> When we pack high-density information into a data table or a complex dashboard we are increasing the visual entropy of the entire system. Forcing the brain to decode intricate, non-universal shapes in a tiny 16-pixel footprint, creates a “cognitive tax” that users pay en masse every time they scan the table. What if it's an icon with a simple shape? How does that compare to noising up the table with long phrases an…
Imagine for example a set of icons that are monochrome, open-ended glyphs comprised of a single stroke with line weight similar to that of the text. This could complicate visual parsing greatly due to high visual similarity to text.
On the other hand, a 16px checkbox control with subtle gradients, shadows, and depth cues looks absolutely nothing like text and is filtered out by the brain almost automatically (unless of course the checkbox state is pertinent to the user's intent). Same goes for a 16px colorful icon with shading like used to be ubiquitous in desktop operating systems.