A rectangle is a blunt instrument
A bounding box tells you roughly where an object is, but it's fundamentally a rectangle, and most objects aren't rectangular. Draw a box around a person standing with their arms slightly out and legs apart, and that box also contains the background visible between their arm and their torso, and the background visible between their legs. The box is honest about location but dishonest about shape, it always includes extra pixels that don't actually belong to the object.
For plenty of tasks that sloppiness doesn't matter. If you just need to know a person is roughly in that region of the frame, a box is fast and good enough. But some tasks need to know exactly which pixels are the object and which aren't, and no amount of squinting at a box gets you there.