Study predicts imminent irreversible planetary collapse

> 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.

> 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.

It is true, you cannot do science without speculation. And it doesn't matter if you speculate before or after gathering a particular set of data, speculation is still required.

I didn't say that you shouldn't test your model, was just pointing out that you can model something based on an incomplete understanding and it can still be considered part of the scientific process before you have gone and checked your results. It can even be science even when there is currently no known way of checking the results, as long as you can point to a reasonable path that might lead you there.

Chaotic models are used in prediction. Regularly. Computational fluid dynamics wouldn't exist as a discipline without this.

And I never claimed that you should draw your final conclusions from a model, just that they can be a fantastically useful tool for exploring the unknown.

They can also be a trap if you trust the model more than the data, but that isn't the fault of modelling.

Those formulas are models. They describe the world and can predict it's behaviour, but can also break down under certain conditions. For example, what happens when I put 240V across a regular 20Ohm resistor? According to V=IR you get 12A running through it - but I doubt you'll get that for very long.

By your argument, V=IR must then be a "model" and is useless, with no predictive power, etc, etc. In practice, the only difference between those formulae and a more complex model is one of degree.

Congratulations, you've stated the theory with far more specificity and a greater degree of testability than the paper's authors managed.