Well, yeah, but that "reversible" hedge is quite significant. Reversible means you have to retain all of your intermediate states, and it is far from clear that you could build a reversible AI that would run for more than a few milliseconds before consuming all of the available resources in its light cone.
> You just need a large enough supply of qubits initialized to |0> for all the copies you plan to do.
Yes, and that produces exactly the same problem: it is far from clear that there are enough qbits in the universe to build a quantum AI this way.
You also have a more fundamental problem, which is that neither quantum computers nor reversible classical computers can do I/O, so how are you going to demonstrate that this thing is in fact an AI? It can't even participate in a Turing test, let alone pass one.
I guess one would have to demonstrate that the program implements an AI by running it on a computer with I/O, then move the same program over to an isolated quantum computer once it's passed that test. Maybe not completely philosophically satisfying, but should work in practice.