Some considerations for a theory of approximate invariants