I think we've got a little mixture of concerns going on here.
I'd expect one can describe some form of (2D or more?) continuum from tuples to DCS/COBRA and if you did so, REST would be sitting far closer to the simple tuple origin than the other. However, REST is an architecture for data exchange/manipulation, tuples are a formal mathematical model. Even at the REST resource level I'd be hard pressed to apply the tuple description to the resources unless you want to regard them as purely opaque blobs? I actually like that view of the REST resources (since, as you point out it allows for loose coupling), but it denies the reality that the utility of REST comes from the fact that the parties do in fact have expectations as to the contents of the resources and the HTTP headers and MIME types shape those expectations.
If I claim to be delivering you XML and you actually get JSON the loose coupling is now a bug, not a feature. (Though SImon seems to imply I should just go ahead and figure out what parser to use and cope with it anyway... ;-)