If, instead of text markup, the server just sent an image of the entire UI remote-desktop style how would it be any different? You could still have a single URL that is entered, all information needed to traverse the service can be encoded in the request/responses, and no knowledge of the entire interface is needed by the client.

But we have diverged pretty far from the point. Software clients using REST APIs cannot use the self-describing nature of REST so to claim they're all using it incorrectly I don't believe is a valid criticism. Browser-like clients, and you can include hyperview in that, can use the self-describing nature because they're pushing all the understanding to the user. But that is such a niche experience (outside of HTML browsers) that it isn't even worth discussing. For hyperview, it's not consuming generic REST APIs -- it's just acting as mobile-specific browser for a single service. That's why I said it's not really interesting.