Almost impossible to explain without throwing in lots of maths and image processing/enhancement jargons

. The first time I saw the default matrix numbers of this filter, it immediately reminds me of a "Mexican hat shape " filter used in Pattern Recognition..kinda convolution process that I used years ago in my research but now it gives me headaches to think about it

. You can blur the image by using a "flat top" filter: matrix elements of equal weighting. You can sharpen the image with a "step" shape function....etc. The scale parameter is kinda normalised scaling factor of the filter size.
Anyway if you play with the scale, magnitude or invert the sign of these numbers, or adding extra elements, you will get all sorts of not-so-subtle unnatural effects.