Eg if you have 20 GPS traces in an area and they all turn at one point, that’s a good place to turn. Or you can assume something has a sidewalk if many people have walked there?
There's still a quality difference between a well-worn path and sidewalk. It can be a great way to find places to build new sidewalks. (There's the classic story of the University that didn't pave sidewalks in its quad until well worn paths in the grass were visible, using essentially crowd judgement/"ant hill optimization".)
You can reasonably assume that a place frequented by road bikers is an acceptable surface for electric scooters, or wheelchairs (the surface only, not necessarily an appropriate setting).
A well worn path will show many walkers and runners, but almost no road bikers.
I'm sure there's all sorts of accurate inferences that can be made from the data.