Yes, especially since the bulk of our content at first will be MARC, transformations from what was in MARC yesterday, today and tomorrow will be there.
Technically, the end requirement is to provide a format that is a credible replacement* for MARC(21) in an RDA context (appendix M of the RDA Test report). However, modeling the semantics of the AACR2 would seem to be a necessary endeavor along the way.
The underlying (conceptual) model of the bibliographic universe ought to be one that can be mapped to all major systems. The properties of these mappings are somewhat complicated. To use a recent subject of discussion as a simple example, if one starts with MARC-21 data that contains a non standard textual string and a coded relator, the mapping into a semantic model might not preserve the non-standard string. MARC-21 data using a standard string might map to the same semantic representation.
Such mapping would not be invertible, since the non-standard string would not have been preserved.
[I have some concerns and suggestions about some of the work that has been done and some work that has not been done under the BIBFRAME which I will explore under separate cover.]