Amen

Newton's laws don't describe a mechanism, however. They allow you to predict the motions of bodies, but as for what makes the bodies move that way is not addressed by Newton. He made that point himself on a few occasions. F=ma and all that allows us to relate how bodies move relative to one another, but it is not a mechanism by which bodies move.

That's why I wondered when science says it has provided an "underlying mechanism" hasn't it provided another set of correlations?

Neither. Kepler's laws certainly describe more than a "correlation'. A correlation would be measured in terms of a correlation coefficient, and applying that to planetary orbits would tell you that there is likely some structure to look for. Kepler's laws tell you what that structure is, mathematically. Newton's laws give you a reason why that structure is there, and tells you where else to look for it-- i.e., Newton's laws make predictions. This is all very basic to how science works, and is well known to any student of scientific history.

While it certainly is true that one might foster a false impression by using the word "mechanism", I'm pretty sure that those using that word just mean "unifying principles that explain why Kepler's laws work and how they will be modified in situations other than where they were first noticed". That is kind of what physicists mean by "mechanism", it is not intended to be taken literally (pulleys and levers and so forth).

[quote=Ken G;1295748]Kepler's laws are quite literally a curve fit to Tycho Brahe's empirical data of planetary motion. I think that is, in popular vernacular. a correlation. It may not be expressed in terms of the square of the cosine of an angle between the deviations of a curve from its mean and the wiggles in the data in some Hilbert space (i.e. r squared), but it is a correlation.

Newton's laws are as much a mechanism as pulleys and ropes. He supplies an explanation as to how a force causes the motion of a mass to change, F = ma. He provides a quantitative description of how that force arises from mass and geometry, F = G*M*m/r^2, and he provided the methods (calculus) to solve the resulting differential equations which provide the geometrical description of the orbits. With pulleys and ropes you still need to describe the notions of tension, why ropes can withstand tension, how that tension in the ropes, combined with the geometry of the pulleys results in a net force applied to a load, etc. Moreover Newton's laws provide an explanation for a great deal more than planetary orbits. They explain the motion of falling objects, the result of elastic collisions, classical kinetic theory of gasses, conservation of energy, conservation of momentum, .....

Newton's laws do not explain how many angels (or philosophers) can dance on the head of a pin or why angels (no one could ever hope to understand the motivation of philosophers) would want to do that. But any mechanistic explanation of anything must start with some undefined concepts. Newton's laws start with fewer such notions than most explanations.

Ahh, let's not worry about angels or philosophers, but on identifying that which is lacking from a team of data miners who successfully predict corn crop yields. Correlation has been pitted against causation or underlying mechanism. I purposely equated the two to motivate others to draw the distinction between them.

So, Newton provided the mechanism underlying Kepler. The word causation or the phrase underlying mechanism can suggest that Newton detailed the physical mechanism or cause of the motion we observe. He himself denied doing so, leaving the cause of gravity unexplained. Others expected to see a mechanical explanation from Newton, something along the lines of Descartes' vortexes. Since Newton provided no such mechanism, some charged him with re-introducing the occult back into physics. Gravity appeared to be an underlying, hidden, or behind-the-scenes mechanism that had the unusual property of action at a distance. Newton denied that and suggested that his theory works regardless of the underlying cause. He attributed its truth to the fact it was derived from experimentation—observing the behavior of pendulums, planets, and so on.

Both Ken and yourself nicely describe science indeed providing a mechanism, however, it apparently is not the mechanism underlying gravity (in the sense that an engine might be said to be part of the mechanism underlying a car's operation), but more like the procedure a scientists follows to successfully make predictions, etc. The active scientist himself seems to be the mechanism here. Or, if a modern scientist finds manual calculation tedious, the mechanism might be said to consist of his calculator or computer programmed with the appropriate algorithm. But that puts him in the same position as the data miners. Both have machines that follow a structured sequence to make successful predictions.

[quote=Joe Durnavich;1295968]...

Quote:

Correlation has been pitted against causation or underlying mechanism. I purposely equated the two to motivate others to draw the distinction between them.

Correlation and causation are quite different. Causal relationships result in 100% correlation. Correlation may have nothing to do with causation. Here is an example of poor reasoning based on confusing the two:

Old people tend to have more money than do young people.
Old people tend to die more frequently than do young people.
Therefore, wealth will kill you.

Equating two disparate notions simply to provoke a discussion is an indication of a desire for discussion rather than a desire for a conclusion. If the objective is simply discussion, then this is likely to be pointless. If it appears to be pointless, I will not pursue it further.

I like to draw a distinction between an explanation of HOW something works as opposed to WHY it works. Science explains how things work. Theology tries to explain why things work. It is generally a bad idea to mix science and theology. Newton's theory does a very fine job of explaining the how.


The active scientist is most certainly not the mechanism. If I step off of a cliff I am going to get hurt, whether or not there is a physicist in attendance.

The calculation has nothing to do with the operation of the phenomena. I fall off the cliff whether or not anyone has calculated the trajectory that ensues.

Not at all. Data mining does not follow a clear principle. It is unlikely to duplicate results accurately given slightly different input data. Nor can one apply logic to that procedure to derive further general consequences. Data mining works purely on correlation, with no underlying principle that could imply causation. Correlation is very dependent on the specific input data set. Physical principles are not.

To be fair, this does show correlation failing, but it does not directly address causation. What would be an example of causation? Perhaps: Arterial plaque buildup in the elderly restricts the arteries increasing the chance of a blockage leading to heart attack and possibly, death. My point is, although it is more of a question, is not whatever we identify as the “cause” going to be itself a correlation or comprised of them? (I use the term “correlation” in a looser, non-formal sense. By it I mean observing behavior and noting relationships but not necessarily in terms of an “underlying mechanism.”)

Very good. I was going to write above that “causation” or “underlying mechanism” is treated almost as a holy phenomena and that scientists were then like priests who have special powers of insight to an occult or hidden world of agents that makes the world behave as it does. I contend that success in life consists of observing behavior and making the best use of it. When you do get around to opening the car's hood to study its inner workings, you are observing more behavior. I thought that disqualifying Vulgar correlation in favor of Sacred causation risked throwing the baby out with the bathwater. What we don't know does not disqualify what we do know and use to our benefit.

What is the mechanism then, if not the active scientist observing a comet's positions and calculating future positions? Kepler's and Newton's laws allow prediction, but are they the motive force behind objects in motion? Does a planet solve Kepler's equations as it moves?

To be clear, nobody is suggesting any sort of subjectivism here. By “active scientist” I mean a person embedded in the world, interacting with it, where achievement depends on what is done as well as the way the world behaves. A person jumping off a cliff doesn't need a physicist to get hurt. The human body and the cliff have that covered. We do need a physicist if we want to arrange our actions around the falling subject. Perhaps we are having him test a parachute for us.

Consider Newton in the act of developing his physical principles: He times a pendulums, observes planetary and cometary motions, develops mathematical procedures to predict the motions he observed, notes discrepancies, goes back to make further observations, revises his equations, checks for discrepancies, and so on until an adequate theory is in hand. His theory is dependent on the input data by virtue of the fact it was repeatedly revised until it could reliably predict such data. True, the corn-crop predictors may not be able to predict anything else, but Newton cannot predict corn crop yields. Finite application scope is common here.

You're not 'accusing me of anything wrong' -- you're just labelling what I say 'folk psychology'.

Frankly, your broad sweeping handwavings on this topic sound rather a lot like what I would describe as folk philosophy. The kind of empty verbosity that misses the mark of the essential by a very long shot, and ends up giving the whole discipline of philosophy an undeserved bad name.

I do think it's telling. It's telling because it shows that we have some experience, however limited, of how science actually works. We know it's not the neat, quiet, ever-widening process of error-screening and elimination that you seem to be describing. In real science, error is often unpredictable, unidentifiable, and impossible to eliminate.
You speak of failure as though it were a shameful thing, to be hidden or rationalised away. But error is as much a part of science as understanding. Saying that science is composed of models is a way of acknowledging this fact.

Science is not about simplifying the data. It's about using the data as a springboard to learn more than what the data contains in itself.

That's simply not true. What matters in science -- what makes science science, I'd say -- is not to group the data you already have in various ways, but to extrapolate from the data you have into the unknown. This is what scientists do every day, in every science.

That's an excellent example, that shows very clearly how science is not about summarising data, but rather about extrapolating from known data into the unknown. Newton experimented with pendulums, but he couldn't do the same kinds of experiments with the planets. Yet when he understood the principle of gravitation for small objects, he immediately generalised it to all bodies in the universe, even though he had not checked, and could not have checked, that gravitation applied as well to the Moon and the Sun and the planets and the stars as it did to pendulums and apples.

And that wider context is precisely what I am calling a model. Is this truly a hard notion to grasp for you, or are you simply hostile to it?

__________________
"A witty saying proves nothing" Voltaire.
"All your bias are belong to us" Ara Pacis.

http://www.nature.com/news/2008/0808...tml?s=news_rss

In a completely on topic post (according to the OP),

__________________
smile, and the Universe smiles with you

What I don't understand is the entire premise of the exercise. What do they even mean by a "solution" to what will happen if you remove a prey? Obviously the answer to that is many things can happen and any particular outcome will be randomly chosen over a wide range of possibilities. The goal, then, should be to associate a probability with each outcome. How do "genetic algorithms" do that? And if one intends instead to apply a "search procedure", then is not the data being searched of equal importance to the procedure? What is the dataset they are talking about searching? I don't know, I can't tell exactly what they are saying but my "silly sensors" are beeping.

Here's a recent post in another forum that supports the idea that science uses models, and what I wrote previously about there always being some amount of error. I would emphasise the following, especially:

__________________
"A witty saying proves nothing" Voltaire.
"All your bias are belong to us" Ara Pacis.

I don't have a clue. But I am familiar with the power of the Google. It is a brave new world.

__________________
smile, and the Universe smiles with you

As I said, folk psychology has its advantages. Modern philosophy appreciates the strategies we employ to get things done. Metaphors work for a reason. That's why I said at the outset that I am not disagreeing with you. The notion of "scientific model" works whether it is like the plastic model of the solar system or not. The analogy to models captures some of the aspects of evolving theories such as being in error to a greater or lesser degree. Some philosophers argue that folk theory should qualify as a legitimate theory. I can see their point.

To be fair, you could have used that paragraph to support your notion of model. It may be true that I just don't get it. But in that case, merely dismissing my views will not help me see the model that you see.

Well said.

I don't mean to imply that science is neat. We latch onto any technique that works. We shift paradigms when a new framework proves more successful than the old.

I don't speak of failure as a shameful thing. I meant that failure most readily prompts the analogy to models. The thinking is along the lines of "there must be a model because whatever we say turns out to be wrong to some degree."

My preferred metaphor for science (or any field of knowledge) is as a dynamic feedback loop that involves the person and the environment. Error feeds back in as a correcting influence. It's like driving and hearing the rumble strips. You nudge the steering wheel and stay on course. The fact that engineers put steering wheels on cars shows that success and failure go hand in hand. It is expected at the outset that there will be error. So we learn to use it to our advantage, to let it guide us on course.

I didn't say that science is about simplifying the data. I spoke of science simplifying as acting economically, of making the most of the data at hand.

I never said that we group the data we already have in various ways. What I said was, "It discovers and teaches techniques that can be applied to many particulars, perhaps an antibiotic that can treat several types of infection. In that sense, science is of the particular, just groups of them." In that example, science is extrapolating a technique to cure multiple types of infection.

I never said that science is about summarizing data. I tend to consider science in the wider context of human achievement. Science is about achieving, about making life better.

I agree with what you are driving at here. Although, Newton did acquire planetary observations from astronomers and kept revising his techniques (that "error signal feedback" analogy) until an adequate theory was in hand. But you can see what I mean by "acting economically." By studying and mastering pendulums and planets, he found techniques that apply much, much more widely.

I wish no ill will on anyone's notion of model. Len expressed angst over whether science's models can only asymptotically approach an inaccessible reality. I wanted to point out the metaphor inherent in that and have him reconsider if the problem he thinks exists really is the case.

I tried to clarify my position on models here in this post:

Scientific Method Obsolete?

The quote describes several possible techniques for calculating features of black holes. What are the models here and in what sense are they models? (E.g., are they like the plastic solar system model?)

This excerpt from the post:

"But note that they do not actually measure rotation; rather, they measure the width of the broad iron k-alpha line, and fit the observations to a model where the line is broad because it is seen in reflection off of a portion of the accretion disk that is affected by frame dragging. That's how they derive a high confidence level for rapid rotation."

desribes treating an accretion disk as if there were a reflection. I'm surely oversimplifying, but here we see a case of gainfully leveraging techniques learned from reflecting objects to accretion disk rotation. In that sense, reflecting objects are used as a model of aspects of the accretion disk.

It seems we have reached a lot of common ground. If you say that the model-reality dichotomy is just one perspective among others about science, I can certainly agree.

I like that perspective, too.

Touché.

To be honest, what we each say seems to have been converging in our latest posts, and that last post of mine did come off more flippant than I would have preferred. I apologise for this.

But when you extrapolate you are moving from the particular to the less-particular, and often to the general. This is why I say that science is about making generalisations. Furthermore, since we cannot access the general completely (say, measure the amount of dark matter in each galaxy of the universe), there will always be uncertainty in what we know about the parts of the world we cannot measure directly. This is why I claim that we can never be sure that our most general descriptions of reality overlap perfectly with reality, and why I insist that they should be viewed as models of reality, not necessarily equal in every respect to reality itself.

To give you a more concrete example, here's a recent thread about whether the universe is finite or infinite. This is a query that pops up frequently in the forum, and whose answer, from what I understand, remains an open question in astrophysics. In other words, we have currently not enough data to decide what should be the answer. One day we may find data that points towards one definite answer, but in the thread Ken G made an excellent point. Paraphrasing him a bit, since we are only human and finite, we will only ever be able to observe finite parts of the universe. Suppose we adhere to a strictly empiricist standard which requires us to believe only that which we can measure (however indirectly). Then even given evidence that points to an infinite universe, it's quite possible that we will never be able to distinguish an actually infinite universe from one that is finite but behaves as though it were infinite within our horizon of sight, and stops shortly outside that horizon. Neither finite nor infinite descriptions of the universe can be more than extrapolations from the part of the universe that is visible/measureable to us. The latter is necessarily finite, regardless of whether the entire universe is finite or not.

The models would be the Schwarzschild solution (non-charged, non-rotating black hole) and the Kerr solution (non-charged but rotating black hole). Either may be the correct model, we're not sure, although physicists seem pretty convinced that, since everything else in the universe rotates, black holes must rotate too, which would make the Kerr model the best. But for many kinds of questions the Schwarzschild solution is close enough to reality to provide an accurate answer. On the other hand, we know that even the Kerr solution is likely not 100% accurate, because it's derived from classical GR, and thus does not take into account the levels where quantum phenomena become relevant.

__________________
"A witty saying proves nothing" Voltaire.
"All your bias are belong to us" Ara Pacis.

When the correlations observed from data mining are strong we get speculation about underlying causal principles, as in astronomy, climate science and any discipline where observations are ordered to try to find their causes. This movement from observed correlations towards general causal principles and laws is a main theme of the advance of science.

That supports the use of data mining as a tool in discovery. I see nothing wrong with that. But data mining and statistical correlation are not the end product of good science.

The move from Tycho Brahe's raw data to the curve fits of Kepler was a significant step too. But the real crown jewel was the development of mechanics by Newton to explain Kepler's laws from a few profound principles.

Yes, no one is debating that honing data mining capabilities might be a useful tool for guiding future developments in science. However, those future developments are still going to look a lot like the scientific method as it has been applied in the past, and the goal is still going to be unification and understanding that no correlation coefficient is going to provide by itself. To me, asking if data mining can replace the need for the scientific method is like asking if the invention of eyeglasses replaces the need for eyesight.

But finding correlations is the beginning of good science.Tycho's data contained the significant correlations observed by Kepler, who in a way was Newton's miner. Newton's understanding was built on hypotheses derived from the strong correlation coefficients of Kepler's ellipses, fitted to Tycho's data. Inductive data mining - including use of statistical correlation - provides the material for deductive reasoning. Scientific method is a combination of the two - applying deductive reason to the material of inductive observation.

No debate there. The OP was suggesting that more powerful correlation-finding capabilities somehow replaces the scientific method, rather than simply enhancing it.

And always has been, exactly. The new capabilities just make us better at it.

__________________
Logic is the grammar of truth.

Meaning and absolute certainty are incompatible, and profound meaning and absolute certainty are profoundly incompatible.

The only thing intelligence is capable of is recognizing itself.

Philosophically speaking, the quote in the OP is utterly ridiculous.

Science, at its base, consists of testing testable propositions.

Finding a correlation and testing it are entirely within the realm of science. One place that the article falls down flat (i.e. is fatally flawed) is where it says that finding a correlation is enough. No, first you find one, and then you test it. Then you analyze it for cofactors, etc. Arguing "it is so, we don't need to know why" is a very familiar statement, but not one usually found in science.

Now, curiously, this still does not mean "correlation means causation", because the quote, again, does not address cofactors at all.

One key to the fallacious nature of the article is in the first paragraph, wherein it states "don't have to settle for models at all", when they are in fact discussing a statistical model. The author does not understand what a model is or can be. I know nothing of the author, but in fact the author makes a basic mistake when he says "... no model" and then proceeds to talk about statistical modelling.

Then, we see a completely mistaken summary of why "correlation does not imply causation", a statement that does not include at any level the idea of confounding factors, or cofactors, what have you. It is ironic that the article makes this mistake when in fact a huge advantage of huge data sets for analysis is that one CAN analyze cofactors, confounding factors, etc, to the limits of resolution of the data set, yet this power of huge data sets is ignored in what appears to be a rush to dismiss science on a completely mistaken basis that shows, again, that the author knows, or perhaps writes as though he knows, absolutely nothing about science at all. Rather than issues of of cofactors and confounding factors, the author tells us that the problem with statistical analysis is statistical confidence, which of course is patently ridiculous, as the statistical analysis hands the confidence to you on a platter.

From there we go to straw men. "massive amounts of data ... replace every other tool ...", for instance, is preposterous, analyzing massive amounts of data is often the start of understanding, and the only difference between Google and Ogg the Caveman discovering fire is that Ogg didn't have to process as much information to realize "ouch, that hurt". The question is purely one of scale.

The article proceeds from rhetorical fallacy, through logical fallacy, to the statement "Correlation is enough", which, of course, it isn't.

Yes, huge computers will be very useful in finding patterns, and in determining the actual correlations between various things, no doubt. This does not deny science, it enables it.

And with that, the article is completely, tragically, and horridly mistaken.

It is wrong.

Ditto that whole post.

In defense of the Wired article (since everyone seems to be dismissing it at the moment), perhaps the author did not mean to claim that all science everywhere is 'obsolete', but only in some areas of research, like those in the examples he gives. And the word 'obsolete' itself may have been a bit of hyperbole. His basic point, which I take to be that 'brute force' statistical methods can get reliable results where more theory-laden methods are less successful, is not without interest.

Having said this, I must also second what jj_0001 just wrote. It's a mistake to think that mining for correlations frees one from theoretical assumptions. A correlation is a model, too.

__________________
"A witty saying proves nothing" Voltaire.
"All your bias are belong to us" Ara Pacis.

First off, I posted an example of how somebody actually used the method discussed in the article. I didn't diss anything.

Second, if you actually read the article, it is clear that what is being discussed is not doing a Google search, but using massive super computers to search impossible amounts of data, with intelligent algorithms, which he pointed out has already occurred.

In regards to models and data, the power of computers, and the massive amounts of data that can be stored and looked at by computers, has indeed changed the way we think of science. Especially considering models.

The high speed new Network being built because of CERN is a good example. Computers are going to be gathering and sending vast amounts of data, to lots of computers. You don't need a model or a theory when you can observe and analyze what actually is.

Or, if you know how things work, a computer can actually do the work of creating what is, based on how things work.

We already do this with virtual wind tunnels, stress factors, and many other engineering marvels, where you don't have just a model, you have a virtual object, and can see what would happen.

__________________
smile, and the Universe smiles with you

Heh, where did I ever say you had "dissed" the article? Where did I ever even address you? And what's with the chip on the shoulder?

__________________
"A witty saying proves nothing" Voltaire.
"All your bias are belong to us" Ara Pacis.

That's not a chip on my shoulder. Take a closer look.

__________________
smile, and the Universe smiles with you

I see now that you posted a reply on August the 8th. This is the 13th. The 8th doesn't fall into my personal definition of 'at the moment' anymore.

__________________
"A witty saying proves nothing" Voltaire.
"All your bias are belong to us" Ara Pacis.

In other words, science changes. That's one of the things that sets science apart from belief systems.
***cough*** They're building a giant statistical model. Yes, it's mega++supercalifragilistic-ginormus.

It's a model. It's a simple model, but that's not a bad thing, necessarily.
No, it's measuring what is, not creating it. This is not to say that measuring it isn't a very good thing, of course, because it is. It's measuring "what is", not creating it.
Uh. You certainly have a mathematical model in (virtual) wind tunnels, stress analysis, etc. You very much most certainly do.

Well then, all is forgiven.

__________________
smile, and the Universe smiles with you

Like I said, I don't know. But Nature printed it, not me. Ask the author.

__________________
smile, and the Universe smiles with you