Aren't the semantics the same?
No for a couple of reasons.
Is <scientificName>mus musculus</scientificName> the same as <scientificName>Mus musculus</scientificName> ?
<Occurrence>
</Occurrence>
This is not the case with the integer id, or a string literal
Also triple and quadstores can handle URI's much more efficiently than plain text or UTF-8 text. (32 bytes per triple)
Lastly, Literals cannot be used as subjects in triples.
Is the current GNI RDF the greatest things since sliced bread?
No, it is a start and a needed part of a system to track name use, name to name relations and name to concept relations etc.
Also the systems are moving toward formatting the human readable views such that the URI is replaced with the rdfs:Label.
You can already see this in this example from Sig.ma where the URI's are replaced in the human view with the names of states etc.