My understanding of the vision is that when all your responses all described using (fielding original) REST API's via RDF, using URI identifiers everywhere -- then a client that has never seen a particular server can still automatically figure out useful things to do with it (per the end-user's commands, expressed to the software in configuration or execution by some UI), solely by understanding enough of the identifiers in use.
You wouldn't need to write new software for each new API or server, even novel servers doing novel things would re-use a lot of the same identifiers, just mixing and matching them in different ways, and the client could still "automatically" make use of them.
I... don't think it's worked out very well. As far as actually getting to that envisioned scenario. I don't think "semantic web" technology is likely to. I am not a fan.
But I think "semantic web" and RDF is where you get when you try to take HATEOS/REST and deal with what you're describing: What do we need to standardize/formalize so the client can know something about the meaning of the response and be able to do something with it other than show it to a human, even for a novel interface? RDF, ostensibly.
The fielding/HATEOS REST, and the origial vision of RDF/semantic web... are deeply aligned technologies, part of the same ideology or vision.