Dual theories: 'Same but different' or 'different but same'?

I argue that, underneath the glitz, dual theories are simply examples of theoretically equivalent descriptions of the same underlying physical content: I distinguish them from cases of genuine underdetermination on the grounds that there is no real incompatibility involved between the descriptions. The incompatibility is at the level of purely unphysical structure. I argue that dual pairs are in fact very strongly analogous to gauge-related solutions even for dual pairs that look the most radically distinct, such as AdS/CFT. However, again by analogy with gauge freedom, I conjecture that dualities always point to a more fundamental (intrinsic) description, namely that in which the representational redundancy is eliminated.