I would expect the resolution of this dilemma to hold no relation whatsoever to AdS/CFT, being it specific to anti de Sitter spaces, which makes it irrelevant to our universe (unless someone proves an analog version for de Sitter spaces, which seems unlikely)
I think the key to the resolution of this dilemma is the false assumption that quantum evolution is unitary everywhere and for all observers.
To explain why the above assumption is false, take for instance the process of measurement of quantum observables. Regardless of the philosophical doctrine of measurement one subscribes to, the fact is that part of the information of the measured state is lost forever to the party that obtained information from the measurement. Yes, when observers do measurements to quantum systems, the global superposition of the system observer-system is still there "somewhere", evolving under perfect unitary evolution. However, some of the information from the measured system that is accessible to the observer is lost forever, even in principle, because of apparent collapse. So, it should be clear in this case that even if evolution is unitary for outside observers, physical observer eigenstates involved in the measurement will observe violations of unitarity.
so, unitarity is clearly not an absolute property of a system evolution, but depends on the "frame" where it is observed, and how the system couples to the observational frame.
Let's speculate how to apply the above in the scenario of the black hole information paradox: an observer outside the black hole sends information inside it, when the information crosses the event horizon, the information is lost forever for him.
However, observers inside the event horizon still see the whole system behaving under unitary evolution; the information sent inside the event horizon is still reachable to them. In particular, all entanglement correlations between states in and out of the horizon are preserved from their point of view, of course, as long as the measurement devices keep sending data inside the horizon for these observers to measure the correlations.
The trouble arises because some people is naively expecting quantum unitary evolution to be an absolute property that is agreed on by all observers, but quantum measurements are a clear example how this is not true in general.
This post imported from StackExchange Physics at 2014-05-14 19:47 (UCT), posted by SE-user lurscher