In metal type, the matrices were designed in different weights, sizes and styles. Each single matrice was designed and cut individually to look right at the intended size (e.g. by thinning or thickening lines). It was not simply a scaled version of a “mother font”: a 24 pt letter scaled down to half its size would not look like the 12 pt letter of that type (and it would probably look a little malformed). Tim Ahrens and Shoko Mugikura write about this.
Leap in time to the digital era: I understand that there are formats which store information about a letter as a “shape of pixels” (bitmap, which makes letters blurry in too large sizes), and there are formats which store that information as a vector function (OpenType, which makes the font scalable).
When writing on a computer, it is possible to choose between a plethora of sizes (even fractions like 10.5 pt). I assume that these are scaled sizes, which means that if you took a print of a 24 pt letter and scaled it down to 12 pt, it would look exactly like the letter that was originally printed in 12 pt.
If they work that way, neither of these two formats take into account what I described in the first paragraph.
Does modern digital typography address this, and if so, how?