Of course they would have to introduce a fixed delay of a phrase (or multiple) and people would have to play accordingly.
So making it possible for people to jam together across such distances is an interesting challenge indeed. Such a delay makes a naive approach incredibly awkward! Good musicians can get around it with practice and using drones, for example.
However, if you're synced to a shared reference /background track, you could play in time to that. The problem is that you can't hear any of the contributions of the other players while you're playing -- the signal is too far away.
But you could imagine sort of controlling an AI mimic which would stand in for you in the ears of the other players, and your contribution would be patched in to the final mix. As a minimal step you could just represent your playing with a couple of loops (A and B pieces) and your improvisations can get auditioned and patched in for later contributions later on in the song.
Some things to think about. It's easier to imagine doing it all in the same area! And you can do in in real space, so that's a plus :D