|
Author
|
Topic: Can intelligence be described scientifically?
|
Micah Sparacio
Member
Member # 6
|
posted 06. March 2003 14:39
Erik writes:
quote:
Generality, accuracy and simplicity are all ends in themselves, and we try to make our models as general, accurate, and simple as possible. To a very good approximation, that's what scientific theorizing is all about. All other things being equal, use the simpler model because that is... well, simpler.
In contemporary philosophy of science, the notion that simplicity is an end in itself is very much in question. In fact, my philosophy of science professor continually brings up the possibility that the principle of simplicity became a virtue in science in large part due to laziness/convenience (Duhem and Quine). Who says that the world should be simpler rather than complex? Does the data stream indicate that simpler theories are preferable?
My only reason for bringing this up is that it seems quite naive to suppose that the data stream is not theory-laden to a certain degree. The mere fact that observation involves both the unconscious and conscious organization/structuring of data (Hanson) seems to indicate that the practice of science requires and involves more than the data stream.
Perhaps I'm not even addressing your point in this thread, as I've only read the last two or three posts. I apologize if my criticism is off target, and will try to read your initial post in the next week (though I'll be camping for the next four days).
[ 06. March 2003, 14:55: Message edited by: Micah Sparacio ]
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 06. March 2003 15:29
Micah,
I think this somewhat misrepresents or misinterprets the reasons for "simplicity" in scientific explanations.
Such simplicity, for example, does not make any effect on how simple actual relationships in science are in the sense that they get as complex as required to represent the underlying reality.
But consider statements like:
God willed that force be proportional to acceleration.
God willed that momentum is conserved.
God willed that energy is conserved.
Now one could test these statements by checking for systems in experiments constructed to demonstrate acceleration and force relationships, and momentum and energy relationships.
So are verifications of those relationships demonstrations of the above sentences? The answer is no, because they don't demonstrate anything about "God".
Now I gave this as an extreme example. But it shows the essential point, the observation has to differentiate all the elements in the proposition. One removes elements that are not necessry to be consistent with the part of observation of the physical world that is not examined in detail with the construct being discussed. Adding extraneous detail to the discussion is not helpful in understanding these physical relationships.
As such the removing extraneous details from the explanation is the notion of "simplicity" that is being dealt with. It does not follow from actual simplicity of the real world, but from the practical consideration of finding understanding that is actually based on observation of the physical world.
Now there are cases of alternate explanations (or descriptions of relationships in) aspects of nature in which there might be different levels of description of the complexity involved, but which are exactly equivalent in terms of having nothing extraneous in the particular description. In this case there is not a strong case for the simpler appearing description, because the equivalence in terms of having nothing extraneous in each does not lead to any way to exclude one description over the other. If they can be translated as with a mechanistic translation process (as in equivalent mathematical formulations) then the two are actually the same explanation. If you have two statements that are in reality the same explanation, but one is simpler to embrace, then naturally the simplest to understand is preferable, but the other is not excluded since it is in fact equivalent. In this case it is simply a matter of good writing practice to get your point across in the best way -- and once again not something that affects the degree of complexity in underlying nature that is being illuminated.
I think these points are important to consider in this thread, which seems to be gettin off topic in my view.
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 06. March 2003 16:54
Micah Sparacio, since you are the second one to misunderstand my point about simplicity, it seems reasonable to conclude that I'm not expressing myself well. Let me therefore quote from the article "Occam's Two Razors: The Sharp and the Blunt" (the link takes you directly to the PDF, click here if you prefer to download it via a visit to the author's home page):
quote:
"Leaving aside for the moment the question of how to measure simplicity, let generalization error of a model be its error rate on unseen examples, and training-set error be its error on the examples it was learned from. Then the formulation that is perhaps closest to Occam's original intent is:
First razor: Given two models with the same generalization error, the simpler one should be preferred because simplicitly is desirable in itself.
On the other hand, within KDD Occam's razor is often used in a quite different sense, that can be stated as:
Second razor: Given two models with the same training-set error, the simpler one should be preferred because it is likely to have lower generalization error."
In Domingos's terms, I am (for the purposes of this thread) only advocating the first razor. Please note that I have not been advocating any particular inference method. Generality, accuracy and simplicity are what scientists should try to achieve, but I have made no statement about how to achieve the mix of these goals. To the best of my knowledge, the first razor is not questioned in contemporary philosophy of science. Regarding the second razor, it can be phrased more generally than Domingos does so that it is not necessarily a statement about the universe. For instance, in the essay "Simplicity is Not Truth-Indicative", Bruce Edmonds* argues that that simple models are not more likely to be correct, but he thinks that certain psychological facts makes a preference for simpler models a surer path to the correct model.
I think the theory-ladeness of observations is a more interesting point. It's getting late where I am and my ability to express my thoughts in words is not at maximum right now, so I'll get back to that later.
Erik
* BTW, some articles on Bruce Edmond's main page seem relevant to the subject of modelling intelligence.
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 06. March 2003 18:59
Cornelias, I seem to have already answered everything you asked, so I'll quote myself to answer your questions.
quote:
You are right that V = iR was within the experimental error, but so were an infinity of other models. Why do you pick out V = iR as your solution?
quote:
Rex: I pick V=iR out of the possible values because it's simple and consistent with the data.
quote:
I'm not clear on how you justify your assumption of uniformity based on past experience. What does that have to do with the future?
quote:
Rex: I have a model, and the model is that the external world exists (which inherently includes concepts like identity), and that induction works. I note that this is consistent--if I'm especially clever, I note that given Godel's incompleteness theorem, I'm likely to be unable to do better than note a lack of obvious inconsistency.
quote:
How does science work without the assumption of uniformity?
quote:
Rex: In the past, inductive models applied to distant-past events predicted near-past events.
And a few more words here. You need not assume uniformity, but uniformity is a low-information model, so it is worth trying early on if you have no reason to expect anything. In the past, that worked.
I would appreciate it if you made a greater effort to actually understand my posts before responding.
Erik, I'm not sure I'd view simplicity as an end in itself, personally. There are arbitrarily many models of arbitrary complexity that are equivalent to simple models to within experimental error. However, I value the simple models because they are more efficient for an entity of my limited intelligence. I really couldn't care less how complicated things were as ends in themselves; I'm much more motivated by what I can do with models, and I can do more with simple models than complex ones, as long as the simple model is good enough. (E.g. I'd use Newton's equations for interplanetary travel, but I'd probably switch over to relativistic equations for intergalactic travel, since I'm not sure the Newtonian equations would be useful enough over the distances/velocities involved.)
Edited attribution into the quotes. It was too confusing.
[ 06. March 2003, 19:00: Message edited by: Rex Kerr ]
IP: Logged
|
|
Cornelius G. Hunter
Member
Member # 81
|
posted 07. March 2003 13:46
Gedanken and Rex:
***************
Gedanken wrote:
But consider statements like:
God willed that force be proportional to acceleration.
God willed that momentum is conserved.
God willed that energy is conserved.
Now one could test these statements by checking for systems in experiments constructed to demonstrate acceleration and force relationships, and momentum and energy relationships.
So are verifications of those relationships demonstrations of the above sentences? The answer is no, because they don't demonstrate anything about "God".
***************
Why is it that they don't demonstrate anything about "God"?
Rex:
I'm sorry you thought I wasn't doing well at actually understanding your posts. Actually, I thought I was on track, so perhaps we have a disconnect. In particular, you felt I asked questions which you had already answered. For example, I asked why you pick out V = iR as your solution from the infinity of possibilities and you say you had already answered that by explaining that it is simple and consistent with the data. Yes, I realized that. Being consistent with the data fails to distinguish V = iR from the other infinity of solutions, so we're left with your criterion of simplicity. I wanted you to justify that criterion. In response to Erik you made your justification, saying: "I can do more with simple models than complex ones." I don't think this is quite right. I love to approximate and the Taylor series, for example, is a great tool. But in science, our use of approximations is carefully undertaken. The approximations are the exceptions to the rule. We know where we have introduced the approximation, and we keep careful track of how the approximation propagates in our solutions. We don't arbitrarily select approximations and publish them in textbooks as exact answers.
Next, I asked about how you justify your assumption of uniformity based on past experience. What does that have to do with the future, and How does science work without the assumption of uniformity? Again, you felt you had already answered. In part, you said: "In the past, inductive models applied to distant-past events predicted near-past events." But again, I was asking you for further justification, for this answer you've given is circular. Do you not see that you are employing the assumption of uniformity to justify your use of the assumption? Yes, you can show that a training set can be used to make good predictions of a test set (both sets in the past), but what does this have to do with the future? If you say the success of the predictions in the test set suggests to us that they will also succeed in future test sets, then you are assuming uniformity.
--Cornelius
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 07. March 2003 14:09
quote:
Why is it that they don't demonstrate anything about "God"?
Actually, given assumption of God as given it is most certainly the case that observations of nature are also observations about God. For example we can observe that God has given us a consistent reality that can be studied in science (once again starting with an assumption that God gave us whatever was observed, whatever the result of those observations).
But what if I had said:
The Great Turtle willed that force be proportional to acceleration.
The Great Turtle willed that momentum is conserved.
The Great Turtle willed that energy is conserved.
What would that be saying about The Great Turtle? Once again starting with an assumption that the Iroquois’ Great Trutle gave us whatever was observed in nature. In fact is would be saying something quite significant about “The Great Turtle”. Some would even make an interpretation that a Christian God was more compatible with observations of reality than the Iroquois vision was compatible with present day scientific observations of reality. But how universal would this be, how many aspects of the Iroquois vision might be more in accordance to certain aspects of observation? That might be a very interesting question.
But there is little in observations of acceleration, momentum, and energy relationships that speaks about “The Great Turtle’s” role in these things. These studies do not, for example, directly make distinctions between philosophical views of God vs. The Great Turtle -- rather these are quite distant issues that can be resolved after we have resolved the issue of what are the relationships of acceleration, momentum, energy.
That is the point, the complication of the creator or any other added aspect are important, but not so important that we should not examine the relationships in absence of those complications. Eric has described important issues in simplifying models to that which can be determined, and Eric and Rex have spoken eloquently on the issue. I’m just demonstrating an aspect in what I hope can be some degree of clarity, and is not intended to show the only dimension or issue of “simplifying” of models or scientific descriptions of reality.
IP: Logged
|
|
Cornelius G. Hunter
Member
Member # 81
|
posted 07. March 2003 14:10
Erik:
I understand you see this discussion of metaphysics in science as not quite so relevant, and that you want to move on. I don't see it this way, and if you'll permit, I'd like to make another response on this subject. You say:
******************
Erik wrote:
A scientific model should describe as much of the data stream as accurately and simply as possible, but it need not be true in any deeper sense than being accurate.
You apparently think that this amounts to the assumption that the world is simple. It does not. I mean that simplicity is an end in itself, not that simplicity is necessarily a good predictor of accuracy. Generality, accuracy and simplicity are all ends in themselves, and we try to make our models as general, accurate, and simple as possible. To a very good approximation, that's what scientific theorizing is all about. All other things being equal, use the simpler model because that is... well, simpler.
…
Furthermore, I did not write that scientific models assume nothing more than the existence of the external world. I wrote that no metaphysical assumptions beyond that are needed in order to do science.
******************
Regardless of how you want to cast your assumption about simplicity (eg, not an assumption about the world) it remains a metaphysical assumption. Metaphysical assumptions are those that lie outside of science. As one philosopher put it, they are the ones that are made before the science begins. I see no problem with your saying simplicity is an end in itself, but how then do you conclude that this metaphysical assumption does not constitute anything beyond assuming the existence of the external world?
--Cornelius
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 07. March 2003 20:15
This is starting to venture rather hopelessly off topic, but I'll make one last point to try to clarify myself. The reason why this might be relevant is that there are people (typically Christian Apologists) who maintain that the only possible comprehensible worldview is one with the God of the Bible; all others are nonsensical, or assume His Divine Intelligence or whatever. In this view, either everyone or no-one needs an explicit model of intelligence. The justification for this is usually complaining about circular arguments until you postulate God as the reason for everything, and the reason for himself. (Circular, but oddly accepted.)
Cornelius wrote:
quote:
We don't arbitrarily select approximations and publish them in textbooks as exact answers.
Yes we do--exactly when the approximations are simplifications that are within error of the best estimate. For example, my physics textbook has the formula pV=nRT in it. It is noted that this is for an "ideal gas", and that there are no such gases, really. Yet it is published as an exact answer.
A recent estimate of the charge on an electron was 1.6022177 33 +- 49 * 10^-19 C. Why was this estimate published instead of 1.6022177 331329571891984985689239517981759813459820658 +- 489699125609160984650918245098162309561098347? Depending on how the data was gathered, and what computerized representation was used to store it, and the fractional expansion of the formulae involved, something like the latter may have been closer to the true value. If you do your calculations to infinite precision in Mathematica, you see stuff like this all the time. Why was the short version selected? Because it's simpler!
We do select approximations and publish them as exact answers--and it is not at all arbitrary; it is precisely because they are simple.
quote:
Do you not see that you are employing the assumption of uniformity to justify your use of the assumption?
Yes, of course I see that. And I explained why I consider that good enough, although I suppose I didn't do it in enough detail. Requoting myself:
quote:
I note that this is consistent--if I'm especially clever, I note that given Godel's incompleteness theorem, I'm likely to be unable to do better than note a lack of obvious inconsistency.
It helps to know what Godel's Incompleteness Theorem says (hereafter GIT). It was formally proven for first-order predicate logic, but work since has shown it to apply to a wide variety of similar systems. Basically, it says that truth is not decidable, in that there exist statements that cannot be shown to be true or false within the system.
The implication is that it is not even in principle possible to know whether a set of axioms is consistent, working within those axioms, because you can't evaluate the truth of some statements and thus you can't tell whether they'd lead to a contradiction.
For this reason, axiom-sets cannot be proven correct or incorrect, and thus demanding such is silly. What you can do is show consistency, that is, look for obvious contradictions. An obvious contradiction to uniformity would be if it had never held in the past. We look and see, aha, there was uniformity. So it seems to be an okay axiom to hold, at least for now.
This also gets back to my point about simplicity. Uniformity is simple. That makes things a lot easier.
Finally, you seem to be drifting in the direction of asking "why"-type questions without pausing to think about whether the question itself is well-posed. Suppose you ask, "why is the universe simple?" If we are asking a causal why, our questions could be of two forms. Either they will be internal to the universe, in which case they will depend on a finite number of other qualities of the universe and hence will be circular, or they will refer to some property of A1 (an additional thing outside the universe). If you continue your quest of asking why, you soon find that A1 is in the same spot; either it is circular or refers to A2. Then if you look at the structure of U+Ai, either there is something circular there, or we never know why.
The point is that you don't distinguish the validity of multiple self-consistent views about the universe by demanding non-circularity. It isn't going to happen. You have to appeal to the universe to throw out the bad ones. Science is the process of taking the most generic set of good ones--the ones we have pretty much all include uniformity--and seeing how far you can get without worrying about all the particulars of each worldview. You're right that there's a bit of an assumption there, but it's usually taken to be as minimal as possible. (Not always; the success of materialism has prompted many to favor materialistic approaches to science. This is not minimal, but is suggested by the minimal adopted set since it contains uniformity.)
IP: Logged
|
|
Cornelius G. Hunter
Member
Member # 81
|
posted 08. March 2003 00:14
Rex:
I said "We don't arbitrarily select approximations and publish them in textbooks as exact answers" and you disagreed, giving the ideal gas formula and the electron charge as examples. Neither example supports your claim. Every first year chemistry student knows the assumptions, limitations, and errors that come along with the ideal gas formula. You say in your textbook it is presented as an exact answer – you must have a pretty bad textbook. And regarding the electron charge, even the example as you give it contains the uncertainty on the value. The value is not even presented as an exact value.
Here's the point. You justified your selection of V = iR from the infinity of possible models from pragmatism. You said "I can do more with simple models than complex ones," as though you don't attach any special weight to this simplest of solutions beyond mere pragmatism. I suspect that, in fact, you do. I know I do.
My contention is that V = iR is thought of by physicists, and presented in textbooks, as a sort of true or ideal answer. I say *sort of true or ideal* because, of course, the relationship doesn't hold over all conditions (eg, temperature ranges) nor for non ideal resistors (ie, those with inductance and capacitance). Never is it presented, nor do I think even thought of, in the pragmatic terms you present.
And indeed, the point I was trying to make is that if you were correct, then we wouldn’t see "V = iR" in textbooks with a box around it, and later repeated in the summary of the chapter, etc. If you remove the simplicity ideal, then what you would see in textbooks would probably be the latest Bureau of Standards fit, as with the electron charge.
Regarding uniformity and my point about using the principle of uniformity to justify uniformity, you say "The point is that you don't distinguish the validity of multiple self-consistent views about the universe by demanding non-circularity. It isn't going to happen." I agree; believe me, I'm not the one here trying to enforce non-circularity. My point was not to force non-circularity but to force the admission that there is more going on in science than merely the assumption that the external world exists. You agree there is a "bit of an assumption" going on here. Good, sounds like we agree.
But then you say that "What you can do is show consistency, that is, look for obvious contradictions. An obvious contradiction to uniformity would be if it had never held in the past. We look and see, aha, there was uniformity." But since, in point of fact, uniformity hasn't held in the past, then I assume by modus tollens you now reject uniformity. You don't? But surely you agree that V = iR (or V = iR' – 0.000000087) has never been found to be true in subsequent experiments. Why then do you say uniformity has been demonstrated?
Finally, as an aside you mention that "the success of materialism has prompted many to favor materialistic approaches to science." What "success of materialism" are you referring to? Since I can't think of any I'll not venture a response. I am curious about what sorts of assumptions one must be making in order to come to such a conclusion.
--Cornelius
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 08. March 2003 01:58
quote:
My contention is that V = iR is thought of by physicists, and presented in textbooks, as a sort of true or ideal answer. I say *sort of true or ideal* because, of course, the relationship doesn't hold over all conditions (eg, temperature ranges) nor for non ideal resistors (ie, those with inductance and capacitance). Never is it presented, nor do I think even thought of, in the pragmatic terms you present.
quote:
But then you say that "What you can do is show consistency, that is, look for obvious contradictions. An obvious contradiction to uniformity would be if it had never held in the past. We look and see, aha, there was uniformity." But since, in point of fact, uniformity hasn't held in the past, then I assume by modus tollens you now reject uniformity. You don't? But surely you agree that V = iR (or V = iR' – 0.000000087) has never been found to be true in subsequent experiments. Why then do you say uniformity has been demonstrated?
I think that the basic question we are getting down to (and is relevant to defining intelligence scientifically) is the issue that is sometimes called “methodological naturalism”. Now I must point out that I don’t accept “methodological naturalism” as being essential to science. What I think is essential is empiricism or basing distinction on simple assumptions of a naïve form of realism and on observation of physical reality that can be shared in a fairly universal manner. (To be shared in a universal manner, one must not start with assumptions of particular religious values, for example. This does not deny the truth, as apparent to some, of those values or constructs, rather it is to point out that they are not shared with sufficient universality and have not been based sufficiently on such naïve concepts of realism combined with observation of physical reality to become shared with considerable universality.)
Now I must point out that I have been shot down in my presentations before on what I might call a distinction between sharable observation from physical reality and what is called “methodological naturalism,” and I wanted to point out that my view and understanding of terms may not be “shared” completely.
I think that what is done in science is constructing models of physical reality, based on that observation combined with naïve assumption of existence of a realism. (And once again in my use of the term “model” I have been challenged -- what I want to do is to use a loose definition of the term “model”.)
Now “models” are abstractions. They are inherently not accurate all the time, precisely because they are abstractions, they don’t take into account everything about physical reality.
Rather than V = iR as a model for a physical equation, I think a better one would be F = mA, along with classical physic’s assumptions of constant mass and Euclidean geometry. Rex has already discussed this. But Cornelius said “An obvious contradiction to uniformity would be if it had never held in the past.” But F = mA is a prime example -- it holds sufficiently accurately to be useful for a tremendous number of cases. It was most emphatically not disproven by quantum mechanics and by relativity!. The view that QM or relativity demonstrates that the model of classical physics is in error is simply a claim that classical physics was not consisting of models (as I have indicated) in the first place! What these new paradigms showed was that cases in which the models did not appear to hold accurately did have more complex models that held in greater numbers of cases. (And of course we have the “correspondence principles” that must apply in such cases, since the simpler model most emphatically does hold accurately in so many cases.)
The essence of science is in part an art. It is the art of modeling, wherein one learns when a model will be accurate and how to deal as best as possible with cases in which it can be expected to not be so accurate. Then to go on and create such models.
(Please understand that I am including “relational” models in my concept here. For example descent with modification. It makes distinct predictions, yet those predictions are in the form of relations rather than narrowly focused expected action as in much of physics.)
They are useful because we can do useful things with them. We can go on and use such models in engineering, and in a great many cases in which we need to know answers to questions of what to expect in the physical world.
But we understand that these models are not the final answer. And that even though as ideals they can be shown to be highly accurate within proper domains of experience (domains learned as art) they are not taken as absolute truths. Because of this, we can make very useful predictions and designs. And society finds the usefulness of these predictions and designs sufficiently important to invest in study of such naïve models of reality. They are ideals and are not cast off simply because they are not accurate in all domains.
When they are cast off is when they are supplanted with models as easy to use (and thus more or less equally “simple”) but which cover larger domains of experience. (This includes being more accurate, as in more confining in relationships.) Thus QM and relativity, which supersede classical mechanics, does not supplant classical mechanics because they are not simpler to use.
We get into problems when we take such ideals as though they were absolutes, as though there was no art to knowing when and to what degree they apply. But we also see problems when people want to challenge that such ideals and models are not highly useful to society, simply because they see others making the first mistake.
[ 08. March 2003, 02:14: Message edited by: gedanken ]
IP: Logged
|
|
Cornelius G. Hunter
Member
Member # 81
|
posted 08. March 2003 03:30
Gedanken:
******************
Gedanken wrote:
But Cornelius said “An obvious contradiction to uniformity would be if it had never held in the past.” But F = mA is a prime example -- it holds sufficiently accurately to be useful for a tremendous number of cases.
******************
Actually, Rex wrote that bit about uniformity potentially being contradicted. I was just quoting him.
Regarding your praise of F = mA, I agree. F = mA has worked very well for us. We have assumed uniformity and simplicity and those assumptions have born fruit very well. Shall we now deny that we ever made them and deny they are necessary for science?
Do you see something of a double standard here? The evolutionist can say things like: "We assume simplicity for its own sake, but it constitutes nothing beyond the mere assumption that the world exists," or, "Yes, we make circular arguments now and then and we have little presuppositions here and there, but science is nothing more than materialism."
Can you imagine the reaction an IDer or creationist would get making statements like these? "Well, we assume God created the species and work from there." He would be shot down, this despite the fact that the biological evidence is against evolution. Anyone who knows about the DNA code, for example, knows that evolution is not "just science."
--Cornelius
[ 08. March 2003, 04:04: Message edited by: Cornelius G. Hunter ]
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 08. March 2003 11:23
Cornelius,
I fully understood that you were quoting Rex in the quote I quoted. I was simply identifying the portion of your post I was commenting on -- and your use of Rex’s point in your own way. But you appear to have ignored the entire sense of my post to focus on details you find important -- I suggest re-reading the major points I made. I shall continue that general topic:
Part two: Science as modeling
quote:
But since, in point of fact, uniformity hasn't held in the past, then I assume by modus tollens you now reject uniformity. You don't? …
How do humans learn about the usefulness of logic? I think that humans learn about this by experience -- that we find consistency in our experience that we can describe logical constructs in language, and find patterns are consistent with shared experience.
In fact I am somewhat skeptical of the arbitrary use of strict deductive logic without considering the art of its use. I am certainly skeptical of human’s ability to use logic consistently.
Just an interesting note, in searching for a clear definition of modus tolens the first Google link was to a discussion about the validity of MT: here.
An example from this site:
quote:
If there is fire here, then there is oxygen here.
There is no oxygen here.
Therefore, there is no fire here.
Now or course this is logically sound. But it can be misinterpreted and the types of misinterpretation point to the difficulty both in modeling reality and in having overconfidence in logic in the abstract.
Consider an area containing molten metal, possibly with heat being caused by a nuclear reaction (as opposed to an oxygen based heat source). Many people might view this as a place of “fire.” (And of course there is no oxygen.) Now strictly they would be wrong -- but this is a matter of degree as many of the aspects of “fire” are present including destructive influences of heat on structure coherence, etc.
What I’m getting at is that there is some art to even the use of logic. Inconsistency, even in appearance of logic itself, does not rule out the usefulness of logic -- quite the opposite. Rather it only shows that one must give greater thought to the problem and not trivialize it.
The usefulness of logic comes from the same notion of our ability to share concepts of the real world that gives us the usefulness of scientific approach. Modeling the real world is an art that can be shared in a fairly universal way, and focusing on that shared experience gives us useful ways to communicate in technical discussions. A very large part of our ability to effectively perform design in the modern world depends on the art of finding universal shared descriptions of how the real world operates -- of scientific modeling.
The original subject of V=iR is a case in which I have many years experience. This model is indeed something that must be taken as an approximation, an ideal. But it is not an ideal to be striven for, rather an ideal in simplicity of description that helps understand what happens when we construct such systems, and wherein the art of knowing what causes deviation from the simple model is essential to the art of design of electronic systems. I assure the reader that the model as written appears to be violated an exceptional number of times, in an incredible variety of ways. Is this cause for electronics designers to stop using that ideal?
I am equally skeptical of use of rules of deductive logic taken without the use of the art of proper application. These are used to argue against science in many ways. But science is an art of providing precision as is possible, not in claiming absolutes. The use of strict logic in mathematical reasoning is based on logical equivalence: if CONDITION then DERIVED-CONDITION. We only have “DERIVED-CONDITION” to the extent that CONDITION was an accurate description in the first place. Since science is inherently about approximation of reality, we rarely have a perfect case of perfect matching the original “CONDITION”. We see such arguments used against descent with modification, for example, in which chain of logical implication that descendent is the same species as ancestor thus “demonstrates” that we could not have had a descendent of a different species. The problem is of course that the original premise (that descendent is the same species as ancestor) is an approximation -- and the chain of slight changes shows the failure of the original claim, not the notion of common descent. And it highlights the art of scientific modeling.
An aspect that I wanted to emphasize (which will apply to our model of intelligence itself) is that all scientific models are creations of our own intelligence. They are descriptions in natural language in terms that we understand of constructs that approximate the way the real world works. We try as best we can to make those constructs accurately reflect the way the world works. But they are creative constructions, and cannot be divested of that character. This is not a reflection so much on the physical reality being modeled, rather it is a reflection of the character of the very processes of thought that are the subject of our discussion.
In my opinion I would like to get back to attempting to define intelligence itself (and doing so in a scientific way) rather than continuing to debate the general nature of science. I think that our discussions of the nature of science itself is important, as it is essential to understand the nature of science in order to make a scientific model or definition.
[ 08. March 2003, 11:30: Message edited by: gedanken ]
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 08. March 2003 13:57
Micah Sparacio, in response to your point that observations are theory-laden, I modify the one metaphysical assumption made from "an external world exists" to
"An external world exists and our (potentially theory-laden) observations are observations of this external world."
Rex Kerr, many physicists would say that their models are simple. In fact, many physicists would say that the theory of general relativity is simpler than Newtonian mechanics, but not that it is easier to use. I believe this sense of "simplicity" is measured subjectively as the ratio the size of the class of phenomena described by the model to the model complexity (where both "quantities" in the ratio are subjective). I think searching for models that are even simpler in this sense is something that scientists do as an end in itself. For instance, whenever someone publishes a paper suggesting a unified framework for describing phenomena previously thought to be distinct, I think this is the consideration that makes the reviewer think it is worth publishing (why have two models for two different classes of phenomena when you can have one model for both?).
Erik
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 08. March 2003 13:59
Cornelius G. Hunter, in your latest response to me, you wrote: "Regardless of how you want to cast your assumption about simplicity (eg, not an assumption about the world) it remains a metaphysical assumption. Metaphysical assumptions are those that lie outside of science. As one philosopher put it, they are the ones that are made before the science begins. I see no problem with your saying simplicity is an end in itself, but how then do you conclude that this metaphysical assumption does not constitute anything beyond assuming the existence of the external world?"
There are at least two fatal problems with your response:
Problem 1: You need to distinguish carefully between goals (aka ends in themselves) and assumptions. For instance, a stock broker has as his goal to make lots of money (because to him making lots of money is an end in itself). One of the more sophisticated ways to try to achieve this goal, is to formulate a model of the stock market. In such a model, it might be assumed that relative price changes are uncorrelated over time periods longer than one hour. In this example, we have
- the goal = making lots of money
- the assumption = relative price changes are uncorrelated over time periods longer than one hour
What we most certainly do not have in this example is the assumption = making lots of money. That's absurd. A goal is a goal, an assumption is an assumption, and a goal is not an assumption.
In the case of doing science, the goal(s) are finding models that are general, accurate, and simple. The models that are formulated may depend on various assumptions like time-translation invariance (perhaps what you call "uniformity"). Although it is difficult to not make any assumptions, there is no fixed assumption that must be made (e.g. not all our models assume time-translation invariance). In this case, we have
- the goal = finding models that are maximally general, accurate and simple
- the assumption = e.g. time-translation invariance (or something else)
What we most certainly do not have is the assumption = finding models that are maximally general, accurate and simple. To sum up, the "assumption" you mention is not an assumption, it is a goal.
Problem 2: You need to distinguish carefully between non-metaphysical assumptions and metaphysical assumption. Contrary to your definition, metaphysics is not defined as that which lies outside of science. For instance, the question "how long time period is needed for 50 % of a gram of carbon-14 to decay?" is a non-metaphysical question. On the other hand, the question "what is time?" is a typical metaphysical question. The question "does Bill Clinton like cigars?" is neither a scientific nor a metaphysical question. Not everything that is outside science is metaphysics. Furthermore, the assumptions that go into scientific models are not metaphysical, as is proved by the fact that metaphysicians have not yet reached any consensus even regarding the metaphysics of phenomena for which very accurate models already exist. For instance, quantum mechanics seen as a description of observable quantities and an algorithm for making predictions is both uncontroversial and extremely accurate. However, the metaphysicians have not managed to reach any non-trivial consensus regarding the metaphysical issues surrounding QM (which are not part of the scientific model itself). The reason is that QM does not make any metaphysical assumptions beyond the one I have already mentioned (if it did, metaphysicians would be forced to agree on much more). The same is true of the rest of science.
We have now exchanged a few posts on this topic and you are no closer to providing an argument for your claim that doing science requires that we have (i) a scientific model of intelligence, and (ii) that we make metaphysical assumptions about intelligence.
Erik
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 09. March 2003 01:23
quote:
My contention is that V = iR is thought of by physicists, and presented in textbooks, as a sort of true or ideal answer.
It is ideal because it is simple. And because it is linear. Counting is linear, too, so this is very handy. V=iR is not true for any real substance. ("true" is an ideal also.)
I said that simplicity was an ideal, so your statement about "if you were correct" makes little sense to me.
quote:
But since, in point of fact, uniformity hasn't held in the past, then I assume by modus tollens you now reject uniformity. You don't? But surely you agree that V = iR (or V = iR' ? 0.000000087) has never been found to be true in subsequent experiments. Why then do you say uniformity has been demonstrated?
Because it repeats within error. Simplicity, again. Not everything is uniform, so by modus tollens I reject that everything is uniform. Isn't that profound?
quote:
What "success of materialism" are you referring to? Since I can't think of any I'll not venture a response.
Motion of planets and the sun, nature of the earth, size of the earth, location of the earth, nature of matter, composition of life, origin of lightning and weather, etc. etc. etc..
There have been non-materialistic explanations for all of these, and none have turned out to be very accurate.
Erik wrote:
quote:
many physicists would say that the theory of general relativity is simpler than Newtonian mechanics, but not that it is easier to use . . . I think searching for models that are even simpler in this sense is something that scientists do as an end in itself.
I agree; it is fun to come up with such models. Science is art, in part, and this is a way in which it is. This isn't what makes it useful. Unless the simpler model happens to also be more accurate.
Gedanken wrote:
quote:
I would like to get back to attempting to define intelligence itself (and doing so in a scientific way) rather than continuing to debate the general nature of science.
I can suggest some necessary characteristics of intelligence.
- The actions/responses of an intelligent entity suggest that the entity has an internal model of its environment.
- Intelligent entities gracefully handle noise.
- Intelligent entities modulate their behavior to attain goal states.
- Intelligent entities can learn near-arbitrary correlations upon demand in a small number of trials.
But this is a rather unsatisfying list, at least to me personally.
IP: Logged
|
|
|
Printer-friendly view of this topic