The idea is that you can actually query stuff inside a torrent; like actually perform an SQL query and only the pieces relevant to the query will be downloaded.
This sort of allows for distributed querying if you think about it, since your query could be satisfied by many different peers that are seeding/leeching.
Imagine a Web built this way. Where sites are served by people that visit them and are not just static sites, but fully queryable as you'd expect in the normal HTTP web.
I wrote more on this here: https://medium.com/@lmatteis/torrentnet-bd4f6dab15e4