We argue that the natural pressure to develop increasingly expressive metadata systems will soon present new and very difficult problems to the metadata community. In particular we draw attention to propositional attitudes, which are important for metadata for trust and provenance, and for cultural objects, but which are extremely difficult to represent in a way that allows all valid, but no invalid, inferences. Both the W3C and the museum informatics community (CIDOC) have deferred addressing this problem, preferring instead to stipulate limitations to their systems. But eventually it must be solved. We believe that this sort of problem will be more important to the community in coming years than the traditional expressiveness vs. efficiency tradeoff.