The trick to all of this is finding gestures (or whatever you want to call them) that allow some sort of direct manipulation in areas where the human mind has no experience. The best example right now is pinch-to-zoom, which has absolutely no physical equivalent, but is extremely intuitive. Compare this to having arrow and +/- magnifying glass buttons.
As software blends more with the rest of the world, the machine-intentioned interaction is breaking down. It's kind of like trying to draw a photorealistic image using an etch-a-sketch, when instead you just need a pen and paper.
(You could argue that pinch-to-zoom isn't very discoverable/learnable, but the reality is that people figure it out incredibly quickly, which is kind of amazing when you think about it. There's certainly something magically compelling about using one's hands in that way. Perhaps it's actually more important for interfaces to be instantly learnable than "intuitive.")