The problem is that the granting of permission isn't strictly tried to that particular feature. So if the app asks you for permission A for "reasonable feature 1" it can then use that permission in the background for "unreasonable creepy feature 2" since you've now granted it unilateral permission.
Sure, like I said there is no way for me to know that after they asked for permission to use GPS when I engaged the "Run" function the first time, they won't spy on my location every time I just use the app (apart from the little GPS activity icon perhaps).
But the context of me switching to the "running" feature and then I'm asked if I want to allow the GPS function is enough for me to not feel watched. It's all about context.
Same if the app asked for permission to see your photos (forever) when you use the custom cover art feature.
For a malicious app this makes no difference, but for those apps we (sort of) trust, the context makes all the difference.
I don't have a solution for location, but for photos, it seems like it would be best to require no permission to present a standard control (i.e. UIImagePickerController), but with the API only actually providing any information when the control completes (photo is selected, or control is cancelled). The user selecting the photo is implicit permission.
Permission would be necessary to actually query for media programmatically, which is the real issue. Currently, they are conflated.
For location, I suppose that the "Google Maps is using your location" bar at the top could be expanded and always present when the location permission is enabled. Tapping it would revoke the permission until it is enabled again. This could get annoying though, some apps legitimately do depend on location all the time.
