize every cause we can think of. In the Caputo example we have not categorized the probability that Caputo rigged the flip/drawing/whatever. Call that event I1 (intelligent cause #1).
So actual analysis of the probability, done without the “eliminatory” requirement, would be P(E) = sum P(C1) + P(C2) + P(C3) + … + P(I1). However the explanatory filter only considers P(E) = P(C1) + P(C2) + P(C3). That is because we don’t consider the “intelligent design” cause. It is also because we have at the moment only been able to think of 3 causes for the event, C1, C2, and C3. That is in part because the “eliminative” explanatory filter explicitly says not to consider the intelligent cause which may be readily at hand. (We don’t consider whether Caputo cheated, we don’t consider whether humans carved and raised the Easter Island figures, we don’t consider whether the programmers of my slot machine put in a special back door that causes strings of “D”s, because that is how the EF is supposed to work.)
So now I am considering some things that are left out. Consider C4, a chance cause that we didn’t think of. Consider I1, the intelligent cause that we were explicitly not thinking of.
What I am doing is classifying those causes that we left out. I1 is category #1, C4 is of category #2.
But we left out I1 because we were explicitly supposed to leave it out, as part of the formal procedure of the explanatory filter. We were not explicitly supposed to leave out C4, that’s just a fact of life in the real world that we did not think of every possibility.
But the key point is that since we were explicitly told not to “think” of the intelligent cause I1 in the analysis – so we did not think about it. That was not particularly a case of not being able to think about it. Or is it? What were potential intelligent causes of a flagellum? (Don’t answer that here – give it its own thread.)
quote:
… Second, your example is not low probability relative to the universal probability bound. Does that impact your conclusions?
(Emphasis added – I assume you mean the slot machine example in question.)
What I am getting at is the ability to use the explanatory filter in the real world. Go back and read the description of the slot machine case again (I mean that for everyone, not speaking to Rex). In the slot machine case we analyzed and re-analyzed possible cases over and over (in the hypothetical scenario of application of the EF.) And we thought that we had eliminated the possibility of the chance failure of the machine code by auxiliary means, such as checksums and analysis of algorithms. And we also thought we had eliminated the possibility of the programmers cheating, once again by various parallel systems of observation and analysis.
But yes – the event E did turn out to not be of low probability by natural causes. That’s what we know now! It’s not what we knew then when we were first doing the analysis. The originally known probability of the event by the “chance” processes was 1/2^500, as each “D” had a 50:50 probability according to all analysis that was done, that was explicitly considered in evaluating the EF.
Let’s consider the “tautological nature” of the EF that Rex’s remark points out.
Of course, now the event E appears to not have low probability. But that is the case every time that an actual physical explanation for any event is discovered. The chances that the “actual” event was discovered, and it was of extremely low probability by class – is of course extremely low, effectively non-existent. So any event, when its causes are known in greater detail, will be likely to make the EF conditions true. It is very unlikely that a specification can be given “independent of the event” for an event that is actually of extremely low probability.
The EF is tautologically correct if we always consider our most recent information. But that does not make it a “reliable” pattern recognition procedure for real world use (reliable in the sense of “no false positives”).
(Cases in which the EF were not “tautologically correct” are cases in which either the specification was not truly independent of the event, or these cases only occur with expected rate of occurrence of the UPB or other threshold used in the analysis.)
What is at issue is the issue of specification and its relation to probability, and that relates to our just previous discussion. If the specification is one that has “motive” for someone to fulfill – such as the string of “D” results – then we have knowledge of a likely human cause for the event. We should of course consider those causes in any comparative analysis, but we leave them out of the EF’s eliminative analysis.
“Motive” is potentially a useful aspect to consider among others, but of course it can be abused as Rex has pointed out. Motive is not directly related to probability, the path is tortuous and the analysis is still difficult. One may still be faced with inability to estimate probabilities. I am suggesting recognition of factors that can possibly lead to improvement of the EF, primarily by recognizing that other factors than strictly “eliminative” might be useful.
[ 08. September 2003, 11:32: Message edited by: gedanken ]
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 08. September 2003 10:57
By the way Mike and Micah’s points about the angle of viewing are quite relevant.
OOPS, I just “saw” something interesting (hypothetical scenario follows).
I just “saw” (in my mind) a suspended configuration of dots in a museum. The dots are red and green, and distributed in three dimensions. And when one walks arround the display, various dot combinations appear. But at one particular angle of viewing something very odd happens. I look at the dots, and a very interesting color pattern emerges. In apparent RED letters I see a message: “INTELLIGENTLY DESIGNED MESSAGE” in the pattern, written in English characters and language as shown. This appears only in the specific angle of viewing because the suspention of colors is distributed so that they no longer align in the pattern when viewed from any other angle!
Another case: A message is encoded on a page but is viewable only with ultraviolet light. This is equivalent to a “special angle”, only is a special viewing condition. Once again eliminated with the viewing “robustness” consideration. In fact the “robustness” of viewing criterion would eliminate all cases of cryptography from ID analysis.
Now consider Mike’s request for robustness of the pattern analysis. Do we eliminate these cases? Of course that would remove additional possibilities of “false positives”. In fact, anything one does to explicitly remove possible “positive” results of the EF will monotonically improve its “reliability” in terms of “no false positives”.
In fact, when we put enough constraints that the EF produces no positive results whatsoever, then it will have maximal “reliability” in terms of no false positives, since it produces no positives at all.
So the answer to Mike’s question is yes – it will either improve the EF or at least remain neutral.
And in the stone figures, we really don’t need to “improve” the EF in that way. That is already built into the probability analysis. There are more chances that a random process can make a figure that meets a pattern when viewed only in a particular viewpoint than a random process can make a figure that meets a pattern when viewed from multiple viewpoints. This is very closely related to my example above, as the pattern events have to occur in nearly associated third dimension to creat the pattern as viewable from multiple angles, rather than just in the two dimensional projection of the view.
Mike’s concerns will already be accounted for in analyzing the EF’s probability that the event was created by random processes – when one does the analysis considering all aspects correctly. Of course “consideriang all aspect correctly” is the rub. We must correctly analyze the probability calculations. One cannot just use calculations that one might wish or desire to give the result.
But this does not resolve the issues I brought up with regard to “specification”. The “specification” of the Easter Island figures given, of “human anatomical form” (approximate statement), is not sufficient. It does not describe that which appears in the Easter Island forms. Or if it is taken to be so general and loose language that it does, it applies to the natural forms as well. We have not seen the “specification” given independently of the event.
And if you iteratively change the specification (adding more complexity) so as to make one that captures the Easter Island figures, but rejects the random stone figures, then you have most certainly not created a specification that is “independent of the events”. In fact you have tailored the specification to the event – created a fabrication. You can’t for example, make use of side knowledge that the Easter Island figures are “designed” in order to refine your specification. You can’t possibly know this if the EF is your only tool of analysis!
The Easter Island and other stone figures are not a case in which we have demonstrated any “false positive” results. Rather they demonstrate difficulties with specification and analysis of probabilities. So far we have not found a specification that simultaneously does two things: a) captures the Easter Island figures, b) rejects the random stone figures. We have cases that reject both, we have cases that accept both. Neither are failures of the EF, because the case that accepts both is known to be high probability and thus the EF does not “infer design” – it is a false negative failure to infer design of the Easter Island figures. And the other case does not capture the Easter Island nor the random stone figures – once again another “false negative”.
Neither case is a failure of the EF, just we have not demonstrated the Easter Island as a useful case of the EF application, nor understood how to create a useful specification that is truly “independent of the event”.
For a “false positive” we have to leave something out of the analysis, as is discussed in just previous post.
Of course there is a problem if one claims that the formal Explanatory Filter procedure has actually made an objective distinction on the Easter Island figures. That has not been demonstrated, and should not be claimed.
[ 08. September 2003, 13:08: Message edited by: gedanken ]
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 09. September 2003 11:15
This post has two sections, first some thinking out loud, and second further development on Mike’s “robustness” idea. I wanted to complement all those who participated in this thread, as I feel it has for me supplied some real “brainstorming”, as my concept has changed considerably from when I started the thread based on what I have learned.
---
(Thinking out loud)
Earlier I made a post “Part 1 – type 1-3 distinction”, which was intended to be a sort of organizing heading for a series of concepts, possibly for extension to a paper. I don’t mean that the post itself was paper contents, rather it and discussion therefrom could outline topics.
What I have discovered is that instead of 2 parts, there should probably be 3 parts:
(The “distinctions” are combinations of projected and “actual” concluded ID logic conditins, found in my second post on page 2)
Part 1 – “Type 1-3 distinction” : discusses how one could have a failure of the EF based on transitions or differences between the EF apparently and erroneously producing, and not producing an “inference of design”, while the actual cause was a non-intelligent natural process. Here the issue is whether or not the EF infers design by apparently meeting the “specification-complexity” criterion. (Does “motive” get smuggled into the decision of meeting the “specification-complexity” criterion?) Also intended under this heading was the aspect of missing information contributing to a mistaken analysis of non-intelligent natural processes, and whether “motive” would help in knowing whether to continue to look for such missing possibilities.
Part 2 – “Type 3-4 distinction” : was to discuss how one could have a failure of the EF cases in which the EF did “infer design”, and on transitions or differences between the actual universally agreed cause being classifiable as “design” or not. We assume that it is clear that the “specification-complexity” criterion is met, and the issue is how we come to universal agreement on the cause being classifiable as “design”. (Does this use “motive” for example in our analysis that is universally agreeable?)
My question is whether we need a “Part 3”, heading which discusses claims that the “specification-complexity” criterion has been met, when in fact they have not been met in the formal procedure given the information at hand. In other words one intuitively perceives “design”, yet the formal procedure of the EF if followed would not actually produce that result – yet is claimed to produce that result. What is not clear to me is whether this case is really under my “part 1” heading, or did it bring in new issues. One issue is changing the meaning of terms in the middle of the argument (a logical fallacy), as in changing the meaning of meeting the “specification” of “the anatomical form of Homo sapiens.” Would projecting such a logical fallacy fall under part 1? Do we use “motivation” apparent in our own perception of the situation, for example, in interpreting our meaning of meeting the specification as an experimenter bias?
---
On Mike’s “robustness” issue:
I was thinking first of how to show the correctness of one aspect of what Dr. Nelson suggested, Mike and Micah and others discussed. But this is the aspect with regard to natural random processes constructing such an apparent pattern only:
Consider a 10 x 10 x 10 (three dimensional) grid. We quantize the grid from continuous 3-space to make a conceptually simpler analysis. First look at a two-dimensional projection:
code:
..........
.....*....
....*.*...
...*...*..
..*.....*.
.*.......*
.*********
.*.......*
.*.......*
.*.......*
In this grid we see an apparent letter “A”. Now consider that this is a three dimensional space, and that the “*”s are observable in that pattern head on from our “Z” axis, but that they are distributed among the 10 different (quantized) planes. (I can’t draw that – wish I could.) Consider this exact bit combination in two dimensions, with a 1 where there is a “*”, and 0 otherwise. It forms a stream of 100 bits per plane, or 1000 bits in all.
Now if the pattern “A” were required to be in a specific plane entirely, planes normal to either X, Y, or Z axes, there would be only 30 combinations of 2^1000 combinations that produce that result. This corresponds to the “A” occurring in the 10 X planes, the 10 Y planes, or the 10 Z planes. (I’ll ignore diagonal planar presentation of the character for simplicity of analysis.) Since the “A” occurs in a plane in this case, it is readily viewable off-angle and will be viewable from a wide cone of angles in each case.
But if the pattern is allowed to randomly occur in depth, yet is still viewable in strictly the Z plane (out of the paper), we could allow a large number of combinations. Each of the 24 “*”s could appear in any of 10 planes, independently of the others, making for 10^24 combinations viewable strictly from the Z axis, as compared to only 30 combinations that are strictly planar in one of 3 axes. There would be even more combinations (slightly less than 3 times) for the other two axes, for about 3*10^24 combinations as compared to 3*10^1 combinations for planar representation. So the chances of the figure arising by chance has increased by a factor of about 10^23.
But we need consider that stone structures must be connected, and that natural processes make much more approximately smooth surfaces. (Extensions are easily knocked off!). So to construct a stone figure in three dimensions, it will be much more constrained and more likely in all aspects to have similar edges in multiple views. However the general notion, as opposed to a quantifiable ratio, is shown above.
NOW we consider the issue of meeting the specification. Nothing in the original specification (here possibly appearance of an English language letter, or other case of matching human figure in some level of abstraction) required viewing to occur in multiple planes or cones of viewing. Why would this be a requirement? Specifically why does “The specification is the anatomical form of Homo sapiens” state at any point that the “anatomical form” should also be viewable from different angles?
It is very important to recognize that it is the probability of meeting the specification that is supposed to be analyzed in the explanatory filter – not the probability of the specific event in all its characteristics! Dr. Nelson has characterized the specifics of the individual cases in his description of the problem of viewing angles – affecting probability of the event as discussed above – and he did not describe a differentiation of the stone faces from the stone figures on Easter Island in his specification!
Also, for example, I have considered the case of cryptography – that the character was intended to only be recognizable from a single angle. But wait – we have just brought up “intention” or “motive”! That the Easter Island figures should be viewable as human forms from various angles is a statement of motive in method of construction of a “representation”. No such analysis of representational motive is allowed in an eliminative inference!
BUT let’s consider further the decision of where we could apply such a “robustness” consideration in the EF analysis.
“Robustness” could be added to the procedural steps of the EF – but as I have pointed out that would eliminate all cryptography applications, and could seriously damage any possible interpretation in biological forms as well.
It could be part of the specification. Now we have complicated the specification – we have for example “robustly viewable appearance of the anatomical form of Homo sapiens.” Where for example do we find that given independently of the event? We are treading on dangerous ground of tailoring the specification to the event, and thus hardly “independent of the event”. We are making the interpretation of the matching of the pattern much more subjective, much more subject to abuse.
But more importantly the Easter Island figures are not “robustly viewable appearance of the anatomical form of Homo sapiens.” The reason is simple, they do not correspond accurately to the anatomical form, except in part and only in certain aspects of the apparently cut edges indicating significant recognizable aspects of human form. Where do we humans have arms without hands, attached all along the side? Where do we have extremely large eye sockets? Extremely large bulb or trapezoid in center of face (where nose should be)? There are no legs and feet – how much of the human figure has to be matched?
A vague generalized specification of “human form” without specifying what portion of the human form allows the Mars face image to qualify quite nicely. Where are the limitations that narrow it to the aspects that are viewable in the Easter Island figures? “Robustness” criterion can be used but will eliminate all these peculiarities of the Easter Island figures from meeting the specification.
And when we widen the specification, we allow the Mars figure and others, thus producing a high probability event – thus again not meeting the “specification-complexity” criterion due in this case to failing the “complexity” aspect.
From everything I have been able to come up with, the EF is better off left alone with regard to “robustness”, just accounting for the probabilities properly using “robustness” as a guide for noting the probabilities of random process construction.
I don’t yet find “robustness” useful as a specificational aspect, unless one is considering “motive”. But when considering “motive”, it is extremely useful. The apparent “motive” of the people who constructed the Easter Island figures was to have a very robustly visible representation of human form in that location (with further aspects of “motive” quite opaque to us). But it is clear that there was some desire for recognition of those figures. This is very useful in a comparative analysis of causation, as the presence and motivation of people to make human representation has been well known – motive, means, and opportunity!
We could write our specification as “robustly viewable apparently representational appearance of a substantial fraction of the anatomical form of Homo sapiens.” Are we treading dangerously into tailoring the specification to the event? Has this improved the situation? We are clearly dealing with motivation of the presentation, and are much more accurately describing a comparative approach by implying a known motive, means and opportunity in our specification.
I make no claim that the explanatory filter is unreliable (no false positives) under all circumstances. It is not at all unreliable in the circumstance that there is substantial evidence of an intelligent agent constructing such an “event”, with for example motive, means, and opportunity. If you construct your “specification” such as to specify the existence of that evidence of motive, means, and opportunity, then the explanatory filter will certainly be reliable. It is in essence a comparative method, with one side of the comparison appearing in the specification. The demonstrating of a competing hypothesis of natural causes as low probability, as compared to (in relative terms) high probability intelligent causation is certainly part of any comparative method – thus the comparative method would not differ in its discussion of the natural process alternatives. The methods of inference of design could also be improved if they explicitly stated such consideration.
[ 09. September 2003, 16:00: Message edited by: gedanken ]
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 09. September 2003 16:30
Paul Nelson wrote (about a particular specification):
quote:
In the case of New Hampshire's "Old Man of the Mountain," for instance (alas the formation collapsed recently), the human-like features of the pattern disappear at nearly all angles other than the one shown in the photograph above.
RBH replied:
quote:
That seems to me to be a dangerous position for ID. It invokes yet another criterion for specification, namely that the whole event that is tested for specification must match the "independent" pattern. Thus, for example, on Nelson's new criterion a portrait painting is also not specified, since viewing the portrait from the side reveals only the edge of the frame of the portrait and the "human-like features of the pattern" disappear from that angle. In other words, if one can find an 'angle' from which the allegedly specified event does not match the "independent" pattern, it ceases to be specified. Is Nelson sure that IDists want to go down that road?
That's not how specifications work. In the case of the statues of the Easter Island, we have an alleged specification based on our knowledge of shape of humans, and this knowledge is apparently "independent" of the observed event in exactly the right sense. This knowledge also somehow uniquely determines a function f on the sample space. The specification for a statue on the Easter Island is then a set
S = {w | c >= f(w)},
where c is chosen is chosen so that S is the minimal set that contains the observed event. The specification is not invalid for the geological formations. The difference is that we may have to choose a larger threshold c in those case, thus making the specification less tight. Assuming that we really followed all of the instructions in constructing the function f, it is actually impossible for an event not to be specified. By choosing the threshold sufficiently large we can make literally any event (in the chosen sample space) specified, but the specification is not necessarily tight. Provided that we accept "specifications" as meaningful, the proper conclusion is therefore that all of the statues and geological formations are specified, but some specifications are tighter than others.
Erik
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 09. September 2003 20:00
Eric points out that we may want to delve into the “Generic Chance Elimination” form of the Explanatory Filter more deeply.
I’ll point out at the outset that I don’t claim that the EF (in various forms) is always unreliable, rather my claim is that its reliability depends on precisely the degree to which the specificational aspects correspond to any available knowledge of motive, means and opportunity. This thread was intended to study the “motive” aspect, but that was intended in a field of possible outside considerations. (I considered “motive” to be an especially relevant external criterion that seems to be smuggled into the examples, though seemingly labeled as not being considered.)
The choice of measure functions must remain independent of the event to be meaningful.
Rewording Eric’s terminology a little so as to be as close to the language in No Free Lunch (chapter 2) :
The specification for a figure (‘event’ E) on the Easter Island is then a set
Tgamma = {w element-of Omega | f(w) >= gamma }
We have lots of features that could be used to identify the match and many-dimensional configurations. Set Omega is the set of all possible stone configurations (including varying sizes and delimited volumes of larger stones). The function f(w) extracts features as a measurement that will have to be defined in some rigorous way, if the probabilities are to be even conceptually defined in a rigorous way. Then the criterion is gamma, which is of course chosen to make “rejection region” R=Tgamma an “extremal set” with smallest probability, wherein R includes E, the event.
I propose that feature extraction function f(w) extracts features of a stone configuration, and produces a feature similarity comparison to “the anatomical form of Homo sapiens”. If we have no rigorous or objectively repeatable measure of similarity to “the anatomical form of Homo sapiens” then we have no way whatsoever to quantify its probability. Rather than the way Eric suggested (which would seem to be in terms of difference), I suggest comparing the number of features or similar increasing value as similarity to whole anatomical form increases. This way we could have increasing function in two distinct cases – for example a whole body would have roughly twice the comparison of half a body, but also greater detail on the same fraction of the “body” would have increasing similarity. With a “difference” comparison, half a body would improve little even when done in more detail since the missing half (difference) would swamp the measure. For example we don’t want an extremely refined bust to have a lower rating that a gross set of cylinder shapes corresponding to body parts.
Now we have the problem of defining f in a manner that is even remotely rigorous, and potentially “independent of the event”. I’m going to discuss it strictly in terms of generic body form information, and not in terms of features of the surroundings or position which are not aspects of “anatomical form”, because the specification was suggested to be strictly in terms of “anatomical form” and not other features. It would be difficult to include environmental features and still claim the measure was “independent of the event”.
Clearly if we define feature extraction function f such that it gives higher value to stone configurations about which we have generalized knowledge as possibilities of representational activities of agents (humans) that have occupied Earth (and potentially Easter Island at the time), then we would have a more reliable explanatory filter than if we did not.
We could, for example, have f count the number of individual body major parts (leg, arm, etc.) that are represented in a clearly anatomically correct manner. How do we deduct for shape anomalies? These are difficult questions. Is f taking account of the positioning of the stone configurations, as in connected or not connected to other stone regions and surrounding terrain? If not, then it misses information that we observe in the Easter Island case. How much emphasis does f give to the concept of an apparently carved edge which could denote a division between body parts, rather than having the body parts distinctly formed and separated (as in having arms, ears, etc. shaped in true three dimensions as opposed to apparently “representational” edges defining them).
What we see is that there are dozens or hundreds of features of the Easter Island figures that correspond to representational elements (eyes sockets, nose, arms), and probably only a handful of more accurately positioned body parts (torso, head, neck). Yet they are not “accurate” in correspondence to the human body.
I will speculate on various feature extraction measures: If f requires extremely faithful comparison to body parts, then it will find few features in any of the random or Easter Island figures. But if it incorporates recognition of the representational features, then it will capture the Easter Island figures with a much higher score. In the latter case I believe extraction function f would rate the Easter Island figures much more highly than the other “geological” figures.
But of course the feature extraction function that rates Easter Island figures highly is one that is tuned with knowledge of human representation techniques. One that requires faithful conformance to body shape will rate them poorly.
Beyond that, speculation on probability densities and integrals would be rather meaningless. This case is not really subject to a rigorous analysis, even though Eric and I have proposed a start on such, because it would be almost impossible to rate configurations in terms of probability of construction.
Unless we also consider the location, position, etc. aspects (not easily part of “specification” independent of the event) then in my opinion we would not have a measure that produces a probability integral of 10^-150 or less for the Easter Island figures by natural geological processes. This is due to limitations on measurement meeting the “specification” (as compared to natural cases), and it not a proper reflection of the fully detailed circumstances of those figures in their location. We are not capturing the free-standing nature of these particular figures, because our measure is based on conformance to specification of human body shape, not optimizing for considerations like free-standing in a field. We are also not capturing the aspect of assembly of four units side by side. Many such aspects are missed due to not being a comparative analysis. (A comparative analysis would not have requirement of independence of event hypothesis for features measured).
I make the claim about probability because of the prevalence of some existing stone figures like the face. The probability of such figures is relatively high, since several of them exist. So the probability of more complete figures existing would be powers of those probabilities. Just remember that the feature extraction needs to accept considerable abstraction or “representation”, and if it does so it will tend to accept other abstractions in other locations. And if it does not accept (or score highly) the rather crude representation elements, then it will have a low score and other locations with low score will rate highly in relative terms. Since the figures are more complete, this is in essence a small multiple on the 10 exponent of the probability (as multiple simultaneous features must be present). This is my basis for estimating that the probability would not reach 10^-150, since the probabilities of partial elements existing is so near unity. Once again a feature extraction that accounted for position, and surroundings would do much better, but is not allowed due to not being “independent of the event”.
In the end, we recognize that the Easter Island figures could have been produced by humans, and not likely by geological events, and that they have characteristics of objects produced by humans. It is not the matching to the human anatomical form, but to the human penchant for representation that gives the best clues.
Now let’s assume that the EF would generate a low probability for the Easter Island figures, given the known geological processes that we might consider for the region (though not necessarily low as UPB). We still have a knowledge that human construction is a likely explanation. That compares favorably with our having missed a possible geological or other natural process that could make such shapes. We are fully justified in concluding “design”.
With the more highly rating continuous functional measurement methodology, we may have captured the information that is needed to predict high relative probability of human construction. The more that feature extraction function f extracts features of human representational carving design, the lower probability will be found for the extramal set in the chance elimination method. We may properly conclude that “design” was category – but we have assured ourselves that the designer activity was possible because we have smuggled in side knowledge of motive, means, and opportunity in generic terms. We know humans desire (are motivated) to construct representational likeness, they use tools and construct large objects, and they existed in approximate time frame and not inconceivable locational access.
Before this I had not realized that comparative methods can capture more information from the scene than eliminative methods, and could do so in a manner that is measurable, because comparative methods are not restricted to being independent of the hypothesized event.
By the way, no adaptation needs to be made for “robustness”, since the “extremal set” step in the generic chance elimination procedure combined with increasing feature extraction increasing match eeks out the most robust conditions available based on independently specified measurement criterion for the probability distribution.
But remember this is all based on the highly questionable ability to construct an objective feature extraction function from stone configurations to similarity (independent of the event), and then even more questionable ability to estimate probabilities of their construction by natural non-intelligent processes of the environment of the scene.
[ 10. September 2003, 15:56: Message edited by: gedanken ]
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 10. September 2003 15:56
Now I am going to demonstrate the application of the newly discovered aspect – that more information can be extracted from the scene in a comparative method than with the eliminative method which must remain independent of the hypothesized event. (Remember that the eliminative method has no hypothesized event for the intelligent agent action and must then necessarily ignore aspects that are very specific to the given scene and its environment.)
Consider our fictional case of the slot machines that generates strings of “D” and “R” symbols. In the end we discovered that it simply got stuck. But all analysis showed that was not possible until a last minute test actually turned up the real cause.
Of course Dr. Dembski has something to say about this: (No Free Lunch pp67-68.)
quote:
Design inferences therefore eliminate chance in the global sense of closing the door to all relevant chance explanations. To be sure, this cannot be done with absolute finality since there is always the possibility that some crucial probability distribution was missed. Nonetheless it is not enough for the design skeptic merely to note that adding a new chance explanation to the mix can upset a design inference. Instead, the design skeptic needs to explicitly propose a new chance explanation and argue for its relevance to the case at hand. Design inferences are falsifiable in the sense that if we are wrong in our assessment of the probability distributions that might characterize an event, then we can be wrong in eliminating chance and inferring design for that event. But the mere possibility of falsifiability is never enough to falsify a claim. Nor does feeding false claims into an argument invalidate that argument. An argument’s validity depends on its logical structure and not on the truth or falsehood of its premises (for the validity of the design inference see section 2.7).
The problem is that “section 2.7” does not take any account of two different probabilities: First it does not account for the “possibility that some crucial probability distribution was missed”. This could be given a rudimentary approximation of probability, it could be given bounds. Nor does it account for probability that an intelligent agent could cause the event – once again something that could be given a rudimentary estimate. If the probability of the event being caused by an intelligent agent is reasonably much greater than the probability that such a “crucial” cause was missed, then we may be confident in the generic chance elimination result when it says that the “chance” hypothesized cases are extremely improbable. (In fact we are confident in that precisely because we have thus made a comparative argument, with the chance elimination and confidence in our having “swept the field clean” as one side, and the known “reasonable” probability of agent action on the other. These are not statistically quantified, but the significance alpha in the generic chance elimination argument has little if anything whatsoever to do with the actual probabilities of the alternatives.)
Now in our case we thought we had “swept the field clean” of possible chance processes. The problem is that we could not find any likely “intelligent design” processes either! Rather than proposing that we now know the answer, we should have continued to research the problem. We should reject both the proposed chance causes as well as the “design” cause! And that is what we “personally” did in our little fictional scenario, and it paid off – even though the generic chance elimination procedure showed “design” and was unreliable.
In our case, the probability of the “intelligent agent” acting was so low (because of the controls) that the possibility of our having missed something in the analysis was a very important probability.
Let’s compare this to the Caputo example. Once again from NFL, p 56:
quote:
The key question therefore before the court was whether Caputo actually put this procedure into practice when he made the ballot selections, or whether he purposely circumvented this procedure in order that the Democrats would consistently come out on top. And since Caputo’s actual drawing of the capsules was obscured to witnesses, it was this question that the court had to answer. …
Motive, means, and opportunity – these are worth considering and are being smuggled into the analysis without acknowledgment in the formal procedure. The eliminative approach makes no use of any of these, since motive, means, and opportunity are all aspects of the “intelligent design” cause, and are thus formally ignored. The probability of the event is treated as though the only processes worth considering were the random processes that Caputo claimed.
In our slot machine example, there was possibly some hidden motive, means, or opportunity. But they did not show up as likely candidates. Making use of this information was very helpful in knowing to continue the research. There was no necessity for “the design skeptic needs to explicitly propose a new chance explanation and argue for its relevance”. Rather what was needed was to know that the job was not done, that research needed to continue. A declaration that the cause was found (“design”), that the job was done, should not have been given.
Yet another word on the tautological nature of the chance elimination argument (in this case) here: Of course now that we discovered another chance distribution that we had not previously known about or thought about, it should be entered into future consideration of the chance elimination procedure. But what was the reliability of the chance elimination procedure given the information at hand? It was not based on the significance level alpha which was the statistical quantity used, rather the reliability was dependent in part on factors that we were explicitly told to ignore—the details of the intelligent agency, even if considered in a most generic manner.
How could we make use of the additional information that must be ignored in the generic chance elimination argument? Here we turn to the second example Dr. Dembski provides, of the “compressability of strings”. According to the compressabilty of strings argument, our occurrence in our case of 500 sequential “D” symbols, compressable with a short algorithm, suggests “intelligent design” since it looks “non-random”. But by using information at hand we can do comparative analysis, which differs from case to case. Comparative analysis does not mean that we simply choose the best result of comparison (such as by Bayesean posterior probability), it also means knowing when to continue doing research.
In the case of our slot machine failure, it turned out to be a “chance” event of reasonably low probability that caused the sequence of “D” symbols—even though such a sequence is highly compressible. The normal sequence of random “D” and “R” symbols was of course generated by the normal 50:50 random process that was expected. So we had multiple chance processes. But the compressability of strings suggested that we should look for “Design” when coming from an algorithmic computer system, similar to the Caputo example. But the compresability should also have been a clue to the failure of the machine – precisely a “chance” event! (Rex intuitively came to this conclusion.)
But now consider a different failure of our machine. Rather our slot machine simply varies the probability just slightly so as to favor the house, but otherwise appears random. This would be a case where the “compressabilty” signature might be very useful, since it could have been a secretly coded algorithm hidden in the machine. The difference in importance is where to look. From our previous post we could use a feature extraction function which accounts for “motive”. Slight advantage for the house is of greater probability as a “designed” action for the slot machine, than the highly “unlikely” outcome of 500 “D”s – precisely the opposite analysis as in the Caputo example, and differing precisely because motive is important to consider. Motive, means and opportunity are worth considering.
Use of a feature extraction function that is tuned to the possible motives is a method that cannot be used in the generic chance elimination argument, because they must be independent of the hypothesized event and characteristics of the hypothesized (even if generalized) agent. But equally compressible strings can be differentiated based on motive, in our example of various slot machine failures. This can produce a superior discriminent. Dembski rejects Sober’s suggestion of comparative methods, when they are in fact superior in many cases, and the chance elimination procedure is unreliable in many cases unrelated by its statistical parameters.
Dr. Dembski says (P114) :
quote:
Still more problematic for these critics, however, is that the likelihood approach—even if it could be made to work—can make sense of design only by presupposing specified complexity.
Unfortunately Dr. Dembski has it backward. “Specified complexity” comes out of generalized intelligent agent action hypotheses. Caputo had motive, means, and opportunity, and we knew it and the court used that information in its considerations. The string of “D”s was most significantly “specified” because it corresponded to Caputo’s motive. But humans also had motive, means, and opportunity to erect the Easter Island figures—just not as clearly defined. Dr. Dembski asks us to completely and absolutely generalize and make abstract the concept of intelligent agent action, but does not appreciate the value of hypotheses that partially generalize and only partially abstract the actions of that agent, while still considering motive, means, and opportunity.
The explanatory filter and chance elimination procedures are most unreliable when faced with an unlikely intelligent agency. The statistical parameter alpha used in the generic chance elimination procedure does not characterize the reliability of the procedure when it is given a low value.
By the way, I like Dr. Dembski’s comments on NFL, page 103, in which he describes an urn experiment. There are two urns, each with ten balls. The first is split evenly 5:5, the second 7:3 with white and black balls respectively. A thousand samples are taken with replacement, but it is not known from which urn, and they all turn out to be white balls. Dr. Dembski shows how the comparative likelihood method would predict that the second urn was the source by a factor of 10^150, but that it seems quite ridiculous to accept that either urn could have been the source. But then of course Dr. Dembski is not willing to use the same criterion in comparing a conclusion of action by an intelligent agent, and action by a specific list of natural non-intelligent processes that were studied. I have no problem with Dembski’s eliminative argument being considered to “eliminate” the hypotheses as tested—my difficulty is with then drawing an unexplored and untested conclusion.
[ 10. September 2003, 17:20: Message edited by: gedanken ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 10. September 2003 18:39
gedanken's remarkable analysis deserves a more thoughtful response than this, but for the moment this is all I have time for. He wrote quote:
I have no problem with Dembski's eliminative argument being considered to "eliminate" the hypotheses as tested - my difficulty is with then drawing an unexplored and untested conclusion.
In NFL and elsewhere Dembski argues strenuously that a general Bayesian approach (estimating the posterior probability distribution over an array of hypotheses conditional upon evidence) is inappropriate, and clings to a Fisherian hypothesis testing approach. The advantage of the latter for ID is that it enables one to "infer" an (unspecified) ID conclusion without ever having even peeked at any (unspecified) ID hypotheses! That is, one does not have to have a specific ID hypothesis nor does one have to evaluate an ID hypothesis in light of the evidence in order to "infer" it.
RBH
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 12. September 2003 00:39
In his article with HTML title shown on the blue browser title bar as Elliott Sober’s Intentional Fallacy (and titled “Elliott Sober’s Independent Evidence Requirement for Design” at top of the article) Dr. Dembski brings up issues that are highly relevant to this discussion. (But I fail to see how Sober presented an “intentional fallacy”, and as we shall see I don’t even think that Sober was wrong at all.)
Dr. Dembski brings up an example in which technology advances to an ability to create significant lifeforms from “scratch”. (And of course that has already been done today in terms of creating a Polio virus, a rather small “lifeform”.) He ponders the question of detecting design of organisms in that particular environment. The key here is the consideration of independent knowledge of a potential “designer” with motive, means, and opportunity to produce the event in question.
The article suggests that in such a case the “independent evidence” would help establish an acceptable level of credibility for the detection of “design” to be considered scientific. This is important because Dr. Dembski is validating comparative methodology here, even though he disputes such comparative methodology in No Free Lunch in discussing Sober’s criticism. (p.101)
But now what about the original biological evolution question that interests ID enthusiast’s? Dr. Dembski suggests that there has been no change in the evidence in moving the discussion to that sort of case.
In “Elliott Sober’s Intentional Fallacy “ Dr. Dembski comments:
quote:
But what if that's not the case? What if our best probabilistic analysis of the biological systems in question tells us that unguided natural processes could not have produced them with anything like reasonable probability? Could the design of life in that case become more probable than a Darwinian explanation (probabilities here being interpreted in a Bayesian or likelihood sense) simply in virtue of there being independent empirical evidence attesting to designers with the capacity to produce biological systems?
and
quote:
But if design is a better explanation simply because of the independent attestation of technologically advanced space aliens, why should it not be a better explanation absent their independent attestation? If Darwinism is so poor an explanation that it would cave the instant space aliens capable of generating living forms in all their complexity could be independently attested, then why should it cease to be a poor explanation absent those space aliens? Again, the facts of biology themselves have not changed.
This gets to the heart of the issue! (But Dr. Dembski’s presentation gets the results exactly backwards.) Here is the problem – the analysis ignores the relative importance of what is missing from the analysls:
The EF methodology, for example, leaves a very important aspect of its own analysis unspoken. Let’s return to the two examples I have discussed previously, Caputo, and our hypothetical “R”/“D” string generating slot machine.
What is unspoken in the EF is that there is a mistake inherent in the analysis of Caputo’s probability of creating the string of “D”s. Left out of the analysis is the probability that Caputo, who had motive, means, and opportunity, should cheat and fake the results. So what is worth comparing is the posterior probabilities that Caputo cheated, and the posterior probabilities that some non-intelligent case of regularity was left out of the original analysis. Clearly Caputo had motive, means, and opportunity, and the probability that he acted is relevant in considering the reliability of the EF method in this case. As discussed at length, our slot machine had no such apparent motive, and we in “fact” (actually fiction) discovered the unexpected and unexamined natural cause.
These two classes of cases, each left out of the original analysis and left unspoken, are what should be considered with respect to reliability of the design inference. It is very important to recognize that both of these were left out, were left unspoken in the formal “design inference” method. Which is the most significant, the reasonableness that Caputo had access and motive, or that some non-intelligent natural cause was missed in the analysis of the case?
Now consider Dembski’s case used to suggest “Sober’s intentional fallacy”. He suggests that independent evidence need not be considered at all, ignoring how the EF procedure depends on that evidence being smuggled into the specifications used in examples used to demonstrate “reliability”.
I realize that incorporation of such aspects into ID recognition procedures like the EF might affect the results in certain favorite applications like biological organism structures like the flagellum.
For example consider the analysis I showed above how the EF essentially leaves out two competing classes of causes. One is the missed details of the “design” cause, the other is the missed details of possible non-intelligent natural processes. We could compare the probability that space aliens had some motivation and access to create flagellum structures in some bacteria, and were present, versus the probability that some natural process that was not found in the initial ID enthusiast’s presentation on the flagellum was in fact the cause.
We can note that both probabilities would easily be above the “universal probability bound” and that the ID enthusiast’s presented “explanation” of natural process was indeed unlikely to be the actual explanation for the flagellum structure. What is important to note is that the significance value alpha, used (for example) in the “generic chance elimination” version of the EF, is not really “significant” at all. Rather what is significant is the comparison of the case of leaving something out of the natural process analysis as compared to leaving out the analysis of the designer actually acting. In the case of the flagellum, for example, it is much more reasonable to find a missing piece of evidence supporting natural cause, than to find support for a designer action. And ID enthusiasts keep demanding that evolution scholars find solid positive detailed evidence for the specific construction of that flagellum, or else they consider the evolutionary scenario to have failed. They are not willing to accept the considerable evidence that shows likely explanations for the flagellum as though they did not affect the reliability of the explanatory filter or variant “design inference”.
The “design inference” is unreliable because these important factors are not considered. What has changed from the Caputo case to the case of the biological system that Dr. Dembski considers is that in the Caputo case motive, means, and opportunity give the “design” cause explanatory power – and give reason that easily swamps the chances of leaving something out of the analysis of the natural processes. This cannot be said for the biology case in which the EF will not be reliable in terms of no false positives because there are no reasons to expect the agent action to be likely.
In the Next post I shall give a mathematical proof of the unreliability of the EF variants, under these sorts of conditions. This will actually take these probabilities of errors and agent actions described in this post into account – not as a comparative method, but to show the reliability of the EF under these conditions.
Then I am next considering generalizing the “design inference” procedure with degree of consideration of aspects like motive (or other elements of side knowledge) to see how the reliability improves when these factors are included. If I can work it out, I will add in the improvement found with removing the restriction on the “independence from the event” in the sense of independence from all knowledge of the alternate hypothesis that includes factors from the potential intelligent agent.
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 12. September 2003 00:50
Gedanken:
quote:
Then I am next considering generalizing the “design inference” procedure with degree of consideration of aspects like motive (or other elements of side knowledge) to see how the reliability improves when these factors are included. If I can work it out, I will add in the improvement found with removing the restriction on the “independence from the event” in the sense of independence from all knowledge of the alternate hypothesis that includes factors from the potential intelligent agent.
If Gedanken's attempts to include motives pan out to be succesful in helping establish a design inference then we may start to understand why design inferences in some case are doable while in other cases they remain likely unapproachable, namely 'motive'. For instance in the SETI case the assumptions are that an sentient intelligence would use the same frequency in which we would be transmitting a signal since this frequency range is least likely to be confused with natural noises. Although it seems that some black holes may be 'singing' in a very low b-flat :-)
What's the chance of a b-flat 57 octaves below the middle C I wonder ?
Criminology, archaeology may benefit from a better understanding of how additional data find its way into the design inference. After all motive, opportunity, and means are commonly used in criminology.
The same seems to apply to the story Paul presented at the Saint Paul United Methodist church, namely the 'Mint Jelly Ridge', a story about someone who is cheating insurance companies by pretending to slip on mint jelly in restaurants. While patterns was certainly relevant, motive must have played a role as well.
Btw I am impressed to hear that Paul seems to consider it too early for ID to be considered a valid scientific alternative to evolution (NCSE reports vol 23 no 2 2003)
I think that starting with Del Ratzsch realization that the design inference is not very well suited to detect new design, combined with the 'motive' driven examples proposed by Gedanken we may be able to extend the design inference to make it workable in scientific inquiry by including positive evidence in the form of motive, opportunity, means etc.
I am looking forward to Gendanken's well reasoned continuation of the exploration of the design inference, it weaknesses and how we may be able to rectify some of these weaknesses.
[ 12. September 2003, 13:00: Message edited by: Pim van Meurs ]
IP: Logged
|
|
Mike Gene
Member
Member # 149
|
posted 13. September 2003 19:09
Gedanken: But this does not resolve the issues I brought up with regard to “specification”. The “specification” of the Easter Island figures given, of “human anatomical form” (approximate statement), is not sufficient. It does not describe that which appears in the Easter Island forms. Or if it is taken to be so general and loose language that it does, it applies to the natural forms as well. We have not seen the “specification” given independently of the event.
I wouldn’t be so sure about this.
Hey, I gave the rock/statue pictures the toddler test. I asked my 3-year-old daughter some questions.
I showed her the picture of the rock (from Wyoming) without the nose (thanks to Photoshop). I asked her, “What is that?” She said, “I don’t know.” The, I carefully scrolled to the New England “Old Man” picture (minus nose) and asked her the same thing. Again, “I don’t know.” Then, I showed her the Easter Island statues (with blurred noses). What’s that? She said, “Three man rocks.” Then, I showed her the originals. She labeled the Wyoming picture (the black and white one) “A giant rock.” She said “I don’t know” with the “Old Man” picture.
That a little child can pick out “man rocks” but not the “faces” suggests the specification of “human anatomical form” may indeed be sufficient. More later.
I should also mention that earlier, I had done the wife test. I showed my wife the Wyoming picture (minus nose). She identified it as “some rock.” The same with the New England rock. She recognized the Easter Island structures (without nose) as statues. Then, I showed her the originals. She still didn’t see the profiles and labeled them as “rocks.” With the Wyoming formation, she said her eyes focuses on the most superior portion of the formation. Once I pointed out the profile, she immediately saw it.
[ 13. September 2003, 19:12: Message edited by: Mike Gene ]
IP: Logged
|
|
Pim van Meurs
Member
Member # 541
|
posted 13. September 2003 20:09
Mike shows that the issue of specification is very subjective making the issue of reliably detecting design based on specifications a bit troublesome.
Especially when we get to deal with trying to detect 'new design' something for which the design inference seems quite unsuitable. One may propose 'front loading' but such merely displaces the issue to the initial conditions failing to help us resolve the issues surrounding design, intelligent design, appearance of design, actual design let alone the 'designer' which may include natural processes, regularity/chance, or supernatural processes or intelligent causes.
It is undeniable that humans are 'well equipped' to see specifications and 'design' where none may actually exist or at least can be identified as such in an objective manner.
If specification is such a problematic issue or in other words, if specification is almost trivial then the argument becomes a pure issue of complexity which is just another term for probability. Seems we may have come a full circle without being able to resolve these issues.
On the other hand, gedanken seems to be developing an extension to the ID inference which may be more workable in areas of 'known design' such as criminology, archaeology etc.
The question whether or not ID can play a relevant role in detecting 'new design' seems to yet to be supported and in fact direct and indirect evidence suggests that the likelihood of such is small.
Nevertheless, as Del I am still looking forward to positive contributions of design. We may take notice of some of the examples of peer reviewed literature Dembski is claiming as relevant to intelligent design.
Perhaps an interesting thread make explore how these papers show research 'supportive of intelligent design'?
[ 13. September 2003, 20:12: Message edited by: Pim van Meurs ]
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 13. September 2003 20:48
Mike--interesting observations.
I happen to know (from psychophysical tests) that what you did was the best way to get the result that people couldn't identify the natural features. You probably didn't do this intentionally, but:
Note that the nose is the most prominent feature in a profile but not of a head-on view. Therefore, removing the noses from the head-on view wasn't really the same type of manipulation. (Removing the eyes is vastly more confusing there--I'm sure the Mars face would have been easily detected with no nose (since it already doesn't have much of one ![[Razz]](tongue.gif) ), but not without eyes.)
Also, people tend to see the same patterns on multiple presentations of a scene, even if the scene has changed. Thus, people who don't see the head the first time are less likely to see it the second time when you've added back the salient features.
That said, I think the Easter Island statues are easier to identify as human-like than the natural formations. I just wanted to point out that the experiments were complicated by well-understood psychological factors. (It's still a fun observation.)
IP: Logged
|
|
|
Printer-friendly view of this topic