[fuoqi]

It's a weird article. It's a matter of priorities, a theory can be as beautiful as anything you have seen, but if it does not have any predictive power, then for "hard" science (i.e. physics, chemistry, biology, etc.) its value is almost zero. "Almost" because interesting ideas sometimes can be developed in unexpected directions and even in different fields.

And "beauty" and "elegance" are certainly important, they just come after an ability to predict. Even if this ability is yet unclear, scientific community is open to investigation of promising "elegant" ideas (e.g. see history of the string theory). In the worst case scenario we get an interesting branch of mathematics.

I prefer to look at it at from the following angle: at extreme we have simply collection of experimental records. By accumulating those records we describe reality in the most straightforward way. "Laws" are the way to compress those observations. As an input of this algorithm we have experimental conditions, and as output experimental results. As a byproduct of this compression we get an ability to predict with a certain confidence what will happen in future experiments. "Beauty", "elegance", "compactness", "unity" are subjective ways to measure quality of this compression. Quality can be measured using different metrics: size of the algorithm (i.e. number of laws and constants in them), information loss (i.e. its precision), computational efficiency (i.e. an ability to compute prediction). This way we can view Newton laws as an efficient lossy algorithm, while quantum mechanics are near lossless, but more complex and computationally difficult.