I am not sure about a practical purpose, but I am completely fascinated by the fact that our senses are curtailed and there is a lot more happening around us that we can detect.
What happen is that most people lose it because they don't use.
When I was younger (about 8 years old) I could do it very easily, back then I had no idea how I did it, or how it worked, I only "knew" stuff happening around me, and I thought it was something supernatural (since most people do not had the same feelings I had).
After a while I stopped caring about it, got busy with other things, and it faded away, then I learned that it was echolocation.
I want to get it back, but I don't have time to retrain myself... :(
But seriously, it was very nice, it made pratically impossible to sneak on me, anyone moving near me while there are lots of sound, like in the middle of a busy city or school, I would "feel" (it is very hard to describe) "movement" behind me, and would look (startling whoever was sneaking).
Oh wow, I'm stupid. In tiredness last night I meant to type electrolocation [1] but ended up typing echolocation. Sorry about that.
You're completely right that we can develop echolocation quite functionally if we throw ourselves at it, as some blind people have demonstrated readily. The best examples are probably Daniel Kish [2] as well as Ben Underwood, who famously was that kid shown rollerblading down the street using it.
Suffice to say, had I actually typed electrolocation I'd look less stupid right now :) It's why I mentioned subterranean animals, e.g.: monotremes such as the echidna and platypus, as well as underwater creatures which tend to hunt in mud and silt. In both cases, echolocation isn't really effective since dirt/mud doesn't tend to let sound created by an animal penetrate too well.
Now, if you guys have examples of where humans have shown electrolocation, I'd certainly like to read about that!
However if I recall correctly, whether this is actually innate ability in humans, of something that happens when you throw large neural network on a problem is still an open question. Seeing what even 100 neurons in artificial nerual network can learn, I'd guess the latter, but that's just a guess.