I believe that we will eventually abandon the idea of sampled instruments entirely, at least as they're approached today. Remember that samples were born out of frustration with the limitations of even the most complex synthesizers, which ultimately offer only a small subset of the many variables that comprise a musical instrument's timbre and behavior. I've always felt they were an interim solution.
In the 60's, students at MIT replicated the sound of a grand piano so believable that when played behind a curtain alongside a real piano, audiences could not distinguish between them. However, that experiment required no less than 80 oscillators and could play only one note. Not one not at a time, one note, period. It simultaneously demonstrated both the potential of synthesis as well as the enormity of the task. I wonder if anybody then considered the possibility of what could be done with unlimited
virtual oscillators created completely in software.
Of course, we're not quite there yet. We can't yet create an unlimited number of oscillators with today's consumer computers. But projecting the trajectory of computer development over the past 60 years into the coming decades, it's clearly just a matter of time.
And it's not just hardware holding us back. Equally important is software modeling technology, but it'll get there, too. This morning it snowed here in Seattle. Most would not have seen it coming, but the National Weather Service's models did, and correctly predicted it. Now if you and I could only afford the computers that the NWS has...