|
|
|
|
|
by ajkjk
518 days ago
|
|
I think it's a big improvement. Stiffness is something you can picture directly, so the data -> conclusions inference "stiffness" -> "mass and short range" follows directly from the facts you know and your model of what they mean. Whereas "particles have mass" -> "short range" requires someone also telling you how the inference step (the ->) works, and you just memorize this as a fact: "somebody told me that mass implies short range". You can't do anything with that (without unpacking it into the math), and it's much harder to pattern-match to other situations, especially non-physical ones. It seems to me like the right criteria for a good model is: * there are as few non-intuitable inferences as possible, so most conclusions can be derived from a small amount of knowledge * and of course, inferences you make with your intuition should not be wrong (I suppose any time you approximate a model with a simpler one---such as the underlying math with a series of atomic notions, as in this case---you have done some simplification and now you might make wrong inferences. But a lot of the wrongness can be "controlled" with just a few more atoms. For instance "you can divide two numbers, unless the denominator is zero" is such a control: division is intuitive, but there's one special case, so you memorize the general rule plus the case, and that's still a good foundation for doing inference with) |
|