I think there is a kernel of truth worth examining in James' statements.
The traditional catalog entry is not metadata or even data but rather is a metadocument -- a document (the catalog record) about a document (the resource). The great conceptual stumbling block for many of us (and I count myself in that number), as humans accustomed to interpreting documents and metadocuments, is to visualize a data architecture that disaggregates our traditional metadocument components into discrete data units which can then be, not just transferred and stored by machines, but interpreted, manipulated, and reconstituted back into metadocuments by them. Further, these resulting metadocuments need not be the same as what was input, but can be profiled to render output in different languages (via linked vocabulary registries), or with less information (say for a mobile display), or more information (by drawing on data from external sources).
I would go so far as to say that the Anglo-American community is particularly hampered by this transition, since our linguistic syntax relies on position to determine the function of the words we read (i.e. the data we intake) -- "Dog bites man" is different than "Man bites dog" -- where inflected syntax relies on declension and conjugation as markup to determine the corresponding word functions. Consequently, we have a stronger tendency to need to "read" our data in documentary form, as presented in the metadocuments that make up our catalog records, rather than drawing on it solely as data.