It's interesting to think about the methodology that could potentially be used to emulate human playback (and by that, I mean
real human playback, not Finale's existing technology
).
I'm thinking the underlying principle could work along the same lines as motion capture does for animation. Load the score for a Bach partita into a computer, set a performer down to play it, then analyse how their interpretation of each individual instruction differs from or is faithful to the written material. If you went through this process with several hundred performers, you might be able to construct a primitive library of "Baroque performance nuances". Then, it would be a simple matter of incorporating that into a sequencer, notation software package, etc.
Given these generalised nuances, it might be possible to get it to think originally with some AI.
Jazz is probably a whole different kettle of fish.