4 ms·

I don't know, but I also want to point out that "often proven to be predictive" is pretty weak cheese. I mean, I have a model for predicting the result of coin

I don't know, but I also want to point out that "often proven to be predictive" is pretty weak cheese. I mean, I have a model for predicting the result of coin flips which is often predictive. It's called "heads".

SwellJoe 14y ago
Are you suggesting that models are not a useful scientific tool?
- bgilroy26 14y ago
  Going from "We can be confident in our models" to "We can be confident in our planetary models" is a big step up in scale and complexity.
- ars 14y ago
  Yes, somewhat. Models are a useful tool to see what happens when the science is known. They work well to check if you understand your theory well: You model it and experiment with it at the same time. If you model matches the real world then you are good to go. But if you try to make a model based only on your observations, without also being able to experiment, then that's not science, that's tautology. Your model will do whatever you want, and you have no way of knowing if it has anything to do with the real world because you can not run multiple experiments. In some cases if you wait long enough you may randomly get enough variation in the real world to be confident in your model (for example stellar evolution). But using a model to predict something you have never seen in the real world? That's not science, that's speculation.
  - anthonyb 14y ago
    I think you're confused about how models work in a scientific sense. They can be used for prediction, and if they couldn't they wouldn't have any value at all. Of course they sometimes don't line up with the real world; why do you think there are a metric shit-ton of scientists in Antarctica? In terms of the article, the theory of climate is pretty well understood at this stage, and the models are sound at the macro scale. The hard parts are now in factoring in relatively small changes to make the numbers line up even better, and working through the implications of their predictions... which is what the article seems to be doing.
    - ars 14y ago
      > They can be used for prediction, and if they couldn't they wouldn't have any value at all. Correct - they can't be used for prediction, and therefor they have no value at all (for prediction - they are useful for validation). > Of course they sometimes don't line up with the real world Then you improve them, but you don't try to predict from them. By improving them you increase your knowledge, but you don't predict from them because the only thing that comes out of a model is what you put into it. You learn nothing about a real world from a model. You learn about your understanding of the real world from the model. (They better you know the real world the better you model - the model teach you about your own knowledge. It has zero value in teaching you about the world since nothing can come out of it that you didn't put into it.) > why do you think there are a metric shit-ton of scientists in Antarctica? To do research about the real world? Is this supposed to be a big revelation? Not sure waht you were trying to imply here. > The hard parts are now in factoring in relatively small changes to make the numbers line up even better Well obviously. The details are always the hard part - the details is also where all the interesting stuff happens. You can make the best model in the world, and then miss the one tiny detail that only happens outside the domain of your model, and all the predictions are worthless.
      - anthonyb 14y ago
        If what you are saying is true, then every scientific and mathematical formula would be worthless, since after all, you just get out of it what you put into it. e=mc^2? f=ma? v=ir? Yep, no bearing on the real world or any predictive power whatsoever.
        ars 14y ago
        You are ktizo both have the same misconception. Perhaps this explains your belief in the value of models. Those formulas are not models! They are exact mathematical representations of a phenomena! A model by it's nature can not include everything, they include everything possible of course, but the world is too complex for them to include everything, so they must estimate. If you have a feedback loop with the real world you can tune your model to make it useful, but you can never get out of it anything you did not put in, since it's impossible to include everything. If a model did include everything then of course it would work perfectly. But it's not possible to do that in the real world.
  - ktizo 14y ago
    You can't do science without speculation. Science often models things it has never seen in the real world. Sometimes, when it becomes possible to make a measurement, the model is found to be accurate. Like in the classic xkcd "Science. It Works, Bitches." cartoon about the microwave background radiation. - http://xkcd.com/54/ http://xkcd.com/54/ Models do not just do whatever you want. Many unexpected behaviours turn up in models and some of them can be almost impossible to know the future behaviour of in advance of running them, even when you know all the input states. In science, models are often what you use when the science isn't known, as you can use them as a guide to pick up on interesting things to go and look at. Engineering is usually where you use models when the science is known. [edit] And economics is where you use models when the science isn't known, and then you worship them and hope that money falls out.
    - ars 14y ago
      > You can't do science without speculation. That's not true. It's a fallacy that the first step in the scientific method is formulating a hypothesis. It's completely unnecessary. The first step is "let's see what happens". You do not need any speculation or hypothesis first. That comes later - after you have collected your data then you try to understand and predict. > Sometimes, when it becomes possible to make a measurement, the model is found to be accurate. And for more often the model is wrong. But you have no idea if it's wrong or right if you can not test the real world. > microwave background radiation That's not a model, and the fact the you think it is makes we wonder. That's an exact mathematical representation of the phenomena. A model is imprecise, it includes as many parameters as possible, but by necessity can not include everything since the world is too complex. > Models do not just do whatever you want. Many unexpected behaviours turn up in models and some of them can be almost impossible to know the future behaviour of in advance of running them, even when you know all the input states. That's called Chaos. And the interesting thing about Chaos is that tiny changes in the input (for example what decimal precision you use) cause large changes in the output. If your model is chaotic then it's utterly useless for any conclusions whatsoever because it's completely impossible for you to enter the input with the same level of precision as the real world. Chaos is fun to look at, but pretty useless for prediction. There's a second thing possible called emergent behavior. But that too can not be modeled without understanding the real world first. What you do is keep changing the model till it matches the real world, then pull out the seemingly simple rules that cause complex behavior. But the model will fail as soon as you go outside the domain it was built in. Just because something acts the same way every time in the limited circumstances you tried does not mean it will keep doing so forever. That's a common extrapolation fallacy. So again, useless for prediction since prediction by definition puts you in a circumstance you have not yet seen. > In science, models are often what you use when the science isn't known, as you can use them as a guide to pick up on interesting things to go and look at. Operative word: To go and look at. Not to draw conclusions from. Engineering uses models to validate the assumptions, Science uses models to verify understanding. In no field are (should) models be used to draw conclusions.
- a5seo 14y ago
  http://www.amazon.com/Models-Behaving-Badly-Confusing-Illusion-Reality-Disaster/dp/1439164983 http://www.amazon.com/Models-Behaving-Badly-Confusing-Illusi... "A model...[is] really much more of a metaphor, an attempt to find an analogy between something you want to understand and something to really do understand, either heuristically or by a theory." from http://www.econtalk.org/archives/2012/03/derman_on_theor.html http://www.econtalk.org/archives/2012/03/derman_on_theor.htm... - very good interview.
ktizo 14y ago
El Niño is the main thing that springs to mind when thinking about predictive powers of (short term) climate modelling. They are getting reasonably good at that. Also, your coin flip is not a predictive model any more than claiming that using the same numbers every week on the lottery is a predictive model. For one thing, it makes no attempt to model anything, and for another, it has no attempt to be predictive, it just attempts to be right nearly half the time (edge), through understanding of the likely odds, which is a different thing altogether.

SwellJoe 14y ago

Are you suggesting that models are not a useful scientific tool?

bgilroy26 14y ago
Going from "We can be confident in our models" to "We can be confident in our planetary models" is a big step up in scale and complexity.
ars 14y ago
Yes, somewhat. Models are a useful tool to see what happens when the science is known. They work well to check if you understand your theory well: You model it and experiment with it at the same time. If you model matches the real world then you are good to go. But if you try to make a model based only on your observations, without also being able to experiment, then that's not science, that's tautology. Your model will do whatever you want, and you have no way of knowing if it has anything to do with the real world because you can not run multiple experiments. In some cases if you wait long enough you may randomly get enough variation in the real world to be confident in your model (for example stellar evolution). But using a model to predict something you have never seen in the real world? That's not science, that's speculation.
- anthonyb 14y ago
  I think you're confused about how models work in a scientific sense. They can be used for prediction, and if they couldn't they wouldn't have any value at all. Of course they sometimes don't line up with the real world; why do you think there are a metric shit-ton of scientists in Antarctica? In terms of the article, the theory of climate is pretty well understood at this stage, and the models are sound at the macro scale. The hard parts are now in factoring in relatively small changes to make the numbers line up even better, and working through the implications of their predictions... which is what the article seems to be doing.
  - ars 14y ago
    > They can be used for prediction, and if they couldn't they wouldn't have any value at all. Correct - they can't be used for prediction, and therefor they have no value at all (for prediction - they are useful for validation). > Of course they sometimes don't line up with the real world Then you improve them, but you don't try to predict from them. By improving them you increase your knowledge, but you don't predict from them because the only thing that comes out of a model is what you put into it. You learn nothing about a real world from a model. You learn about your understanding of the real world from the model. (They better you know the real world the better you model - the model teach you about your own knowledge. It has zero value in teaching you about the world since nothing can come out of it that you didn't put into it.) > why do you think there are a metric shit-ton of scientists in Antarctica? To do research about the real world? Is this supposed to be a big revelation? Not sure waht you were trying to imply here. > The hard parts are now in factoring in relatively small changes to make the numbers line up even better Well obviously. The details are always the hard part - the details is also where all the interesting stuff happens. You can make the best model in the world, and then miss the one tiny detail that only happens outside the domain of your model, and all the predictions are worthless.
    - anthonyb 14y ago
      If what you are saying is true, then every scientific and mathematical formula would be worthless, since after all, you just get out of it what you put into it. e=mc^2? f=ma? v=ir? Yep, no bearing on the real world or any predictive power whatsoever.
      - ars 14y ago
        You are ktizo both have the same misconception. Perhaps this explains your belief in the value of models. Those formulas are not models! They are exact mathematical representations of a phenomena! A model by it's nature can not include everything, they include everything possible of course, but the world is too complex for them to include everything, so they must estimate. If you have a feedback loop with the real world you can tune your model to make it useful, but you can never get out of it anything you did not put in, since it's impossible to include everything. If a model did include everything then of course it would work perfectly. But it's not possible to do that in the real world.
- ktizo 14y ago
  You can't do science without speculation. Science often models things it has never seen in the real world. Sometimes, when it becomes possible to make a measurement, the model is found to be accurate. Like in the classic xkcd "Science. It Works, Bitches." cartoon about the microwave background radiation. - http://xkcd.com/54/ http://xkcd.com/54/ Models do not just do whatever you want. Many unexpected behaviours turn up in models and some of them can be almost impossible to know the future behaviour of in advance of running them, even when you know all the input states. In science, models are often what you use when the science isn't known, as you can use them as a guide to pick up on interesting things to go and look at. Engineering is usually where you use models when the science is known. [edit] And economics is where you use models when the science isn't known, and then you worship them and hope that money falls out.
  - ars 14y ago
    > You can't do science without speculation. That's not true. It's a fallacy that the first step in the scientific method is formulating a hypothesis. It's completely unnecessary. The first step is "let's see what happens". You do not need any speculation or hypothesis first. That comes later - after you have collected your data then you try to understand and predict. > Sometimes, when it becomes possible to make a measurement, the model is found to be accurate. And for more often the model is wrong. But you have no idea if it's wrong or right if you can not test the real world. > microwave background radiation That's not a model, and the fact the you think it is makes we wonder. That's an exact mathematical representation of the phenomena. A model is imprecise, it includes as many parameters as possible, but by necessity can not include everything since the world is too complex. > Models do not just do whatever you want. Many unexpected behaviours turn up in models and some of them can be almost impossible to know the future behaviour of in advance of running them, even when you know all the input states. That's called Chaos. And the interesting thing about Chaos is that tiny changes in the input (for example what decimal precision you use) cause large changes in the output. If your model is chaotic then it's utterly useless for any conclusions whatsoever because it's completely impossible for you to enter the input with the same level of precision as the real world. Chaos is fun to look at, but pretty useless for prediction. There's a second thing possible called emergent behavior. But that too can not be modeled without understanding the real world first. What you do is keep changing the model till it matches the real world, then pull out the seemingly simple rules that cause complex behavior. But the model will fail as soon as you go outside the domain it was built in. Just because something acts the same way every time in the limited circumstances you tried does not mean it will keep doing so forever. That's a common extrapolation fallacy. So again, useless for prediction since prediction by definition puts you in a circumstance you have not yet seen. > In science, models are often what you use when the science isn't known, as you can use them as a guide to pick up on interesting things to go and look at. Operative word: To go and look at. Not to draw conclusions from. Engineering uses models to validate the assumptions, Science uses models to verify understanding. In no field are (should) models be used to draw conclusions.
a5seo 14y ago
http://www.amazon.com/Models-Behaving-Badly-Confusing-Illusion-Reality-Disaster/dp/1439164983 http://www.amazon.com/Models-Behaving-Badly-Confusing-Illusi... "A model...[is] really much more of a metaphor, an attempt to find an analogy between something you want to understand and something to really do understand, either heuristically or by a theory." from http://www.econtalk.org/archives/2012/03/derman_on_theor.html http://www.econtalk.org/archives/2012/03/derman_on_theor.htm... - very good interview.

ktizo 14y ago

El Niño is the main thing that springs to mind when thinking about predictive powers of (short term) climate modelling. They are getting reasonably good at that. Also, your coin flip is not a predictive model any more than claiming that using the same numbers every week on the lottery is a predictive model. For one thing, it makes no attempt to model anything, and for another, it has no attempt to be predictive, it just attempts to be right nearly half the time (edge), through understanding of the likely odds, which is a different thing altogether.