Semantic UI

Approaches

The general approach is to describe both input and output in a way that leaves some freedom to implementations – that is, focusing on semantics, or on the model, possibly providing some hints for concrete interfaces. Examples of those, as well as of related or potentially useful for it technologies, are below.

HTML: While it's a rather bad example in how it is used in practice, and rather restrictive by design, HTML tries to rely on semantic markup, can be accompanied by RDFa, and CSS can be seen as a representation recommendation. And it's the most widely used one.
XMPP Data Forms
XForms: Form descriptions targeting thin clients.
Fresnel: An RDF display vocabulary, which provides hints on how to present RDF data.
Semantic forms: RDF-based form generators.
Plain RDF
Other combinations of XML, XSLT, RDF, CSS: Generic RDF browsers and editors can be used as well, XML with associated XSLT can suggest representations.
SCXML: A generic FSM description language, apparently intended for voice interfaces. Seems handy, though those can be embedded into types/propositions as well (e.g., Idris ST).
Declarative GUI libraries: There is a variety of libraries that try to achieve usable declarative GUIs. A recent paper on the topic is Declarative GUIs: Simple, Consistent, and Verified, which also focuses on state machines and relies on dependent types, and there's more in the "Related Work" section. Being libraries, usually they are language-dependent, but they still seem nicer than more common approaches. Common GUI libraries, such as GTK, also support XML-based UI definitions (e.g., GtkBuilder).
swagger2 with Swagger UI (and perhaps servant-swagger)
Coalpit
Python Fire: Haskell (and Python; didn't check it, but looks similar) packages generating interfaces out of types: JSON with OpenAPI in case of Swagger, command-line options and DSV with usage strings in case of Coalpit. These allow to easily expose complex structures (and functions operating on those) to other programs and users, and can potentially be extended to use RDF, presentation hints, and so on, and would get considerably more descriptive with dependent types.
My toy/prototype project: Doesn't even have a proper name, but still serves as an example: semantic and textual UI descriptions are read as input and rendered, user actions produce output.
DBMS interfaces: While most of the web applications seem to be little more than eye-candy and restrictive interfaces to databases, and there are generic interfaces to those, perhaps this is worth mentioning. Those are primarily about storage and retrieval, but business logic can be implemented in a DBMS as well.
Sequent calculus
Other proof calculi: A handy way to deal with propositions in general, and may be useful for zooming user interfaces.

The mentioned projects can be grouped into those focusing on fuzzy description logic, on intuitionistic logic, and the less generic, extensible, and/or semantic ones, which don't focus on a formal logic. Unsurprisingly, propositions are helpful to carry semantics.

An interface outline

Aiming exchange of propositions in different kinds of logic, perhaps UI should mostly deal with sequent calculus, while the program it communicates with would handle inference rules, read user-composed sequents, compose new ones. A loose analogy can be made with sequent calculus corresponding to a terminal, logic – to a shell, program – to streams of propositions (though probably in most cases it would be more handy to define custom inference rules in a program itself, perhaps plugging some kind of logic as a library), escape sequences – to rendering hints.

There are nice opportunities for UI zomming with sequent calculus or other proof calculi, and the system may be rather simple/minimal and generic with it. But a bit of experimentation is needed to see how usable it would be, and it may be just a fun thing to play with.