Not a professional, but studied it in college. It's mostly to either fill in gaps or loud noises.

One thing you can often do is get a "noise print" of the room, and you can isolate someone's audio basically perfectly. From there you can create a room tone and slap it under the entire track. Now if you need to mute or something you just cut the talking track and the room noise carries over.

If you don't get a good room tone, say you want to use someone looking at the camera, but the director was talking. If you try to filter out the directors voice, it's likely going to sound weird because some of the tones overlap with the room. So you mute it and slap the room tone over and you're good. They often get too much, because room tones vary ever so slightly. If you get a tiny half second sample, unless you get very lucky you'll pick up that something is repeating or sounds weird. If you have 10-20 seconds you can loop that no problem.