What is a computer model?
It is a process where you program a computer to simulate what you know about reality. Or to simulate how you think something might work.
For instance, you can program a computer to simulate the behaviour of a ball rolling down a ramp. Since you likely understand this quite well, your computer simulation will be extremely accurate and the behaviour of the virtual ball will very closely model the behaviour of a real ball.
This kind of model is usually useful for either educational purposes or for predicting the behaviour of things. That is, predicting behaviour which is very well understood but which the computer can model more efficiently then you would be able to by other means.
Perhaps you could get a pen and paper, calculate what will happen and draw it out. But it is more efficient to program a computer to do this for you and tell you what might happen.
Perhaps you run a computer model of how an expensive jet engine might work. You want to try wrinkle out any huge flaws in your design before you go to the great expense of building a jet engine. This way you might be able to find out if it has any serious problems which might result in immediate and costly failure.
Or perhaps you can model how a building will perform under stresses or how an engine design will work under certain conditions. Things you probably could do by hand, but which are far more efficient to get a computer to do for you.
It might be cheaper than building the wall just to do tests that can be done with some accuracy on your computer.
Is a computer model proof of anything?
No, it is not. A computer model simulates the behaviour of something. But it is still simply a computer program. It will only do what you tell it to do and no more. So, that means that the result of a computer model can only be as good as the programming put into it.
For a computer model to emulate the behaviour of something, the programmer must program in how he thinks that something should behave in various situations. He has to make assumptions or use prior knowledge about how that thing should act in those contexts.
This is the first way in which computer models can go wrong. The programmer’s assumptions have to be correct and what they know (or think they know) about the things they are modelling have to be correct.
If the assumptions he programs into the model are wrong, then the results of the model will certainly be less than completely reliable.
But, even if all of their assumptions are correct, the results have to match reality. The program has to be able to produce the correct result. And the programmer must show that this is indeed the right result for it to be of any use.
Meaning that you already know the result. Or can at least check that the results you get are consistent with what you know to be true.
Surely if all the assumptions are correct, the outcome will match reality?
Maybe not. Maybe the model does not account for all of the significant factors in a real-life scenario. This is another way in which computer models can be unreliable; they are prone to over-simplification and to ignoring key factors.
But as we have seen, this is the main issue with computer models; they prove nothing about how the things they model actually work. Yes, they are useful for helping to identify potential areas of research. But their results do not themselves prove anything. They are not replacements for research and experiments to learn about how reality works.
They are nothing more than automated thought experiments and prove no more than would a thought experiment. If you conduct a thought experiment in your mind, it proves nothing about external reality. If you want to learn about how reality works, you have to study nature by performing experiments and collecting data.
The computer will probably get more reliable results than your unaided brain will, but it is still just a very detailed thought experiment.
So, does that mean that we should ignore computer models and conclude nothing from them? No.
Computer models can be very useful.
In as far as they indicate potential outcomes that can be shown to closely match real-world results, then they are certainly useful in predicting what might happen under those circumstances. They can help identify unexpected results that perhaps should be accounted for.
For instance, suppose our model of the jet engine does blow up. Well, we should ask ourselves why it did that. Perhaps we look into our model and find out that putting this piece here causes the entire thing to be unstable.
Well, that is something worth looking at. We look into that and we find out that, yes, that piece there would make the whole thing unstable. And therefore we should not put that there.
Or, take models that purport to represent bacterial growth over time. We might look at that model and predict interesting changes in bacteria populations and how they might adapt to their environment. Thus helping us identify real-world patterns and facts.
Or finding planets using computer models of gravitational lensing for the purposes of finding planets (something my wife, Ashna Sharan used in her Masters of Astrophysics thesis).
This is why computer models are far from useless. In as far as they help identify facts of reality, then they are certainly useful. However, they are no replacement for experiments and other forms of analyzing what actually happens in physical reality. As opposed to the thought experiment inside the computer.
Experiments with physical objects can make for great science. Watching the Matrix? Not so much …