RDF can totally be validated, with
RDF schema. Doesn't adhere to the schema, it's not valid. Oh, you mean
DTD validation? RDF exists because of places where
DTD validation isn't possible.
Furthermore, RDF != XML. You have totally misunderstood the spec. RDF is an data structure, a graph consisting of nodes with labelled edges. RDF can be serialized as XML, but it can also be serialized as p3 (a new effort to make RDF more understandable), or even as rows of (subject,verb,object) tuples.
RDF is derived from an AI/expert system project called CYC, which utilized a format called KIF. It inherited some useful concepts like reification which make frames of logic more practical. The usual example is that with RDF, a computer can know that Dracula lives in Transylvania, Dracula is a vampire, and vampires do not exist, and not come to the conclusion that things that don't exist live in Transylvania.
RDF by itself is not all that useful as it does not present an API for manipulating data. In addition, some aspects of RDF like anonymous resources could do with some more explanation. aboutEach is pretty useless. And they try to get cute with things like typed nodes, which is an XML serialization trick that doesn't change the internal model but makes RDF Descriptions look like they are plain ol' XML elements. When they're not.